KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
/*
|
|
|
|
|
* This program is free software; you can redistribute it and/or modify
|
|
|
|
|
* it under the terms of the GNU General Public License, version 2, as
|
|
|
|
|
* published by the Free Software Foundation.
|
|
|
|
|
*
|
|
|
|
|
* This program is distributed in the hope that it will be useful,
|
|
|
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
|
* GNU General Public License for more details.
|
|
|
|
|
*
|
|
|
|
|
* Copyright 2011 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
|
|
|
|
|
*
|
|
|
|
|
* Derived from book3s_rmhandlers.S and other files, which are:
|
|
|
|
|
*
|
|
|
|
|
* Copyright SUSE Linux Products GmbH 2009
|
|
|
|
|
*
|
|
|
|
|
* Authors: Alexander Graf <agraf@suse.de>
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
#include <asm/ppc_asm.h>
|
|
|
|
|
#include <asm/kvm_asm.h>
|
|
|
|
|
#include <asm/reg.h>
|
2011-07-23 15:41:11 +08:00
|
|
|
#include <asm/mmu.h>
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
#include <asm/page.h>
|
2011-07-23 15:41:11 +08:00
|
|
|
#include <asm/ptrace.h>
|
|
|
|
|
#include <asm/hvcall.h>
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
#include <asm/asm-offsets.h>
|
|
|
|
|
#include <asm/exception-64s.h>
|
2012-02-03 08:54:17 +08:00
|
|
|
#include <asm/kvm_book3s_asm.h>
|
2016-03-01 15:29:20 +08:00
|
|
|
#include <asm/book3s/64/mmu-hash.h>
|
2018-10-08 13:30:51 +08:00
|
|
|
#include <asm/export.h>
|
2014-03-25 07:47:02 +08:00
|
|
|
#include <asm/tm.h>
|
KVM: PPC: Book3S HV: Fix TB corruption in guest exit path on HMI interrupt
When a guest is assigned to a core it converts the host Timebase (TB)
into guest TB by adding guest timebase offset before entering into
guest. During guest exit it restores the guest TB to host TB. This means
under certain conditions (Guest migration) host TB and guest TB can differ.
When we get an HMI for TB related issues the opal HMI handler would
try fixing errors and restore the correct host TB value. With no guest
running, we don't have any issues. But with guest running on the core
we run into TB corruption issues.
If we get an HMI while in the guest, the current HMI handler invokes opal
hmi handler before forcing guest to exit. The guest exit path subtracts
the guest TB offset from the current TB value which may have already
been restored with host value by opal hmi handler. This leads to incorrect
host and guest TB values.
With split-core, things become more complex. With split-core, TB also gets
split and each subcore gets its own TB register. When a hmi handler fixes
a TB error and restores the TB value, it affects all the TB values of
sibling subcores on the same core. On TB errors all the thread in the core
gets HMI. With existing code, the individual threads call opal hmi handle
independently which can easily throw TB out of sync if we have guest
running on subcores. Hence we will need to co-ordinate with all the
threads before making opal hmi handler call followed by TB resync.
This patch introduces a sibling subcore state structure (shared by all
threads in the core) in paca which holds information about whether sibling
subcores are in Guest mode or host mode. An array in_guest[] of size
MAX_SUBCORE_PER_CORE=4 is used to maintain the state of each subcore.
The subcore id is used as index into in_guest[] array. Only primary
thread entering/exiting the guest is responsible to set/unset its
designated array element.
On TB error, we get HMI interrupt on every thread on the core. Upon HMI,
this patch will now force guest to vacate the core/subcore. Primary
thread from each subcore will then turn off its respective bit
from the above bitmap during the guest exit path just after the
guest->host partition switch is complete.
All other threads that have just exited the guest OR were already in host
will wait until all other subcores clears their respective bit.
Once all the subcores turn off their respective bit, all threads will
will make call to opal hmi handler.
It is not necessary that opal hmi handler would resync the TB value for
every HMI interrupts. It would do so only for the HMI caused due to
TB errors. For rest, it would not touch TB value. Hence to make things
simpler, primary thread would call TB resync explicitly once for each
core immediately after opal hmi handler instead of subtracting guest
offset from TB. TB resync call will restore the TB with host value.
Thus we can be sure about the TB state.
One of the primary threads exiting the guest will take up the
responsibility of calling TB resync. It will use one of the top bits
(bit 63) from subcore state flags bitmap to make the decision. The first
primary thread (among the subcores) that is able to set the bit will
have to call the TB resync. Rest all other threads will wait until TB
resync is complete. Once TB resync is complete all threads will then
proceed.
Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-05-15 12:14:26 +08:00
|
|
|
#include <asm/opal.h>
|
2017-04-05 15:54:56 +08:00
|
|
|
#include <asm/xive-regs.h>
|
KVM: PPC: Book3S HV: Handle unexpected interrupts better
At present, if an interrupt (i.e. an exception or trap) occurs in the
code where KVM is switching the MMU to or from guest context, we jump
to kvmppc_bad_host_intr, where we simply spin with interrupts disabled.
In this situation, it is hard to debug what happened because we get no
indication as to which interrupt occurred or where. Typically we get
a cascade of stall and soft lockup warnings from other CPUs.
In order to get more information for debugging, this adds code to
create a stack frame on the emergency stack and save register values
to it. We start half-way down the emergency stack in order to give
ourselves some chance of being able to do a stack trace on secondary
threads that are already on the emergency stack.
On POWER7 or POWER8, we then just spin, as before, because we don't
know what state the MMU context is in or what other threads are doing,
and we can't switch back to host context without coordinating with
other threads. On POWER9 we can do better; there we load up the host
MMU context and jump to C code, which prints an oops message to the
console and panics.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-01 14:17:27 +08:00
|
|
|
#include <asm/thread_info.h>
|
2018-07-06 00:24:57 +08:00
|
|
|
#include <asm/asm-compat.h>
|
2018-07-06 00:25:01 +08:00
|
|
|
#include <asm/feature-fixups.h>
|
2014-03-25 07:47:02 +08:00
|
|
|
|
2017-05-22 14:25:14 +08:00
|
|
|
/* Sign-extend HDEC if not on POWER9 */
|
|
|
|
|
#define EXTEND_HDEC(reg) \
|
|
|
|
|
BEGIN_FTR_SECTION; \
|
|
|
|
|
extsw reg, reg; \
|
|
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
/* Values in HSTATE_NAPPING(r13) */
|
|
|
|
|
#define NAPPING_CEDE 1
|
|
|
|
|
#define NAPPING_NOVCPU 2
|
|
|
|
|
|
2017-06-16 09:53:19 +08:00
|
|
|
/* Stack frame offsets for kvmppc_hv_entry */
|
2018-10-08 13:30:55 +08:00
|
|
|
#define SFS 208
|
2017-06-16 09:53:19 +08:00
|
|
|
#define STACK_SLOT_TRAP (SFS-4)
|
2018-10-08 13:30:55 +08:00
|
|
|
#define STACK_SLOT_SHORT_PATH (SFS-8)
|
2017-06-16 09:53:19 +08:00
|
|
|
#define STACK_SLOT_TID (SFS-16)
|
|
|
|
|
#define STACK_SLOT_PSSCR (SFS-24)
|
|
|
|
|
#define STACK_SLOT_PID (SFS-32)
|
|
|
|
|
#define STACK_SLOT_IAMR (SFS-40)
|
|
|
|
|
#define STACK_SLOT_CIABR (SFS-48)
|
|
|
|
|
#define STACK_SLOT_DAWR (SFS-56)
|
|
|
|
|
#define STACK_SLOT_DAWRX (SFS-64)
|
2017-02-15 11:30:17 +08:00
|
|
|
#define STACK_SLOT_HFSCR (SFS-72)
|
powerpc/kvm: Save and restore host AMR/IAMR/UAMOR
When the hash MMU is active the AMR, IAMR and UAMOR are used for
pkeys. The AMR is directly writable by user space, and the UAMOR masks
those writes, meaning both registers are effectively user register
state. The IAMR is used to create an execute only key.
Also we must maintain the value of at least the AMR when running in
process context, so that any memory accesses done by the kernel on
behalf of the process are correctly controlled by the AMR.
Although we are correctly switching all registers when going into a
guest, on returning to the host we just write 0 into all regs, except
on Power9 where we restore the IAMR correctly.
This could be observed by a user process if it writes the AMR, then
runs a guest and we then return immediately to it without
rescheduling. Because we have written 0 to the AMR that would have the
effect of granting read/write permission to pages that the process was
trying to protect.
In addition, when using the Radix MMU, the AMR can prevent inadvertent
kernel access to userspace data, writing 0 to the AMR disables that
protection.
So save and restore AMR, IAMR and UAMOR.
Fixes: cf43d3b26452 ("powerpc: Enable pkey subsystem")
Cc: stable@vger.kernel.org # v4.16+
Signed-off-by: Russell Currey <ruscur@russell.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Paul Mackerras <paulus@ozlabs.org>
2019-02-22 10:22:08 +08:00
|
|
|
#define STACK_SLOT_AMR (SFS-80)
|
|
|
|
|
#define STACK_SLOT_UAMOR (SFS-88)
|
2018-10-08 13:30:55 +08:00
|
|
|
/* the following is used by the P9 short path */
|
|
|
|
|
#define STACK_SLOT_NVGPRS (SFS-152) /* 18 gprs */
|
2017-06-16 09:53:19 +08:00
|
|
|
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
/*
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
* Call kvmppc_hv_entry in real mode.
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
* Must be called with interrupts hard-disabled.
|
|
|
|
|
*
|
|
|
|
|
* Input Registers:
|
|
|
|
|
*
|
|
|
|
|
* LR = return address to continue at after eventually re-enabling MMU
|
|
|
|
|
*/
|
2014-06-12 16:16:53 +08:00
|
|
|
_GLOBAL_TOC(kvmppc_hv_entry_trampoline)
|
2013-09-06 11:23:44 +08:00
|
|
|
mflr r0
|
|
|
|
|
std r0, PPC_LR_STKOFF(r1)
|
|
|
|
|
stdu r1, -112(r1)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
mfmsr r10
|
KVM: PPC: Book3S HV: Close race with testing for signals on guest entry
At present, interrupts are hard-disabled fairly late in the guest
entry path, in the assembly code. Since we check for pending signals
for the vCPU(s) task(s) earlier in the guest entry path, it is
possible for a signal to be delivered before we enter the guest but
not be noticed until after we exit the guest for some other reason.
Similarly, it is possible for the scheduler to request a reschedule
while we are in the guest entry path, and we won't notice until after
we have run the guest, potentially for a whole timeslice.
Furthermore, with a radix guest on POWER9, we can take the interrupt
with the MMU on. In this case we end up leaving interrupts
hard-disabled after the guest exit, and they are likely to stay
hard-disabled until we exit to userspace or context-switch to
another process. This was masking the fact that we were also not
setting the RI (recoverable interrupt) bit in the MSR, meaning
that if we had taken an interrupt, it would have crashed the host
kernel with an unrecoverable interrupt message.
To close these races, we need to check for signals and reschedule
requests after hard-disabling interrupts, and then keep interrupts
hard-disabled until we enter the guest. If there is a signal or a
reschedule request from another CPU, it will send an IPI, which will
cause a guest exit.
This puts the interrupt disabling before we call kvmppc_start_thread()
for all the secondary threads of this core that are going to run vCPUs.
The reason for that is that once we have started the secondary threads
there is no easy way to back out without going through at least part
of the guest entry path. However, kvmppc_start_thread() includes some
code for radix guests which needs to call smp_call_function(), which
must be called with interrupts enabled. To solve this problem, this
patch moves that code into a separate function that is called earlier.
When the guest exit is caused by an external interrupt, a hypervisor
doorbell or a hypervisor maintenance interrupt, we now handle these
using the replay facility. __kvmppc_vcore_entry() now returns the
trap number that caused the exit on this thread, and instead of the
assembly code jumping to the handler entry, we return to C code with
interrupts still hard-disabled and set the irq_happened flag in the
PACA, so that when we do local_irq_enable() the appropriate handler
gets called.
With all this, we now have the interrupt soft-enable flag clear while
we are in the guest. This is useful because code in the real-mode
hypercall handlers that checks whether interrupts are enabled will
now see that they are disabled, which is correct, since interrupts
are hard-disabled in the real-mode code.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-06-26 13:45:51 +08:00
|
|
|
std r10, HSTATE_HOST_MSR(r13)
|
2013-09-06 11:23:44 +08:00
|
|
|
LOAD_REG_ADDR(r5, kvmppc_call_hv_entry)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
li r0,MSR_RI
|
|
|
|
|
andc r0,r10,r0
|
|
|
|
|
li r6,MSR_IR | MSR_DR
|
|
|
|
|
andc r6,r10,r6
|
|
|
|
|
mtmsrd r0,1 /* clear RI in MSR */
|
|
|
|
|
mtsrr0 r5
|
|
|
|
|
mtsrr1 r6
|
2018-01-10 00:07:15 +08:00
|
|
|
RFI_TO_KERNEL
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
|
2013-09-06 11:23:44 +08:00
|
|
|
kvmppc_call_hv_entry:
|
KVM: PPC: Book3S HV: Run HPT guests on POWER9 radix hosts
This patch removes the restriction that a radix host can only run
radix guests, allowing us to run HPT (hashed page table) guests as
well. This is useful because it provides a way to run old guest
kernels that know about POWER8 but not POWER9.
Unfortunately, POWER9 currently has a restriction that all threads
in a given code must either all be in HPT mode, or all in radix mode.
This means that when entering a HPT guest, we have to obtain control
of all 4 threads in the core and get them to switch their LPIDR and
LPCR registers, even if they are not going to run a guest. On guest
exit we also have to get all threads to switch LPIDR and LPCR back
to host values.
To make this feasible, we require that KVM not be in the "independent
threads" mode, and that the CPU cores be in single-threaded mode from
the host kernel's perspective (only thread 0 online; threads 1, 2 and
3 offline). That allows us to use the same code as on POWER8 for
obtaining control of the secondary threads.
To manage the LPCR/LPIDR changes required, we extend the kvm_split_info
struct to contain the information needed by the secondary threads.
All threads perform a barrier synchronization (where all threads wait
for every other thread to reach the synchronization point) on guest
entry, both before and after loading LPCR and LPIDR. On guest exit,
they all once again perform a barrier synchronization both before
and after loading host values into LPCR and LPIDR.
Finally, it is also currently necessary to flush the entire TLB every
time we enter a HPT guest on a radix host. We do this on thread 0
with a loop of tlbiel instructions.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-19 11:11:23 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
/* On P9, do LPCR setting, if necessary */
|
|
|
|
|
ld r3, HSTATE_SPLIT_MODE(r13)
|
|
|
|
|
cmpdi r3, 0
|
|
|
|
|
beq 46f
|
|
|
|
|
lwz r4, KVM_SPLIT_DO_SET(r3)
|
|
|
|
|
cmpwi r4, 0
|
|
|
|
|
beq 46f
|
|
|
|
|
bl kvmhv_p9_set_lpcr
|
|
|
|
|
nop
|
|
|
|
|
46:
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
2013-09-06 11:23:44 +08:00
|
|
|
bl kvmppc_hv_entry
|
|
|
|
|
|
|
|
|
|
/* Back from guest - restore host state and return to caller */
|
|
|
|
|
|
2014-01-08 18:25:19 +08:00
|
|
|
BEGIN_FTR_SECTION
|
2013-09-06 11:23:44 +08:00
|
|
|
/* Restore host DABR and DABRX */
|
|
|
|
|
ld r5,HSTATE_DABR(r13)
|
|
|
|
|
li r6,7
|
|
|
|
|
mtspr SPRN_DABR,r5
|
|
|
|
|
mtspr SPRN_DABRX,r6
|
2014-01-08 18:25:19 +08:00
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
|
2013-09-06 11:23:44 +08:00
|
|
|
|
|
|
|
|
/* Restore SPRG3 */
|
2014-03-11 06:29:38 +08:00
|
|
|
ld r3,PACA_SPRG_VDSO(r13)
|
|
|
|
|
mtspr SPRN_SPRG_VDSO_WRITE,r3
|
2013-09-06 11:23:44 +08:00
|
|
|
|
|
|
|
|
/* Reload the host's PMU registers */
|
2018-10-08 13:30:51 +08:00
|
|
|
bl kvmhv_load_host_pmu
|
2013-09-06 11:23:44 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
/*
|
|
|
|
|
* Reload DEC. HDEC interrupts were disabled when
|
|
|
|
|
* we reloaded the host's LPCR value.
|
|
|
|
|
*/
|
|
|
|
|
ld r3, HSTATE_DECEXP(r13)
|
|
|
|
|
mftb r4
|
|
|
|
|
subf r4, r4, r3
|
|
|
|
|
mtspr SPRN_DEC, r4
|
|
|
|
|
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
/* hwthread_req may have got set by cede or no vcpu, so clear it */
|
|
|
|
|
li r0, 0
|
|
|
|
|
stb r0, HSTATE_HWTHREAD_REQ(r13)
|
|
|
|
|
|
2013-09-06 11:23:44 +08:00
|
|
|
/*
|
2017-05-11 19:03:37 +08:00
|
|
|
* For external interrupts we need to call the Linux
|
|
|
|
|
* handler to process the interrupt. We do that by jumping
|
|
|
|
|
* to absolute address 0x500 for external interrupts.
|
|
|
|
|
* The [h]rfid at the end of the handler will return to
|
|
|
|
|
* the book3s_hv_interrupts.S code. For other interrupts
|
|
|
|
|
* we do the rfid to get back to the book3s_hv_interrupts.S
|
|
|
|
|
* code here.
|
2013-09-06 11:23:44 +08:00
|
|
|
*/
|
|
|
|
|
ld r8, 112+PPC_LR_STKOFF(r1)
|
|
|
|
|
addi r1, r1, 112
|
|
|
|
|
ld r7, HSTATE_HOST_MSR(r13)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Close race with testing for signals on guest entry
At present, interrupts are hard-disabled fairly late in the guest
entry path, in the assembly code. Since we check for pending signals
for the vCPU(s) task(s) earlier in the guest entry path, it is
possible for a signal to be delivered before we enter the guest but
not be noticed until after we exit the guest for some other reason.
Similarly, it is possible for the scheduler to request a reschedule
while we are in the guest entry path, and we won't notice until after
we have run the guest, potentially for a whole timeslice.
Furthermore, with a radix guest on POWER9, we can take the interrupt
with the MMU on. In this case we end up leaving interrupts
hard-disabled after the guest exit, and they are likely to stay
hard-disabled until we exit to userspace or context-switch to
another process. This was masking the fact that we were also not
setting the RI (recoverable interrupt) bit in the MSR, meaning
that if we had taken an interrupt, it would have crashed the host
kernel with an unrecoverable interrupt message.
To close these races, we need to check for signals and reschedule
requests after hard-disabling interrupts, and then keep interrupts
hard-disabled until we enter the guest. If there is a signal or a
reschedule request from another CPU, it will send an IPI, which will
cause a guest exit.
This puts the interrupt disabling before we call kvmppc_start_thread()
for all the secondary threads of this core that are going to run vCPUs.
The reason for that is that once we have started the secondary threads
there is no easy way to back out without going through at least part
of the guest entry path. However, kvmppc_start_thread() includes some
code for radix guests which needs to call smp_call_function(), which
must be called with interrupts enabled. To solve this problem, this
patch moves that code into a separate function that is called earlier.
When the guest exit is caused by an external interrupt, a hypervisor
doorbell or a hypervisor maintenance interrupt, we now handle these
using the replay facility. __kvmppc_vcore_entry() now returns the
trap number that caused the exit on this thread, and instead of the
assembly code jumping to the handler entry, we return to C code with
interrupts still hard-disabled and set the irq_happened flag in the
PACA, so that when we do local_irq_enable() the appropriate handler
gets called.
With all this, we now have the interrupt soft-enable flag clear while
we are in the guest. This is useful because code in the real-mode
hypercall handlers that checks whether interrupts are enabled will
now see that they are disabled, which is correct, since interrupts
are hard-disabled in the real-mode code.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-06-26 13:45:51 +08:00
|
|
|
/* Return the trap number on this thread as the return value */
|
|
|
|
|
mr r3, r12
|
|
|
|
|
|
2017-01-30 18:21:51 +08:00
|
|
|
/*
|
|
|
|
|
* If we came back from the guest via a relocation-on interrupt,
|
|
|
|
|
* we will be in virtual mode at this point, which makes it a
|
|
|
|
|
* little easier to get back to the caller.
|
|
|
|
|
*/
|
|
|
|
|
mfmsr r0
|
|
|
|
|
andi. r0, r0, MSR_IR /* in real mode? */
|
|
|
|
|
bne .Lvirt_return
|
|
|
|
|
|
KVM: PPC: Book3S HV: Close race with testing for signals on guest entry
At present, interrupts are hard-disabled fairly late in the guest
entry path, in the assembly code. Since we check for pending signals
for the vCPU(s) task(s) earlier in the guest entry path, it is
possible for a signal to be delivered before we enter the guest but
not be noticed until after we exit the guest for some other reason.
Similarly, it is possible for the scheduler to request a reschedule
while we are in the guest entry path, and we won't notice until after
we have run the guest, potentially for a whole timeslice.
Furthermore, with a radix guest on POWER9, we can take the interrupt
with the MMU on. In this case we end up leaving interrupts
hard-disabled after the guest exit, and they are likely to stay
hard-disabled until we exit to userspace or context-switch to
another process. This was masking the fact that we were also not
setting the RI (recoverable interrupt) bit in the MSR, meaning
that if we had taken an interrupt, it would have crashed the host
kernel with an unrecoverable interrupt message.
To close these races, we need to check for signals and reschedule
requests after hard-disabling interrupts, and then keep interrupts
hard-disabled until we enter the guest. If there is a signal or a
reschedule request from another CPU, it will send an IPI, which will
cause a guest exit.
This puts the interrupt disabling before we call kvmppc_start_thread()
for all the secondary threads of this core that are going to run vCPUs.
The reason for that is that once we have started the secondary threads
there is no easy way to back out without going through at least part
of the guest entry path. However, kvmppc_start_thread() includes some
code for radix guests which needs to call smp_call_function(), which
must be called with interrupts enabled. To solve this problem, this
patch moves that code into a separate function that is called earlier.
When the guest exit is caused by an external interrupt, a hypervisor
doorbell or a hypervisor maintenance interrupt, we now handle these
using the replay facility. __kvmppc_vcore_entry() now returns the
trap number that caused the exit on this thread, and instead of the
assembly code jumping to the handler entry, we return to C code with
interrupts still hard-disabled and set the irq_happened flag in the
PACA, so that when we do local_irq_enable() the appropriate handler
gets called.
With all this, we now have the interrupt soft-enable flag clear while
we are in the guest. This is useful because code in the real-mode
hypercall handlers that checks whether interrupts are enabled will
now see that they are disabled, which is correct, since interrupts
are hard-disabled in the real-mode code.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-06-26 13:45:51 +08:00
|
|
|
/* RFI into the highmem handler */
|
2013-09-06 11:23:44 +08:00
|
|
|
mfmsr r6
|
|
|
|
|
li r0, MSR_RI
|
|
|
|
|
andc r6, r6, r0
|
|
|
|
|
mtmsrd r6, 1 /* Clear RI in MSR */
|
|
|
|
|
mtsrr0 r8
|
|
|
|
|
mtsrr1 r7
|
2018-01-10 00:07:15 +08:00
|
|
|
RFI_TO_KERNEL
|
2013-09-06 11:23:44 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Close race with testing for signals on guest entry
At present, interrupts are hard-disabled fairly late in the guest
entry path, in the assembly code. Since we check for pending signals
for the vCPU(s) task(s) earlier in the guest entry path, it is
possible for a signal to be delivered before we enter the guest but
not be noticed until after we exit the guest for some other reason.
Similarly, it is possible for the scheduler to request a reschedule
while we are in the guest entry path, and we won't notice until after
we have run the guest, potentially for a whole timeslice.
Furthermore, with a radix guest on POWER9, we can take the interrupt
with the MMU on. In this case we end up leaving interrupts
hard-disabled after the guest exit, and they are likely to stay
hard-disabled until we exit to userspace or context-switch to
another process. This was masking the fact that we were also not
setting the RI (recoverable interrupt) bit in the MSR, meaning
that if we had taken an interrupt, it would have crashed the host
kernel with an unrecoverable interrupt message.
To close these races, we need to check for signals and reschedule
requests after hard-disabling interrupts, and then keep interrupts
hard-disabled until we enter the guest. If there is a signal or a
reschedule request from another CPU, it will send an IPI, which will
cause a guest exit.
This puts the interrupt disabling before we call kvmppc_start_thread()
for all the secondary threads of this core that are going to run vCPUs.
The reason for that is that once we have started the secondary threads
there is no easy way to back out without going through at least part
of the guest entry path. However, kvmppc_start_thread() includes some
code for radix guests which needs to call smp_call_function(), which
must be called with interrupts enabled. To solve this problem, this
patch moves that code into a separate function that is called earlier.
When the guest exit is caused by an external interrupt, a hypervisor
doorbell or a hypervisor maintenance interrupt, we now handle these
using the replay facility. __kvmppc_vcore_entry() now returns the
trap number that caused the exit on this thread, and instead of the
assembly code jumping to the handler entry, we return to C code with
interrupts still hard-disabled and set the irq_happened flag in the
PACA, so that when we do local_irq_enable() the appropriate handler
gets called.
With all this, we now have the interrupt soft-enable flag clear while
we are in the guest. This is useful because code in the real-mode
hypercall handlers that checks whether interrupts are enabled will
now see that they are disabled, which is correct, since interrupts
are hard-disabled in the real-mode code.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-06-26 13:45:51 +08:00
|
|
|
/* Virtual-mode return */
|
2017-01-30 18:21:51 +08:00
|
|
|
.Lvirt_return:
|
KVM: PPC: Book3S HV: Close race with testing for signals on guest entry
At present, interrupts are hard-disabled fairly late in the guest
entry path, in the assembly code. Since we check for pending signals
for the vCPU(s) task(s) earlier in the guest entry path, it is
possible for a signal to be delivered before we enter the guest but
not be noticed until after we exit the guest for some other reason.
Similarly, it is possible for the scheduler to request a reschedule
while we are in the guest entry path, and we won't notice until after
we have run the guest, potentially for a whole timeslice.
Furthermore, with a radix guest on POWER9, we can take the interrupt
with the MMU on. In this case we end up leaving interrupts
hard-disabled after the guest exit, and they are likely to stay
hard-disabled until we exit to userspace or context-switch to
another process. This was masking the fact that we were also not
setting the RI (recoverable interrupt) bit in the MSR, meaning
that if we had taken an interrupt, it would have crashed the host
kernel with an unrecoverable interrupt message.
To close these races, we need to check for signals and reschedule
requests after hard-disabling interrupts, and then keep interrupts
hard-disabled until we enter the guest. If there is a signal or a
reschedule request from another CPU, it will send an IPI, which will
cause a guest exit.
This puts the interrupt disabling before we call kvmppc_start_thread()
for all the secondary threads of this core that are going to run vCPUs.
The reason for that is that once we have started the secondary threads
there is no easy way to back out without going through at least part
of the guest entry path. However, kvmppc_start_thread() includes some
code for radix guests which needs to call smp_call_function(), which
must be called with interrupts enabled. To solve this problem, this
patch moves that code into a separate function that is called earlier.
When the guest exit is caused by an external interrupt, a hypervisor
doorbell or a hypervisor maintenance interrupt, we now handle these
using the replay facility. __kvmppc_vcore_entry() now returns the
trap number that caused the exit on this thread, and instead of the
assembly code jumping to the handler entry, we return to C code with
interrupts still hard-disabled and set the irq_happened flag in the
PACA, so that when we do local_irq_enable() the appropriate handler
gets called.
With all this, we now have the interrupt soft-enable flag clear while
we are in the guest. This is useful because code in the real-mode
hypercall handlers that checks whether interrupts are enabled will
now see that they are disabled, which is correct, since interrupts
are hard-disabled in the real-mode code.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-06-26 13:45:51 +08:00
|
|
|
mtlr r8
|
2017-01-30 18:21:51 +08:00
|
|
|
blr
|
|
|
|
|
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
kvmppc_primary_no_guest:
|
|
|
|
|
/* We handle this much like a ceded vcpu */
|
2015-03-28 11:21:08 +08:00
|
|
|
/* put the HDEC into the DEC, since HDEC interrupts don't wake us */
|
2017-05-22 14:25:14 +08:00
|
|
|
/* HDEC may be larger than DEC for arch >= v3.00, but since the */
|
|
|
|
|
/* HDEC value came from DEC in the first place, it will fit */
|
2015-03-28 11:21:08 +08:00
|
|
|
mfspr r3, SPRN_HDEC
|
|
|
|
|
mtspr SPRN_DEC, r3
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
/*
|
|
|
|
|
* Make sure the primary has finished the MMU switch.
|
|
|
|
|
* We should never get here on a secondary thread, but
|
|
|
|
|
* check it for robustness' sake.
|
|
|
|
|
*/
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
65: lbz r0, VCORE_IN_GUEST(r5)
|
|
|
|
|
cmpwi r0, 0
|
|
|
|
|
beq 65b
|
|
|
|
|
/* Set LPCR. */
|
|
|
|
|
ld r8,VCORE_LPCR(r5)
|
|
|
|
|
mtspr SPRN_LPCR,r8
|
|
|
|
|
isync
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
/* set our bit in napping_threads */
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
lbz r7, HSTATE_PTID(r13)
|
|
|
|
|
li r0, 1
|
|
|
|
|
sld r0, r0, r7
|
|
|
|
|
addi r6, r5, VCORE_NAPPING_THREADS
|
|
|
|
|
1: lwarx r3, 0, r6
|
|
|
|
|
or r3, r3, r0
|
|
|
|
|
stwcx. r3, 0, r6
|
|
|
|
|
bne 1b
|
2015-03-28 11:21:09 +08:00
|
|
|
/* order napping_threads update vs testing entry_exit_map */
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
isync
|
|
|
|
|
li r12, 0
|
|
|
|
|
lwz r7, VCORE_ENTRY_EXIT(r5)
|
|
|
|
|
cmpwi r7, 0x100
|
|
|
|
|
bge kvm_novcpu_exit /* another thread already exiting */
|
|
|
|
|
li r3, NAPPING_NOVCPU
|
|
|
|
|
stb r3, HSTATE_NAPPING(r13)
|
|
|
|
|
|
2015-03-28 11:21:07 +08:00
|
|
|
li r3, 0 /* Don't wake on privileged (OS) doorbell */
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
b kvm_do_nap
|
|
|
|
|
|
2016-08-19 13:35:46 +08:00
|
|
|
/*
|
|
|
|
|
* kvm_novcpu_wakeup
|
|
|
|
|
* Entered from kvm_start_guest if kvm_hstate.napping is set
|
|
|
|
|
* to NAPPING_NOVCPU
|
|
|
|
|
* r2 = kernel TOC
|
|
|
|
|
* r13 = paca
|
|
|
|
|
*/
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
kvm_novcpu_wakeup:
|
|
|
|
|
ld r1, HSTATE_HOST_R1(r13)
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
li r0, 0
|
|
|
|
|
stb r0, HSTATE_NAPPING(r13)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
/* check the wake reason */
|
|
|
|
|
bl kvmppc_check_wake_reason
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
|
2016-08-19 13:35:46 +08:00
|
|
|
/*
|
|
|
|
|
* Restore volatile registers since we could have called
|
|
|
|
|
* a C routine in kvmppc_check_wake_reason.
|
|
|
|
|
* r5 = VCORE
|
|
|
|
|
*/
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
/* see if any other thread is already exiting */
|
|
|
|
|
lwz r0, VCORE_ENTRY_EXIT(r5)
|
|
|
|
|
cmpwi r0, 0x100
|
|
|
|
|
bge kvm_novcpu_exit
|
|
|
|
|
|
|
|
|
|
/* clear our bit in napping_threads */
|
|
|
|
|
lbz r7, HSTATE_PTID(r13)
|
|
|
|
|
li r0, 1
|
|
|
|
|
sld r0, r0, r7
|
|
|
|
|
addi r6, r5, VCORE_NAPPING_THREADS
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
4: lwarx r7, 0, r6
|
|
|
|
|
andc r7, r7, r0
|
|
|
|
|
stwcx. r7, 0, r6
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
bne 4b
|
|
|
|
|
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
/* See if the wake reason means we need to exit */
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
cmpdi r3, 0
|
|
|
|
|
bge kvm_novcpu_exit
|
|
|
|
|
|
2015-03-28 11:21:08 +08:00
|
|
|
/* See if our timeslice has expired (HDEC is negative) */
|
|
|
|
|
mfspr r0, SPRN_HDEC
|
2017-05-22 14:25:14 +08:00
|
|
|
EXTEND_HDEC(r0)
|
2015-03-28 11:21:08 +08:00
|
|
|
li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
|
2017-05-22 14:25:14 +08:00
|
|
|
cmpdi r0, 0
|
2015-03-28 11:21:08 +08:00
|
|
|
blt kvm_novcpu_exit
|
|
|
|
|
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
/* Got an IPI but other vcpus aren't yet exiting, must be a latecomer */
|
|
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
cmpdi r4, 0
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
beq kvmppc_primary_no_guest
|
|
|
|
|
|
|
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
|
|
|
|
addi r3, r4, VCPU_TB_RMENTRY
|
|
|
|
|
bl kvmhv_start_timing
|
|
|
|
|
#endif
|
|
|
|
|
b kvmppc_got_guest
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
|
|
|
|
|
kvm_novcpu_exit:
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
|
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
cmpdi r4, 0
|
|
|
|
|
beq 13f
|
|
|
|
|
addi r3, r4, VCPU_TB_RMEXIT
|
|
|
|
|
bl kvmhv_accumulate_time
|
|
|
|
|
#endif
|
2015-03-28 11:21:11 +08:00
|
|
|
13: mr r3, r12
|
2017-06-16 09:53:19 +08:00
|
|
|
stw r12, STACK_SLOT_TRAP(r1)
|
2015-03-28 11:21:11 +08:00
|
|
|
bl kvmhv_commence_exit
|
|
|
|
|
nop
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
b kvmhv_switch_to_host
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
/*
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
* We come in here when wakened from nap mode.
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
* Relocation is off and most register values are lost.
|
|
|
|
|
* r13 points to the PACA.
|
2017-06-13 21:05:51 +08:00
|
|
|
* r3 contains the SRR1 wakeup value, SRR1 is trashed.
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
*/
|
|
|
|
|
.globl kvm_start_guest
|
|
|
|
|
kvm_start_guest:
|
2014-04-11 18:31:58 +08:00
|
|
|
/* Set runlatch bit the minute you wake up from nap */
|
2015-03-28 11:21:04 +08:00
|
|
|
mfspr r0, SPRN_CTRLF
|
|
|
|
|
ori r0, r0, 1
|
|
|
|
|
mtspr SPRN_CTRLT, r0
|
2014-04-11 18:31:58 +08:00
|
|
|
|
2017-06-13 21:05:51 +08:00
|
|
|
/*
|
|
|
|
|
* Could avoid this and pass it through in r3. For now,
|
|
|
|
|
* code expects it to be in SRR1.
|
|
|
|
|
*/
|
|
|
|
|
mtspr SPRN_SRR1,r3
|
|
|
|
|
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
ld r2,PACATOC(r13)
|
|
|
|
|
|
2018-04-19 15:04:05 +08:00
|
|
|
li r0,0
|
|
|
|
|
stb r0,PACA_FTRACE_ENABLED(r13)
|
|
|
|
|
|
2012-02-03 08:54:17 +08:00
|
|
|
li r0,KVM_HWTHREAD_IN_KVM
|
|
|
|
|
stb r0,HSTATE_HWTHREAD_STATE(r13)
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
|
2012-02-03 08:54:17 +08:00
|
|
|
/* NV GPR values from power7_idle() will no longer be valid */
|
|
|
|
|
li r0,1
|
|
|
|
|
stb r0,PACA_NAPSTATELOST(r13)
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
|
2013-04-18 04:31:41 +08:00
|
|
|
/* were we napping due to cede? */
|
|
|
|
|
lbz r0,HSTATE_NAPPING(r13)
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
cmpwi r0,NAPPING_CEDE
|
|
|
|
|
beq kvm_end_cede
|
|
|
|
|
cmpwi r0,NAPPING_NOVCPU
|
|
|
|
|
beq kvm_novcpu_wakeup
|
|
|
|
|
|
|
|
|
|
ld r1,PACAEMERGSP(r13)
|
|
|
|
|
subi r1,r1,STACK_FRAME_OVERHEAD
|
2013-04-18 04:31:41 +08:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* We weren't napping due to cede, so this must be a secondary
|
|
|
|
|
* thread being woken up to run a guest, or being woken up due
|
|
|
|
|
* to a stray IPI. (Or due to some machine check or hypervisor
|
|
|
|
|
* maintenance interrupt while the core is in KVM.)
|
|
|
|
|
*/
|
2012-02-03 08:54:17 +08:00
|
|
|
|
|
|
|
|
/* Check the wake reason in SRR1 to see why we got here */
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
bl kvmppc_check_wake_reason
|
2016-08-19 13:35:46 +08:00
|
|
|
/*
|
|
|
|
|
* kvmppc_check_wake_reason could invoke a C routine, but we
|
|
|
|
|
* have no volatile registers to restore when we return.
|
|
|
|
|
*/
|
|
|
|
|
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
cmpdi r3, 0
|
|
|
|
|
bge kvm_no_guest
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
/* get vcore pointer, NULL if we have nothing to run */
|
|
|
|
|
ld r5,HSTATE_KVM_VCORE(r13)
|
|
|
|
|
cmpdi r5,0
|
|
|
|
|
/* if we have no vcore to run, go back to sleep */
|
2012-10-15 09:16:14 +08:00
|
|
|
beq kvm_no_guest
|
2012-02-03 08:54:17 +08:00
|
|
|
|
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
When a secondary hardware thread has finished running a KVM guest, we
currently put that thread into nap mode using a nap instruction in
the KVM code. This changes the code so that instead of doing a nap
instruction directly, we instead cause the call to power7_nap() that
put the thread into nap mode to return. The reason for doing this is
to avoid having the KVM code having to know what low-power mode to
put the thread into.
In the case of a secondary thread used to run a KVM guest, the thread
will be offline from the point of view of the host kernel, and the
relevant power7_nap() call is the one in pnv_smp_cpu_disable().
In this case we don't want to clear pending IPIs in the offline loop
in that function, since that might cause us to miss the wakeup for
the next time the thread needs to run a guest. To tell whether or
not to clear the interrupt, we use the SRR1 value returned from
power7_nap(), and check if it indicates an external interrupt. We
arrange that the return from power7_nap() when we have finished running
a guest returns 0, so pending interrupts don't get flushed in that
case.
Note that it is important a secondary thread that has finished
executing in the guest, or that didn't have a guest to run, should
not return to power7_nap's caller while the kvm_hstate.hwthread_req
flag in the PACA is non-zero, because the return from power7_nap
will reenable the MMU, and the MMU might still be in guest context.
In this situation we spin at low priority in real mode waiting for
hwthread_req to become zero.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-12-03 11:48:40 +08:00
|
|
|
kvm_secondary_got_guest:
|
|
|
|
|
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
/* Set HSTATE_DSCR(r13) to something sensible */
|
2015-05-21 14:43:03 +08:00
|
|
|
ld r6, PACA_DSCR_DEFAULT(r13)
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
std r6, HSTATE_DSCR(r13)
|
2011-12-06 03:47:26 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
/* On thread 0 of a subcore, set HDEC to max */
|
|
|
|
|
lbz r4, HSTATE_PTID(r13)
|
|
|
|
|
cmpwi r4, 0
|
|
|
|
|
bne 63f
|
2017-05-22 14:25:14 +08:00
|
|
|
LOAD_REG_ADDR(r6, decrementer_max)
|
|
|
|
|
ld r6, 0(r6)
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
mtspr SPRN_HDEC, r6
|
|
|
|
|
/* and set per-LPAR registers, if doing dynamic micro-threading */
|
|
|
|
|
ld r6, HSTATE_SPLIT_MODE(r13)
|
|
|
|
|
cmpdi r6, 0
|
|
|
|
|
beq 63f
|
KVM: PPC: Book3S HV: Run HPT guests on POWER9 radix hosts
This patch removes the restriction that a radix host can only run
radix guests, allowing us to run HPT (hashed page table) guests as
well. This is useful because it provides a way to run old guest
kernels that know about POWER8 but not POWER9.
Unfortunately, POWER9 currently has a restriction that all threads
in a given code must either all be in HPT mode, or all in radix mode.
This means that when entering a HPT guest, we have to obtain control
of all 4 threads in the core and get them to switch their LPIDR and
LPCR registers, even if they are not going to run a guest. On guest
exit we also have to get all threads to switch LPIDR and LPCR back
to host values.
To make this feasible, we require that KVM not be in the "independent
threads" mode, and that the CPU cores be in single-threaded mode from
the host kernel's perspective (only thread 0 online; threads 1, 2 and
3 offline). That allows us to use the same code as on POWER8 for
obtaining control of the secondary threads.
To manage the LPCR/LPIDR changes required, we extend the kvm_split_info
struct to contain the information needed by the secondary threads.
All threads perform a barrier synchronization (where all threads wait
for every other thread to reach the synchronization point) on guest
entry, both before and after loading LPCR and LPIDR. On guest exit,
they all once again perform a barrier synchronization both before
and after loading host values into LPCR and LPIDR.
Finally, it is also currently necessary to flush the entire TLB every
time we enter a HPT guest on a radix host. We do this on thread 0
with a loop of tlbiel instructions.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-19 11:11:23 +08:00
|
|
|
BEGIN_FTR_SECTION
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
ld r0, KVM_SPLIT_RPR(r6)
|
|
|
|
|
mtspr SPRN_RPR, r0
|
|
|
|
|
ld r0, KVM_SPLIT_PMMAR(r6)
|
|
|
|
|
mtspr SPRN_PMMAR, r0
|
|
|
|
|
ld r0, KVM_SPLIT_LDBAR(r6)
|
|
|
|
|
mtspr SPRN_LDBAR, r0
|
|
|
|
|
isync
|
KVM: PPC: Book3S HV: Run HPT guests on POWER9 radix hosts
This patch removes the restriction that a radix host can only run
radix guests, allowing us to run HPT (hashed page table) guests as
well. This is useful because it provides a way to run old guest
kernels that know about POWER8 but not POWER9.
Unfortunately, POWER9 currently has a restriction that all threads
in a given code must either all be in HPT mode, or all in radix mode.
This means that when entering a HPT guest, we have to obtain control
of all 4 threads in the core and get them to switch their LPIDR and
LPCR registers, even if they are not going to run a guest. On guest
exit we also have to get all threads to switch LPIDR and LPCR back
to host values.
To make this feasible, we require that KVM not be in the "independent
threads" mode, and that the CPU cores be in single-threaded mode from
the host kernel's perspective (only thread 0 online; threads 1, 2 and
3 offline). That allows us to use the same code as on POWER8 for
obtaining control of the secondary threads.
To manage the LPCR/LPIDR changes required, we extend the kvm_split_info
struct to contain the information needed by the secondary threads.
All threads perform a barrier synchronization (where all threads wait
for every other thread to reach the synchronization point) on guest
entry, both before and after loading LPCR and LPIDR. On guest exit,
they all once again perform a barrier synchronization both before
and after loading host values into LPCR and LPIDR.
Finally, it is also currently necessary to flush the entire TLB every
time we enter a HPT guest on a radix host. We do this on thread 0
with a loop of tlbiel instructions.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-19 11:11:23 +08:00
|
|
|
FTR_SECTION_ELSE
|
|
|
|
|
/* On P9 we use the split_info for coordinating LPCR changes */
|
|
|
|
|
lwz r4, KVM_SPLIT_DO_SET(r6)
|
|
|
|
|
cmpwi r4, 0
|
2018-02-09 01:38:53 +08:00
|
|
|
beq 1f
|
KVM: PPC: Book3S HV: Run HPT guests on POWER9 radix hosts
This patch removes the restriction that a radix host can only run
radix guests, allowing us to run HPT (hashed page table) guests as
well. This is useful because it provides a way to run old guest
kernels that know about POWER8 but not POWER9.
Unfortunately, POWER9 currently has a restriction that all threads
in a given code must either all be in HPT mode, or all in radix mode.
This means that when entering a HPT guest, we have to obtain control
of all 4 threads in the core and get them to switch their LPIDR and
LPCR registers, even if they are not going to run a guest. On guest
exit we also have to get all threads to switch LPIDR and LPCR back
to host values.
To make this feasible, we require that KVM not be in the "independent
threads" mode, and that the CPU cores be in single-threaded mode from
the host kernel's perspective (only thread 0 online; threads 1, 2 and
3 offline). That allows us to use the same code as on POWER8 for
obtaining control of the secondary threads.
To manage the LPCR/LPIDR changes required, we extend the kvm_split_info
struct to contain the information needed by the secondary threads.
All threads perform a barrier synchronization (where all threads wait
for every other thread to reach the synchronization point) on guest
entry, both before and after loading LPCR and LPIDR. On guest exit,
they all once again perform a barrier synchronization both before
and after loading host values into LPCR and LPIDR.
Finally, it is also currently necessary to flush the entire TLB every
time we enter a HPT guest on a radix host. We do this on thread 0
with a loop of tlbiel instructions.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-19 11:11:23 +08:00
|
|
|
mr r3, r6
|
|
|
|
|
bl kvmhv_p9_set_lpcr
|
|
|
|
|
nop
|
2018-02-09 01:38:53 +08:00
|
|
|
1:
|
KVM: PPC: Book3S HV: Run HPT guests on POWER9 radix hosts
This patch removes the restriction that a radix host can only run
radix guests, allowing us to run HPT (hashed page table) guests as
well. This is useful because it provides a way to run old guest
kernels that know about POWER8 but not POWER9.
Unfortunately, POWER9 currently has a restriction that all threads
in a given code must either all be in HPT mode, or all in radix mode.
This means that when entering a HPT guest, we have to obtain control
of all 4 threads in the core and get them to switch their LPIDR and
LPCR registers, even if they are not going to run a guest. On guest
exit we also have to get all threads to switch LPIDR and LPCR back
to host values.
To make this feasible, we require that KVM not be in the "independent
threads" mode, and that the CPU cores be in single-threaded mode from
the host kernel's perspective (only thread 0 online; threads 1, 2 and
3 offline). That allows us to use the same code as on POWER8 for
obtaining control of the secondary threads.
To manage the LPCR/LPIDR changes required, we extend the kvm_split_info
struct to contain the information needed by the secondary threads.
All threads perform a barrier synchronization (where all threads wait
for every other thread to reach the synchronization point) on guest
entry, both before and after loading LPCR and LPIDR. On guest exit,
they all once again perform a barrier synchronization both before
and after loading host values into LPCR and LPIDR.
Finally, it is also currently necessary to flush the entire TLB every
time we enter a HPT guest on a radix host. We do this on thread 0
with a loop of tlbiel instructions.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-19 11:11:23 +08:00
|
|
|
ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
63:
|
|
|
|
|
/* Order load of vcpu after load of vcore */
|
2015-03-28 11:21:06 +08:00
|
|
|
lwsync
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
bl kvmppc_hv_entry
|
2013-09-06 11:23:44 +08:00
|
|
|
|
|
|
|
|
/* Back from the guest, go back to nap */
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
/* Clear our vcpu and vcore pointers so we don't come back in early */
|
2013-09-06 11:23:44 +08:00
|
|
|
li r0, 0
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
std r0, HSTATE_KVM_VCPU(r13)
|
2013-11-16 14:46:03 +08:00
|
|
|
/*
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
* Once we clear HSTATE_KVM_VCORE(r13), the code in
|
2015-03-28 11:21:06 +08:00
|
|
|
* kvmppc_run_core() is going to assume that all our vcpu
|
|
|
|
|
* state is visible in memory. This lwsync makes sure
|
|
|
|
|
* that that is true.
|
2013-11-16 14:46:03 +08:00
|
|
|
*/
|
2013-09-06 11:23:44 +08:00
|
|
|
lwsync
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
std r0, HSTATE_KVM_VCORE(r13)
|
2013-09-06 11:23:44 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Fix TB corruption in guest exit path on HMI interrupt
When a guest is assigned to a core it converts the host Timebase (TB)
into guest TB by adding guest timebase offset before entering into
guest. During guest exit it restores the guest TB to host TB. This means
under certain conditions (Guest migration) host TB and guest TB can differ.
When we get an HMI for TB related issues the opal HMI handler would
try fixing errors and restore the correct host TB value. With no guest
running, we don't have any issues. But with guest running on the core
we run into TB corruption issues.
If we get an HMI while in the guest, the current HMI handler invokes opal
hmi handler before forcing guest to exit. The guest exit path subtracts
the guest TB offset from the current TB value which may have already
been restored with host value by opal hmi handler. This leads to incorrect
host and guest TB values.
With split-core, things become more complex. With split-core, TB also gets
split and each subcore gets its own TB register. When a hmi handler fixes
a TB error and restores the TB value, it affects all the TB values of
sibling subcores on the same core. On TB errors all the thread in the core
gets HMI. With existing code, the individual threads call opal hmi handle
independently which can easily throw TB out of sync if we have guest
running on subcores. Hence we will need to co-ordinate with all the
threads before making opal hmi handler call followed by TB resync.
This patch introduces a sibling subcore state structure (shared by all
threads in the core) in paca which holds information about whether sibling
subcores are in Guest mode or host mode. An array in_guest[] of size
MAX_SUBCORE_PER_CORE=4 is used to maintain the state of each subcore.
The subcore id is used as index into in_guest[] array. Only primary
thread entering/exiting the guest is responsible to set/unset its
designated array element.
On TB error, we get HMI interrupt on every thread on the core. Upon HMI,
this patch will now force guest to vacate the core/subcore. Primary
thread from each subcore will then turn off its respective bit
from the above bitmap during the guest exit path just after the
guest->host partition switch is complete.
All other threads that have just exited the guest OR were already in host
will wait until all other subcores clears their respective bit.
Once all the subcores turn off their respective bit, all threads will
will make call to opal hmi handler.
It is not necessary that opal hmi handler would resync the TB value for
every HMI interrupts. It would do so only for the HMI caused due to
TB errors. For rest, it would not touch TB value. Hence to make things
simpler, primary thread would call TB resync explicitly once for each
core immediately after opal hmi handler instead of subtracting guest
offset from TB. TB resync call will restore the TB with host value.
Thus we can be sure about the TB state.
One of the primary threads exiting the guest will take up the
responsibility of calling TB resync. It will use one of the top bits
(bit 63) from subcore state flags bitmap to make the decision. The first
primary thread (among the subcores) that is able to set the bit will
have to call the TB resync. Rest all other threads will wait until TB
resync is complete. Once TB resync is complete all threads will then
proceed.
Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-05-15 12:14:26 +08:00
|
|
|
/*
|
|
|
|
|
* All secondaries exiting guest will fall through this path.
|
|
|
|
|
* Before proceeding, just check for HMI interrupt and
|
|
|
|
|
* invoke opal hmi handler. By now we are sure that the
|
|
|
|
|
* primary thread on this core/subcore has already made partition
|
|
|
|
|
* switch/TB resync and we are good to call opal hmi handler.
|
|
|
|
|
*/
|
|
|
|
|
cmpwi r12, BOOK3S_INTERRUPT_HMI
|
|
|
|
|
bne kvm_no_guest
|
|
|
|
|
|
|
|
|
|
li r3,0 /* NULL argument */
|
|
|
|
|
bl hmi_exception_realmode
|
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
When a secondary hardware thread has finished running a KVM guest, we
currently put that thread into nap mode using a nap instruction in
the KVM code. This changes the code so that instead of doing a nap
instruction directly, we instead cause the call to power7_nap() that
put the thread into nap mode to return. The reason for doing this is
to avoid having the KVM code having to know what low-power mode to
put the thread into.
In the case of a secondary thread used to run a KVM guest, the thread
will be offline from the point of view of the host kernel, and the
relevant power7_nap() call is the one in pnv_smp_cpu_disable().
In this case we don't want to clear pending IPIs in the offline loop
in that function, since that might cause us to miss the wakeup for
the next time the thread needs to run a guest. To tell whether or
not to clear the interrupt, we use the SRR1 value returned from
power7_nap(), and check if it indicates an external interrupt. We
arrange that the return from power7_nap() when we have finished running
a guest returns 0, so pending interrupts don't get flushed in that
case.
Note that it is important a secondary thread that has finished
executing in the guest, or that didn't have a guest to run, should
not return to power7_nap's caller while the kvm_hstate.hwthread_req
flag in the PACA is non-zero, because the return from power7_nap
will reenable the MMU, and the MMU might still be in guest context.
In this situation we spin at low priority in real mode waiting for
hwthread_req to become zero.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-12-03 11:48:40 +08:00
|
|
|
/*
|
|
|
|
|
* At this point we have finished executing in the guest.
|
|
|
|
|
* We need to wait for hwthread_req to become zero, since
|
|
|
|
|
* we may not turn on the MMU while hwthread_req is non-zero.
|
|
|
|
|
* While waiting we also need to check if we get given a vcpu to run.
|
|
|
|
|
*/
|
2013-09-06 11:23:44 +08:00
|
|
|
kvm_no_guest:
|
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
When a secondary hardware thread has finished running a KVM guest, we
currently put that thread into nap mode using a nap instruction in
the KVM code. This changes the code so that instead of doing a nap
instruction directly, we instead cause the call to power7_nap() that
put the thread into nap mode to return. The reason for doing this is
to avoid having the KVM code having to know what low-power mode to
put the thread into.
In the case of a secondary thread used to run a KVM guest, the thread
will be offline from the point of view of the host kernel, and the
relevant power7_nap() call is the one in pnv_smp_cpu_disable().
In this case we don't want to clear pending IPIs in the offline loop
in that function, since that might cause us to miss the wakeup for
the next time the thread needs to run a guest. To tell whether or
not to clear the interrupt, we use the SRR1 value returned from
power7_nap(), and check if it indicates an external interrupt. We
arrange that the return from power7_nap() when we have finished running
a guest returns 0, so pending interrupts don't get flushed in that
case.
Note that it is important a secondary thread that has finished
executing in the guest, or that didn't have a guest to run, should
not return to power7_nap's caller while the kvm_hstate.hwthread_req
flag in the PACA is non-zero, because the return from power7_nap
will reenable the MMU, and the MMU might still be in guest context.
In this situation we spin at low priority in real mode waiting for
hwthread_req to become zero.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-12-03 11:48:40 +08:00
|
|
|
lbz r3, HSTATE_HWTHREAD_REQ(r13)
|
|
|
|
|
cmpwi r3, 0
|
|
|
|
|
bne 53f
|
|
|
|
|
HMT_MEDIUM
|
|
|
|
|
li r0, KVM_HWTHREAD_IN_KERNEL
|
2013-09-06 11:23:44 +08:00
|
|
|
stb r0, HSTATE_HWTHREAD_STATE(r13)
|
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
When a secondary hardware thread has finished running a KVM guest, we
currently put that thread into nap mode using a nap instruction in
the KVM code. This changes the code so that instead of doing a nap
instruction directly, we instead cause the call to power7_nap() that
put the thread into nap mode to return. The reason for doing this is
to avoid having the KVM code having to know what low-power mode to
put the thread into.
In the case of a secondary thread used to run a KVM guest, the thread
will be offline from the point of view of the host kernel, and the
relevant power7_nap() call is the one in pnv_smp_cpu_disable().
In this case we don't want to clear pending IPIs in the offline loop
in that function, since that might cause us to miss the wakeup for
the next time the thread needs to run a guest. To tell whether or
not to clear the interrupt, we use the SRR1 value returned from
power7_nap(), and check if it indicates an external interrupt. We
arrange that the return from power7_nap() when we have finished running
a guest returns 0, so pending interrupts don't get flushed in that
case.
Note that it is important a secondary thread that has finished
executing in the guest, or that didn't have a guest to run, should
not return to power7_nap's caller while the kvm_hstate.hwthread_req
flag in the PACA is non-zero, because the return from power7_nap
will reenable the MMU, and the MMU might still be in guest context.
In this situation we spin at low priority in real mode waiting for
hwthread_req to become zero.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-12-03 11:48:40 +08:00
|
|
|
/* need to recheck hwthread_req after a barrier, to avoid race */
|
|
|
|
|
sync
|
|
|
|
|
lbz r3, HSTATE_HWTHREAD_REQ(r13)
|
|
|
|
|
cmpwi r3, 0
|
|
|
|
|
bne 54f
|
|
|
|
|
/*
|
2016-07-08 14:20:46 +08:00
|
|
|
* We jump to pnv_wakeup_loss, which will return to the caller
|
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
When a secondary hardware thread has finished running a KVM guest, we
currently put that thread into nap mode using a nap instruction in
the KVM code. This changes the code so that instead of doing a nap
instruction directly, we instead cause the call to power7_nap() that
put the thread into nap mode to return. The reason for doing this is
to avoid having the KVM code having to know what low-power mode to
put the thread into.
In the case of a secondary thread used to run a KVM guest, the thread
will be offline from the point of view of the host kernel, and the
relevant power7_nap() call is the one in pnv_smp_cpu_disable().
In this case we don't want to clear pending IPIs in the offline loop
in that function, since that might cause us to miss the wakeup for
the next time the thread needs to run a guest. To tell whether or
not to clear the interrupt, we use the SRR1 value returned from
power7_nap(), and check if it indicates an external interrupt. We
arrange that the return from power7_nap() when we have finished running
a guest returns 0, so pending interrupts don't get flushed in that
case.
Note that it is important a secondary thread that has finished
executing in the guest, or that didn't have a guest to run, should
not return to power7_nap's caller while the kvm_hstate.hwthread_req
flag in the PACA is non-zero, because the return from power7_nap
will reenable the MMU, and the MMU might still be in guest context.
In this situation we spin at low priority in real mode waiting for
hwthread_req to become zero.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-12-03 11:48:40 +08:00
|
|
|
* of power7_nap in the powernv cpu offline loop. The value we
|
2017-06-13 21:05:51 +08:00
|
|
|
* put in r3 becomes the return value for power7_nap. pnv_wakeup_loss
|
|
|
|
|
* requires SRR1 in r12.
|
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
When a secondary hardware thread has finished running a KVM guest, we
currently put that thread into nap mode using a nap instruction in
the KVM code. This changes the code so that instead of doing a nap
instruction directly, we instead cause the call to power7_nap() that
put the thread into nap mode to return. The reason for doing this is
to avoid having the KVM code having to know what low-power mode to
put the thread into.
In the case of a secondary thread used to run a KVM guest, the thread
will be offline from the point of view of the host kernel, and the
relevant power7_nap() call is the one in pnv_smp_cpu_disable().
In this case we don't want to clear pending IPIs in the offline loop
in that function, since that might cause us to miss the wakeup for
the next time the thread needs to run a guest. To tell whether or
not to clear the interrupt, we use the SRR1 value returned from
power7_nap(), and check if it indicates an external interrupt. We
arrange that the return from power7_nap() when we have finished running
a guest returns 0, so pending interrupts don't get flushed in that
case.
Note that it is important a secondary thread that has finished
executing in the guest, or that didn't have a guest to run, should
not return to power7_nap's caller while the kvm_hstate.hwthread_req
flag in the PACA is non-zero, because the return from power7_nap
will reenable the MMU, and the MMU might still be in guest context.
In this situation we spin at low priority in real mode waiting for
hwthread_req to become zero.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-12-03 11:48:40 +08:00
|
|
|
*/
|
2013-09-06 11:23:44 +08:00
|
|
|
li r3, LPCR_PECE0
|
|
|
|
|
mfspr r4, SPRN_LPCR
|
|
|
|
|
rlwimi r4, r3, 0, LPCR_PECE0 | LPCR_PECE1
|
|
|
|
|
mtspr SPRN_LPCR, r4
|
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
When a secondary hardware thread has finished running a KVM guest, we
currently put that thread into nap mode using a nap instruction in
the KVM code. This changes the code so that instead of doing a nap
instruction directly, we instead cause the call to power7_nap() that
put the thread into nap mode to return. The reason for doing this is
to avoid having the KVM code having to know what low-power mode to
put the thread into.
In the case of a secondary thread used to run a KVM guest, the thread
will be offline from the point of view of the host kernel, and the
relevant power7_nap() call is the one in pnv_smp_cpu_disable().
In this case we don't want to clear pending IPIs in the offline loop
in that function, since that might cause us to miss the wakeup for
the next time the thread needs to run a guest. To tell whether or
not to clear the interrupt, we use the SRR1 value returned from
power7_nap(), and check if it indicates an external interrupt. We
arrange that the return from power7_nap() when we have finished running
a guest returns 0, so pending interrupts don't get flushed in that
case.
Note that it is important a secondary thread that has finished
executing in the guest, or that didn't have a guest to run, should
not return to power7_nap's caller while the kvm_hstate.hwthread_req
flag in the PACA is non-zero, because the return from power7_nap
will reenable the MMU, and the MMU might still be in guest context.
In this situation we spin at low priority in real mode waiting for
hwthread_req to become zero.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-12-03 11:48:40 +08:00
|
|
|
li r3, 0
|
2017-06-13 21:05:51 +08:00
|
|
|
mfspr r12,SPRN_SRR1
|
2016-07-08 14:20:46 +08:00
|
|
|
b pnv_wakeup_loss
|
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
When a secondary hardware thread has finished running a KVM guest, we
currently put that thread into nap mode using a nap instruction in
the KVM code. This changes the code so that instead of doing a nap
instruction directly, we instead cause the call to power7_nap() that
put the thread into nap mode to return. The reason for doing this is
to avoid having the KVM code having to know what low-power mode to
put the thread into.
In the case of a secondary thread used to run a KVM guest, the thread
will be offline from the point of view of the host kernel, and the
relevant power7_nap() call is the one in pnv_smp_cpu_disable().
In this case we don't want to clear pending IPIs in the offline loop
in that function, since that might cause us to miss the wakeup for
the next time the thread needs to run a guest. To tell whether or
not to clear the interrupt, we use the SRR1 value returned from
power7_nap(), and check if it indicates an external interrupt. We
arrange that the return from power7_nap() when we have finished running
a guest returns 0, so pending interrupts don't get flushed in that
case.
Note that it is important a secondary thread that has finished
executing in the guest, or that didn't have a guest to run, should
not return to power7_nap's caller while the kvm_hstate.hwthread_req
flag in the PACA is non-zero, because the return from power7_nap
will reenable the MMU, and the MMU might still be in guest context.
In this situation we spin at low priority in real mode waiting for
hwthread_req to become zero.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-12-03 11:48:40 +08:00
|
|
|
|
|
|
|
|
53: HMT_LOW
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
cmpdi r5, 0
|
|
|
|
|
bne 60f
|
|
|
|
|
ld r3, HSTATE_SPLIT_MODE(r13)
|
|
|
|
|
cmpdi r3, 0
|
|
|
|
|
beq kvm_no_guest
|
KVM: PPC: Book3S HV: Run HPT guests on POWER9 radix hosts
This patch removes the restriction that a radix host can only run
radix guests, allowing us to run HPT (hashed page table) guests as
well. This is useful because it provides a way to run old guest
kernels that know about POWER8 but not POWER9.
Unfortunately, POWER9 currently has a restriction that all threads
in a given code must either all be in HPT mode, or all in radix mode.
This means that when entering a HPT guest, we have to obtain control
of all 4 threads in the core and get them to switch their LPIDR and
LPCR registers, even if they are not going to run a guest. On guest
exit we also have to get all threads to switch LPIDR and LPCR back
to host values.
To make this feasible, we require that KVM not be in the "independent
threads" mode, and that the CPU cores be in single-threaded mode from
the host kernel's perspective (only thread 0 online; threads 1, 2 and
3 offline). That allows us to use the same code as on POWER8 for
obtaining control of the secondary threads.
To manage the LPCR/LPIDR changes required, we extend the kvm_split_info
struct to contain the information needed by the secondary threads.
All threads perform a barrier synchronization (where all threads wait
for every other thread to reach the synchronization point) on guest
entry, both before and after loading LPCR and LPIDR. On guest exit,
they all once again perform a barrier synchronization both before
and after loading host values into LPCR and LPIDR.
Finally, it is also currently necessary to flush the entire TLB every
time we enter a HPT guest on a radix host. We do this on thread 0
with a loop of tlbiel instructions.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-19 11:11:23 +08:00
|
|
|
lwz r0, KVM_SPLIT_DO_SET(r3)
|
|
|
|
|
cmpwi r0, 0
|
|
|
|
|
bne kvmhv_do_set
|
|
|
|
|
lwz r0, KVM_SPLIT_DO_RESTORE(r3)
|
|
|
|
|
cmpwi r0, 0
|
|
|
|
|
bne kvmhv_do_restore
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
lbz r0, KVM_SPLIT_DO_NAP(r3)
|
|
|
|
|
cmpwi r0, 0
|
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
When a secondary hardware thread has finished running a KVM guest, we
currently put that thread into nap mode using a nap instruction in
the KVM code. This changes the code so that instead of doing a nap
instruction directly, we instead cause the call to power7_nap() that
put the thread into nap mode to return. The reason for doing this is
to avoid having the KVM code having to know what low-power mode to
put the thread into.
In the case of a secondary thread used to run a KVM guest, the thread
will be offline from the point of view of the host kernel, and the
relevant power7_nap() call is the one in pnv_smp_cpu_disable().
In this case we don't want to clear pending IPIs in the offline loop
in that function, since that might cause us to miss the wakeup for
the next time the thread needs to run a guest. To tell whether or
not to clear the interrupt, we use the SRR1 value returned from
power7_nap(), and check if it indicates an external interrupt. We
arrange that the return from power7_nap() when we have finished running
a guest returns 0, so pending interrupts don't get flushed in that
case.
Note that it is important a secondary thread that has finished
executing in the guest, or that didn't have a guest to run, should
not return to power7_nap's caller while the kvm_hstate.hwthread_req
flag in the PACA is non-zero, because the return from power7_nap
will reenable the MMU, and the MMU might still be in guest context.
In this situation we spin at low priority in real mode waiting for
hwthread_req to become zero.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-12-03 11:48:40 +08:00
|
|
|
beq kvm_no_guest
|
|
|
|
|
HMT_MEDIUM
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
b kvm_unsplit_nap
|
|
|
|
|
60: HMT_MEDIUM
|
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
When a secondary hardware thread has finished running a KVM guest, we
currently put that thread into nap mode using a nap instruction in
the KVM code. This changes the code so that instead of doing a nap
instruction directly, we instead cause the call to power7_nap() that
put the thread into nap mode to return. The reason for doing this is
to avoid having the KVM code having to know what low-power mode to
put the thread into.
In the case of a secondary thread used to run a KVM guest, the thread
will be offline from the point of view of the host kernel, and the
relevant power7_nap() call is the one in pnv_smp_cpu_disable().
In this case we don't want to clear pending IPIs in the offline loop
in that function, since that might cause us to miss the wakeup for
the next time the thread needs to run a guest. To tell whether or
not to clear the interrupt, we use the SRR1 value returned from
power7_nap(), and check if it indicates an external interrupt. We
arrange that the return from power7_nap() when we have finished running
a guest returns 0, so pending interrupts don't get flushed in that
case.
Note that it is important a secondary thread that has finished
executing in the guest, or that didn't have a guest to run, should
not return to power7_nap's caller while the kvm_hstate.hwthread_req
flag in the PACA is non-zero, because the return from power7_nap
will reenable the MMU, and the MMU might still be in guest context.
In this situation we spin at low priority in real mode waiting for
hwthread_req to become zero.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-12-03 11:48:40 +08:00
|
|
|
b kvm_secondary_got_guest
|
|
|
|
|
|
|
|
|
|
54: li r0, KVM_HWTHREAD_IN_KVM
|
|
|
|
|
stb r0, HSTATE_HWTHREAD_STATE(r13)
|
|
|
|
|
b kvm_no_guest
|
2013-09-06 11:23:44 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Run HPT guests on POWER9 radix hosts
This patch removes the restriction that a radix host can only run
radix guests, allowing us to run HPT (hashed page table) guests as
well. This is useful because it provides a way to run old guest
kernels that know about POWER8 but not POWER9.
Unfortunately, POWER9 currently has a restriction that all threads
in a given code must either all be in HPT mode, or all in radix mode.
This means that when entering a HPT guest, we have to obtain control
of all 4 threads in the core and get them to switch their LPIDR and
LPCR registers, even if they are not going to run a guest. On guest
exit we also have to get all threads to switch LPIDR and LPCR back
to host values.
To make this feasible, we require that KVM not be in the "independent
threads" mode, and that the CPU cores be in single-threaded mode from
the host kernel's perspective (only thread 0 online; threads 1, 2 and
3 offline). That allows us to use the same code as on POWER8 for
obtaining control of the secondary threads.
To manage the LPCR/LPIDR changes required, we extend the kvm_split_info
struct to contain the information needed by the secondary threads.
All threads perform a barrier synchronization (where all threads wait
for every other thread to reach the synchronization point) on guest
entry, both before and after loading LPCR and LPIDR. On guest exit,
they all once again perform a barrier synchronization both before
and after loading host values into LPCR and LPIDR.
Finally, it is also currently necessary to flush the entire TLB every
time we enter a HPT guest on a radix host. We do this on thread 0
with a loop of tlbiel instructions.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-19 11:11:23 +08:00
|
|
|
kvmhv_do_set:
|
|
|
|
|
/* Set LPCR, LPIDR etc. on P9 */
|
|
|
|
|
HMT_MEDIUM
|
|
|
|
|
bl kvmhv_p9_set_lpcr
|
|
|
|
|
nop
|
|
|
|
|
b kvm_no_guest
|
|
|
|
|
|
|
|
|
|
kvmhv_do_restore:
|
|
|
|
|
HMT_MEDIUM
|
|
|
|
|
bl kvmhv_p9_restore_lpcr
|
|
|
|
|
nop
|
|
|
|
|
b kvm_no_guest
|
|
|
|
|
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
/*
|
|
|
|
|
* Here the primary thread is trying to return the core to
|
|
|
|
|
* whole-core mode, so we need to nap.
|
|
|
|
|
*/
|
|
|
|
|
kvm_unsplit_nap:
|
KVM: PPC: Book3S HV: Fix TB corruption in guest exit path on HMI interrupt
When a guest is assigned to a core it converts the host Timebase (TB)
into guest TB by adding guest timebase offset before entering into
guest. During guest exit it restores the guest TB to host TB. This means
under certain conditions (Guest migration) host TB and guest TB can differ.
When we get an HMI for TB related issues the opal HMI handler would
try fixing errors and restore the correct host TB value. With no guest
running, we don't have any issues. But with guest running on the core
we run into TB corruption issues.
If we get an HMI while in the guest, the current HMI handler invokes opal
hmi handler before forcing guest to exit. The guest exit path subtracts
the guest TB offset from the current TB value which may have already
been restored with host value by opal hmi handler. This leads to incorrect
host and guest TB values.
With split-core, things become more complex. With split-core, TB also gets
split and each subcore gets its own TB register. When a hmi handler fixes
a TB error and restores the TB value, it affects all the TB values of
sibling subcores on the same core. On TB errors all the thread in the core
gets HMI. With existing code, the individual threads call opal hmi handle
independently which can easily throw TB out of sync if we have guest
running on subcores. Hence we will need to co-ordinate with all the
threads before making opal hmi handler call followed by TB resync.
This patch introduces a sibling subcore state structure (shared by all
threads in the core) in paca which holds information about whether sibling
subcores are in Guest mode or host mode. An array in_guest[] of size
MAX_SUBCORE_PER_CORE=4 is used to maintain the state of each subcore.
The subcore id is used as index into in_guest[] array. Only primary
thread entering/exiting the guest is responsible to set/unset its
designated array element.
On TB error, we get HMI interrupt on every thread on the core. Upon HMI,
this patch will now force guest to vacate the core/subcore. Primary
thread from each subcore will then turn off its respective bit
from the above bitmap during the guest exit path just after the
guest->host partition switch is complete.
All other threads that have just exited the guest OR were already in host
will wait until all other subcores clears their respective bit.
Once all the subcores turn off their respective bit, all threads will
will make call to opal hmi handler.
It is not necessary that opal hmi handler would resync the TB value for
every HMI interrupts. It would do so only for the HMI caused due to
TB errors. For rest, it would not touch TB value. Hence to make things
simpler, primary thread would call TB resync explicitly once for each
core immediately after opal hmi handler instead of subtracting guest
offset from TB. TB resync call will restore the TB with host value.
Thus we can be sure about the TB state.
One of the primary threads exiting the guest will take up the
responsibility of calling TB resync. It will use one of the top bits
(bit 63) from subcore state flags bitmap to make the decision. The first
primary thread (among the subcores) that is able to set the bit will
have to call the TB resync. Rest all other threads will wait until TB
resync is complete. Once TB resync is complete all threads will then
proceed.
Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-05-15 12:14:26 +08:00
|
|
|
/*
|
|
|
|
|
* When secondaries are napping in kvm_unsplit_nap() with
|
|
|
|
|
* hwthread_req = 1, HMI goes ignored even though subcores are
|
|
|
|
|
* already exited the guest. Hence HMI keeps waking up secondaries
|
|
|
|
|
* from nap in a loop and secondaries always go back to nap since
|
|
|
|
|
* no vcore is assigned to them. This makes impossible for primary
|
|
|
|
|
* thread to get hold of secondary threads resulting into a soft
|
|
|
|
|
* lockup in KVM path.
|
|
|
|
|
*
|
|
|
|
|
* Let us check if HMI is pending and handle it before we go to nap.
|
|
|
|
|
*/
|
|
|
|
|
cmpwi r12, BOOK3S_INTERRUPT_HMI
|
|
|
|
|
bne 55f
|
|
|
|
|
li r3, 0 /* NULL argument */
|
|
|
|
|
bl hmi_exception_realmode
|
|
|
|
|
55:
|
2015-09-03 00:18:58 +08:00
|
|
|
/*
|
|
|
|
|
* Ensure that secondary doesn't nap when it has
|
|
|
|
|
* its vcore pointer set.
|
|
|
|
|
*/
|
|
|
|
|
sync /* matches smp_mb() before setting split_info.do_nap */
|
|
|
|
|
ld r0, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
cmpdi r0, 0
|
|
|
|
|
bne kvm_no_guest
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
/* clear any pending message */
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
lis r6, (PPC_DBELL_SERVER << (63-36))@h
|
|
|
|
|
PPC_MSGCLR(6)
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
|
|
|
|
/* Set kvm_split_mode.napped[tid] = 1 */
|
|
|
|
|
ld r3, HSTATE_SPLIT_MODE(r13)
|
|
|
|
|
li r0, 1
|
KVM: PPC: Book3S HV: Run HPT guests on POWER9 radix hosts
This patch removes the restriction that a radix host can only run
radix guests, allowing us to run HPT (hashed page table) guests as
well. This is useful because it provides a way to run old guest
kernels that know about POWER8 but not POWER9.
Unfortunately, POWER9 currently has a restriction that all threads
in a given code must either all be in HPT mode, or all in radix mode.
This means that when entering a HPT guest, we have to obtain control
of all 4 threads in the core and get them to switch their LPIDR and
LPCR registers, even if they are not going to run a guest. On guest
exit we also have to get all threads to switch LPIDR and LPCR back
to host values.
To make this feasible, we require that KVM not be in the "independent
threads" mode, and that the CPU cores be in single-threaded mode from
the host kernel's perspective (only thread 0 online; threads 1, 2 and
3 offline). That allows us to use the same code as on POWER8 for
obtaining control of the secondary threads.
To manage the LPCR/LPIDR changes required, we extend the kvm_split_info
struct to contain the information needed by the secondary threads.
All threads perform a barrier synchronization (where all threads wait
for every other thread to reach the synchronization point) on guest
entry, both before and after loading LPCR and LPIDR. On guest exit,
they all once again perform a barrier synchronization both before
and after loading host values into LPCR and LPIDR.
Finally, it is also currently necessary to flush the entire TLB every
time we enter a HPT guest on a radix host. We do this on thread 0
with a loop of tlbiel instructions.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-19 11:11:23 +08:00
|
|
|
lbz r4, HSTATE_TID(r13)
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
addi r4, r4, KVM_SPLIT_NAPPED
|
|
|
|
|
stbx r0, r3, r4
|
|
|
|
|
/* Check the do_nap flag again after setting napped[] */
|
|
|
|
|
sync
|
|
|
|
|
lbz r0, KVM_SPLIT_DO_NAP(r3)
|
|
|
|
|
cmpwi r0, 0
|
|
|
|
|
beq 57f
|
|
|
|
|
li r3, (LPCR_PECEDH | LPCR_PECE0) >> 4
|
2016-11-18 11:34:07 +08:00
|
|
|
mfspr r5, SPRN_LPCR
|
|
|
|
|
rlwimi r5, r3, 4, (LPCR_PECEDP | LPCR_PECEDH | LPCR_PECE0 | LPCR_PECE1)
|
|
|
|
|
b kvm_nap_sequence
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
|
|
|
|
|
57: li r0, 0
|
|
|
|
|
stbx r0, r3, r4
|
|
|
|
|
b kvm_no_guest
|
|
|
|
|
|
2013-09-06 11:23:44 +08:00
|
|
|
/******************************************************************************
|
|
|
|
|
* *
|
|
|
|
|
* Entry code *
|
|
|
|
|
* *
|
|
|
|
|
*****************************************************************************/
|
|
|
|
|
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
.global kvmppc_hv_entry
|
|
|
|
|
kvmppc_hv_entry:
|
|
|
|
|
|
|
|
|
|
/* Required state:
|
|
|
|
|
*
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
* R4 = vcpu pointer (or NULL)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
* MSR = ~IR|DR
|
|
|
|
|
* R13 = PACA
|
|
|
|
|
* R1 = host R1
|
2014-08-19 12:59:30 +08:00
|
|
|
* R2 = TOC
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
* all other volatile GPRS = free
|
2017-01-30 18:21:45 +08:00
|
|
|
* Does not preserve non-volatile GPRs or CR fields
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
*/
|
|
|
|
|
mflr r0
|
2013-09-06 11:23:44 +08:00
|
|
|
std r0, PPC_LR_STKOFF(r1)
|
2017-06-16 09:53:19 +08:00
|
|
|
stdu r1, -SFS(r1)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
|
|
|
|
|
/* Save R1 in the PACA */
|
|
|
|
|
std r1, HSTATE_HOST_R1(r13)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Better handling of exceptions that happen in real mode
When an interrupt or exception happens in the guest that comes to the
host, the CPU goes to hypervisor real mode (MMU off) to handle the
exception but doesn't change the MMU context. After saving a few
registers, we then clear the "in guest" flag. If, for any reason,
we get an exception in the real-mode code, that then gets handled
by the normal kernel exception handlers, which turn the MMU on. This
is disastrous if the MMU is still set to the guest context, since we
end up executing instructions from random places in the guest kernel
with hypervisor privilege.
In order to catch this situation, we define a new value for the "in guest"
flag, KVM_GUEST_MODE_HOST_HV, to indicate that we are in hypervisor real
mode with guest MMU context. If the "in guest" flag is set to this value,
we branch off to an emergency handler. For the moment, this just does
a branch to self to stop the CPU from doing anything further.
While we're here, we define another new flag value to indicate that we
are in a HV guest, as distinct from a PR guest. This will be useful
when we have a kernel that can support both PR and HV guests concurrently.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-10-04 19:45:04 +08:00
|
|
|
li r6, KVM_GUEST_MODE_HOST_HV
|
|
|
|
|
stb r6, HSTATE_IN_GUEST(r13)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
|
|
|
|
/* Store initial timestamp */
|
|
|
|
|
cmpdi r4, 0
|
|
|
|
|
beq 1f
|
|
|
|
|
addi r3, r4, VCPU_TB_RMENTRY
|
|
|
|
|
bl kvmhv_start_timing
|
|
|
|
|
1:
|
|
|
|
|
#endif
|
2017-01-30 18:21:45 +08:00
|
|
|
|
|
|
|
|
/* Use cr7 as an indication of radix mode */
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
ld r9, VCORE_KVM(r5) /* pointer to struct kvm */
|
|
|
|
|
lbz r0, KVM_RADIX(r9)
|
|
|
|
|
cmpwi cr7, r0, 0
|
|
|
|
|
|
KVM: PPC: book3s_hv: Add support for PPC970-family processors
This adds support for running KVM guests in supervisor mode on those
PPC970 processors that have a usable hypervisor mode. Unfortunately,
Apple G5 machines have supervisor mode disabled (MSR[HV] is forced to
1), but the YDL PowerStation does have a usable hypervisor mode.
There are several differences between the PPC970 and POWER7 in how
guests are managed. These differences are accommodated using the
CPU_FTR_ARCH_201 (PPC970) and CPU_FTR_ARCH_206 (POWER7) CPU feature
bits. Notably, on PPC970:
* The LPCR, LPID or RMOR registers don't exist, and the functions of
those registers are provided by bits in HID4 and one bit in HID0.
* External interrupts can be directed to the hypervisor, but unlike
POWER7 they are masked by MSR[EE] in non-hypervisor modes and use
SRR0/1 not HSRR0/1.
* There is no virtual RMA (VRMA) mode; the guest must use an RMO
(real mode offset) area.
* The TLB entries are not tagged with the LPID, so it is necessary to
flush the whole TLB on partition switch. Furthermore, when switching
partitions we have to ensure that no other CPU is executing the tlbie
or tlbsync instructions in either the old or the new partition,
otherwise undefined behaviour can occur.
* The PMU has 8 counters (PMC registers) rather than 6.
* The DSCR, PURR, SPURR, AMR, AMOR, UAMOR registers don't exist.
* The SLB has 64 entries rather than 32.
* There is no mediated external interrupt facility, so if we switch to
a guest that has a virtual external interrupt pending but the guest
has MSR[EE] = 0, we have to arrange to have an interrupt pending for
it so that we can get control back once it re-enables interrupts. We
do that by sending ourselves an IPI with smp_send_reschedule after
hard-disabling interrupts.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:40:08 +08:00
|
|
|
/*
|
2014-12-03 10:30:38 +08:00
|
|
|
* POWER7/POWER8 host -> guest partition switch code.
|
KVM: PPC: book3s_hv: Add support for PPC970-family processors
This adds support for running KVM guests in supervisor mode on those
PPC970 processors that have a usable hypervisor mode. Unfortunately,
Apple G5 machines have supervisor mode disabled (MSR[HV] is forced to
1), but the YDL PowerStation does have a usable hypervisor mode.
There are several differences between the PPC970 and POWER7 in how
guests are managed. These differences are accommodated using the
CPU_FTR_ARCH_201 (PPC970) and CPU_FTR_ARCH_206 (POWER7) CPU feature
bits. Notably, on PPC970:
* The LPCR, LPID or RMOR registers don't exist, and the functions of
those registers are provided by bits in HID4 and one bit in HID0.
* External interrupts can be directed to the hypervisor, but unlike
POWER7 they are masked by MSR[EE] in non-hypervisor modes and use
SRR0/1 not HSRR0/1.
* There is no virtual RMA (VRMA) mode; the guest must use an RMO
(real mode offset) area.
* The TLB entries are not tagged with the LPID, so it is necessary to
flush the whole TLB on partition switch. Furthermore, when switching
partitions we have to ensure that no other CPU is executing the tlbie
or tlbsync instructions in either the old or the new partition,
otherwise undefined behaviour can occur.
* The PMU has 8 counters (PMC registers) rather than 6.
* The DSCR, PURR, SPURR, AMR, AMOR, UAMOR registers don't exist.
* The SLB has 64 entries rather than 32.
* There is no mediated external interrupt facility, so if we switch to
a guest that has a virtual external interrupt pending but the guest
has MSR[EE] = 0, we have to arrange to have an interrupt pending for
it so that we can get control back once it re-enables interrupts. We
do that by sending ourselves an IPI with smp_send_reschedule after
hard-disabling interrupts.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:40:08 +08:00
|
|
|
* We don't have to lock against concurrent tlbies,
|
|
|
|
|
* but we do have to coordinate across hardware threads.
|
|
|
|
|
*/
|
2015-03-28 11:21:09 +08:00
|
|
|
/* Set bit in entry map iff exit map is zero. */
|
|
|
|
|
li r7, 1
|
|
|
|
|
lbz r6, HSTATE_PTID(r13)
|
|
|
|
|
sld r7, r7, r6
|
2017-01-30 18:21:45 +08:00
|
|
|
addi r8, r5, VCORE_ENTRY_EXIT
|
|
|
|
|
21: lwarx r3, 0, r8
|
2015-03-28 11:21:09 +08:00
|
|
|
cmpwi r3, 0x100 /* any threads starting to exit? */
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
bge secondary_too_late /* if so we're too late to the party */
|
2015-03-28 11:21:09 +08:00
|
|
|
or r3, r3, r7
|
2017-01-30 18:21:45 +08:00
|
|
|
stwcx. r3, 0, r8
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
bne 21b
|
|
|
|
|
|
|
|
|
|
/* Primary thread switches to guest partition. */
|
|
|
|
|
cmpwi r6,0
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
bne 10f
|
2018-05-17 15:06:29 +08:00
|
|
|
|
|
|
|
|
/* Radix has already switched LPID and flushed core TLB */
|
|
|
|
|
bne cr7, 22f
|
|
|
|
|
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
lwz r7,KVM_LPID(r9)
|
2016-11-16 19:25:20 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r6,KVM_SDR1(r9)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
li r0,LPID_RSVD /* switch to reserved LPID */
|
|
|
|
|
mtspr SPRN_LPID,r0
|
|
|
|
|
ptesync
|
|
|
|
|
mtspr SPRN_SDR1,r6 /* switch to partition page table */
|
2016-11-16 19:25:20 +08:00
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
mtspr SPRN_LPID,r7
|
|
|
|
|
isync
|
KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
When we change or remove a HPT (hashed page table) entry, we can do
either a global TLB invalidation (tlbie) that works across the whole
machine, or a local invalidation (tlbiel) that only affects this core.
Currently we do local invalidations if the VM has only one vcpu or if
the guest requests it with the H_LOCAL flag, though the guest Linux
kernel currently doesn't ever use H_LOCAL. Then, to cope with the
possibility that vcpus moving around to different physical cores might
expose stale TLB entries, there is some code in kvmppc_hv_entry to
flush the whole TLB of entries for this VM if either this vcpu is now
running on a different physical core from where it last ran, or if this
physical core last ran a different vcpu.
There are a number of problems on POWER7 with this as it stands:
- The TLB invalidation is done per thread, whereas it only needs to be
done per core, since the TLB is shared between the threads.
- With the possibility of the host paging out guest pages, the use of
H_LOCAL by an SMP guest is dangerous since the guest could possibly
retain and use a stale TLB entry pointing to a page that had been
removed from the guest.
- The TLB invalidations that we do when a vcpu moves from one physical
core to another are unnecessary in the case of an SMP guest that isn't
using H_LOCAL.
- The optimization of using local invalidations rather than global should
apply to guests with one virtual core, not just one vcpu.
(None of this applies on PPC970, since there we always have to
invalidate the whole TLB when entering and leaving the guest, and we
can't support paging out guest memory.)
To fix these problems and simplify the code, we now maintain a simple
cpumask of which cpus need to flush the TLB on entry to the guest.
(This is indexed by cpu, though we only ever use the bits for thread
0 of each core.) Whenever we do a local TLB invalidation, we set the
bits for every cpu except the bit for thread 0 of the core that we're
currently running on. Whenever we enter a guest, we test and clear the
bit for our core, and flush the TLB if it was set.
On initial startup of the VM, and when resetting the HPT, we set all the
bits in the need_tlb_flush cpumask, since any core could potentially have
stale TLB entries from the previous VM to use the same LPID, or the
previous contents of the HPT.
Then, we maintain a count of the number of online virtual cores, and use
that when deciding whether to use a local invalidation rather than the
number of online vcpus. The code to make that decision is extracted out
into a new function, global_invalidates(). For multi-core guests on
POWER7 (i.e. when we are using mmu notifiers), we now never do local
invalidations regardless of the H_LOCAL flag.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-22 07:28:08 +08:00
|
|
|
|
2018-05-17 15:06:29 +08:00
|
|
|
/* See if we need to flush the TLB. Hash has to be done in RM */
|
KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
When we change or remove a HPT (hashed page table) entry, we can do
either a global TLB invalidation (tlbie) that works across the whole
machine, or a local invalidation (tlbiel) that only affects this core.
Currently we do local invalidations if the VM has only one vcpu or if
the guest requests it with the H_LOCAL flag, though the guest Linux
kernel currently doesn't ever use H_LOCAL. Then, to cope with the
possibility that vcpus moving around to different physical cores might
expose stale TLB entries, there is some code in kvmppc_hv_entry to
flush the whole TLB of entries for this VM if either this vcpu is now
running on a different physical core from where it last ran, or if this
physical core last ran a different vcpu.
There are a number of problems on POWER7 with this as it stands:
- The TLB invalidation is done per thread, whereas it only needs to be
done per core, since the TLB is shared between the threads.
- With the possibility of the host paging out guest pages, the use of
H_LOCAL by an SMP guest is dangerous since the guest could possibly
retain and use a stale TLB entry pointing to a page that had been
removed from the guest.
- The TLB invalidations that we do when a vcpu moves from one physical
core to another are unnecessary in the case of an SMP guest that isn't
using H_LOCAL.
- The optimization of using local invalidations rather than global should
apply to guests with one virtual core, not just one vcpu.
(None of this applies on PPC970, since there we always have to
invalidate the whole TLB when entering and leaving the guest, and we
can't support paging out guest memory.)
To fix these problems and simplify the code, we now maintain a simple
cpumask of which cpus need to flush the TLB on entry to the guest.
(This is indexed by cpu, though we only ever use the bits for thread
0 of each core.) Whenever we do a local TLB invalidation, we set the
bits for every cpu except the bit for thread 0 of the core that we're
currently running on. Whenever we enter a guest, we test and clear the
bit for our core, and flush the TLB if it was set.
On initial startup of the VM, and when resetting the HPT, we set all the
bits in the need_tlb_flush cpumask, since any core could potentially have
stale TLB entries from the previous VM to use the same LPID, or the
previous contents of the HPT.
Then, we maintain a count of the number of online virtual cores, and use
that when deciding whether to use a local invalidation rather than the
number of online vcpus. The code to make that decision is extracted out
into a new function, global_invalidates(). For multi-core guests on
POWER7 (i.e. when we are using mmu notifiers), we now never do local
invalidations regardless of the H_LOCAL flag.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-22 07:28:08 +08:00
|
|
|
lhz r6,PACAPACAINDEX(r13) /* test_bit(cpu, need_tlb_flush) */
|
KVM: PPC: Book3S HV: Invalidate TLB on radix guest vcpu movement
With radix, the guest can do TLB invalidations itself using the tlbie
(global) and tlbiel (local) TLB invalidation instructions. Linux guests
use local TLB invalidations for translations that have only ever been
accessed on one vcpu. However, that doesn't mean that the translations
have only been accessed on one physical cpu (pcpu) since vcpus can move
around from one pcpu to another. Thus a tlbiel might leave behind stale
TLB entries on a pcpu where the vcpu previously ran, and if that task
then moves back to that previous pcpu, it could see those stale TLB
entries and thus access memory incorrectly. The usual symptom of this
is random segfaults in userspace programs in the guest.
To cope with this, we detect when a vcpu is about to start executing on
a thread in a core that is a different core from the last time it
executed. If that is the case, then we mark the core as needing a
TLB flush and then send an interrupt to any thread in the core that is
currently running a vcpu from the same guest. This will get those vcpus
out of the guest, and the first one to re-enter the guest will do the
TLB flush. The reason for interrupting the vcpus executing on the old
core is to cope with the following scenario:
CPU 0 CPU 1 CPU 4
(core 0) (core 0) (core 1)
VCPU 0 runs task X VCPU 1 runs
core 0 TLB gets
entries from task X
VCPU 0 moves to CPU 4
VCPU 0 runs task X
Unmap pages of task X
tlbiel
(still VCPU 1) task X moves to VCPU 1
task X runs
task X sees stale TLB
entries
That is, as soon as the VCPU starts executing on the new core, it
could unmap and tlbiel some page table entries, and then the task
could migrate to one of the VCPUs running on the old core and
potentially see stale TLB entries.
Since the TLB is shared between all the threads in a core, we only
use the bit of kvm->arch.need_tlb_flush corresponding to the first
thread in the core. To ensure that we don't have a window where we
can miss a flush, this moves the clearing of the bit from before the
actual flush to after it. This way, two threads might both do the
flush, but we prevent the situation where one thread can enter the
guest before the flush is finished.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-01-30 18:21:50 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
/*
|
|
|
|
|
* On POWER9, individual threads can come in here, but the
|
|
|
|
|
* TLB is shared between the 4 threads in a core, hence
|
|
|
|
|
* invalidating on one thread invalidates for all.
|
|
|
|
|
* Thus we make all 4 threads use the same bit here.
|
|
|
|
|
*/
|
|
|
|
|
clrrdi r6,r6,2
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
When we change or remove a HPT (hashed page table) entry, we can do
either a global TLB invalidation (tlbie) that works across the whole
machine, or a local invalidation (tlbiel) that only affects this core.
Currently we do local invalidations if the VM has only one vcpu or if
the guest requests it with the H_LOCAL flag, though the guest Linux
kernel currently doesn't ever use H_LOCAL. Then, to cope with the
possibility that vcpus moving around to different physical cores might
expose stale TLB entries, there is some code in kvmppc_hv_entry to
flush the whole TLB of entries for this VM if either this vcpu is now
running on a different physical core from where it last ran, or if this
physical core last ran a different vcpu.
There are a number of problems on POWER7 with this as it stands:
- The TLB invalidation is done per thread, whereas it only needs to be
done per core, since the TLB is shared between the threads.
- With the possibility of the host paging out guest pages, the use of
H_LOCAL by an SMP guest is dangerous since the guest could possibly
retain and use a stale TLB entry pointing to a page that had been
removed from the guest.
- The TLB invalidations that we do when a vcpu moves from one physical
core to another are unnecessary in the case of an SMP guest that isn't
using H_LOCAL.
- The optimization of using local invalidations rather than global should
apply to guests with one virtual core, not just one vcpu.
(None of this applies on PPC970, since there we always have to
invalidate the whole TLB when entering and leaving the guest, and we
can't support paging out guest memory.)
To fix these problems and simplify the code, we now maintain a simple
cpumask of which cpus need to flush the TLB on entry to the guest.
(This is indexed by cpu, though we only ever use the bits for thread
0 of each core.) Whenever we do a local TLB invalidation, we set the
bits for every cpu except the bit for thread 0 of the core that we're
currently running on. Whenever we enter a guest, we test and clear the
bit for our core, and flush the TLB if it was set.
On initial startup of the VM, and when resetting the HPT, we set all the
bits in the need_tlb_flush cpumask, since any core could potentially have
stale TLB entries from the previous VM to use the same LPID, or the
previous contents of the HPT.
Then, we maintain a count of the number of online virtual cores, and use
that when deciding whether to use a local invalidation rather than the
number of online vcpus. The code to make that decision is extracted out
into a new function, global_invalidates(). For multi-core guests on
POWER7 (i.e. when we are using mmu notifiers), we now never do local
invalidations regardless of the H_LOCAL flag.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-22 07:28:08 +08:00
|
|
|
clrldi r7,r6,64-6 /* extract bit number (6 bits) */
|
|
|
|
|
srdi r6,r6,6 /* doubleword number */
|
|
|
|
|
sldi r6,r6,3 /* address offset */
|
|
|
|
|
add r6,r6,r9
|
|
|
|
|
addi r6,r6,KVM_NEED_FLUSH /* dword in kvm->arch.need_tlb_flush */
|
KVM: PPC: Book3S HV: Invalidate TLB on radix guest vcpu movement
With radix, the guest can do TLB invalidations itself using the tlbie
(global) and tlbiel (local) TLB invalidation instructions. Linux guests
use local TLB invalidations for translations that have only ever been
accessed on one vcpu. However, that doesn't mean that the translations
have only been accessed on one physical cpu (pcpu) since vcpus can move
around from one pcpu to another. Thus a tlbiel might leave behind stale
TLB entries on a pcpu where the vcpu previously ran, and if that task
then moves back to that previous pcpu, it could see those stale TLB
entries and thus access memory incorrectly. The usual symptom of this
is random segfaults in userspace programs in the guest.
To cope with this, we detect when a vcpu is about to start executing on
a thread in a core that is a different core from the last time it
executed. If that is the case, then we mark the core as needing a
TLB flush and then send an interrupt to any thread in the core that is
currently running a vcpu from the same guest. This will get those vcpus
out of the guest, and the first one to re-enter the guest will do the
TLB flush. The reason for interrupting the vcpus executing on the old
core is to cope with the following scenario:
CPU 0 CPU 1 CPU 4
(core 0) (core 0) (core 1)
VCPU 0 runs task X VCPU 1 runs
core 0 TLB gets
entries from task X
VCPU 0 moves to CPU 4
VCPU 0 runs task X
Unmap pages of task X
tlbiel
(still VCPU 1) task X moves to VCPU 1
task X runs
task X sees stale TLB
entries
That is, as soon as the VCPU starts executing on the new core, it
could unmap and tlbiel some page table entries, and then the task
could migrate to one of the VCPUs running on the old core and
potentially see stale TLB entries.
Since the TLB is shared between all the threads in a core, we only
use the bit of kvm->arch.need_tlb_flush corresponding to the first
thread in the core. To ensure that we don't have a window where we
can miss a flush, this moves the clearing of the bit from before the
actual flush to after it. This way, two threads might both do the
flush, but we prevent the situation where one thread can enter the
guest before the flush is finished.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-01-30 18:21:50 +08:00
|
|
|
li r8,1
|
|
|
|
|
sld r8,r8,r7
|
KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
When we change or remove a HPT (hashed page table) entry, we can do
either a global TLB invalidation (tlbie) that works across the whole
machine, or a local invalidation (tlbiel) that only affects this core.
Currently we do local invalidations if the VM has only one vcpu or if
the guest requests it with the H_LOCAL flag, though the guest Linux
kernel currently doesn't ever use H_LOCAL. Then, to cope with the
possibility that vcpus moving around to different physical cores might
expose stale TLB entries, there is some code in kvmppc_hv_entry to
flush the whole TLB of entries for this VM if either this vcpu is now
running on a different physical core from where it last ran, or if this
physical core last ran a different vcpu.
There are a number of problems on POWER7 with this as it stands:
- The TLB invalidation is done per thread, whereas it only needs to be
done per core, since the TLB is shared between the threads.
- With the possibility of the host paging out guest pages, the use of
H_LOCAL by an SMP guest is dangerous since the guest could possibly
retain and use a stale TLB entry pointing to a page that had been
removed from the guest.
- The TLB invalidations that we do when a vcpu moves from one physical
core to another are unnecessary in the case of an SMP guest that isn't
using H_LOCAL.
- The optimization of using local invalidations rather than global should
apply to guests with one virtual core, not just one vcpu.
(None of this applies on PPC970, since there we always have to
invalidate the whole TLB when entering and leaving the guest, and we
can't support paging out guest memory.)
To fix these problems and simplify the code, we now maintain a simple
cpumask of which cpus need to flush the TLB on entry to the guest.
(This is indexed by cpu, though we only ever use the bits for thread
0 of each core.) Whenever we do a local TLB invalidation, we set the
bits for every cpu except the bit for thread 0 of the core that we're
currently running on. Whenever we enter a guest, we test and clear the
bit for our core, and flush the TLB if it was set.
On initial startup of the VM, and when resetting the HPT, we set all the
bits in the need_tlb_flush cpumask, since any core could potentially have
stale TLB entries from the previous VM to use the same LPID, or the
previous contents of the HPT.
Then, we maintain a count of the number of online virtual cores, and use
that when deciding whether to use a local invalidation rather than the
number of online vcpus. The code to make that decision is extracted out
into a new function, global_invalidates(). For multi-core guests on
POWER7 (i.e. when we are using mmu notifiers), we now never do local
invalidations regardless of the H_LOCAL flag.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-22 07:28:08 +08:00
|
|
|
ld r7,0(r6)
|
KVM: PPC: Book3S HV: Invalidate TLB on radix guest vcpu movement
With radix, the guest can do TLB invalidations itself using the tlbie
(global) and tlbiel (local) TLB invalidation instructions. Linux guests
use local TLB invalidations for translations that have only ever been
accessed on one vcpu. However, that doesn't mean that the translations
have only been accessed on one physical cpu (pcpu) since vcpus can move
around from one pcpu to another. Thus a tlbiel might leave behind stale
TLB entries on a pcpu where the vcpu previously ran, and if that task
then moves back to that previous pcpu, it could see those stale TLB
entries and thus access memory incorrectly. The usual symptom of this
is random segfaults in userspace programs in the guest.
To cope with this, we detect when a vcpu is about to start executing on
a thread in a core that is a different core from the last time it
executed. If that is the case, then we mark the core as needing a
TLB flush and then send an interrupt to any thread in the core that is
currently running a vcpu from the same guest. This will get those vcpus
out of the guest, and the first one to re-enter the guest will do the
TLB flush. The reason for interrupting the vcpus executing on the old
core is to cope with the following scenario:
CPU 0 CPU 1 CPU 4
(core 0) (core 0) (core 1)
VCPU 0 runs task X VCPU 1 runs
core 0 TLB gets
entries from task X
VCPU 0 moves to CPU 4
VCPU 0 runs task X
Unmap pages of task X
tlbiel
(still VCPU 1) task X moves to VCPU 1
task X runs
task X sees stale TLB
entries
That is, as soon as the VCPU starts executing on the new core, it
could unmap and tlbiel some page table entries, and then the task
could migrate to one of the VCPUs running on the old core and
potentially see stale TLB entries.
Since the TLB is shared between all the threads in a core, we only
use the bit of kvm->arch.need_tlb_flush corresponding to the first
thread in the core. To ensure that we don't have a window where we
can miss a flush, this moves the clearing of the bit from before the
actual flush to after it. This way, two threads might both do the
flush, but we prevent the situation where one thread can enter the
guest before the flush is finished.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-01-30 18:21:50 +08:00
|
|
|
and. r7,r7,r8
|
KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
When we change or remove a HPT (hashed page table) entry, we can do
either a global TLB invalidation (tlbie) that works across the whole
machine, or a local invalidation (tlbiel) that only affects this core.
Currently we do local invalidations if the VM has only one vcpu or if
the guest requests it with the H_LOCAL flag, though the guest Linux
kernel currently doesn't ever use H_LOCAL. Then, to cope with the
possibility that vcpus moving around to different physical cores might
expose stale TLB entries, there is some code in kvmppc_hv_entry to
flush the whole TLB of entries for this VM if either this vcpu is now
running on a different physical core from where it last ran, or if this
physical core last ran a different vcpu.
There are a number of problems on POWER7 with this as it stands:
- The TLB invalidation is done per thread, whereas it only needs to be
done per core, since the TLB is shared between the threads.
- With the possibility of the host paging out guest pages, the use of
H_LOCAL by an SMP guest is dangerous since the guest could possibly
retain and use a stale TLB entry pointing to a page that had been
removed from the guest.
- The TLB invalidations that we do when a vcpu moves from one physical
core to another are unnecessary in the case of an SMP guest that isn't
using H_LOCAL.
- The optimization of using local invalidations rather than global should
apply to guests with one virtual core, not just one vcpu.
(None of this applies on PPC970, since there we always have to
invalidate the whole TLB when entering and leaving the guest, and we
can't support paging out guest memory.)
To fix these problems and simplify the code, we now maintain a simple
cpumask of which cpus need to flush the TLB on entry to the guest.
(This is indexed by cpu, though we only ever use the bits for thread
0 of each core.) Whenever we do a local TLB invalidation, we set the
bits for every cpu except the bit for thread 0 of the core that we're
currently running on. Whenever we enter a guest, we test and clear the
bit for our core, and flush the TLB if it was set.
On initial startup of the VM, and when resetting the HPT, we set all the
bits in the need_tlb_flush cpumask, since any core could potentially have
stale TLB entries from the previous VM to use the same LPID, or the
previous contents of the HPT.
Then, we maintain a count of the number of online virtual cores, and use
that when deciding whether to use a local invalidation rather than the
number of online vcpus. The code to make that decision is extracted out
into a new function, global_invalidates(). For multi-core guests on
POWER7 (i.e. when we are using mmu notifiers), we now never do local
invalidations regardless of the H_LOCAL flag.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-22 07:28:08 +08:00
|
|
|
beq 22f
|
2014-01-08 18:25:22 +08:00
|
|
|
/* Flush the TLB of any entries for this LPID */
|
KVM: PPC: Book3S HV: Invalidate TLB on radix guest vcpu movement
With radix, the guest can do TLB invalidations itself using the tlbie
(global) and tlbiel (local) TLB invalidation instructions. Linux guests
use local TLB invalidations for translations that have only ever been
accessed on one vcpu. However, that doesn't mean that the translations
have only been accessed on one physical cpu (pcpu) since vcpus can move
around from one pcpu to another. Thus a tlbiel might leave behind stale
TLB entries on a pcpu where the vcpu previously ran, and if that task
then moves back to that previous pcpu, it could see those stale TLB
entries and thus access memory incorrectly. The usual symptom of this
is random segfaults in userspace programs in the guest.
To cope with this, we detect when a vcpu is about to start executing on
a thread in a core that is a different core from the last time it
executed. If that is the case, then we mark the core as needing a
TLB flush and then send an interrupt to any thread in the core that is
currently running a vcpu from the same guest. This will get those vcpus
out of the guest, and the first one to re-enter the guest will do the
TLB flush. The reason for interrupting the vcpus executing on the old
core is to cope with the following scenario:
CPU 0 CPU 1 CPU 4
(core 0) (core 0) (core 1)
VCPU 0 runs task X VCPU 1 runs
core 0 TLB gets
entries from task X
VCPU 0 moves to CPU 4
VCPU 0 runs task X
Unmap pages of task X
tlbiel
(still VCPU 1) task X moves to VCPU 1
task X runs
task X sees stale TLB
entries
That is, as soon as the VCPU starts executing on the new core, it
could unmap and tlbiel some page table entries, and then the task
could migrate to one of the VCPUs running on the old core and
potentially see stale TLB entries.
Since the TLB is shared between all the threads in a core, we only
use the bit of kvm->arch.need_tlb_flush corresponding to the first
thread in the core. To ensure that we don't have a window where we
can miss a flush, this moves the clearing of the bit from before the
actual flush to after it. This way, two threads might both do the
flush, but we prevent the situation where one thread can enter the
guest before the flush is finished.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-01-30 18:21:50 +08:00
|
|
|
lwz r0,KVM_TLB_SETS(r9)
|
|
|
|
|
mtctr r0
|
KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
When we change or remove a HPT (hashed page table) entry, we can do
either a global TLB invalidation (tlbie) that works across the whole
machine, or a local invalidation (tlbiel) that only affects this core.
Currently we do local invalidations if the VM has only one vcpu or if
the guest requests it with the H_LOCAL flag, though the guest Linux
kernel currently doesn't ever use H_LOCAL. Then, to cope with the
possibility that vcpus moving around to different physical cores might
expose stale TLB entries, there is some code in kvmppc_hv_entry to
flush the whole TLB of entries for this VM if either this vcpu is now
running on a different physical core from where it last ran, or if this
physical core last ran a different vcpu.
There are a number of problems on POWER7 with this as it stands:
- The TLB invalidation is done per thread, whereas it only needs to be
done per core, since the TLB is shared between the threads.
- With the possibility of the host paging out guest pages, the use of
H_LOCAL by an SMP guest is dangerous since the guest could possibly
retain and use a stale TLB entry pointing to a page that had been
removed from the guest.
- The TLB invalidations that we do when a vcpu moves from one physical
core to another are unnecessary in the case of an SMP guest that isn't
using H_LOCAL.
- The optimization of using local invalidations rather than global should
apply to guests with one virtual core, not just one vcpu.
(None of this applies on PPC970, since there we always have to
invalidate the whole TLB when entering and leaving the guest, and we
can't support paging out guest memory.)
To fix these problems and simplify the code, we now maintain a simple
cpumask of which cpus need to flush the TLB on entry to the guest.
(This is indexed by cpu, though we only ever use the bits for thread
0 of each core.) Whenever we do a local TLB invalidation, we set the
bits for every cpu except the bit for thread 0 of the core that we're
currently running on. Whenever we enter a guest, we test and clear the
bit for our core, and flush the TLB if it was set.
On initial startup of the VM, and when resetting the HPT, we set all the
bits in the need_tlb_flush cpumask, since any core could potentially have
stale TLB entries from the previous VM to use the same LPID, or the
previous contents of the HPT.
Then, we maintain a count of the number of online virtual cores, and use
that when deciding whether to use a local invalidation rather than the
number of online vcpus. The code to make that decision is extracted out
into a new function, global_invalidates(). For multi-core guests on
POWER7 (i.e. when we are using mmu notifiers), we now never do local
invalidations regardless of the H_LOCAL flag.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-22 07:28:08 +08:00
|
|
|
li r7,0x800 /* IS field = 0b10 */
|
|
|
|
|
ptesync
|
KVM: PPC: Book3S HV: Invalidate TLB on radix guest vcpu movement
With radix, the guest can do TLB invalidations itself using the tlbie
(global) and tlbiel (local) TLB invalidation instructions. Linux guests
use local TLB invalidations for translations that have only ever been
accessed on one vcpu. However, that doesn't mean that the translations
have only been accessed on one physical cpu (pcpu) since vcpus can move
around from one pcpu to another. Thus a tlbiel might leave behind stale
TLB entries on a pcpu where the vcpu previously ran, and if that task
then moves back to that previous pcpu, it could see those stale TLB
entries and thus access memory incorrectly. The usual symptom of this
is random segfaults in userspace programs in the guest.
To cope with this, we detect when a vcpu is about to start executing on
a thread in a core that is a different core from the last time it
executed. If that is the case, then we mark the core as needing a
TLB flush and then send an interrupt to any thread in the core that is
currently running a vcpu from the same guest. This will get those vcpus
out of the guest, and the first one to re-enter the guest will do the
TLB flush. The reason for interrupting the vcpus executing on the old
core is to cope with the following scenario:
CPU 0 CPU 1 CPU 4
(core 0) (core 0) (core 1)
VCPU 0 runs task X VCPU 1 runs
core 0 TLB gets
entries from task X
VCPU 0 moves to CPU 4
VCPU 0 runs task X
Unmap pages of task X
tlbiel
(still VCPU 1) task X moves to VCPU 1
task X runs
task X sees stale TLB
entries
That is, as soon as the VCPU starts executing on the new core, it
could unmap and tlbiel some page table entries, and then the task
could migrate to one of the VCPUs running on the old core and
potentially see stale TLB entries.
Since the TLB is shared between all the threads in a core, we only
use the bit of kvm->arch.need_tlb_flush corresponding to the first
thread in the core. To ensure that we don't have a window where we
can miss a flush, this moves the clearing of the bit from before the
actual flush to after it. This way, two threads might both do the
flush, but we prevent the situation where one thread can enter the
guest before the flush is finished.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-01-30 18:21:50 +08:00
|
|
|
li r0,0 /* RS for P9 version of tlbiel */
|
|
|
|
|
28: tlbiel r7 /* On P9, rs=0, RIC=0, PRS=0, R=0 */
|
KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
When we change or remove a HPT (hashed page table) entry, we can do
either a global TLB invalidation (tlbie) that works across the whole
machine, or a local invalidation (tlbiel) that only affects this core.
Currently we do local invalidations if the VM has only one vcpu or if
the guest requests it with the H_LOCAL flag, though the guest Linux
kernel currently doesn't ever use H_LOCAL. Then, to cope with the
possibility that vcpus moving around to different physical cores might
expose stale TLB entries, there is some code in kvmppc_hv_entry to
flush the whole TLB of entries for this VM if either this vcpu is now
running on a different physical core from where it last ran, or if this
physical core last ran a different vcpu.
There are a number of problems on POWER7 with this as it stands:
- The TLB invalidation is done per thread, whereas it only needs to be
done per core, since the TLB is shared between the threads.
- With the possibility of the host paging out guest pages, the use of
H_LOCAL by an SMP guest is dangerous since the guest could possibly
retain and use a stale TLB entry pointing to a page that had been
removed from the guest.
- The TLB invalidations that we do when a vcpu moves from one physical
core to another are unnecessary in the case of an SMP guest that isn't
using H_LOCAL.
- The optimization of using local invalidations rather than global should
apply to guests with one virtual core, not just one vcpu.
(None of this applies on PPC970, since there we always have to
invalidate the whole TLB when entering and leaving the guest, and we
can't support paging out guest memory.)
To fix these problems and simplify the code, we now maintain a simple
cpumask of which cpus need to flush the TLB on entry to the guest.
(This is indexed by cpu, though we only ever use the bits for thread
0 of each core.) Whenever we do a local TLB invalidation, we set the
bits for every cpu except the bit for thread 0 of the core that we're
currently running on. Whenever we enter a guest, we test and clear the
bit for our core, and flush the TLB if it was set.
On initial startup of the VM, and when resetting the HPT, we set all the
bits in the need_tlb_flush cpumask, since any core could potentially have
stale TLB entries from the previous VM to use the same LPID, or the
previous contents of the HPT.
Then, we maintain a count of the number of online virtual cores, and use
that when deciding whether to use a local invalidation rather than the
number of online vcpus. The code to make that decision is extracted out
into a new function, global_invalidates(). For multi-core guests on
POWER7 (i.e. when we are using mmu notifiers), we now never do local
invalidations regardless of the H_LOCAL flag.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-22 07:28:08 +08:00
|
|
|
addi r7,r7,0x1000
|
|
|
|
|
bdnz 28b
|
2018-05-17 15:06:29 +08:00
|
|
|
ptesync
|
KVM: PPC: Book3S HV: Invalidate TLB on radix guest vcpu movement
With radix, the guest can do TLB invalidations itself using the tlbie
(global) and tlbiel (local) TLB invalidation instructions. Linux guests
use local TLB invalidations for translations that have only ever been
accessed on one vcpu. However, that doesn't mean that the translations
have only been accessed on one physical cpu (pcpu) since vcpus can move
around from one pcpu to another. Thus a tlbiel might leave behind stale
TLB entries on a pcpu where the vcpu previously ran, and if that task
then moves back to that previous pcpu, it could see those stale TLB
entries and thus access memory incorrectly. The usual symptom of this
is random segfaults in userspace programs in the guest.
To cope with this, we detect when a vcpu is about to start executing on
a thread in a core that is a different core from the last time it
executed. If that is the case, then we mark the core as needing a
TLB flush and then send an interrupt to any thread in the core that is
currently running a vcpu from the same guest. This will get those vcpus
out of the guest, and the first one to re-enter the guest will do the
TLB flush. The reason for interrupting the vcpus executing on the old
core is to cope with the following scenario:
CPU 0 CPU 1 CPU 4
(core 0) (core 0) (core 1)
VCPU 0 runs task X VCPU 1 runs
core 0 TLB gets
entries from task X
VCPU 0 moves to CPU 4
VCPU 0 runs task X
Unmap pages of task X
tlbiel
(still VCPU 1) task X moves to VCPU 1
task X runs
task X sees stale TLB
entries
That is, as soon as the VCPU starts executing on the new core, it
could unmap and tlbiel some page table entries, and then the task
could migrate to one of the VCPUs running on the old core and
potentially see stale TLB entries.
Since the TLB is shared between all the threads in a core, we only
use the bit of kvm->arch.need_tlb_flush corresponding to the first
thread in the core. To ensure that we don't have a window where we
can miss a flush, this moves the clearing of the bit from before the
actual flush to after it. This way, two threads might both do the
flush, but we prevent the situation where one thread can enter the
guest before the flush is finished.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-01-30 18:21:50 +08:00
|
|
|
23: ldarx r7,0,r6 /* clear the bit after TLB flushed */
|
|
|
|
|
andc r7,r7,r8
|
|
|
|
|
stdcx. r7,0,r6
|
|
|
|
|
bne 23b
|
KVM: PPC: Book3S HV: Improve handling of local vs. global TLB invalidations
When we change or remove a HPT (hashed page table) entry, we can do
either a global TLB invalidation (tlbie) that works across the whole
machine, or a local invalidation (tlbiel) that only affects this core.
Currently we do local invalidations if the VM has only one vcpu or if
the guest requests it with the H_LOCAL flag, though the guest Linux
kernel currently doesn't ever use H_LOCAL. Then, to cope with the
possibility that vcpus moving around to different physical cores might
expose stale TLB entries, there is some code in kvmppc_hv_entry to
flush the whole TLB of entries for this VM if either this vcpu is now
running on a different physical core from where it last ran, or if this
physical core last ran a different vcpu.
There are a number of problems on POWER7 with this as it stands:
- The TLB invalidation is done per thread, whereas it only needs to be
done per core, since the TLB is shared between the threads.
- With the possibility of the host paging out guest pages, the use of
H_LOCAL by an SMP guest is dangerous since the guest could possibly
retain and use a stale TLB entry pointing to a page that had been
removed from the guest.
- The TLB invalidations that we do when a vcpu moves from one physical
core to another are unnecessary in the case of an SMP guest that isn't
using H_LOCAL.
- The optimization of using local invalidations rather than global should
apply to guests with one virtual core, not just one vcpu.
(None of this applies on PPC970, since there we always have to
invalidate the whole TLB when entering and leaving the guest, and we
can't support paging out guest memory.)
To fix these problems and simplify the code, we now maintain a simple
cpumask of which cpus need to flush the TLB on entry to the guest.
(This is indexed by cpu, though we only ever use the bits for thread
0 of each core.) Whenever we do a local TLB invalidation, we set the
bits for every cpu except the bit for thread 0 of the core that we're
currently running on. Whenever we enter a guest, we test and clear the
bit for our core, and flush the TLB if it was set.
On initial startup of the VM, and when resetting the HPT, we set all the
bits in the need_tlb_flush cpumask, since any core could potentially have
stale TLB entries from the previous VM to use the same LPID, or the
previous contents of the HPT.
Then, we maintain a count of the number of online virtual cores, and use
that when deciding whether to use a local invalidation rather than the
number of online vcpus. The code to make that decision is extracted out
into a new function, global_invalidates(). For multi-core guests on
POWER7 (i.e. when we are using mmu notifiers), we now never do local
invalidations regardless of the H_LOCAL flag.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-22 07:28:08 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Implement timebase offset for guests
This allows guests to have a different timebase origin from the host.
This is needed for migration, where a guest can migrate from one host
to another and the two hosts might have a different timebase origin.
However, the timebase seen by the guest must not go backwards, and
should go forwards only by a small amount corresponding to the time
taken for the migration.
Therefore this provides a new per-vcpu value accessed via the one_reg
interface using the new KVM_REG_PPC_TB_OFFSET identifier. This value
defaults to 0 and is not modified by KVM. On entering the guest, this
value is added onto the timebase, and on exiting the guest, it is
subtracted from the timebase.
This is only supported for recent POWER hardware which has the TBU40
(timebase upper 40 bits) register. Writing to the TBU40 register only
alters the upper 40 bits of the timebase, leaving the lower 24 bits
unchanged. This provides a way to modify the timebase for guest
migration without disturbing the synchronization of the timebase
registers across CPU cores. The kernel rounds up the value given
to a multiple of 2^24.
Timebase values stored in KVM structures (struct kvm_vcpu, struct
kvmppc_vcore, etc.) are stored as host timebase values. The timebase
values in the dispatch trace log need to be guest timebase values,
however, since that is read directly by the guest. This moves the
setting of vcpu->arch.dec_expires on guest exit to a point after we
have restored the host timebase so that vcpu->arch.dec_expires is a
host timebase value.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-09-06 11:17:46 +08:00
|
|
|
/* Add timebase offset onto timebase */
|
|
|
|
|
22: ld r8,VCORE_TB_OFFSET(r5)
|
|
|
|
|
cmpdi r8,0
|
|
|
|
|
beq 37f
|
KVM: PPC: Book3S HV: Snapshot timebase offset on guest entry
Currently, the HV KVM guest entry/exit code adds the timebase offset
from the vcore struct to the timebase on guest entry, and subtracts
it on guest exit. Which is fine, except that it is possible for
userspace to change the offset using the SET_ONE_REG interface while
the vcore is running, as there is only one timebase offset per vcore
but potentially multiple VCPUs in the vcore. If that were to happen,
KVM would subtract a different offset on guest exit from that which
it had added on guest entry, leading to the timebase being out of sync
between cores in the host, which then leads to bad things happening
such as hangs and spurious watchdog timeouts.
To fix this, we add a new field 'tb_offset_applied' to the vcore struct
which stores the offset that is currently applied to the timebase.
This value is set from the vcore tb_offset field on guest entry, and
is what is subtracted from the timebase on guest exit. Since it is
zero when the timebase offset is not applied, we can simplify the
logic in kvmhv_start_timing and kvmhv_accumulate_time.
In addition, we had secondary threads reading the timebase while
running concurrently with code on the primary thread which would
eventually add or subtract the timebase offset from the timebase.
This occurred while saving or restoring the DEC register value on
the secondary threads. Although no specific incorrect behaviour has
been observed, this is a race which should be fixed. To fix it, we
move the DEC saving code to just before we call kvmhv_commence_exit,
and the DEC restoring code to after the point where we have waited
for the primary thread to switch the MMU context and add the timebase
offset. That way we are sure that the timebase contains the guest
timebase value in both cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-04-20 20:51:11 +08:00
|
|
|
std r8, VCORE_TB_OFFSET_APPL(r5)
|
KVM: PPC: Book3S HV: Implement timebase offset for guests
This allows guests to have a different timebase origin from the host.
This is needed for migration, where a guest can migrate from one host
to another and the two hosts might have a different timebase origin.
However, the timebase seen by the guest must not go backwards, and
should go forwards only by a small amount corresponding to the time
taken for the migration.
Therefore this provides a new per-vcpu value accessed via the one_reg
interface using the new KVM_REG_PPC_TB_OFFSET identifier. This value
defaults to 0 and is not modified by KVM. On entering the guest, this
value is added onto the timebase, and on exiting the guest, it is
subtracted from the timebase.
This is only supported for recent POWER hardware which has the TBU40
(timebase upper 40 bits) register. Writing to the TBU40 register only
alters the upper 40 bits of the timebase, leaving the lower 24 bits
unchanged. This provides a way to modify the timebase for guest
migration without disturbing the synchronization of the timebase
registers across CPU cores. The kernel rounds up the value given
to a multiple of 2^24.
Timebase values stored in KVM structures (struct kvm_vcpu, struct
kvmppc_vcore, etc.) are stored as host timebase values. The timebase
values in the dispatch trace log need to be guest timebase values,
however, since that is read directly by the guest. This moves the
setting of vcpu->arch.dec_expires on guest exit to a point after we
have restored the host timebase so that vcpu->arch.dec_expires is a
host timebase value.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-09-06 11:17:46 +08:00
|
|
|
mftb r6 /* current host timebase */
|
|
|
|
|
add r8,r8,r6
|
|
|
|
|
mtspr SPRN_TBU40,r8 /* update upper 40 bits */
|
|
|
|
|
mftb r7 /* check if lower 24 bits overflowed */
|
|
|
|
|
clrldi r6,r6,40
|
|
|
|
|
clrldi r7,r7,40
|
|
|
|
|
cmpld r7,r6
|
|
|
|
|
bge 37f
|
|
|
|
|
addis r8,r8,0x100 /* if so, increment upper 40 bits */
|
|
|
|
|
mtspr SPRN_TBU40,r8
|
|
|
|
|
|
2013-09-21 12:35:02 +08:00
|
|
|
/* Load guest PCR value to select appropriate compat mode */
|
|
|
|
|
37: ld r7, VCORE_PCR(r5)
|
|
|
|
|
cmpdi r7, 0
|
|
|
|
|
beq 38f
|
|
|
|
|
mtspr SPRN_PCR, r7
|
|
|
|
|
38:
|
2014-01-08 18:25:21 +08:00
|
|
|
|
|
|
|
|
BEGIN_FTR_SECTION
|
2016-09-15 11:42:52 +08:00
|
|
|
/* DPDES and VTB are shared between threads */
|
2014-01-08 18:25:21 +08:00
|
|
|
ld r8, VCORE_DPDES(r5)
|
2016-09-15 11:42:52 +08:00
|
|
|
ld r7, VCORE_VTB(r5)
|
2014-01-08 18:25:21 +08:00
|
|
|
mtspr SPRN_DPDES, r8
|
2016-09-15 11:42:52 +08:00
|
|
|
mtspr SPRN_VTB, r7
|
2014-01-08 18:25:21 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Fix TB corruption in guest exit path on HMI interrupt
When a guest is assigned to a core it converts the host Timebase (TB)
into guest TB by adding guest timebase offset before entering into
guest. During guest exit it restores the guest TB to host TB. This means
under certain conditions (Guest migration) host TB and guest TB can differ.
When we get an HMI for TB related issues the opal HMI handler would
try fixing errors and restore the correct host TB value. With no guest
running, we don't have any issues. But with guest running on the core
we run into TB corruption issues.
If we get an HMI while in the guest, the current HMI handler invokes opal
hmi handler before forcing guest to exit. The guest exit path subtracts
the guest TB offset from the current TB value which may have already
been restored with host value by opal hmi handler. This leads to incorrect
host and guest TB values.
With split-core, things become more complex. With split-core, TB also gets
split and each subcore gets its own TB register. When a hmi handler fixes
a TB error and restores the TB value, it affects all the TB values of
sibling subcores on the same core. On TB errors all the thread in the core
gets HMI. With existing code, the individual threads call opal hmi handle
independently which can easily throw TB out of sync if we have guest
running on subcores. Hence we will need to co-ordinate with all the
threads before making opal hmi handler call followed by TB resync.
This patch introduces a sibling subcore state structure (shared by all
threads in the core) in paca which holds information about whether sibling
subcores are in Guest mode or host mode. An array in_guest[] of size
MAX_SUBCORE_PER_CORE=4 is used to maintain the state of each subcore.
The subcore id is used as index into in_guest[] array. Only primary
thread entering/exiting the guest is responsible to set/unset its
designated array element.
On TB error, we get HMI interrupt on every thread on the core. Upon HMI,
this patch will now force guest to vacate the core/subcore. Primary
thread from each subcore will then turn off its respective bit
from the above bitmap during the guest exit path just after the
guest->host partition switch is complete.
All other threads that have just exited the guest OR were already in host
will wait until all other subcores clears their respective bit.
Once all the subcores turn off their respective bit, all threads will
will make call to opal hmi handler.
It is not necessary that opal hmi handler would resync the TB value for
every HMI interrupts. It would do so only for the HMI caused due to
TB errors. For rest, it would not touch TB value. Hence to make things
simpler, primary thread would call TB resync explicitly once for each
core immediately after opal hmi handler instead of subtracting guest
offset from TB. TB resync call will restore the TB with host value.
Thus we can be sure about the TB state.
One of the primary threads exiting the guest will take up the
responsibility of calling TB resync. It will use one of the top bits
(bit 63) from subcore state flags bitmap to make the decision. The first
primary thread (among the subcores) that is able to set the bit will
have to call the TB resync. Rest all other threads will wait until TB
resync is complete. Once TB resync is complete all threads will then
proceed.
Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-05-15 12:14:26 +08:00
|
|
|
/* Mark the subcore state as inside guest */
|
|
|
|
|
bl kvmppc_subcore_enter_guest
|
|
|
|
|
nop
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
2013-09-21 12:35:02 +08:00
|
|
|
li r0,1
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
stb r0,VCORE_IN_GUEST(r5) /* signal secondaries to continue */
|
KVM: PPC: book3s_hv: Add support for PPC970-family processors
This adds support for running KVM guests in supervisor mode on those
PPC970 processors that have a usable hypervisor mode. Unfortunately,
Apple G5 machines have supervisor mode disabled (MSR[HV] is forced to
1), but the YDL PowerStation does have a usable hypervisor mode.
There are several differences between the PPC970 and POWER7 in how
guests are managed. These differences are accommodated using the
CPU_FTR_ARCH_201 (PPC970) and CPU_FTR_ARCH_206 (POWER7) CPU feature
bits. Notably, on PPC970:
* The LPCR, LPID or RMOR registers don't exist, and the functions of
those registers are provided by bits in HID4 and one bit in HID0.
* External interrupts can be directed to the hypervisor, but unlike
POWER7 they are masked by MSR[EE] in non-hypervisor modes and use
SRR0/1 not HSRR0/1.
* There is no virtual RMA (VRMA) mode; the guest must use an RMO
(real mode offset) area.
* The TLB entries are not tagged with the LPID, so it is necessary to
flush the whole TLB on partition switch. Furthermore, when switching
partitions we have to ensure that no other CPU is executing the tlbie
or tlbsync instructions in either the old or the new partition,
otherwise undefined behaviour can occur.
* The PMU has 8 counters (PMC registers) rather than 6.
* The DSCR, PURR, SPURR, AMR, AMOR, UAMOR registers don't exist.
* The SLB has 64 entries rather than 32.
* There is no mediated external interrupt facility, so if we switch to
a guest that has a virtual external interrupt pending but the guest
has MSR[EE] = 0, we have to arrange to have an interrupt pending for
it so that we can get control back once it re-enables interrupts. We
do that by sending ourselves an IPI with smp_send_reschedule after
hard-disabling interrupts.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:40:08 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
/* Do we have a guest vcpu to run? */
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
10: cmpdi r4, 0
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
beq kvmppc_primary_no_guest
|
|
|
|
|
kvmppc_got_guest:
|
|
|
|
|
/* Increment yield count if they have a VPA */
|
|
|
|
|
ld r3, VCPU_VPA(r4)
|
|
|
|
|
cmpdi r3, 0
|
|
|
|
|
beq 25f
|
2014-06-11 16:36:17 +08:00
|
|
|
li r6, LPPACA_YIELDCOUNT
|
|
|
|
|
LWZX_BE r5, r3, r6
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
addi r5, r5, 1
|
2014-06-11 16:36:17 +08:00
|
|
|
STWX_BE r5, r3, r6
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
li r6, 1
|
|
|
|
|
stb r6, VCPU_VPA_DIRTY(r4)
|
|
|
|
|
25:
|
|
|
|
|
|
|
|
|
|
/* Save purr/spurr */
|
|
|
|
|
mfspr r5,SPRN_PURR
|
|
|
|
|
mfspr r6,SPRN_SPURR
|
|
|
|
|
std r5,HSTATE_PURR(r13)
|
|
|
|
|
std r6,HSTATE_SPURR(r13)
|
|
|
|
|
ld r7,VCPU_PURR(r4)
|
|
|
|
|
ld r8,VCPU_SPURR(r4)
|
|
|
|
|
mtspr SPRN_PURR,r7
|
|
|
|
|
mtspr SPRN_SPURR,r8
|
|
|
|
|
|
2016-11-18 10:11:42 +08:00
|
|
|
/* Save host values of some registers */
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
mfspr r5, SPRN_TIDR
|
|
|
|
|
mfspr r6, SPRN_PSSCR
|
2017-01-30 18:21:45 +08:00
|
|
|
mfspr r7, SPRN_PID
|
2016-11-18 10:11:42 +08:00
|
|
|
std r5, STACK_SLOT_TID(r1)
|
|
|
|
|
std r6, STACK_SLOT_PSSCR(r1)
|
2017-01-30 18:21:45 +08:00
|
|
|
std r7, STACK_SLOT_PID(r1)
|
2017-02-15 11:30:17 +08:00
|
|
|
mfspr r5, SPRN_HFSCR
|
|
|
|
|
std r5, STACK_SLOT_HFSCR(r1)
|
2016-11-18 10:11:42 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
2017-06-16 09:53:19 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
mfspr r5, SPRN_CIABR
|
|
|
|
|
mfspr r6, SPRN_DAWR
|
|
|
|
|
mfspr r7, SPRN_DAWRX
|
powerpc/kvm: Save and restore host AMR/IAMR/UAMOR
When the hash MMU is active the AMR, IAMR and UAMOR are used for
pkeys. The AMR is directly writable by user space, and the UAMOR masks
those writes, meaning both registers are effectively user register
state. The IAMR is used to create an execute only key.
Also we must maintain the value of at least the AMR when running in
process context, so that any memory accesses done by the kernel on
behalf of the process are correctly controlled by the AMR.
Although we are correctly switching all registers when going into a
guest, on returning to the host we just write 0 into all regs, except
on Power9 where we restore the IAMR correctly.
This could be observed by a user process if it writes the AMR, then
runs a guest and we then return immediately to it without
rescheduling. Because we have written 0 to the AMR that would have the
effect of granting read/write permission to pages that the process was
trying to protect.
In addition, when using the Radix MMU, the AMR can prevent inadvertent
kernel access to userspace data, writing 0 to the AMR disables that
protection.
So save and restore AMR, IAMR and UAMOR.
Fixes: cf43d3b26452 ("powerpc: Enable pkey subsystem")
Cc: stable@vger.kernel.org # v4.16+
Signed-off-by: Russell Currey <ruscur@russell.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Paul Mackerras <paulus@ozlabs.org>
2019-02-22 10:22:08 +08:00
|
|
|
mfspr r8, SPRN_IAMR
|
2017-06-16 09:53:19 +08:00
|
|
|
std r5, STACK_SLOT_CIABR(r1)
|
|
|
|
|
std r6, STACK_SLOT_DAWR(r1)
|
|
|
|
|
std r7, STACK_SLOT_DAWRX(r1)
|
powerpc/kvm: Save and restore host AMR/IAMR/UAMOR
When the hash MMU is active the AMR, IAMR and UAMOR are used for
pkeys. The AMR is directly writable by user space, and the UAMOR masks
those writes, meaning both registers are effectively user register
state. The IAMR is used to create an execute only key.
Also we must maintain the value of at least the AMR when running in
process context, so that any memory accesses done by the kernel on
behalf of the process are correctly controlled by the AMR.
Although we are correctly switching all registers when going into a
guest, on returning to the host we just write 0 into all regs, except
on Power9 where we restore the IAMR correctly.
This could be observed by a user process if it writes the AMR, then
runs a guest and we then return immediately to it without
rescheduling. Because we have written 0 to the AMR that would have the
effect of granting read/write permission to pages that the process was
trying to protect.
In addition, when using the Radix MMU, the AMR can prevent inadvertent
kernel access to userspace data, writing 0 to the AMR disables that
protection.
So save and restore AMR, IAMR and UAMOR.
Fixes: cf43d3b26452 ("powerpc: Enable pkey subsystem")
Cc: stable@vger.kernel.org # v4.16+
Signed-off-by: Russell Currey <ruscur@russell.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Paul Mackerras <paulus@ozlabs.org>
2019-02-22 10:22:08 +08:00
|
|
|
std r8, STACK_SLOT_IAMR(r1)
|
2017-06-16 09:53:19 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
2016-11-18 10:11:42 +08:00
|
|
|
|
powerpc/kvm: Save and restore host AMR/IAMR/UAMOR
When the hash MMU is active the AMR, IAMR and UAMOR are used for
pkeys. The AMR is directly writable by user space, and the UAMOR masks
those writes, meaning both registers are effectively user register
state. The IAMR is used to create an execute only key.
Also we must maintain the value of at least the AMR when running in
process context, so that any memory accesses done by the kernel on
behalf of the process are correctly controlled by the AMR.
Although we are correctly switching all registers when going into a
guest, on returning to the host we just write 0 into all regs, except
on Power9 where we restore the IAMR correctly.
This could be observed by a user process if it writes the AMR, then
runs a guest and we then return immediately to it without
rescheduling. Because we have written 0 to the AMR that would have the
effect of granting read/write permission to pages that the process was
trying to protect.
In addition, when using the Radix MMU, the AMR can prevent inadvertent
kernel access to userspace data, writing 0 to the AMR disables that
protection.
So save and restore AMR, IAMR and UAMOR.
Fixes: cf43d3b26452 ("powerpc: Enable pkey subsystem")
Cc: stable@vger.kernel.org # v4.16+
Signed-off-by: Russell Currey <ruscur@russell.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Paul Mackerras <paulus@ozlabs.org>
2019-02-22 10:22:08 +08:00
|
|
|
mfspr r5, SPRN_AMR
|
|
|
|
|
std r5, STACK_SLOT_AMR(r1)
|
|
|
|
|
mfspr r6, SPRN_UAMOR
|
|
|
|
|
std r6, STACK_SLOT_UAMOR(r1)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
/* Set partition DABR */
|
|
|
|
|
/* Do this before re-enabling PMU to avoid P7 DABR corruption bug */
|
KVM: PPC: Book3S HV: Add support for DABRX register on POWER7
The DABRX (DABR extension) register on POWER7 processors provides finer
control over which accesses cause a data breakpoint interrupt. It
contains 3 bits which indicate whether to enable accesses in user,
kernel and hypervisor modes respectively to cause data breakpoint
interrupts, plus one bit that enables both real mode and virtual mode
accesses to cause interrupts. Currently, KVM sets DABRX to allow
both kernel and user accesses to cause interrupts while in the guest.
This adds support for the guest to specify other values for DABRX.
PAPR defines a H_SET_XDABR hcall to allow the guest to set both DABR
and DABRX with one call. This adds a real-mode implementation of
H_SET_XDABR, which shares most of its code with the existing H_SET_DABR
implementation. To support this, we add a per-vcpu field to store the
DABRX value plus code to get and set it via the ONE_REG interface.
For Linux guests to use this new hcall, userspace needs to add
"hcall-xdabr" to the set of strings in the /chosen/hypertas-functions
property in the device tree. If userspace does this and then migrates
the guest to a host where the kernel doesn't include this patch, then
userspace will need to implement H_SET_XDABR by writing the specified
DABR value to the DABR using the ONE_REG interface. In that case, the
old kernel will set DABRX to DABRX_USER | DABRX_KERNEL. That should
still work correctly, at least for Linux guests, since Linux guests
cope with getting data breakpoint interrupts in modes that weren't
requested by just ignoring the interrupt, and Linux guests never set
DABRX_BTI.
The other thing this does is to make H_SET_DABR and H_SET_XDABR work
on POWER8, which has the DAWR and DAWRX instead of DABR/X. Guests that
know about POWER8 should use H_SET_MODE rather than H_SET_[X]DABR, but
guests running in POWER7 compatibility mode will still use H_SET_[X]DABR.
For them, this adds the logic to convert DABR/X values into DAWR/X values
on POWER8.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:29 +08:00
|
|
|
lwz r5,VCPU_DABRX(r4)
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
ld r6,VCPU_DABR(r4)
|
|
|
|
|
mtspr SPRN_DABRX,r5
|
|
|
|
|
mtspr SPRN_DABR,r6
|
|
|
|
|
isync
|
|
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
|
|
|
|
|
|
2014-03-25 07:47:02 +08:00
|
|
|
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
/*
|
|
|
|
|
* Branch around the call if both CPU_FTR_TM and
|
|
|
|
|
* CPU_FTR_P9_TM_HV_ASSIST are off.
|
|
|
|
|
*/
|
2014-03-25 07:47:02 +08:00
|
|
|
BEGIN_FTR_SECTION
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
b 91f
|
|
|
|
|
END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
|
KVM: PPC: Book3S HV: Fix bug causing host SLB to be restored incorrectly
Aneesh Kumar reported seeing host crashes when running recent kernels
on POWER8. The symptom was an oops like this:
Unable to handle kernel paging request for data at address 0xf00000000786c620
Faulting instruction address: 0xc00000000030e1e4
Oops: Kernel access of bad area, sig: 11 [#1]
LE SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in: powernv_op_panel
CPU: 24 PID: 6663 Comm: qemu-system-ppc Tainted: G W 4.13.0-rc7-43932-gfc36c59 #2
task: c000000fdeadfe80 task.stack: c000000fdeb68000
NIP: c00000000030e1e4 LR: c00000000030de6c CTR: c000000000103620
REGS: c000000fdeb6b450 TRAP: 0300 Tainted: G W (4.13.0-rc7-43932-gfc36c59)
MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE> CR: 24044428 XER: 20000000
CFAR: c00000000030e134 DAR: f00000000786c620 DSISR: 40000000 SOFTE: 0
GPR00: 0000000000000000 c000000fdeb6b6d0 c0000000010bd000 000000000000e1b0
GPR04: c00000000115e168 c000001fffa6e4b0 c00000000115d000 c000001e1b180386
GPR08: f000000000000000 c000000f9a8913e0 f00000000786c600 00007fff587d0000
GPR12: c000000fdeb68000 c00000000fb0f000 0000000000000001 00007fff587cffff
GPR16: 0000000000000000 c000000000000000 00000000003fffff c000000fdebfe1f8
GPR20: 0000000000000004 c000000fdeb6b8a8 0000000000000001 0008000000000040
GPR24: 07000000000000c0 00007fff587cffff c000000fdec20bf8 00007fff587d0000
GPR28: c000000fdeca9ac0 00007fff587d0000 00007fff587c0000 00007fff587d0000
NIP [c00000000030e1e4] __get_user_pages_fast+0x434/0x1070
LR [c00000000030de6c] __get_user_pages_fast+0xbc/0x1070
Call Trace:
[c000000fdeb6b6d0] [c00000000139dab8] lock_classes+0x0/0x35fe50 (unreliable)
[c000000fdeb6b7e0] [c00000000030ef38] get_user_pages_fast+0xf8/0x120
[c000000fdeb6b830] [c000000000112318] kvmppc_book3s_hv_page_fault+0x308/0xf30
[c000000fdeb6b960] [c00000000010e10c] kvmppc_vcpu_run_hv+0xfdc/0x1f00
[c000000fdeb6bb20] [c0000000000e915c] kvmppc_vcpu_run+0x2c/0x40
[c000000fdeb6bb40] [c0000000000e5650] kvm_arch_vcpu_ioctl_run+0x110/0x300
[c000000fdeb6bbe0] [c0000000000d6468] kvm_vcpu_ioctl+0x528/0x900
[c000000fdeb6bd40] [c0000000003bc04c] do_vfs_ioctl+0xcc/0x950
[c000000fdeb6bde0] [c0000000003bc930] SyS_ioctl+0x60/0x100
[c000000fdeb6be30] [c00000000000b96c] system_call+0x58/0x6c
Instruction dump:
7ca81a14 2fa50000 41de0010 7cc8182a 68c60002 78c6ffe2 0b060000 3cc2000a
794a3664 390610d8 e9080000 7d485214 <e90a0020> 7d435378 790507e1 408202f0
---[ end trace fad4a342d0414aa2 ]---
It turns out that what has happened is that the SLB entry for the
vmmemap region hasn't been reloaded on exit from a guest, and it has
the wrong page size. Then, when the host next accesses the vmemmap
region, it gets a page fault.
Commit a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with
KVM", 2017-07-24) modified the guest exit code so that it now only clears
out the SLB for hash guest. The code tests the radix flag and puts the
result in a non-volatile CR field, CR2, and later branches based on CR2.
Unfortunately, the kvmppc_save_tm function, which gets called between
those two points, modifies all the user-visible registers in the case
where the guest was in transactional or suspended state, except for a
few which it restores (namely r1, r2, r9 and r13). Thus the hash/radix indication in CR2 gets corrupted.
This fixes the problem by re-doing the comparison just before the
result is needed. For good measure, this also adds comments next to
the call sites of kvmppc_save_tm and kvmppc_restore_tm pointing out
that non-volatile register state will be lost.
Cc: stable@vger.kernel.org # v4.13
Fixes: a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with KVM")
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-12 11:47:23 +08:00
|
|
|
/*
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
* NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
|
KVM: PPC: Book3S HV: Fix bug causing host SLB to be restored incorrectly
Aneesh Kumar reported seeing host crashes when running recent kernels
on POWER8. The symptom was an oops like this:
Unable to handle kernel paging request for data at address 0xf00000000786c620
Faulting instruction address: 0xc00000000030e1e4
Oops: Kernel access of bad area, sig: 11 [#1]
LE SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in: powernv_op_panel
CPU: 24 PID: 6663 Comm: qemu-system-ppc Tainted: G W 4.13.0-rc7-43932-gfc36c59 #2
task: c000000fdeadfe80 task.stack: c000000fdeb68000
NIP: c00000000030e1e4 LR: c00000000030de6c CTR: c000000000103620
REGS: c000000fdeb6b450 TRAP: 0300 Tainted: G W (4.13.0-rc7-43932-gfc36c59)
MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE> CR: 24044428 XER: 20000000
CFAR: c00000000030e134 DAR: f00000000786c620 DSISR: 40000000 SOFTE: 0
GPR00: 0000000000000000 c000000fdeb6b6d0 c0000000010bd000 000000000000e1b0
GPR04: c00000000115e168 c000001fffa6e4b0 c00000000115d000 c000001e1b180386
GPR08: f000000000000000 c000000f9a8913e0 f00000000786c600 00007fff587d0000
GPR12: c000000fdeb68000 c00000000fb0f000 0000000000000001 00007fff587cffff
GPR16: 0000000000000000 c000000000000000 00000000003fffff c000000fdebfe1f8
GPR20: 0000000000000004 c000000fdeb6b8a8 0000000000000001 0008000000000040
GPR24: 07000000000000c0 00007fff587cffff c000000fdec20bf8 00007fff587d0000
GPR28: c000000fdeca9ac0 00007fff587d0000 00007fff587c0000 00007fff587d0000
NIP [c00000000030e1e4] __get_user_pages_fast+0x434/0x1070
LR [c00000000030de6c] __get_user_pages_fast+0xbc/0x1070
Call Trace:
[c000000fdeb6b6d0] [c00000000139dab8] lock_classes+0x0/0x35fe50 (unreliable)
[c000000fdeb6b7e0] [c00000000030ef38] get_user_pages_fast+0xf8/0x120
[c000000fdeb6b830] [c000000000112318] kvmppc_book3s_hv_page_fault+0x308/0xf30
[c000000fdeb6b960] [c00000000010e10c] kvmppc_vcpu_run_hv+0xfdc/0x1f00
[c000000fdeb6bb20] [c0000000000e915c] kvmppc_vcpu_run+0x2c/0x40
[c000000fdeb6bb40] [c0000000000e5650] kvm_arch_vcpu_ioctl_run+0x110/0x300
[c000000fdeb6bbe0] [c0000000000d6468] kvm_vcpu_ioctl+0x528/0x900
[c000000fdeb6bd40] [c0000000003bc04c] do_vfs_ioctl+0xcc/0x950
[c000000fdeb6bde0] [c0000000003bc930] SyS_ioctl+0x60/0x100
[c000000fdeb6be30] [c00000000000b96c] system_call+0x58/0x6c
Instruction dump:
7ca81a14 2fa50000 41de0010 7cc8182a 68c60002 78c6ffe2 0b060000 3cc2000a
794a3664 390610d8 e9080000 7d485214 <e90a0020> 7d435378 790507e1 408202f0
---[ end trace fad4a342d0414aa2 ]---
It turns out that what has happened is that the SLB entry for the
vmmemap region hasn't been reloaded on exit from a guest, and it has
the wrong page size. Then, when the host next accesses the vmemmap
region, it gets a page fault.
Commit a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with
KVM", 2017-07-24) modified the guest exit code so that it now only clears
out the SLB for hash guest. The code tests the radix flag and puts the
result in a non-volatile CR field, CR2, and later branches based on CR2.
Unfortunately, the kvmppc_save_tm function, which gets called between
those two points, modifies all the user-visible registers in the case
where the guest was in transactional or suspended state, except for a
few which it restores (namely r1, r2, r9 and r13). Thus the hash/radix indication in CR2 gets corrupted.
This fixes the problem by re-doing the comparison just before the
result is needed. For good measure, this also adds comments next to
the call sites of kvmppc_save_tm and kvmppc_restore_tm pointing out
that non-volatile register state will be lost.
Cc: stable@vger.kernel.org # v4.13
Fixes: a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with KVM")
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-12 11:47:23 +08:00
|
|
|
*/
|
2018-05-23 15:01:48 +08:00
|
|
|
mr r3, r4
|
|
|
|
|
ld r4, VCPU_MSR(r3)
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
li r5, 0 /* don't preserve non-vol regs */
|
2018-05-30 18:07:52 +08:00
|
|
|
bl kvmppc_restore_tm_hv
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
nop
|
2018-05-23 15:01:48 +08:00
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
91:
|
2014-03-25 07:47:02 +08:00
|
|
|
#endif
|
|
|
|
|
|
2018-10-08 13:30:51 +08:00
|
|
|
/* Load guest PMU registers; r4 = vcpu pointer here */
|
|
|
|
|
mr r3, r4
|
|
|
|
|
bl kvmhv_load_guest_pmu
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
|
|
|
|
|
/* Load up FP, VMX and VSX registers */
|
2018-10-08 13:30:51 +08:00
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
bl kvmppc_load_fp
|
|
|
|
|
|
|
|
|
|
ld r14, VCPU_GPR(R14)(r4)
|
|
|
|
|
ld r15, VCPU_GPR(R15)(r4)
|
|
|
|
|
ld r16, VCPU_GPR(R16)(r4)
|
|
|
|
|
ld r17, VCPU_GPR(R17)(r4)
|
|
|
|
|
ld r18, VCPU_GPR(R18)(r4)
|
|
|
|
|
ld r19, VCPU_GPR(R19)(r4)
|
|
|
|
|
ld r20, VCPU_GPR(R20)(r4)
|
|
|
|
|
ld r21, VCPU_GPR(R21)(r4)
|
|
|
|
|
ld r22, VCPU_GPR(R22)(r4)
|
|
|
|
|
ld r23, VCPU_GPR(R23)(r4)
|
|
|
|
|
ld r24, VCPU_GPR(R24)(r4)
|
|
|
|
|
ld r25, VCPU_GPR(R25)(r4)
|
|
|
|
|
ld r26, VCPU_GPR(R26)(r4)
|
|
|
|
|
ld r27, VCPU_GPR(R27)(r4)
|
|
|
|
|
ld r28, VCPU_GPR(R28)(r4)
|
|
|
|
|
ld r29, VCPU_GPR(R29)(r4)
|
|
|
|
|
ld r30, VCPU_GPR(R30)(r4)
|
|
|
|
|
ld r31, VCPU_GPR(R31)(r4)
|
|
|
|
|
|
|
|
|
|
/* Switch DSCR to guest value */
|
|
|
|
|
ld r5, VCPU_DSCR(r4)
|
|
|
|
|
mtspr SPRN_DSCR, r5
|
|
|
|
|
|
2014-01-08 18:25:21 +08:00
|
|
|
BEGIN_FTR_SECTION
|
2014-12-03 10:30:38 +08:00
|
|
|
/* Skip next section on POWER7 */
|
2014-01-08 18:25:21 +08:00
|
|
|
b 8f
|
|
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
|
|
|
|
|
/* Load up POWER8-specific registers */
|
|
|
|
|
ld r5, VCPU_IAMR(r4)
|
|
|
|
|
lwz r6, VCPU_PSPB(r4)
|
|
|
|
|
ld r7, VCPU_FSCR(r4)
|
|
|
|
|
mtspr SPRN_IAMR, r5
|
|
|
|
|
mtspr SPRN_PSPB, r6
|
|
|
|
|
mtspr SPRN_FSCR, r7
|
|
|
|
|
ld r5, VCPU_DAWR(r4)
|
|
|
|
|
ld r6, VCPU_DAWRX(r4)
|
|
|
|
|
ld r7, VCPU_CIABR(r4)
|
|
|
|
|
ld r8, VCPU_TAR(r4)
|
2018-03-27 12:37:22 +08:00
|
|
|
/*
|
|
|
|
|
* Handle broken DAWR case by not writing it. This means we
|
|
|
|
|
* can still store the DAWR register for migration.
|
|
|
|
|
*/
|
|
|
|
|
BEGIN_FTR_SECTION
|
2014-01-08 18:25:21 +08:00
|
|
|
mtspr SPRN_DAWR, r5
|
|
|
|
|
mtspr SPRN_DAWRX, r6
|
2018-03-27 12:37:22 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_DAWR)
|
2014-01-08 18:25:21 +08:00
|
|
|
mtspr SPRN_CIABR, r7
|
|
|
|
|
mtspr SPRN_TAR, r8
|
|
|
|
|
ld r5, VCPU_IC(r4)
|
2014-01-08 18:25:32 +08:00
|
|
|
ld r8, VCPU_EBBHR(r4)
|
2016-09-15 11:42:52 +08:00
|
|
|
mtspr SPRN_IC, r5
|
2014-01-08 18:25:21 +08:00
|
|
|
mtspr SPRN_EBBHR, r8
|
|
|
|
|
ld r5, VCPU_EBBRR(r4)
|
|
|
|
|
ld r6, VCPU_BESCR(r4)
|
2016-11-16 19:33:27 +08:00
|
|
|
lwz r7, VCPU_GUEST_PID(r4)
|
|
|
|
|
ld r8, VCPU_WORT(r4)
|
2014-01-08 18:25:21 +08:00
|
|
|
mtspr SPRN_EBBRR, r5
|
|
|
|
|
mtspr SPRN_BESCR, r6
|
2016-11-16 19:33:27 +08:00
|
|
|
mtspr SPRN_PID, r7
|
|
|
|
|
mtspr SPRN_WORT, r8
|
|
|
|
|
BEGIN_FTR_SECTION
|
2016-11-18 10:11:42 +08:00
|
|
|
/* POWER8-only registers */
|
2014-01-08 18:25:21 +08:00
|
|
|
ld r5, VCPU_TCSCR(r4)
|
|
|
|
|
ld r6, VCPU_ACOP(r4)
|
2016-11-16 19:33:27 +08:00
|
|
|
ld r7, VCPU_CSIGR(r4)
|
|
|
|
|
ld r8, VCPU_TACR(r4)
|
2014-01-08 18:25:21 +08:00
|
|
|
mtspr SPRN_TCSCR, r5
|
|
|
|
|
mtspr SPRN_ACOP, r6
|
2016-11-16 19:33:27 +08:00
|
|
|
mtspr SPRN_CSIGR, r7
|
|
|
|
|
mtspr SPRN_TACR, r8
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
nop
|
2016-11-18 10:11:42 +08:00
|
|
|
FTR_SECTION_ELSE
|
|
|
|
|
/* POWER9-only registers */
|
|
|
|
|
ld r5, VCPU_TID(r4)
|
|
|
|
|
ld r6, VCPU_PSSCR(r4)
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
lbz r8, HSTATE_FAKE_SUSPEND(r13)
|
2016-11-18 10:11:42 +08:00
|
|
|
oris r6, r6, PSSCR_EC@h /* This makes stop trap to HV */
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
rldimi r6, r8, PSSCR_FAKE_SUSPEND_LG, 63 - PSSCR_FAKE_SUSPEND_LG
|
2017-02-15 11:30:17 +08:00
|
|
|
ld r7, VCPU_HFSCR(r4)
|
2016-11-18 10:11:42 +08:00
|
|
|
mtspr SPRN_TIDR, r5
|
|
|
|
|
mtspr SPRN_PSSCR, r6
|
2017-02-15 11:30:17 +08:00
|
|
|
mtspr SPRN_HFSCR, r7
|
2016-11-18 10:11:42 +08:00
|
|
|
ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
|
2014-01-08 18:25:21 +08:00
|
|
|
8:
|
|
|
|
|
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
ld r5, VCPU_SPRG0(r4)
|
|
|
|
|
ld r6, VCPU_SPRG1(r4)
|
|
|
|
|
ld r7, VCPU_SPRG2(r4)
|
|
|
|
|
ld r8, VCPU_SPRG3(r4)
|
|
|
|
|
mtspr SPRN_SPRG0, r5
|
|
|
|
|
mtspr SPRN_SPRG1, r6
|
|
|
|
|
mtspr SPRN_SPRG2, r7
|
|
|
|
|
mtspr SPRN_SPRG3, r8
|
|
|
|
|
|
|
|
|
|
/* Load up DAR and DSISR */
|
|
|
|
|
ld r5, VCPU_DAR(r4)
|
|
|
|
|
lwz r6, VCPU_DSISR(r4)
|
|
|
|
|
mtspr SPRN_DAR, r5
|
|
|
|
|
mtspr SPRN_DSISR, r6
|
|
|
|
|
|
|
|
|
|
/* Restore AMR and UAMOR, set AMOR to all 1s */
|
|
|
|
|
ld r5,VCPU_AMR(r4)
|
|
|
|
|
ld r6,VCPU_UAMOR(r4)
|
|
|
|
|
li r7,-1
|
|
|
|
|
mtspr SPRN_AMR,r5
|
|
|
|
|
mtspr SPRN_UAMOR,r6
|
|
|
|
|
mtspr SPRN_AMOR,r7
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
|
|
|
|
|
/* Restore state of CTRL run bit; assume 1 on entry */
|
|
|
|
|
lwz r5,VCPU_CTRL(r4)
|
|
|
|
|
andi. r5,r5,1
|
|
|
|
|
bne 4f
|
|
|
|
|
mfspr r6,SPRN_CTRLF
|
|
|
|
|
clrrdi r6,r6,1
|
|
|
|
|
mtspr SPRN_CTRLT,r6
|
|
|
|
|
4:
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
/* Secondary threads wait for primary to have done partition switch */
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
lbz r6, HSTATE_PTID(r13)
|
|
|
|
|
cmpwi r6, 0
|
|
|
|
|
beq 21f
|
|
|
|
|
lbz r0, VCORE_IN_GUEST(r5)
|
|
|
|
|
cmpwi r0, 0
|
|
|
|
|
bne 21f
|
|
|
|
|
HMT_LOW
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
20: lwz r3, VCORE_ENTRY_EXIT(r5)
|
|
|
|
|
cmpwi r3, 0x100
|
|
|
|
|
bge no_switch_exit
|
|
|
|
|
lbz r0, VCORE_IN_GUEST(r5)
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
cmpwi r0, 0
|
|
|
|
|
beq 20b
|
|
|
|
|
HMT_MEDIUM
|
|
|
|
|
21:
|
|
|
|
|
/* Set LPCR. */
|
|
|
|
|
ld r8,VCORE_LPCR(r5)
|
|
|
|
|
mtspr SPRN_LPCR,r8
|
|
|
|
|
isync
|
|
|
|
|
|
KVM: PPC: Book3S HV: Snapshot timebase offset on guest entry
Currently, the HV KVM guest entry/exit code adds the timebase offset
from the vcore struct to the timebase on guest entry, and subtracts
it on guest exit. Which is fine, except that it is possible for
userspace to change the offset using the SET_ONE_REG interface while
the vcore is running, as there is only one timebase offset per vcore
but potentially multiple VCPUs in the vcore. If that were to happen,
KVM would subtract a different offset on guest exit from that which
it had added on guest entry, leading to the timebase being out of sync
between cores in the host, which then leads to bad things happening
such as hangs and spurious watchdog timeouts.
To fix this, we add a new field 'tb_offset_applied' to the vcore struct
which stores the offset that is currently applied to the timebase.
This value is set from the vcore tb_offset field on guest entry, and
is what is subtracted from the timebase on guest exit. Since it is
zero when the timebase offset is not applied, we can simplify the
logic in kvmhv_start_timing and kvmhv_accumulate_time.
In addition, we had secondary threads reading the timebase while
running concurrently with code on the primary thread which would
eventually add or subtract the timebase offset from the timebase.
This occurred while saving or restoring the DEC register value on
the secondary threads. Although no specific incorrect behaviour has
been observed, this is a race which should be fixed. To fix it, we
move the DEC saving code to just before we call kvmhv_commence_exit,
and the DEC restoring code to after the point where we have waited
for the primary thread to switch the MMU context and add the timebase
offset. That way we are sure that the timebase contains the guest
timebase value in both cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-04-20 20:51:11 +08:00
|
|
|
/*
|
|
|
|
|
* Set the decrementer to the guest decrementer.
|
|
|
|
|
*/
|
|
|
|
|
ld r8,VCPU_DEC_EXPIRES(r4)
|
|
|
|
|
/* r8 is a host timebase value here, convert to guest TB */
|
|
|
|
|
ld r5,HSTATE_KVM_VCORE(r13)
|
|
|
|
|
ld r6,VCORE_TB_OFFSET_APPL(r5)
|
|
|
|
|
add r8,r8,r6
|
|
|
|
|
mftb r7
|
|
|
|
|
subf r3,r7,r8
|
|
|
|
|
mtspr SPRN_DEC,r3
|
|
|
|
|
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
/* Check if HDEC expires soon */
|
|
|
|
|
mfspr r3, SPRN_HDEC
|
2017-05-22 14:25:14 +08:00
|
|
|
EXTEND_HDEC(r3)
|
|
|
|
|
cmpdi r3, 512 /* 1 microsecond */
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
blt hdec_soon
|
|
|
|
|
|
2018-01-11 11:51:02 +08:00
|
|
|
/* For hash guest, clear out and reload the SLB */
|
|
|
|
|
ld r6, VCPU_KVM(r4)
|
|
|
|
|
lbz r0, KVM_RADIX(r6)
|
|
|
|
|
cmpwi r0, 0
|
|
|
|
|
bne 9f
|
|
|
|
|
li r6, 0
|
|
|
|
|
slbmte r6, r6
|
|
|
|
|
slbia
|
|
|
|
|
ptesync
|
|
|
|
|
|
|
|
|
|
/* Load up guest SLB entries (N.B. slb_max will be 0 for radix) */
|
|
|
|
|
lwz r5,VCPU_SLB_MAX(r4)
|
|
|
|
|
cmpwi r5,0
|
|
|
|
|
beq 9f
|
|
|
|
|
mtctr r5
|
|
|
|
|
addi r6,r4,VCPU_SLB
|
|
|
|
|
1: ld r8,VCPU_SLB_E(r6)
|
|
|
|
|
ld r9,VCPU_SLB_V(r6)
|
|
|
|
|
slbmte r9,r8
|
|
|
|
|
addi r6,r6,VCPU_SLB_SIZE
|
|
|
|
|
bdnz 1b
|
|
|
|
|
9:
|
|
|
|
|
|
2017-04-05 15:54:56 +08:00
|
|
|
#ifdef CONFIG_KVM_XICS
|
|
|
|
|
/* We are entering the guest on that thread, push VCPU to XIVE */
|
|
|
|
|
ld r10, HSTATE_XIVE_TIMA_PHYS(r13)
|
2017-08-15 12:37:01 +08:00
|
|
|
cmpldi cr0, r10, 0
|
2017-04-05 15:54:56 +08:00
|
|
|
beq no_xive
|
|
|
|
|
ld r11, VCPU_XIVE_SAVED_STATE(r4)
|
|
|
|
|
li r9, TM_QW1_OS
|
|
|
|
|
eieio
|
KVM: PPC: Book3S HV: Add more barriers in XIVE load/unload code
On POWER9 systems, we push the VCPU context onto the XIVE (eXternal
Interrupt Virtualization Engine) hardware when entering a guest,
and pull the context off the XIVE when exiting the guest. The push
is done with cache-inhibited stores, and the pull with cache-inhibited
loads.
Testing has revealed that it is possible (though very rare) for
the stores to get reordered with the loads so that we end up with the
guest VCPU context still loaded on the XIVE after we have exited the
guest. When that happens, it is possible for the same VCPU context
to then get loaded on another CPU, which causes the machine to
checkstop.
To fix this, we add I/O barrier instructions (eieio) before and
after the push and pull operations. As partial compensation for the
potential slowdown caused by the extra barriers, we remove the eieio
instructions between the two stores in the push operation, and between
the two loads in the pull operation. (The architecture requires
loads to cache-inhibited, guarded storage to be kept in order, and
requires stores to cache-inhibited, guarded storage likewise to be
kept in order, but allows such loads and stores to be reordered with
respect to each other.)
Reported-by: Carol L Soto <clsoto@us.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-16 05:37:54 +08:00
|
|
|
stdcix r11,r9,r10
|
2017-04-05 15:54:56 +08:00
|
|
|
lwz r11, VCPU_XIVE_CAM_WORD(r4)
|
|
|
|
|
li r9, TM_QW1_OS + TM_WORD2
|
|
|
|
|
stwcix r11,r9,r10
|
|
|
|
|
li r9, 1
|
2018-01-12 10:37:15 +08:00
|
|
|
stb r9, VCPU_XIVE_PUSHED(r4)
|
KVM: PPC: Book3S HV: Add more barriers in XIVE load/unload code
On POWER9 systems, we push the VCPU context onto the XIVE (eXternal
Interrupt Virtualization Engine) hardware when entering a guest,
and pull the context off the XIVE when exiting the guest. The push
is done with cache-inhibited stores, and the pull with cache-inhibited
loads.
Testing has revealed that it is possible (though very rare) for
the stores to get reordered with the loads so that we end up with the
guest VCPU context still loaded on the XIVE after we have exited the
guest. When that happens, it is possible for the same VCPU context
to then get loaded on another CPU, which causes the machine to
checkstop.
To fix this, we add I/O barrier instructions (eieio) before and
after the push and pull operations. As partial compensation for the
potential slowdown caused by the extra barriers, we remove the eieio
instructions between the two stores in the push operation, and between
the two loads in the pull operation. (The architecture requires
loads to cache-inhibited, guarded storage to be kept in order, and
requires stores to cache-inhibited, guarded storage likewise to be
kept in order, but allows such loads and stores to be reordered with
respect to each other.)
Reported-by: Carol L Soto <clsoto@us.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-16 05:37:54 +08:00
|
|
|
eieio
|
2018-01-12 10:37:13 +08:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* We clear the irq_pending flag. There is a small chance of a
|
|
|
|
|
* race vs. the escalation interrupt happening on another
|
|
|
|
|
* processor setting it again, but the only consequence is to
|
|
|
|
|
* cause a spurrious wakeup on the next H_CEDE which is not an
|
|
|
|
|
* issue.
|
|
|
|
|
*/
|
|
|
|
|
li r0,0
|
|
|
|
|
stb r0, VCPU_IRQ_PENDING(r4)
|
2018-01-12 10:37:16 +08:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* In single escalation mode, if the escalation interrupt is
|
|
|
|
|
* on, we mask it.
|
|
|
|
|
*/
|
|
|
|
|
lbz r0, VCPU_XIVE_ESC_ON(r4)
|
|
|
|
|
cmpwi r0,0
|
|
|
|
|
beq 1f
|
|
|
|
|
ld r10, VCPU_XIVE_ESC_RADDR(r4)
|
|
|
|
|
li r9, XIVE_ESB_SET_PQ_01
|
|
|
|
|
ldcix r0, r10, r9
|
|
|
|
|
sync
|
|
|
|
|
|
|
|
|
|
/* We have a possible subtle race here: The escalation interrupt might
|
|
|
|
|
* have fired and be on its way to the host queue while we mask it,
|
|
|
|
|
* and if we unmask it early enough (re-cede right away), there is
|
|
|
|
|
* a theorical possibility that it fires again, thus landing in the
|
|
|
|
|
* target queue more than once which is a big no-no.
|
|
|
|
|
*
|
|
|
|
|
* Fortunately, solving this is rather easy. If the above load setting
|
|
|
|
|
* PQ to 01 returns a previous value where P is set, then we know the
|
|
|
|
|
* escalation interrupt is somewhere on its way to the host. In that
|
|
|
|
|
* case we simply don't clear the xive_esc_on flag below. It will be
|
|
|
|
|
* eventually cleared by the handler for the escalation interrupt.
|
|
|
|
|
*
|
|
|
|
|
* Then, when doing a cede, we check that flag again before re-enabling
|
|
|
|
|
* the escalation interrupt, and if set, we abort the cede.
|
|
|
|
|
*/
|
|
|
|
|
andi. r0, r0, XIVE_ESB_VAL_P
|
|
|
|
|
bne- 1f
|
|
|
|
|
|
|
|
|
|
/* Now P is 0, we can clear the flag */
|
|
|
|
|
li r0, 0
|
|
|
|
|
stb r0, VCPU_XIVE_ESC_ON(r4)
|
|
|
|
|
1:
|
2017-04-05 15:54:56 +08:00
|
|
|
no_xive:
|
|
|
|
|
#endif /* CONFIG_KVM_XICS */
|
|
|
|
|
|
2018-10-08 13:30:55 +08:00
|
|
|
li r0, 0
|
|
|
|
|
stw r0, STACK_SLOT_SHORT_PATH(r1)
|
|
|
|
|
|
2018-10-08 13:30:52 +08:00
|
|
|
deliver_guest_interrupt: /* r4 = vcpu, r13 = paca */
|
2018-10-08 13:30:50 +08:00
|
|
|
/* Check if we can deliver an external or decrementer interrupt now */
|
|
|
|
|
ld r0, VCPU_PENDING_EXC(r4)
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
/* On POWER9, also check for emulated doorbell interrupt */
|
|
|
|
|
lbz r3, VCPU_DBELL_REQ(r4)
|
|
|
|
|
or r0, r0, r3
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
|
|
|
|
cmpdi r0, 0
|
|
|
|
|
beq 71f
|
|
|
|
|
mr r3, r4
|
|
|
|
|
bl kvmppc_guest_entry_inject_int
|
|
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
71:
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
ld r6, VCPU_SRR0(r4)
|
|
|
|
|
ld r7, VCPU_SRR1(r4)
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
mtspr SPRN_SRR0, r6
|
|
|
|
|
mtspr SPRN_SRR1, r7
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
|
2018-10-08 13:30:55 +08:00
|
|
|
fast_guest_entry_c:
|
|
|
|
|
ld r10, VCPU_PC(r4)
|
|
|
|
|
ld r11, VCPU_MSR(r4)
|
2013-04-18 04:31:41 +08:00
|
|
|
/* r11 = vcpu->arch.msr & ~MSR_HV */
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
rldicl r11, r11, 63 - MSR_HV_LG, 1
|
|
|
|
|
rotldi r11, r11, 1 + MSR_HV_LG
|
|
|
|
|
ori r11, r11, MSR_ME
|
|
|
|
|
|
2018-10-08 13:30:50 +08:00
|
|
|
ld r6, VCPU_CTR(r4)
|
|
|
|
|
ld r7, VCPU_XER(r4)
|
|
|
|
|
mtctr r6
|
|
|
|
|
mtxer r7
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
|
2013-11-19 14:12:48 +08:00
|
|
|
/*
|
|
|
|
|
* Required state:
|
|
|
|
|
* R4 = vcpu
|
|
|
|
|
* R10: value for HSRR0
|
|
|
|
|
* R11: value for HSRR1
|
|
|
|
|
* R13 = PACA
|
|
|
|
|
*/
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
fast_guest_return:
|
2013-04-18 04:31:41 +08:00
|
|
|
li r0,0
|
|
|
|
|
stb r0,VCPU_CEDED(r4) /* cancel cede */
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
mtspr SPRN_HSRR0,r10
|
|
|
|
|
mtspr SPRN_HSRR1,r11
|
|
|
|
|
|
|
|
|
|
/* Activate guest mode, so faults get handled by KVM */
|
KVM: PPC: Book3S HV: Better handling of exceptions that happen in real mode
When an interrupt or exception happens in the guest that comes to the
host, the CPU goes to hypervisor real mode (MMU off) to handle the
exception but doesn't change the MMU context. After saving a few
registers, we then clear the "in guest" flag. If, for any reason,
we get an exception in the real-mode code, that then gets handled
by the normal kernel exception handlers, which turn the MMU on. This
is disastrous if the MMU is still set to the guest context, since we
end up executing instructions from random places in the guest kernel
with hypervisor privilege.
In order to catch this situation, we define a new value for the "in guest"
flag, KVM_GUEST_MODE_HOST_HV, to indicate that we are in hypervisor real
mode with guest MMU context. If the "in guest" flag is set to this value,
we branch off to an emergency handler. For the moment, this just does
a branch to self to stop the CPU from doing anything further.
While we're here, we define another new flag value to indicate that we
are in a HV guest, as distinct from a PR guest. This will be useful
when we have a kernel that can support both PR and HV guests concurrently.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-10-04 19:45:04 +08:00
|
|
|
li r9, KVM_GUEST_MODE_GUEST_HV
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
stb r9, HSTATE_IN_GUEST(r13)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
|
|
|
|
/* Accumulate timing */
|
|
|
|
|
addi r3, r4, VCPU_TB_GUEST
|
|
|
|
|
bl kvmhv_accumulate_time
|
|
|
|
|
#endif
|
|
|
|
|
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
/* Enter guest */
|
|
|
|
|
|
2013-02-05 02:10:51 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r5, VCPU_CFAR(r4)
|
|
|
|
|
mtspr SPRN_CFAR, r5
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
|
2013-09-20 12:52:39 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r0, VCPU_PPR(r4)
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
|
2013-02-05 02:10:51 +08:00
|
|
|
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
ld r5, VCPU_LR(r4)
|
2018-10-08 13:30:58 +08:00
|
|
|
ld r6, VCPU_CR(r4)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
mtlr r5
|
|
|
|
|
mtcr r6
|
|
|
|
|
|
2012-06-25 21:33:10 +08:00
|
|
|
ld r1, VCPU_GPR(R1)(r4)
|
|
|
|
|
ld r2, VCPU_GPR(R2)(r4)
|
|
|
|
|
ld r3, VCPU_GPR(R3)(r4)
|
|
|
|
|
ld r5, VCPU_GPR(R5)(r4)
|
|
|
|
|
ld r6, VCPU_GPR(R6)(r4)
|
|
|
|
|
ld r7, VCPU_GPR(R7)(r4)
|
|
|
|
|
ld r8, VCPU_GPR(R8)(r4)
|
|
|
|
|
ld r9, VCPU_GPR(R9)(r4)
|
|
|
|
|
ld r10, VCPU_GPR(R10)(r4)
|
|
|
|
|
ld r11, VCPU_GPR(R11)(r4)
|
|
|
|
|
ld r12, VCPU_GPR(R12)(r4)
|
|
|
|
|
ld r13, VCPU_GPR(R13)(r4)
|
|
|
|
|
|
2013-09-20 12:52:39 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
mtspr SPRN_PPR, r0
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
|
2017-09-15 13:26:14 +08:00
|
|
|
|
|
|
|
|
/* Move canary into DSISR to check for later */
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
li r0, 0x7fff
|
|
|
|
|
mtspr SPRN_HDSISR, r0
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
|
|
|
|
|
2013-09-20 12:52:39 +08:00
|
|
|
ld r0, VCPU_GPR(R0)(r4)
|
2012-06-25 21:33:10 +08:00
|
|
|
ld r4, VCPU_GPR(R4)(r4)
|
2018-01-10 00:07:15 +08:00
|
|
|
HRFI_TO_GUEST
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
b .
|
|
|
|
|
|
2018-10-08 13:30:55 +08:00
|
|
|
/*
|
|
|
|
|
* Enter the guest on a P9 or later system where we have exactly
|
|
|
|
|
* one vcpu per vcore and we don't need to go to real mode
|
|
|
|
|
* (which implies that host and guest are both using radix MMU mode).
|
|
|
|
|
* r3 = vcpu pointer
|
|
|
|
|
* Most SPRs and all the VSRs have been loaded already.
|
|
|
|
|
*/
|
|
|
|
|
_GLOBAL(__kvmhv_vcpu_entry_p9)
|
|
|
|
|
EXPORT_SYMBOL_GPL(__kvmhv_vcpu_entry_p9)
|
|
|
|
|
mflr r0
|
|
|
|
|
std r0, PPC_LR_STKOFF(r1)
|
|
|
|
|
stdu r1, -SFS(r1)
|
|
|
|
|
|
|
|
|
|
li r0, 1
|
|
|
|
|
stw r0, STACK_SLOT_SHORT_PATH(r1)
|
|
|
|
|
|
|
|
|
|
std r3, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
mfcr r4
|
|
|
|
|
stw r4, SFS+8(r1)
|
|
|
|
|
|
|
|
|
|
std r1, HSTATE_HOST_R1(r13)
|
|
|
|
|
|
|
|
|
|
reg = 14
|
|
|
|
|
.rept 18
|
|
|
|
|
std reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
|
|
|
|
|
reg = reg + 1
|
|
|
|
|
.endr
|
|
|
|
|
|
|
|
|
|
reg = 14
|
|
|
|
|
.rept 18
|
|
|
|
|
ld reg, __VCPU_GPR(reg)(r3)
|
|
|
|
|
reg = reg + 1
|
|
|
|
|
.endr
|
|
|
|
|
|
|
|
|
|
mfmsr r10
|
|
|
|
|
std r10, HSTATE_HOST_MSR(r13)
|
|
|
|
|
|
|
|
|
|
mr r4, r3
|
|
|
|
|
b fast_guest_entry_c
|
|
|
|
|
guest_exit_short_path:
|
|
|
|
|
|
|
|
|
|
li r0, KVM_GUEST_MODE_NONE
|
|
|
|
|
stb r0, HSTATE_IN_GUEST(r13)
|
|
|
|
|
|
|
|
|
|
reg = 14
|
|
|
|
|
.rept 18
|
|
|
|
|
std reg, __VCPU_GPR(reg)(r9)
|
|
|
|
|
reg = reg + 1
|
|
|
|
|
.endr
|
|
|
|
|
|
|
|
|
|
reg = 14
|
|
|
|
|
.rept 18
|
|
|
|
|
ld reg, STACK_SLOT_NVGPRS + ((reg - 14) * 8)(r1)
|
|
|
|
|
reg = reg + 1
|
|
|
|
|
.endr
|
|
|
|
|
|
|
|
|
|
lwz r4, SFS+8(r1)
|
|
|
|
|
mtcr r4
|
|
|
|
|
|
|
|
|
|
mr r3, r12 /* trap number */
|
|
|
|
|
|
|
|
|
|
addi r1, r1, SFS
|
|
|
|
|
ld r0, PPC_LR_STKOFF(r1)
|
|
|
|
|
mtlr r0
|
|
|
|
|
|
|
|
|
|
/* If we are in real mode, do a rfid to get back to the caller */
|
|
|
|
|
mfmsr r4
|
|
|
|
|
andi. r5, r4, MSR_IR
|
|
|
|
|
bnelr
|
|
|
|
|
rldicl r5, r4, 64 - MSR_TS_S_LG, 62 /* extract TS field */
|
|
|
|
|
mtspr SPRN_SRR0, r0
|
|
|
|
|
ld r10, HSTATE_HOST_MSR(r13)
|
|
|
|
|
rldimi r10, r5, MSR_TS_S_LG, 63 - MSR_TS_T_LG
|
|
|
|
|
mtspr SPRN_SRR1, r10
|
|
|
|
|
RFI_TO_KERNEL
|
|
|
|
|
b .
|
|
|
|
|
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
secondary_too_late:
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
li r12, 0
|
2018-03-07 19:17:20 +08:00
|
|
|
stw r12, STACK_SLOT_TRAP(r1)
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
cmpdi r4, 0
|
|
|
|
|
beq 11f
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
stw r12, VCPU_TRAP(r4)
|
|
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
addi r3, r4, VCPU_TB_RMEXIT
|
|
|
|
|
bl kvmhv_accumulate_time
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
#endif
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
11: b kvmhv_switch_to_host
|
|
|
|
|
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
no_switch_exit:
|
|
|
|
|
HMT_MEDIUM
|
|
|
|
|
li r12, 0
|
|
|
|
|
b 12f
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
hdec_soon:
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
li r12, BOOK3S_INTERRUPT_HV_DECREMENTER
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
12: stw r12, VCPU_TRAP(r4)
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
mr r9, r4
|
|
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
addi r3, r4, VCPU_TB_RMEXIT
|
|
|
|
|
bl kvmhv_accumulate_time
|
|
|
|
|
#endif
|
2018-01-11 11:51:02 +08:00
|
|
|
b guest_bypass
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
/******************************************************************************
|
|
|
|
|
* *
|
|
|
|
|
* Exit code *
|
|
|
|
|
* *
|
|
|
|
|
*****************************************************************************/
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* We come here from the first-level interrupt handlers.
|
|
|
|
|
*/
|
2013-10-08 00:47:55 +08:00
|
|
|
.globl kvmppc_interrupt_hv
|
|
|
|
|
kvmppc_interrupt_hv:
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
/*
|
|
|
|
|
* Register contents:
|
2016-12-22 02:29:25 +08:00
|
|
|
* R12 = (guest CR << 32) | interrupt vector
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
* R13 = PACA
|
2016-12-22 02:29:25 +08:00
|
|
|
* guest R12 saved in shadow VCPU SCRATCH0
|
2017-01-27 12:00:34 +08:00
|
|
|
* guest CTR saved in shadow VCPU SCRATCH1 if RELOCATABLE
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
* guest R13 saved in SPRN_SCRATCH0
|
|
|
|
|
*/
|
2017-01-27 12:00:34 +08:00
|
|
|
std r9, HSTATE_SCRATCH2(r13)
|
KVM: PPC: Book3S HV: Better handling of exceptions that happen in real mode
When an interrupt or exception happens in the guest that comes to the
host, the CPU goes to hypervisor real mode (MMU off) to handle the
exception but doesn't change the MMU context. After saving a few
registers, we then clear the "in guest" flag. If, for any reason,
we get an exception in the real-mode code, that then gets handled
by the normal kernel exception handlers, which turn the MMU on. This
is disastrous if the MMU is still set to the guest context, since we
end up executing instructions from random places in the guest kernel
with hypervisor privilege.
In order to catch this situation, we define a new value for the "in guest"
flag, KVM_GUEST_MODE_HOST_HV, to indicate that we are in hypervisor real
mode with guest MMU context. If the "in guest" flag is set to this value,
we branch off to an emergency handler. For the moment, this just does
a branch to self to stop the CPU from doing anything further.
While we're here, we define another new flag value to indicate that we
are in a HV guest, as distinct from a PR guest. This will be useful
when we have a kernel that can support both PR and HV guests concurrently.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-10-04 19:45:04 +08:00
|
|
|
lbz r9, HSTATE_IN_GUEST(r13)
|
|
|
|
|
cmpwi r9, KVM_GUEST_MODE_HOST_HV
|
|
|
|
|
beq kvmppc_bad_host_intr
|
2013-10-08 00:47:55 +08:00
|
|
|
#ifdef CONFIG_KVM_BOOK3S_PR_POSSIBLE
|
|
|
|
|
cmpwi r9, KVM_GUEST_MODE_GUEST
|
2017-01-27 12:00:34 +08:00
|
|
|
ld r9, HSTATE_SCRATCH2(r13)
|
2013-10-08 00:47:55 +08:00
|
|
|
beq kvmppc_interrupt_pr
|
|
|
|
|
#endif
|
KVM: PPC: Book3S HV: Better handling of exceptions that happen in real mode
When an interrupt or exception happens in the guest that comes to the
host, the CPU goes to hypervisor real mode (MMU off) to handle the
exception but doesn't change the MMU context. After saving a few
registers, we then clear the "in guest" flag. If, for any reason,
we get an exception in the real-mode code, that then gets handled
by the normal kernel exception handlers, which turn the MMU on. This
is disastrous if the MMU is still set to the guest context, since we
end up executing instructions from random places in the guest kernel
with hypervisor privilege.
In order to catch this situation, we define a new value for the "in guest"
flag, KVM_GUEST_MODE_HOST_HV, to indicate that we are in hypervisor real
mode with guest MMU context. If the "in guest" flag is set to this value,
we branch off to an emergency handler. For the moment, this just does
a branch to self to stop the CPU from doing anything further.
While we're here, we define another new flag value to indicate that we
are in a HV guest, as distinct from a PR guest. This will be useful
when we have a kernel that can support both PR and HV guests concurrently.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-10-04 19:45:04 +08:00
|
|
|
/* We're now back in the host but in guest MMU context */
|
|
|
|
|
li r9, KVM_GUEST_MODE_HOST_HV
|
|
|
|
|
stb r9, HSTATE_IN_GUEST(r13)
|
|
|
|
|
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
|
|
|
|
|
/* Save registers */
|
|
|
|
|
|
2012-06-25 21:33:10 +08:00
|
|
|
std r0, VCPU_GPR(R0)(r9)
|
|
|
|
|
std r1, VCPU_GPR(R1)(r9)
|
|
|
|
|
std r2, VCPU_GPR(R2)(r9)
|
|
|
|
|
std r3, VCPU_GPR(R3)(r9)
|
|
|
|
|
std r4, VCPU_GPR(R4)(r9)
|
|
|
|
|
std r5, VCPU_GPR(R5)(r9)
|
|
|
|
|
std r6, VCPU_GPR(R6)(r9)
|
|
|
|
|
std r7, VCPU_GPR(R7)(r9)
|
|
|
|
|
std r8, VCPU_GPR(R8)(r9)
|
2017-01-27 12:00:34 +08:00
|
|
|
ld r0, HSTATE_SCRATCH2(r13)
|
2012-06-25 21:33:10 +08:00
|
|
|
std r0, VCPU_GPR(R9)(r9)
|
|
|
|
|
std r10, VCPU_GPR(R10)(r9)
|
|
|
|
|
std r11, VCPU_GPR(R11)(r9)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
ld r3, HSTATE_SCRATCH0(r13)
|
2012-06-25 21:33:10 +08:00
|
|
|
std r3, VCPU_GPR(R12)(r9)
|
2016-12-22 02:29:25 +08:00
|
|
|
/* CR is in the high half of r12 */
|
|
|
|
|
srdi r4, r12, 32
|
2018-10-08 13:30:58 +08:00
|
|
|
std r4, VCPU_CR(r9)
|
2013-02-05 02:10:51 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r3, HSTATE_CFAR(r13)
|
|
|
|
|
std r3, VCPU_CFAR(r9)
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_CFAR)
|
2013-09-20 12:52:39 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r4, HSTATE_PPR(r13)
|
|
|
|
|
std r4, VCPU_PPR(r9)
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_HAS_PPR)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
|
|
|
|
|
/* Restore R1/R2 so we can handle faults */
|
|
|
|
|
ld r1, HSTATE_HOST_R1(r13)
|
|
|
|
|
ld r2, PACATOC(r13)
|
|
|
|
|
|
|
|
|
|
mfspr r10, SPRN_SRR0
|
|
|
|
|
mfspr r11, SPRN_SRR1
|
|
|
|
|
std r10, VCPU_SRR0(r9)
|
|
|
|
|
std r11, VCPU_SRR1(r9)
|
2016-12-22 02:29:25 +08:00
|
|
|
/* trap is in the low half of r12, clear CR from the high half */
|
|
|
|
|
clrldi r12, r12, 32
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
andi. r0, r12, 2 /* need to read HSRR0/1? */
|
|
|
|
|
beq 1f
|
|
|
|
|
mfspr r10, SPRN_HSRR0
|
|
|
|
|
mfspr r11, SPRN_HSRR1
|
|
|
|
|
clrrdi r12, r12, 2
|
|
|
|
|
1: std r10, VCPU_PC(r9)
|
|
|
|
|
std r11, VCPU_MSR(r9)
|
|
|
|
|
|
|
|
|
|
GET_SCRATCH0(r3)
|
|
|
|
|
mflr r4
|
2012-06-25 21:33:10 +08:00
|
|
|
std r3, VCPU_GPR(R13)(r9)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
std r4, VCPU_LR(r9)
|
|
|
|
|
|
|
|
|
|
stw r12,VCPU_TRAP(r9)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Close race with testing for signals on guest entry
At present, interrupts are hard-disabled fairly late in the guest
entry path, in the assembly code. Since we check for pending signals
for the vCPU(s) task(s) earlier in the guest entry path, it is
possible for a signal to be delivered before we enter the guest but
not be noticed until after we exit the guest for some other reason.
Similarly, it is possible for the scheduler to request a reschedule
while we are in the guest entry path, and we won't notice until after
we have run the guest, potentially for a whole timeslice.
Furthermore, with a radix guest on POWER9, we can take the interrupt
with the MMU on. In this case we end up leaving interrupts
hard-disabled after the guest exit, and they are likely to stay
hard-disabled until we exit to userspace or context-switch to
another process. This was masking the fact that we were also not
setting the RI (recoverable interrupt) bit in the MSR, meaning
that if we had taken an interrupt, it would have crashed the host
kernel with an unrecoverable interrupt message.
To close these races, we need to check for signals and reschedule
requests after hard-disabling interrupts, and then keep interrupts
hard-disabled until we enter the guest. If there is a signal or a
reschedule request from another CPU, it will send an IPI, which will
cause a guest exit.
This puts the interrupt disabling before we call kvmppc_start_thread()
for all the secondary threads of this core that are going to run vCPUs.
The reason for that is that once we have started the secondary threads
there is no easy way to back out without going through at least part
of the guest entry path. However, kvmppc_start_thread() includes some
code for radix guests which needs to call smp_call_function(), which
must be called with interrupts enabled. To solve this problem, this
patch moves that code into a separate function that is called earlier.
When the guest exit is caused by an external interrupt, a hypervisor
doorbell or a hypervisor maintenance interrupt, we now handle these
using the replay facility. __kvmppc_vcore_entry() now returns the
trap number that caused the exit on this thread, and instead of the
assembly code jumping to the handler entry, we return to C code with
interrupts still hard-disabled and set the irq_happened flag in the
PACA, so that when we do local_irq_enable() the appropriate handler
gets called.
With all this, we now have the interrupt soft-enable flag clear while
we are in the guest. This is useful because code in the real-mode
hypercall handlers that checks whether interrupts are enabled will
now see that they are disabled, which is correct, since interrupts
are hard-disabled in the real-mode code.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-06-26 13:45:51 +08:00
|
|
|
/*
|
|
|
|
|
* Now that we have saved away SRR0/1 and HSRR0/1,
|
|
|
|
|
* interrupts are recoverable in principle, so set MSR_RI.
|
|
|
|
|
* This becomes important for relocation-on interrupts from
|
|
|
|
|
* the guest, which we can get in radix mode on POWER9.
|
|
|
|
|
*/
|
|
|
|
|
li r0, MSR_RI
|
|
|
|
|
mtmsrd r0, 1
|
|
|
|
|
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
|
|
|
|
addi r3, r9, VCPU_TB_RMINTR
|
|
|
|
|
mr r4, r9
|
|
|
|
|
bl kvmhv_accumulate_time
|
|
|
|
|
ld r5, VCPU_GPR(R5)(r9)
|
|
|
|
|
ld r6, VCPU_GPR(R6)(r9)
|
|
|
|
|
ld r7, VCPU_GPR(R7)(r9)
|
|
|
|
|
ld r8, VCPU_GPR(R8)(r9)
|
|
|
|
|
#endif
|
|
|
|
|
|
KVM: PPC: Book3S HV: Fix endianness of instruction obtained from HEIR register
There are two ways in which a guest instruction can be obtained from
the guest in the guest exit code in book3s_hv_rmhandlers.S. If the
exit was caused by a Hypervisor Emulation interrupt (i.e. an illegal
instruction), the offending instruction is in the HEIR register
(Hypervisor Emulation Instruction Register). If the exit was caused
by a load or store to an emulated MMIO device, we load the instruction
from the guest by turning data relocation on and loading the instruction
with an lwz instruction.
Unfortunately, in the case where the guest has opposite endianness to
the host, these two methods give results of different endianness, but
both get put into vcpu->arch.last_inst. The HEIR value has been loaded
using guest endianness, whereas the lwz will load the instruction using
host endianness. The rest of the code that uses vcpu->arch.last_inst
assumes it was loaded using host endianness.
To fix this, we define a new vcpu field to store the HEIR value. Then,
in kvmppc_handle_exit_hv(), we transfer the value from this new field to
vcpu->arch.last_inst, doing a byte-swap if the guest and host endianness
differ.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-12-03 10:30:39 +08:00
|
|
|
/* Save HEIR (HV emulation assist reg) in emul_inst
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
if this is an HEI (HV emulation interrupt, e40) */
|
|
|
|
|
li r3,KVM_INST_FETCH_FAILED
|
2015-03-20 17:39:40 +08:00
|
|
|
stw r3,VCPU_LAST_INST(r9)
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
cmpwi r12,BOOK3S_INTERRUPT_H_EMUL_ASSIST
|
|
|
|
|
bne 11f
|
|
|
|
|
mfspr r3,SPRN_HEIR
|
KVM: PPC: Book3S HV: Fix endianness of instruction obtained from HEIR register
There are two ways in which a guest instruction can be obtained from
the guest in the guest exit code in book3s_hv_rmhandlers.S. If the
exit was caused by a Hypervisor Emulation interrupt (i.e. an illegal
instruction), the offending instruction is in the HEIR register
(Hypervisor Emulation Instruction Register). If the exit was caused
by a load or store to an emulated MMIO device, we load the instruction
from the guest by turning data relocation on and loading the instruction
with an lwz instruction.
Unfortunately, in the case where the guest has opposite endianness to
the host, these two methods give results of different endianness, but
both get put into vcpu->arch.last_inst. The HEIR value has been loaded
using guest endianness, whereas the lwz will load the instruction using
host endianness. The rest of the code that uses vcpu->arch.last_inst
assumes it was loaded using host endianness.
To fix this, we define a new vcpu field to store the HEIR value. Then,
in kvmppc_handle_exit_hv(), we transfer the value from this new field to
vcpu->arch.last_inst, doing a byte-swap if the guest and host endianness
differ.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-12-03 10:30:39 +08:00
|
|
|
11: stw r3,VCPU_HEIR(r9)
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
|
|
|
|
|
/* these are volatile across C function calls */
|
2017-01-27 12:00:34 +08:00
|
|
|
#ifdef CONFIG_RELOCATABLE
|
|
|
|
|
ld r3, HSTATE_SCRATCH1(r13)
|
|
|
|
|
mtctr r3
|
|
|
|
|
#else
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
mfctr r3
|
2017-01-27 12:00:34 +08:00
|
|
|
#endif
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
mfxer r4
|
|
|
|
|
std r3, VCPU_CTR(r9)
|
2015-05-27 07:56:57 +08:00
|
|
|
std r4, VCPU_XER(r9)
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
|
2018-10-08 13:30:52 +08:00
|
|
|
/* Save more register state */
|
|
|
|
|
mfdar r3
|
|
|
|
|
mfdsisr r4
|
|
|
|
|
std r3, VCPU_DAR(r9)
|
|
|
|
|
stw r4, VCPU_DSISR(r9)
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
/* If this is a page table miss then see if it's theirs or ours */
|
|
|
|
|
cmpwi r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
|
|
|
|
|
beq kvmppc_hdsi
|
2018-10-08 13:30:52 +08:00
|
|
|
std r3, VCPU_FAULT_DAR(r9)
|
|
|
|
|
stw r4, VCPU_FAULT_DSISR(r9)
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
cmpwi r12, BOOK3S_INTERRUPT_H_INST_STORAGE
|
|
|
|
|
beq kvmppc_hisi
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
|
2018-10-08 13:30:52 +08:00
|
|
|
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
|
|
|
|
|
/* For softpatch interrupt, go off and do TM instruction emulation */
|
|
|
|
|
cmpwi r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
|
|
|
|
|
beq kvmppc_tm_emul
|
|
|
|
|
#endif
|
|
|
|
|
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
/* See if this is a leftover HDEC interrupt */
|
|
|
|
|
cmpwi r12,BOOK3S_INTERRUPT_HV_DECREMENTER
|
|
|
|
|
bne 2f
|
|
|
|
|
mfspr r3,SPRN_HDEC
|
2017-08-25 17:52:12 +08:00
|
|
|
EXTEND_HDEC(r3)
|
|
|
|
|
cmpdi r3,0
|
2015-03-28 11:21:04 +08:00
|
|
|
mr r4,r9
|
|
|
|
|
bge fast_guest_return
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
2:
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
/* See if this is an hcall we can handle in real mode */
|
2011-06-29 08:22:05 +08:00
|
|
|
cmpwi r12,BOOK3S_INTERRUPT_SYSCALL
|
|
|
|
|
beq hcall_try_real_mode
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Use msgsnd for signalling threads on POWER8
This uses msgsnd where possible for signalling other threads within
the same core on POWER8 systems, rather than IPIs through the XICS
interrupt controller. This includes waking secondary threads to run
the guest, the interrupts generated by the virtual XICS, and the
interrupts to bring the other threads out of the guest when exiting.
Aggregated statistics from debugfs across vcpus for a guest with 32
vcpus, 8 threads/vcore, running on a POWER8, show this before the
change:
rm_entry: 3387.6ns (228 - 86600, 1008969 samples)
rm_exit: 4561.5ns (12 - 3477452, 1009402 samples)
rm_intr: 1660.0ns (12 - 553050, 3600051 samples)
and this after the change:
rm_entry: 3060.1ns (212 - 65138, 953873 samples)
rm_exit: 4244.1ns (12 - 9693408, 954331 samples)
rm_intr: 1342.3ns (12 - 1104718, 3405326 samples)
for a test of booting Fedora 20 big-endian to the login prompt.
The time taken for a H_PROD hcall (which is handled in the host
kernel) went down from about 35 microseconds to about 16 microseconds
with this change.
The noinline added to kvmppc_run_core turned out to be necessary for
good performance, at least with gcc 4.9.2 as packaged with Fedora 21
and a little-endian POWER8 host.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:12 +08:00
|
|
|
/* Hypervisor doorbell - exit only if host IPI flag set */
|
|
|
|
|
cmpwi r12, BOOK3S_INTERRUPT_H_DOORBELL
|
|
|
|
|
bne 3f
|
2017-03-13 01:03:49 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
PPC_MSGSYNC
|
2017-10-10 18:18:28 +08:00
|
|
|
lwsync
|
2018-10-08 13:31:04 +08:00
|
|
|
/* always exit if we're running a nested guest */
|
|
|
|
|
ld r0, VCPU_NESTED(r9)
|
|
|
|
|
cmpdi r0, 0
|
|
|
|
|
bne guest_exit_cont
|
2017-03-13 01:03:49 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
KVM: PPC: Book3S HV: Use msgsnd for signalling threads on POWER8
This uses msgsnd where possible for signalling other threads within
the same core on POWER8 systems, rather than IPIs through the XICS
interrupt controller. This includes waking secondary threads to run
the guest, the interrupts generated by the virtual XICS, and the
interrupts to bring the other threads out of the guest when exiting.
Aggregated statistics from debugfs across vcpus for a guest with 32
vcpus, 8 threads/vcore, running on a POWER8, show this before the
change:
rm_entry: 3387.6ns (228 - 86600, 1008969 samples)
rm_exit: 4561.5ns (12 - 3477452, 1009402 samples)
rm_intr: 1660.0ns (12 - 553050, 3600051 samples)
and this after the change:
rm_entry: 3060.1ns (212 - 65138, 953873 samples)
rm_exit: 4244.1ns (12 - 9693408, 954331 samples)
rm_intr: 1342.3ns (12 - 1104718, 3405326 samples)
for a test of booting Fedora 20 big-endian to the login prompt.
The time taken for a H_PROD hcall (which is handled in the host
kernel) went down from about 35 microseconds to about 16 microseconds
with this change.
The noinline added to kvmppc_run_core turned out to be necessary for
good performance, at least with gcc 4.9.2 as packaged with Fedora 21
and a little-endian POWER8 host.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:12 +08:00
|
|
|
lbz r0, HSTATE_HOST_IPI(r13)
|
2015-08-07 20:11:20 +08:00
|
|
|
cmpwi r0, 0
|
2018-10-08 13:30:52 +08:00
|
|
|
beq maybe_reenter_guest
|
KVM: PPC: Book3S HV: Use msgsnd for signalling threads on POWER8
This uses msgsnd where possible for signalling other threads within
the same core on POWER8 systems, rather than IPIs through the XICS
interrupt controller. This includes waking secondary threads to run
the guest, the interrupts generated by the virtual XICS, and the
interrupts to bring the other threads out of the guest when exiting.
Aggregated statistics from debugfs across vcpus for a guest with 32
vcpus, 8 threads/vcore, running on a POWER8, show this before the
change:
rm_entry: 3387.6ns (228 - 86600, 1008969 samples)
rm_exit: 4561.5ns (12 - 3477452, 1009402 samples)
rm_intr: 1660.0ns (12 - 553050, 3600051 samples)
and this after the change:
rm_entry: 3060.1ns (212 - 65138, 953873 samples)
rm_exit: 4244.1ns (12 - 9693408, 954331 samples)
rm_intr: 1342.3ns (12 - 1104718, 3405326 samples)
for a test of booting Fedora 20 big-endian to the login prompt.
The time taken for a H_PROD hcall (which is handled in the host
kernel) went down from about 35 microseconds to about 16 microseconds
with this change.
The noinline added to kvmppc_run_core turned out to be necessary for
good performance, at least with gcc 4.9.2 as packaged with Fedora 21
and a little-endian POWER8 host.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:12 +08:00
|
|
|
b guest_exit_cont
|
|
|
|
|
3:
|
2017-02-15 11:30:17 +08:00
|
|
|
/* If it's a hypervisor facility unavailable interrupt, save HFSCR */
|
|
|
|
|
cmpwi r12, BOOK3S_INTERRUPT_H_FAC_UNAVAIL
|
|
|
|
|
bne 14f
|
|
|
|
|
mfspr r3, SPRN_HFSCR
|
|
|
|
|
std r3, VCPU_HFSCR(r9)
|
|
|
|
|
b guest_exit_cont
|
|
|
|
|
14:
|
2013-04-18 04:30:50 +08:00
|
|
|
/* External interrupt ? */
|
|
|
|
|
cmpwi r12, BOOK3S_INTERRUPT_EXTERNAL
|
2018-10-08 13:30:52 +08:00
|
|
|
beq kvmppc_guest_external
|
2018-01-11 11:31:43 +08:00
|
|
|
/* See if it is a machine check */
|
|
|
|
|
cmpwi r12, BOOK3S_INTERRUPT_MACHINE_CHECK
|
|
|
|
|
beq machine_check_realmode
|
2018-10-08 13:30:52 +08:00
|
|
|
/* Or a hypervisor maintenance interrupt */
|
|
|
|
|
cmpwi r12, BOOK3S_INTERRUPT_HMI
|
|
|
|
|
beq hmi_realmode
|
|
|
|
|
|
|
|
|
|
guest_exit_cont: /* r9 = vcpu, r12 = trap, r13 = paca */
|
|
|
|
|
|
2018-01-11 11:31:43 +08:00
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
|
|
|
|
addi r3, r9, VCPU_TB_RMEXIT
|
|
|
|
|
mr r4, r9
|
|
|
|
|
bl kvmhv_accumulate_time
|
|
|
|
|
#endif
|
2017-04-05 15:54:56 +08:00
|
|
|
#ifdef CONFIG_KVM_XICS
|
|
|
|
|
/* We are exiting, pull the VP from the XIVE */
|
2018-01-12 10:37:15 +08:00
|
|
|
lbz r0, VCPU_XIVE_PUSHED(r9)
|
2017-04-05 15:54:56 +08:00
|
|
|
cmpwi cr0, r0, 0
|
|
|
|
|
beq 1f
|
|
|
|
|
li r7, TM_SPC_PULL_OS_CTX
|
|
|
|
|
li r6, TM_QW1_OS
|
|
|
|
|
mfmsr r0
|
2018-01-12 10:37:14 +08:00
|
|
|
andi. r0, r0, MSR_DR /* in real mode? */
|
2017-04-05 15:54:56 +08:00
|
|
|
beq 2f
|
|
|
|
|
ld r10, HSTATE_XIVE_TIMA_VIRT(r13)
|
|
|
|
|
cmpldi cr0, r10, 0
|
|
|
|
|
beq 1f
|
|
|
|
|
/* First load to pull the context, we ignore the value */
|
|
|
|
|
eieio
|
KVM: PPC: Book3S HV: Add more barriers in XIVE load/unload code
On POWER9 systems, we push the VCPU context onto the XIVE (eXternal
Interrupt Virtualization Engine) hardware when entering a guest,
and pull the context off the XIVE when exiting the guest. The push
is done with cache-inhibited stores, and the pull with cache-inhibited
loads.
Testing has revealed that it is possible (though very rare) for
the stores to get reordered with the loads so that we end up with the
guest VCPU context still loaded on the XIVE after we have exited the
guest. When that happens, it is possible for the same VCPU context
to then get loaded on another CPU, which causes the machine to
checkstop.
To fix this, we add I/O barrier instructions (eieio) before and
after the push and pull operations. As partial compensation for the
potential slowdown caused by the extra barriers, we remove the eieio
instructions between the two stores in the push operation, and between
the two loads in the pull operation. (The architecture requires
loads to cache-inhibited, guarded storage to be kept in order, and
requires stores to cache-inhibited, guarded storage likewise to be
kept in order, but allows such loads and stores to be reordered with
respect to each other.)
Reported-by: Carol L Soto <clsoto@us.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-16 05:37:54 +08:00
|
|
|
lwzx r11, r7, r10
|
2017-04-05 15:54:56 +08:00
|
|
|
/* Second load to recover the context state (Words 0 and 1) */
|
|
|
|
|
ldx r11, r6, r10
|
|
|
|
|
b 3f
|
|
|
|
|
2: ld r10, HSTATE_XIVE_TIMA_PHYS(r13)
|
|
|
|
|
cmpldi cr0, r10, 0
|
|
|
|
|
beq 1f
|
|
|
|
|
/* First load to pull the context, we ignore the value */
|
|
|
|
|
eieio
|
KVM: PPC: Book3S HV: Add more barriers in XIVE load/unload code
On POWER9 systems, we push the VCPU context onto the XIVE (eXternal
Interrupt Virtualization Engine) hardware when entering a guest,
and pull the context off the XIVE when exiting the guest. The push
is done with cache-inhibited stores, and the pull with cache-inhibited
loads.
Testing has revealed that it is possible (though very rare) for
the stores to get reordered with the loads so that we end up with the
guest VCPU context still loaded on the XIVE after we have exited the
guest. When that happens, it is possible for the same VCPU context
to then get loaded on another CPU, which causes the machine to
checkstop.
To fix this, we add I/O barrier instructions (eieio) before and
after the push and pull operations. As partial compensation for the
potential slowdown caused by the extra barriers, we remove the eieio
instructions between the two stores in the push operation, and between
the two loads in the pull operation. (The architecture requires
loads to cache-inhibited, guarded storage to be kept in order, and
requires stores to cache-inhibited, guarded storage likewise to be
kept in order, but allows such loads and stores to be reordered with
respect to each other.)
Reported-by: Carol L Soto <clsoto@us.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-16 05:37:54 +08:00
|
|
|
lwzcix r11, r7, r10
|
2017-04-05 15:54:56 +08:00
|
|
|
/* Second load to recover the context state (Words 0 and 1) */
|
|
|
|
|
ldcix r11, r6, r10
|
|
|
|
|
3: std r11, VCPU_XIVE_SAVED_STATE(r9)
|
|
|
|
|
/* Fixup some of the state for the next load */
|
|
|
|
|
li r10, 0
|
|
|
|
|
li r0, 0xff
|
2018-01-12 10:37:15 +08:00
|
|
|
stb r10, VCPU_XIVE_PUSHED(r9)
|
2017-04-05 15:54:56 +08:00
|
|
|
stb r10, (VCPU_XIVE_SAVED_STATE+3)(r9)
|
|
|
|
|
stb r0, (VCPU_XIVE_SAVED_STATE+4)(r9)
|
KVM: PPC: Book3S HV: Add more barriers in XIVE load/unload code
On POWER9 systems, we push the VCPU context onto the XIVE (eXternal
Interrupt Virtualization Engine) hardware when entering a guest,
and pull the context off the XIVE when exiting the guest. The push
is done with cache-inhibited stores, and the pull with cache-inhibited
loads.
Testing has revealed that it is possible (though very rare) for
the stores to get reordered with the loads so that we end up with the
guest VCPU context still loaded on the XIVE after we have exited the
guest. When that happens, it is possible for the same VCPU context
to then get loaded on another CPU, which causes the machine to
checkstop.
To fix this, we add I/O barrier instructions (eieio) before and
after the push and pull operations. As partial compensation for the
potential slowdown caused by the extra barriers, we remove the eieio
instructions between the two stores in the push operation, and between
the two loads in the pull operation. (The architecture requires
loads to cache-inhibited, guarded storage to be kept in order, and
requires stores to cache-inhibited, guarded storage likewise to be
kept in order, but allows such loads and stores to be reordered with
respect to each other.)
Reported-by: Carol L Soto <clsoto@us.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-16 05:37:54 +08:00
|
|
|
eieio
|
2017-04-05 15:54:56 +08:00
|
|
|
1:
|
|
|
|
|
#endif /* CONFIG_KVM_XICS */
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
|
2018-10-08 13:30:55 +08:00
|
|
|
/* If we came in through the P9 short path, go back out to C now */
|
|
|
|
|
lwz r0, STACK_SLOT_SHORT_PATH(r1)
|
|
|
|
|
cmpwi r0, 0
|
|
|
|
|
bne guest_exit_short_path
|
|
|
|
|
|
2018-01-11 11:51:02 +08:00
|
|
|
/* For hash guest, read the guest SLB and save it away */
|
2017-01-30 18:21:45 +08:00
|
|
|
ld r5, VCPU_KVM(r9)
|
|
|
|
|
lbz r0, KVM_RADIX(r5)
|
|
|
|
|
li r5, 0
|
2018-01-11 11:51:02 +08:00
|
|
|
cmpwi r0, 0
|
|
|
|
|
bne 3f /* for radix, save 0 entries */
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
lwz r0,VCPU_SLB_NR(r9) /* number of entries in SLB */
|
|
|
|
|
mtctr r0
|
|
|
|
|
li r6,0
|
|
|
|
|
addi r7,r9,VCPU_SLB
|
|
|
|
|
1: slbmfee r8,r6
|
|
|
|
|
andis. r0,r8,SLB_ESID_V@h
|
|
|
|
|
beq 2f
|
|
|
|
|
add r8,r8,r6 /* put index in */
|
|
|
|
|
slbmfev r3,r6
|
|
|
|
|
std r8,VCPU_SLB_E(r7)
|
|
|
|
|
std r3,VCPU_SLB_V(r7)
|
|
|
|
|
addi r7,r7,VCPU_SLB_SIZE
|
|
|
|
|
addi r5,r5,1
|
|
|
|
|
2: addi r6,r6,1
|
|
|
|
|
bdnz 1b
|
2018-01-11 11:51:02 +08:00
|
|
|
/* Finally clear out the SLB */
|
|
|
|
|
li r0,0
|
|
|
|
|
slbmte r0,r0
|
|
|
|
|
slbia
|
|
|
|
|
ptesync
|
2017-01-30 18:21:45 +08:00
|
|
|
3: stw r5,VCPU_SLB_MAX(r9)
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
|
2018-03-22 06:48:54 +08:00
|
|
|
/* load host SLB entries */
|
|
|
|
|
BEGIN_MMU_FTR_SECTION
|
|
|
|
|
b 0f
|
|
|
|
|
END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
|
|
|
|
|
ld r8,PACA_SLBSHADOWPTR(r13)
|
|
|
|
|
|
|
|
|
|
.rept SLB_NUM_BOLTED
|
|
|
|
|
li r3, SLBSHADOW_SAVEAREA
|
|
|
|
|
LDX_BE r5, r8, r3
|
|
|
|
|
addi r3, r3, 8
|
|
|
|
|
LDX_BE r6, r8, r3
|
|
|
|
|
andis. r7,r5,SLB_ESID_V@h
|
|
|
|
|
beq 1f
|
|
|
|
|
slbmte r6,r5
|
|
|
|
|
1: addi r8,r8,16
|
|
|
|
|
.endr
|
|
|
|
|
0:
|
|
|
|
|
|
2018-01-11 11:51:02 +08:00
|
|
|
guest_bypass:
|
2018-03-07 19:17:20 +08:00
|
|
|
stw r12, STACK_SLOT_TRAP(r1)
|
KVM: PPC: Book3S HV: Snapshot timebase offset on guest entry
Currently, the HV KVM guest entry/exit code adds the timebase offset
from the vcore struct to the timebase on guest entry, and subtracts
it on guest exit. Which is fine, except that it is possible for
userspace to change the offset using the SET_ONE_REG interface while
the vcore is running, as there is only one timebase offset per vcore
but potentially multiple VCPUs in the vcore. If that were to happen,
KVM would subtract a different offset on guest exit from that which
it had added on guest entry, leading to the timebase being out of sync
between cores in the host, which then leads to bad things happening
such as hangs and spurious watchdog timeouts.
To fix this, we add a new field 'tb_offset_applied' to the vcore struct
which stores the offset that is currently applied to the timebase.
This value is set from the vcore tb_offset field on guest entry, and
is what is subtracted from the timebase on guest exit. Since it is
zero when the timebase offset is not applied, we can simplify the
logic in kvmhv_start_timing and kvmhv_accumulate_time.
In addition, we had secondary threads reading the timebase while
running concurrently with code on the primary thread which would
eventually add or subtract the timebase offset from the timebase.
This occurred while saving or restoring the DEC register value on
the secondary threads. Although no specific incorrect behaviour has
been observed, this is a race which should be fixed. To fix it, we
move the DEC saving code to just before we call kvmhv_commence_exit,
and the DEC restoring code to after the point where we have waited
for the primary thread to switch the MMU context and add the timebase
offset. That way we are sure that the timebase contains the guest
timebase value in both cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-04-20 20:51:11 +08:00
|
|
|
|
|
|
|
|
/* Save DEC */
|
|
|
|
|
/* Do this before kvmhv_commence_exit so we know TB is guest TB */
|
|
|
|
|
ld r3, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
mfspr r5,SPRN_DEC
|
|
|
|
|
mftb r6
|
|
|
|
|
/* On P9, if the guest has large decr enabled, don't sign extend */
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r4, VCORE_LPCR(r3)
|
|
|
|
|
andis. r4, r4, LPCR_LD@h
|
|
|
|
|
bne 16f
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
|
|
|
|
extsw r5,r5
|
|
|
|
|
16: add r5,r5,r6
|
|
|
|
|
/* r5 is a guest timebase value here, convert to host TB */
|
|
|
|
|
ld r4,VCORE_TB_OFFSET_APPL(r3)
|
|
|
|
|
subf r5,r4,r5
|
|
|
|
|
std r5,VCPU_DEC_EXPIRES(r9)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
/* Increment exit count, poke other threads to exit */
|
KVM: PPC: Book3S HV: Snapshot timebase offset on guest entry
Currently, the HV KVM guest entry/exit code adds the timebase offset
from the vcore struct to the timebase on guest entry, and subtracts
it on guest exit. Which is fine, except that it is possible for
userspace to change the offset using the SET_ONE_REG interface while
the vcore is running, as there is only one timebase offset per vcore
but potentially multiple VCPUs in the vcore. If that were to happen,
KVM would subtract a different offset on guest exit from that which
it had added on guest entry, leading to the timebase being out of sync
between cores in the host, which then leads to bad things happening
such as hangs and spurious watchdog timeouts.
To fix this, we add a new field 'tb_offset_applied' to the vcore struct
which stores the offset that is currently applied to the timebase.
This value is set from the vcore tb_offset field on guest entry, and
is what is subtracted from the timebase on guest exit. Since it is
zero when the timebase offset is not applied, we can simplify the
logic in kvmhv_start_timing and kvmhv_accumulate_time.
In addition, we had secondary threads reading the timebase while
running concurrently with code on the primary thread which would
eventually add or subtract the timebase offset from the timebase.
This occurred while saving or restoring the DEC register value on
the secondary threads. Although no specific incorrect behaviour has
been observed, this is a race which should be fixed. To fix it, we
move the DEC saving code to just before we call kvmhv_commence_exit,
and the DEC restoring code to after the point where we have waited
for the primary thread to switch the MMU context and add the timebase
offset. That way we are sure that the timebase contains the guest
timebase value in both cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-04-20 20:51:11 +08:00
|
|
|
mr r3, r12
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
bl kvmhv_commence_exit
|
2015-03-28 11:21:11 +08:00
|
|
|
nop
|
|
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Make use of unused threads when running guests
When running a virtual core of a guest that is configured with fewer
threads per core than the physical cores have, the extra physical
threads are currently unused. This makes it possible to use them to
run one or more other virtual cores from the same guest when certain
conditions are met. This applies on POWER7, and on POWER8 to guests
with one thread per virtual core. (It doesn't apply to POWER8 guests
with multiple threads per vcore because they require a 1-1 virtual to
physical thread mapping in order to be able to use msgsndp and the
TIR.)
The idea is that we maintain a list of preempted vcores for each
physical cpu (i.e. each core, since the host runs single-threaded).
Then, when a vcore is about to run, it checks to see if there are
any vcores on the list for its physical cpu that could be
piggybacked onto this vcore's execution. If so, those additional
vcores are put into state VCORE_PIGGYBACK and their runnable VCPU
threads are started as well as the original vcore, which is called
the master vcore.
After the vcores have exited the guest, the extra ones are put back
onto the preempted list if any of their VCPUs are still runnable and
not idle.
This means that vcpu->arch.ptid is no longer necessarily the same as
the physical thread that the vcpu runs on. In order to make it easier
for code that wants to send an IPI to know which CPU to target, we
now store that in a new field in struct vcpu_arch, called thread_cpu.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Tested-by: Laurent Vivier <lvivier@redhat.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-06-24 19:18:03 +08:00
|
|
|
/* Stop others sending VCPU interrupts to this physical CPU */
|
|
|
|
|
li r0, -1
|
|
|
|
|
stw r0, VCPU_CPU(r9)
|
|
|
|
|
stw r0, VCPU_THREAD_CPU(r9)
|
|
|
|
|
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
/* Save guest CTRL register, set runlatch to 1 */
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
mfspr r6,SPRN_CTRLF
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
stw r6,VCPU_CTRL(r9)
|
|
|
|
|
andi. r0,r6,1
|
|
|
|
|
bne 4f
|
|
|
|
|
ori r6,r6,1
|
|
|
|
|
mtspr SPRN_CTRLT,r6
|
|
|
|
|
4:
|
|
|
|
|
/*
|
|
|
|
|
* Save the guest PURR/SPURR
|
|
|
|
|
*/
|
|
|
|
|
mfspr r5,SPRN_PURR
|
|
|
|
|
mfspr r6,SPRN_SPURR
|
|
|
|
|
ld r7,VCPU_PURR(r9)
|
|
|
|
|
ld r8,VCPU_SPURR(r9)
|
|
|
|
|
std r5,VCPU_PURR(r9)
|
|
|
|
|
std r6,VCPU_SPURR(r9)
|
|
|
|
|
subf r5,r7,r5
|
|
|
|
|
subf r6,r8,r6
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Restore host PURR/SPURR and add guest times
|
|
|
|
|
* so that the time in the guest gets accounted.
|
|
|
|
|
*/
|
|
|
|
|
ld r3,HSTATE_PURR(r13)
|
|
|
|
|
ld r4,HSTATE_SPURR(r13)
|
|
|
|
|
add r3,r3,r5
|
|
|
|
|
add r4,r4,r6
|
|
|
|
|
mtspr SPRN_PURR,r3
|
|
|
|
|
mtspr SPRN_SPURR,r4
|
|
|
|
|
|
2014-01-08 18:25:21 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
b 8f
|
|
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_207S)
|
|
|
|
|
/* Save POWER8-specific registers */
|
|
|
|
|
mfspr r5, SPRN_IAMR
|
|
|
|
|
mfspr r6, SPRN_PSPB
|
|
|
|
|
mfspr r7, SPRN_FSCR
|
|
|
|
|
std r5, VCPU_IAMR(r9)
|
|
|
|
|
stw r6, VCPU_PSPB(r9)
|
|
|
|
|
std r7, VCPU_FSCR(r9)
|
|
|
|
|
mfspr r5, SPRN_IC
|
|
|
|
|
mfspr r7, SPRN_TAR
|
|
|
|
|
std r5, VCPU_IC(r9)
|
|
|
|
|
std r7, VCPU_TAR(r9)
|
2014-01-08 18:25:32 +08:00
|
|
|
mfspr r8, SPRN_EBBHR
|
2014-01-08 18:25:21 +08:00
|
|
|
std r8, VCPU_EBBHR(r9)
|
|
|
|
|
mfspr r5, SPRN_EBBRR
|
|
|
|
|
mfspr r6, SPRN_BESCR
|
2016-11-16 19:33:27 +08:00
|
|
|
mfspr r7, SPRN_PID
|
|
|
|
|
mfspr r8, SPRN_WORT
|
2014-01-08 18:25:21 +08:00
|
|
|
std r5, VCPU_EBBRR(r9)
|
|
|
|
|
std r6, VCPU_BESCR(r9)
|
2016-11-16 19:33:27 +08:00
|
|
|
stw r7, VCPU_GUEST_PID(r9)
|
|
|
|
|
std r8, VCPU_WORT(r9)
|
|
|
|
|
BEGIN_FTR_SECTION
|
2014-01-08 18:25:21 +08:00
|
|
|
mfspr r5, SPRN_TCSCR
|
|
|
|
|
mfspr r6, SPRN_ACOP
|
2016-11-16 19:33:27 +08:00
|
|
|
mfspr r7, SPRN_CSIGR
|
|
|
|
|
mfspr r8, SPRN_TACR
|
2014-01-08 18:25:21 +08:00
|
|
|
std r5, VCPU_TCSCR(r9)
|
|
|
|
|
std r6, VCPU_ACOP(r9)
|
2016-11-16 19:33:27 +08:00
|
|
|
std r7, VCPU_CSIGR(r9)
|
|
|
|
|
std r8, VCPU_TACR(r9)
|
2016-11-18 10:11:42 +08:00
|
|
|
FTR_SECTION_ELSE
|
|
|
|
|
mfspr r5, SPRN_TIDR
|
|
|
|
|
mfspr r6, SPRN_PSSCR
|
|
|
|
|
std r5, VCPU_TID(r9)
|
|
|
|
|
rldicl r6, r6, 4, 50 /* r6 &= PSSCR_GUEST_VIS */
|
|
|
|
|
rotldi r6, r6, 60
|
|
|
|
|
std r6, VCPU_PSSCR(r9)
|
2017-02-15 11:30:17 +08:00
|
|
|
/* Restore host HFSCR value */
|
|
|
|
|
ld r7, STACK_SLOT_HFSCR(r1)
|
|
|
|
|
mtspr SPRN_HFSCR, r7
|
2016-11-18 10:11:42 +08:00
|
|
|
ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
|
2016-03-05 16:34:39 +08:00
|
|
|
/*
|
|
|
|
|
* Restore various registers to 0, where non-zero values
|
|
|
|
|
* set by the guest could disrupt the host.
|
|
|
|
|
*/
|
|
|
|
|
li r0, 0
|
2017-06-15 13:43:17 +08:00
|
|
|
mtspr SPRN_PSPB, r0
|
2016-03-05 16:34:39 +08:00
|
|
|
mtspr SPRN_WORT, r0
|
2016-11-16 19:33:27 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
mtspr SPRN_TCSCR, r0
|
2016-03-05 16:34:39 +08:00
|
|
|
/* Set MMCRS to 1<<31 to freeze and disable the SPMC counters */
|
|
|
|
|
li r0, 1
|
|
|
|
|
sldi r0, r0, 31
|
|
|
|
|
mtspr SPRN_MMCRS, r0
|
2016-11-16 19:33:27 +08:00
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
|
2014-01-08 18:25:21 +08:00
|
|
|
|
powerpc/kvm: Save and restore host AMR/IAMR/UAMOR
When the hash MMU is active the AMR, IAMR and UAMOR are used for
pkeys. The AMR is directly writable by user space, and the UAMOR masks
those writes, meaning both registers are effectively user register
state. The IAMR is used to create an execute only key.
Also we must maintain the value of at least the AMR when running in
process context, so that any memory accesses done by the kernel on
behalf of the process are correctly controlled by the AMR.
Although we are correctly switching all registers when going into a
guest, on returning to the host we just write 0 into all regs, except
on Power9 where we restore the IAMR correctly.
This could be observed by a user process if it writes the AMR, then
runs a guest and we then return immediately to it without
rescheduling. Because we have written 0 to the AMR that would have the
effect of granting read/write permission to pages that the process was
trying to protect.
In addition, when using the Radix MMU, the AMR can prevent inadvertent
kernel access to userspace data, writing 0 to the AMR disables that
protection.
So save and restore AMR, IAMR and UAMOR.
Fixes: cf43d3b26452 ("powerpc: Enable pkey subsystem")
Cc: stable@vger.kernel.org # v4.16+
Signed-off-by: Russell Currey <ruscur@russell.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Paul Mackerras <paulus@ozlabs.org>
2019-02-22 10:22:08 +08:00
|
|
|
/* Save and restore AMR, IAMR and UAMOR before turning on the MMU */
|
|
|
|
|
ld r8, STACK_SLOT_IAMR(r1)
|
|
|
|
|
mtspr SPRN_IAMR, r8
|
|
|
|
|
|
|
|
|
|
8: /* Power7 jumps back in here */
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
mfspr r5,SPRN_AMR
|
|
|
|
|
mfspr r6,SPRN_UAMOR
|
|
|
|
|
std r5,VCPU_AMR(r9)
|
|
|
|
|
std r6,VCPU_UAMOR(r9)
|
powerpc/kvm: Save and restore host AMR/IAMR/UAMOR
When the hash MMU is active the AMR, IAMR and UAMOR are used for
pkeys. The AMR is directly writable by user space, and the UAMOR masks
those writes, meaning both registers are effectively user register
state. The IAMR is used to create an execute only key.
Also we must maintain the value of at least the AMR when running in
process context, so that any memory accesses done by the kernel on
behalf of the process are correctly controlled by the AMR.
Although we are correctly switching all registers when going into a
guest, on returning to the host we just write 0 into all regs, except
on Power9 where we restore the IAMR correctly.
This could be observed by a user process if it writes the AMR, then
runs a guest and we then return immediately to it without
rescheduling. Because we have written 0 to the AMR that would have the
effect of granting read/write permission to pages that the process was
trying to protect.
In addition, when using the Radix MMU, the AMR can prevent inadvertent
kernel access to userspace data, writing 0 to the AMR disables that
protection.
So save and restore AMR, IAMR and UAMOR.
Fixes: cf43d3b26452 ("powerpc: Enable pkey subsystem")
Cc: stable@vger.kernel.org # v4.16+
Signed-off-by: Russell Currey <ruscur@russell.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Acked-by: Paul Mackerras <paulus@ozlabs.org>
2019-02-22 10:22:08 +08:00
|
|
|
ld r5,STACK_SLOT_AMR(r1)
|
|
|
|
|
ld r6,STACK_SLOT_UAMOR(r1)
|
|
|
|
|
mtspr SPRN_AMR, r5
|
2017-06-15 13:43:17 +08:00
|
|
|
mtspr SPRN_UAMOR, r6
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
|
|
|
|
|
/* Switch DSCR back to host value */
|
|
|
|
|
mfspr r8, SPRN_DSCR
|
|
|
|
|
ld r7, HSTATE_DSCR(r13)
|
|
|
|
|
std r8, VCPU_DSCR(r9)
|
|
|
|
|
mtspr SPRN_DSCR, r7
|
|
|
|
|
|
|
|
|
|
/* Save non-volatile GPRs */
|
|
|
|
|
std r14, VCPU_GPR(R14)(r9)
|
|
|
|
|
std r15, VCPU_GPR(R15)(r9)
|
|
|
|
|
std r16, VCPU_GPR(R16)(r9)
|
|
|
|
|
std r17, VCPU_GPR(R17)(r9)
|
|
|
|
|
std r18, VCPU_GPR(R18)(r9)
|
|
|
|
|
std r19, VCPU_GPR(R19)(r9)
|
|
|
|
|
std r20, VCPU_GPR(R20)(r9)
|
|
|
|
|
std r21, VCPU_GPR(R21)(r9)
|
|
|
|
|
std r22, VCPU_GPR(R22)(r9)
|
|
|
|
|
std r23, VCPU_GPR(R23)(r9)
|
|
|
|
|
std r24, VCPU_GPR(R24)(r9)
|
|
|
|
|
std r25, VCPU_GPR(R25)(r9)
|
|
|
|
|
std r26, VCPU_GPR(R26)(r9)
|
|
|
|
|
std r27, VCPU_GPR(R27)(r9)
|
|
|
|
|
std r28, VCPU_GPR(R28)(r9)
|
|
|
|
|
std r29, VCPU_GPR(R29)(r9)
|
|
|
|
|
std r30, VCPU_GPR(R30)(r9)
|
|
|
|
|
std r31, VCPU_GPR(R31)(r9)
|
|
|
|
|
|
|
|
|
|
/* Save SPRGs */
|
|
|
|
|
mfspr r3, SPRN_SPRG0
|
|
|
|
|
mfspr r4, SPRN_SPRG1
|
|
|
|
|
mfspr r5, SPRN_SPRG2
|
|
|
|
|
mfspr r6, SPRN_SPRG3
|
|
|
|
|
std r3, VCPU_SPRG0(r9)
|
|
|
|
|
std r4, VCPU_SPRG1(r9)
|
|
|
|
|
std r5, VCPU_SPRG2(r9)
|
|
|
|
|
std r6, VCPU_SPRG3(r9)
|
|
|
|
|
|
|
|
|
|
/* save FP state */
|
|
|
|
|
mr r3, r9
|
|
|
|
|
bl kvmppc_save_fp
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
|
2014-04-14 06:56:26 +08:00
|
|
|
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
/*
|
|
|
|
|
* Branch around the call if both CPU_FTR_TM and
|
|
|
|
|
* CPU_FTR_P9_TM_HV_ASSIST are off.
|
|
|
|
|
*/
|
2014-04-14 06:56:26 +08:00
|
|
|
BEGIN_FTR_SECTION
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
b 91f
|
|
|
|
|
END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
|
KVM: PPC: Book3S HV: Fix bug causing host SLB to be restored incorrectly
Aneesh Kumar reported seeing host crashes when running recent kernels
on POWER8. The symptom was an oops like this:
Unable to handle kernel paging request for data at address 0xf00000000786c620
Faulting instruction address: 0xc00000000030e1e4
Oops: Kernel access of bad area, sig: 11 [#1]
LE SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in: powernv_op_panel
CPU: 24 PID: 6663 Comm: qemu-system-ppc Tainted: G W 4.13.0-rc7-43932-gfc36c59 #2
task: c000000fdeadfe80 task.stack: c000000fdeb68000
NIP: c00000000030e1e4 LR: c00000000030de6c CTR: c000000000103620
REGS: c000000fdeb6b450 TRAP: 0300 Tainted: G W (4.13.0-rc7-43932-gfc36c59)
MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE> CR: 24044428 XER: 20000000
CFAR: c00000000030e134 DAR: f00000000786c620 DSISR: 40000000 SOFTE: 0
GPR00: 0000000000000000 c000000fdeb6b6d0 c0000000010bd000 000000000000e1b0
GPR04: c00000000115e168 c000001fffa6e4b0 c00000000115d000 c000001e1b180386
GPR08: f000000000000000 c000000f9a8913e0 f00000000786c600 00007fff587d0000
GPR12: c000000fdeb68000 c00000000fb0f000 0000000000000001 00007fff587cffff
GPR16: 0000000000000000 c000000000000000 00000000003fffff c000000fdebfe1f8
GPR20: 0000000000000004 c000000fdeb6b8a8 0000000000000001 0008000000000040
GPR24: 07000000000000c0 00007fff587cffff c000000fdec20bf8 00007fff587d0000
GPR28: c000000fdeca9ac0 00007fff587d0000 00007fff587c0000 00007fff587d0000
NIP [c00000000030e1e4] __get_user_pages_fast+0x434/0x1070
LR [c00000000030de6c] __get_user_pages_fast+0xbc/0x1070
Call Trace:
[c000000fdeb6b6d0] [c00000000139dab8] lock_classes+0x0/0x35fe50 (unreliable)
[c000000fdeb6b7e0] [c00000000030ef38] get_user_pages_fast+0xf8/0x120
[c000000fdeb6b830] [c000000000112318] kvmppc_book3s_hv_page_fault+0x308/0xf30
[c000000fdeb6b960] [c00000000010e10c] kvmppc_vcpu_run_hv+0xfdc/0x1f00
[c000000fdeb6bb20] [c0000000000e915c] kvmppc_vcpu_run+0x2c/0x40
[c000000fdeb6bb40] [c0000000000e5650] kvm_arch_vcpu_ioctl_run+0x110/0x300
[c000000fdeb6bbe0] [c0000000000d6468] kvm_vcpu_ioctl+0x528/0x900
[c000000fdeb6bd40] [c0000000003bc04c] do_vfs_ioctl+0xcc/0x950
[c000000fdeb6bde0] [c0000000003bc930] SyS_ioctl+0x60/0x100
[c000000fdeb6be30] [c00000000000b96c] system_call+0x58/0x6c
Instruction dump:
7ca81a14 2fa50000 41de0010 7cc8182a 68c60002 78c6ffe2 0b060000 3cc2000a
794a3664 390610d8 e9080000 7d485214 <e90a0020> 7d435378 790507e1 408202f0
---[ end trace fad4a342d0414aa2 ]---
It turns out that what has happened is that the SLB entry for the
vmmemap region hasn't been reloaded on exit from a guest, and it has
the wrong page size. Then, when the host next accesses the vmemmap
region, it gets a page fault.
Commit a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with
KVM", 2017-07-24) modified the guest exit code so that it now only clears
out the SLB for hash guest. The code tests the radix flag and puts the
result in a non-volatile CR field, CR2, and later branches based on CR2.
Unfortunately, the kvmppc_save_tm function, which gets called between
those two points, modifies all the user-visible registers in the case
where the guest was in transactional or suspended state, except for a
few which it restores (namely r1, r2, r9 and r13). Thus the hash/radix indication in CR2 gets corrupted.
This fixes the problem by re-doing the comparison just before the
result is needed. For good measure, this also adds comments next to
the call sites of kvmppc_save_tm and kvmppc_restore_tm pointing out
that non-volatile register state will be lost.
Cc: stable@vger.kernel.org # v4.13
Fixes: a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with KVM")
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-12 11:47:23 +08:00
|
|
|
/*
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
* NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
|
KVM: PPC: Book3S HV: Fix bug causing host SLB to be restored incorrectly
Aneesh Kumar reported seeing host crashes when running recent kernels
on POWER8. The symptom was an oops like this:
Unable to handle kernel paging request for data at address 0xf00000000786c620
Faulting instruction address: 0xc00000000030e1e4
Oops: Kernel access of bad area, sig: 11 [#1]
LE SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in: powernv_op_panel
CPU: 24 PID: 6663 Comm: qemu-system-ppc Tainted: G W 4.13.0-rc7-43932-gfc36c59 #2
task: c000000fdeadfe80 task.stack: c000000fdeb68000
NIP: c00000000030e1e4 LR: c00000000030de6c CTR: c000000000103620
REGS: c000000fdeb6b450 TRAP: 0300 Tainted: G W (4.13.0-rc7-43932-gfc36c59)
MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE> CR: 24044428 XER: 20000000
CFAR: c00000000030e134 DAR: f00000000786c620 DSISR: 40000000 SOFTE: 0
GPR00: 0000000000000000 c000000fdeb6b6d0 c0000000010bd000 000000000000e1b0
GPR04: c00000000115e168 c000001fffa6e4b0 c00000000115d000 c000001e1b180386
GPR08: f000000000000000 c000000f9a8913e0 f00000000786c600 00007fff587d0000
GPR12: c000000fdeb68000 c00000000fb0f000 0000000000000001 00007fff587cffff
GPR16: 0000000000000000 c000000000000000 00000000003fffff c000000fdebfe1f8
GPR20: 0000000000000004 c000000fdeb6b8a8 0000000000000001 0008000000000040
GPR24: 07000000000000c0 00007fff587cffff c000000fdec20bf8 00007fff587d0000
GPR28: c000000fdeca9ac0 00007fff587d0000 00007fff587c0000 00007fff587d0000
NIP [c00000000030e1e4] __get_user_pages_fast+0x434/0x1070
LR [c00000000030de6c] __get_user_pages_fast+0xbc/0x1070
Call Trace:
[c000000fdeb6b6d0] [c00000000139dab8] lock_classes+0x0/0x35fe50 (unreliable)
[c000000fdeb6b7e0] [c00000000030ef38] get_user_pages_fast+0xf8/0x120
[c000000fdeb6b830] [c000000000112318] kvmppc_book3s_hv_page_fault+0x308/0xf30
[c000000fdeb6b960] [c00000000010e10c] kvmppc_vcpu_run_hv+0xfdc/0x1f00
[c000000fdeb6bb20] [c0000000000e915c] kvmppc_vcpu_run+0x2c/0x40
[c000000fdeb6bb40] [c0000000000e5650] kvm_arch_vcpu_ioctl_run+0x110/0x300
[c000000fdeb6bbe0] [c0000000000d6468] kvm_vcpu_ioctl+0x528/0x900
[c000000fdeb6bd40] [c0000000003bc04c] do_vfs_ioctl+0xcc/0x950
[c000000fdeb6bde0] [c0000000003bc930] SyS_ioctl+0x60/0x100
[c000000fdeb6be30] [c00000000000b96c] system_call+0x58/0x6c
Instruction dump:
7ca81a14 2fa50000 41de0010 7cc8182a 68c60002 78c6ffe2 0b060000 3cc2000a
794a3664 390610d8 e9080000 7d485214 <e90a0020> 7d435378 790507e1 408202f0
---[ end trace fad4a342d0414aa2 ]---
It turns out that what has happened is that the SLB entry for the
vmmemap region hasn't been reloaded on exit from a guest, and it has
the wrong page size. Then, when the host next accesses the vmemmap
region, it gets a page fault.
Commit a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with
KVM", 2017-07-24) modified the guest exit code so that it now only clears
out the SLB for hash guest. The code tests the radix flag and puts the
result in a non-volatile CR field, CR2, and later branches based on CR2.
Unfortunately, the kvmppc_save_tm function, which gets called between
those two points, modifies all the user-visible registers in the case
where the guest was in transactional or suspended state, except for a
few which it restores (namely r1, r2, r9 and r13). Thus the hash/radix indication in CR2 gets corrupted.
This fixes the problem by re-doing the comparison just before the
result is needed. For good measure, this also adds comments next to
the call sites of kvmppc_save_tm and kvmppc_restore_tm pointing out
that non-volatile register state will be lost.
Cc: stable@vger.kernel.org # v4.13
Fixes: a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with KVM")
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-12 11:47:23 +08:00
|
|
|
*/
|
2018-05-23 15:01:48 +08:00
|
|
|
mr r3, r9
|
|
|
|
|
ld r4, VCPU_MSR(r3)
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
li r5, 0 /* don't preserve non-vol regs */
|
2018-05-30 18:07:52 +08:00
|
|
|
bl kvmppc_save_tm_hv
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
nop
|
2018-05-23 15:01:48 +08:00
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
91:
|
2014-04-14 06:56:26 +08:00
|
|
|
#endif
|
|
|
|
|
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
/* Increment yield count if they have a VPA */
|
|
|
|
|
ld r8, VCPU_VPA(r9) /* do they have a VPA? */
|
|
|
|
|
cmpdi r8, 0
|
|
|
|
|
beq 25f
|
2014-06-11 16:36:17 +08:00
|
|
|
li r4, LPPACA_YIELDCOUNT
|
|
|
|
|
LWZX_BE r3, r8, r4
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
addi r3, r3, 1
|
2014-06-11 16:36:17 +08:00
|
|
|
STWX_BE r3, r8, r4
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
li r3, 1
|
|
|
|
|
stb r3, VCPU_VPA_DIRTY(r9)
|
|
|
|
|
25:
|
|
|
|
|
/* Save PMU registers if requested */
|
|
|
|
|
/* r8 and cr0.eq are live here */
|
2018-10-08 13:30:51 +08:00
|
|
|
mr r3, r9
|
|
|
|
|
li r4, 1
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
beq 21f /* if no VPA, save PMU stuff anyway */
|
2018-10-08 13:30:51 +08:00
|
|
|
lbz r4, LPPACA_PMCINUSE(r8)
|
|
|
|
|
21: bl kvmhv_save_guest_pmu
|
|
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
|
2016-11-18 10:11:42 +08:00
|
|
|
/* Restore host values of some registers */
|
2017-06-16 09:53:19 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r5, STACK_SLOT_CIABR(r1)
|
|
|
|
|
ld r6, STACK_SLOT_DAWR(r1)
|
|
|
|
|
ld r7, STACK_SLOT_DAWRX(r1)
|
|
|
|
|
mtspr SPRN_CIABR, r5
|
2018-03-27 12:37:22 +08:00
|
|
|
/*
|
|
|
|
|
* If the DAWR doesn't work, it's ok to write these here as
|
|
|
|
|
* this value should always be zero
|
|
|
|
|
*/
|
2017-06-16 09:53:19 +08:00
|
|
|
mtspr SPRN_DAWR, r6
|
|
|
|
|
mtspr SPRN_DAWRX, r7
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
2016-11-18 10:11:42 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r5, STACK_SLOT_TID(r1)
|
|
|
|
|
ld r6, STACK_SLOT_PSSCR(r1)
|
2017-01-30 18:21:45 +08:00
|
|
|
ld r7, STACK_SLOT_PID(r1)
|
2016-11-18 10:11:42 +08:00
|
|
|
mtspr SPRN_TIDR, r5
|
|
|
|
|
mtspr SPRN_PSSCR, r6
|
2017-01-30 18:21:45 +08:00
|
|
|
mtspr SPRN_PID, r7
|
2016-11-18 10:11:42 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
powerpc/mm/radix: Workaround prefetch issue with KVM
There's a somewhat architectural issue with Radix MMU and KVM.
When coming out of a guest with AIL (Alternate Interrupt Location, ie,
MMU enabled), we start executing hypervisor code with the PID register
still containing whatever the guest has been using.
The problem is that the CPU can (and will) then start prefetching or
speculatively load from whatever host context has that same PID (if
any), thus bringing translations for that context into the TLB, which
Linux doesn't know about.
This can cause stale translations and subsequent crashes.
Fixing this in a way that is neither racy nor a huge performance
impact is difficult. We could just make the host invalidations always
use broadcast forms but that would hurt single threaded programs for
example.
We chose to fix it instead by partitioning the PID space between guest
and host. This is possible because today Linux only use 19 out of the
20 bits of PID space, so existing guests will work if we make the host
use the top half of the 20 bits space.
We additionally add support for a property to indicate to Linux the
size of the PID register which will be useful if we eventually have
processors with a larger PID space available.
There is still an issue with malicious guests purposefully setting the
PID register to a value in the hosts PID range. Hopefully future HW
can prevent that, but in the meantime, we handle it with a pair of
kludges:
- On the way out of a guest, before we clear the current VCPU in the
PACA, we check the PID and if it's outside of the permitted range
we flush the TLB for that PID.
- When context switching, if the mm is "new" on that CPU (the
corresponding bit was set for the first time in the mm cpumask), we
check if any sibling thread is in KVM (has a non-NULL VCPU pointer
in the PACA). If that is the case, we also flush the PID for that
CPU (core).
This second part is needed to handle the case where a process is
migrated (or starts a new pthread) on a sibling thread of the CPU
coming out of KVM, as there's a window where stale translations can
exist before we detect it and flush them out.
A future optimization could be added by keeping track of whether the
PID has ever been used and avoid doing that for completely fresh PIDs.
We could similarily mark PIDs that have been the subject of a global
invalidation as "fresh". But for now this will do.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
[mpe: Rework the asm to build with CONFIG_PPC_RADIX_MMU=n, drop
unneeded include of kvm_book3s_asm.h]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-24 12:26:06 +08:00
|
|
|
|
|
|
|
|
#ifdef CONFIG_PPC_RADIX_MMU
|
|
|
|
|
/*
|
|
|
|
|
* Are we running hash or radix ?
|
|
|
|
|
*/
|
KVM: PPC: Book3S HV: Fix bug causing host SLB to be restored incorrectly
Aneesh Kumar reported seeing host crashes when running recent kernels
on POWER8. The symptom was an oops like this:
Unable to handle kernel paging request for data at address 0xf00000000786c620
Faulting instruction address: 0xc00000000030e1e4
Oops: Kernel access of bad area, sig: 11 [#1]
LE SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in: powernv_op_panel
CPU: 24 PID: 6663 Comm: qemu-system-ppc Tainted: G W 4.13.0-rc7-43932-gfc36c59 #2
task: c000000fdeadfe80 task.stack: c000000fdeb68000
NIP: c00000000030e1e4 LR: c00000000030de6c CTR: c000000000103620
REGS: c000000fdeb6b450 TRAP: 0300 Tainted: G W (4.13.0-rc7-43932-gfc36c59)
MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE> CR: 24044428 XER: 20000000
CFAR: c00000000030e134 DAR: f00000000786c620 DSISR: 40000000 SOFTE: 0
GPR00: 0000000000000000 c000000fdeb6b6d0 c0000000010bd000 000000000000e1b0
GPR04: c00000000115e168 c000001fffa6e4b0 c00000000115d000 c000001e1b180386
GPR08: f000000000000000 c000000f9a8913e0 f00000000786c600 00007fff587d0000
GPR12: c000000fdeb68000 c00000000fb0f000 0000000000000001 00007fff587cffff
GPR16: 0000000000000000 c000000000000000 00000000003fffff c000000fdebfe1f8
GPR20: 0000000000000004 c000000fdeb6b8a8 0000000000000001 0008000000000040
GPR24: 07000000000000c0 00007fff587cffff c000000fdec20bf8 00007fff587d0000
GPR28: c000000fdeca9ac0 00007fff587d0000 00007fff587c0000 00007fff587d0000
NIP [c00000000030e1e4] __get_user_pages_fast+0x434/0x1070
LR [c00000000030de6c] __get_user_pages_fast+0xbc/0x1070
Call Trace:
[c000000fdeb6b6d0] [c00000000139dab8] lock_classes+0x0/0x35fe50 (unreliable)
[c000000fdeb6b7e0] [c00000000030ef38] get_user_pages_fast+0xf8/0x120
[c000000fdeb6b830] [c000000000112318] kvmppc_book3s_hv_page_fault+0x308/0xf30
[c000000fdeb6b960] [c00000000010e10c] kvmppc_vcpu_run_hv+0xfdc/0x1f00
[c000000fdeb6bb20] [c0000000000e915c] kvmppc_vcpu_run+0x2c/0x40
[c000000fdeb6bb40] [c0000000000e5650] kvm_arch_vcpu_ioctl_run+0x110/0x300
[c000000fdeb6bbe0] [c0000000000d6468] kvm_vcpu_ioctl+0x528/0x900
[c000000fdeb6bd40] [c0000000003bc04c] do_vfs_ioctl+0xcc/0x950
[c000000fdeb6bde0] [c0000000003bc930] SyS_ioctl+0x60/0x100
[c000000fdeb6be30] [c00000000000b96c] system_call+0x58/0x6c
Instruction dump:
7ca81a14 2fa50000 41de0010 7cc8182a 68c60002 78c6ffe2 0b060000 3cc2000a
794a3664 390610d8 e9080000 7d485214 <e90a0020> 7d435378 790507e1 408202f0
---[ end trace fad4a342d0414aa2 ]---
It turns out that what has happened is that the SLB entry for the
vmmemap region hasn't been reloaded on exit from a guest, and it has
the wrong page size. Then, when the host next accesses the vmemmap
region, it gets a page fault.
Commit a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with
KVM", 2017-07-24) modified the guest exit code so that it now only clears
out the SLB for hash guest. The code tests the radix flag and puts the
result in a non-volatile CR field, CR2, and later branches based on CR2.
Unfortunately, the kvmppc_save_tm function, which gets called between
those two points, modifies all the user-visible registers in the case
where the guest was in transactional or suspended state, except for a
few which it restores (namely r1, r2, r9 and r13). Thus the hash/radix indication in CR2 gets corrupted.
This fixes the problem by re-doing the comparison just before the
result is needed. For good measure, this also adds comments next to
the call sites of kvmppc_save_tm and kvmppc_restore_tm pointing out
that non-volatile register state will be lost.
Cc: stable@vger.kernel.org # v4.13
Fixes: a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with KVM")
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-12 11:47:23 +08:00
|
|
|
ld r5, VCPU_KVM(r9)
|
|
|
|
|
lbz r0, KVM_RADIX(r5)
|
|
|
|
|
cmpwi cr2, r0, 0
|
2018-07-05 16:47:00 +08:00
|
|
|
beq cr2, 2f
|
powerpc/mm/radix: Workaround prefetch issue with KVM
There's a somewhat architectural issue with Radix MMU and KVM.
When coming out of a guest with AIL (Alternate Interrupt Location, ie,
MMU enabled), we start executing hypervisor code with the PID register
still containing whatever the guest has been using.
The problem is that the CPU can (and will) then start prefetching or
speculatively load from whatever host context has that same PID (if
any), thus bringing translations for that context into the TLB, which
Linux doesn't know about.
This can cause stale translations and subsequent crashes.
Fixing this in a way that is neither racy nor a huge performance
impact is difficult. We could just make the host invalidations always
use broadcast forms but that would hurt single threaded programs for
example.
We chose to fix it instead by partitioning the PID space between guest
and host. This is possible because today Linux only use 19 out of the
20 bits of PID space, so existing guests will work if we make the host
use the top half of the 20 bits space.
We additionally add support for a property to indicate to Linux the
size of the PID register which will be useful if we eventually have
processors with a larger PID space available.
There is still an issue with malicious guests purposefully setting the
PID register to a value in the hosts PID range. Hopefully future HW
can prevent that, but in the meantime, we handle it with a pair of
kludges:
- On the way out of a guest, before we clear the current VCPU in the
PACA, we check the PID and if it's outside of the permitted range
we flush the TLB for that PID.
- When context switching, if the mm is "new" on that CPU (the
corresponding bit was set for the first time in the mm cpumask), we
check if any sibling thread is in KVM (has a non-NULL VCPU pointer
in the PACA). If that is the case, we also flush the PID for that
CPU (core).
This second part is needed to handle the case where a process is
migrated (or starts a new pthread) on a sibling thread of the CPU
coming out of KVM, as there's a window where stale translations can
exist before we detect it and flush them out.
A future optimization could be added by keeping track of whether the
PID has ever been used and avoid doing that for completely fresh PIDs.
We could similarily mark PIDs that have been the subject of a global
invalidation as "fresh". But for now this will do.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
[mpe: Rework the asm to build with CONFIG_PPC_RADIX_MMU=n, drop
unneeded include of kvm_book3s_asm.h]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-24 12:26:06 +08:00
|
|
|
|
2018-05-17 12:47:59 +08:00
|
|
|
/*
|
|
|
|
|
* Radix: do eieio; tlbsync; ptesync sequence in case we
|
|
|
|
|
* interrupted the guest between a tlbie and a ptesync.
|
|
|
|
|
*/
|
|
|
|
|
eieio
|
|
|
|
|
tlbsync
|
|
|
|
|
ptesync
|
|
|
|
|
|
powerpc/mm/radix: Workaround prefetch issue with KVM
There's a somewhat architectural issue with Radix MMU and KVM.
When coming out of a guest with AIL (Alternate Interrupt Location, ie,
MMU enabled), we start executing hypervisor code with the PID register
still containing whatever the guest has been using.
The problem is that the CPU can (and will) then start prefetching or
speculatively load from whatever host context has that same PID (if
any), thus bringing translations for that context into the TLB, which
Linux doesn't know about.
This can cause stale translations and subsequent crashes.
Fixing this in a way that is neither racy nor a huge performance
impact is difficult. We could just make the host invalidations always
use broadcast forms but that would hurt single threaded programs for
example.
We chose to fix it instead by partitioning the PID space between guest
and host. This is possible because today Linux only use 19 out of the
20 bits of PID space, so existing guests will work if we make the host
use the top half of the 20 bits space.
We additionally add support for a property to indicate to Linux the
size of the PID register which will be useful if we eventually have
processors with a larger PID space available.
There is still an issue with malicious guests purposefully setting the
PID register to a value in the hosts PID range. Hopefully future HW
can prevent that, but in the meantime, we handle it with a pair of
kludges:
- On the way out of a guest, before we clear the current VCPU in the
PACA, we check the PID and if it's outside of the permitted range
we flush the TLB for that PID.
- When context switching, if the mm is "new" on that CPU (the
corresponding bit was set for the first time in the mm cpumask), we
check if any sibling thread is in KVM (has a non-NULL VCPU pointer
in the PACA). If that is the case, we also flush the PID for that
CPU (core).
This second part is needed to handle the case where a process is
migrated (or starts a new pthread) on a sibling thread of the CPU
coming out of KVM, as there's a window where stale translations can
exist before we detect it and flush them out.
A future optimization could be added by keeping track of whether the
PID has ever been used and avoid doing that for completely fresh PIDs.
We could similarily mark PIDs that have been the subject of a global
invalidation as "fresh". But for now this will do.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
[mpe: Rework the asm to build with CONFIG_PPC_RADIX_MMU=n, drop
unneeded include of kvm_book3s_asm.h]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-24 12:26:06 +08:00
|
|
|
/* Radix: Handle the case where the guest used an illegal PID */
|
|
|
|
|
LOAD_REG_ADDR(r4, mmu_base_pid)
|
|
|
|
|
lwz r3, VCPU_GUEST_PID(r9)
|
|
|
|
|
lwz r5, 0(r4)
|
|
|
|
|
cmpw cr0,r3,r5
|
|
|
|
|
blt 2f
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Illegal PID, the HW might have prefetched and cached in the TLB
|
|
|
|
|
* some translations for the LPID 0 / guest PID combination which
|
|
|
|
|
* Linux doesn't know about, so we need to flush that PID out of
|
|
|
|
|
* the TLB. First we need to set LPIDR to 0 so tlbiel applies to
|
|
|
|
|
* the right context.
|
|
|
|
|
*/
|
|
|
|
|
li r0,0
|
|
|
|
|
mtspr SPRN_LPID,r0
|
|
|
|
|
isync
|
|
|
|
|
|
|
|
|
|
/* Then do a congruence class local flush */
|
|
|
|
|
ld r6,VCPU_KVM(r9)
|
|
|
|
|
lwz r0,KVM_TLB_SETS(r6)
|
|
|
|
|
mtctr r0
|
|
|
|
|
li r7,0x400 /* IS field = 0b01 */
|
|
|
|
|
ptesync
|
|
|
|
|
sldi r0,r3,32 /* RS has PID */
|
|
|
|
|
1: PPC_TLBIEL(7,0,2,1,1) /* RIC=2, PRS=1, R=1 */
|
|
|
|
|
addi r7,r7,0x1000
|
|
|
|
|
bdnz 1b
|
|
|
|
|
ptesync
|
|
|
|
|
|
2018-07-05 16:47:00 +08:00
|
|
|
2:
|
powerpc/mm/radix: Workaround prefetch issue with KVM
There's a somewhat architectural issue with Radix MMU and KVM.
When coming out of a guest with AIL (Alternate Interrupt Location, ie,
MMU enabled), we start executing hypervisor code with the PID register
still containing whatever the guest has been using.
The problem is that the CPU can (and will) then start prefetching or
speculatively load from whatever host context has that same PID (if
any), thus bringing translations for that context into the TLB, which
Linux doesn't know about.
This can cause stale translations and subsequent crashes.
Fixing this in a way that is neither racy nor a huge performance
impact is difficult. We could just make the host invalidations always
use broadcast forms but that would hurt single threaded programs for
example.
We chose to fix it instead by partitioning the PID space between guest
and host. This is possible because today Linux only use 19 out of the
20 bits of PID space, so existing guests will work if we make the host
use the top half of the 20 bits space.
We additionally add support for a property to indicate to Linux the
size of the PID register which will be useful if we eventually have
processors with a larger PID space available.
There is still an issue with malicious guests purposefully setting the
PID register to a value in the hosts PID range. Hopefully future HW
can prevent that, but in the meantime, we handle it with a pair of
kludges:
- On the way out of a guest, before we clear the current VCPU in the
PACA, we check the PID and if it's outside of the permitted range
we flush the TLB for that PID.
- When context switching, if the mm is "new" on that CPU (the
corresponding bit was set for the first time in the mm cpumask), we
check if any sibling thread is in KVM (has a non-NULL VCPU pointer
in the PACA). If that is the case, we also flush the PID for that
CPU (core).
This second part is needed to handle the case where a process is
migrated (or starts a new pthread) on a sibling thread of the CPU
coming out of KVM, as there's a window where stale translations can
exist before we detect it and flush them out.
A future optimization could be added by keeping track of whether the
PID has ever been used and avoid doing that for completely fresh PIDs.
We could similarily mark PIDs that have been the subject of a global
invalidation as "fresh". But for now this will do.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
[mpe: Rework the asm to build with CONFIG_PPC_RADIX_MMU=n, drop
unneeded include of kvm_book3s_asm.h]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-07-24 12:26:06 +08:00
|
|
|
#endif /* CONFIG_PPC_RADIX_MMU */
|
2016-11-18 10:11:42 +08:00
|
|
|
|
KVM: PPC: book3s_hv: Add support for PPC970-family processors
This adds support for running KVM guests in supervisor mode on those
PPC970 processors that have a usable hypervisor mode. Unfortunately,
Apple G5 machines have supervisor mode disabled (MSR[HV] is forced to
1), but the YDL PowerStation does have a usable hypervisor mode.
There are several differences between the PPC970 and POWER7 in how
guests are managed. These differences are accommodated using the
CPU_FTR_ARCH_201 (PPC970) and CPU_FTR_ARCH_206 (POWER7) CPU feature
bits. Notably, on PPC970:
* The LPCR, LPID or RMOR registers don't exist, and the functions of
those registers are provided by bits in HID4 and one bit in HID0.
* External interrupts can be directed to the hypervisor, but unlike
POWER7 they are masked by MSR[EE] in non-hypervisor modes and use
SRR0/1 not HSRR0/1.
* There is no virtual RMA (VRMA) mode; the guest must use an RMO
(real mode offset) area.
* The TLB entries are not tagged with the LPID, so it is necessary to
flush the whole TLB on partition switch. Furthermore, when switching
partitions we have to ensure that no other CPU is executing the tlbie
or tlbsync instructions in either the old or the new partition,
otherwise undefined behaviour can occur.
* The PMU has 8 counters (PMC registers) rather than 6.
* The DSCR, PURR, SPURR, AMR, AMOR, UAMOR registers don't exist.
* The SLB has 64 entries rather than 32.
* There is no mediated external interrupt facility, so if we switch to
a guest that has a virtual external interrupt pending but the guest
has MSR[EE] = 0, we have to arrange to have an interrupt pending for
it so that we can get control back once it re-enables interrupts. We
do that by sending ourselves an IPI with smp_send_reschedule after
hard-disabling interrupts.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:40:08 +08:00
|
|
|
/*
|
2014-12-03 10:30:38 +08:00
|
|
|
* POWER7/POWER8 guest -> host partition switch code.
|
KVM: PPC: book3s_hv: Add support for PPC970-family processors
This adds support for running KVM guests in supervisor mode on those
PPC970 processors that have a usable hypervisor mode. Unfortunately,
Apple G5 machines have supervisor mode disabled (MSR[HV] is forced to
1), but the YDL PowerStation does have a usable hypervisor mode.
There are several differences between the PPC970 and POWER7 in how
guests are managed. These differences are accommodated using the
CPU_FTR_ARCH_201 (PPC970) and CPU_FTR_ARCH_206 (POWER7) CPU feature
bits. Notably, on PPC970:
* The LPCR, LPID or RMOR registers don't exist, and the functions of
those registers are provided by bits in HID4 and one bit in HID0.
* External interrupts can be directed to the hypervisor, but unlike
POWER7 they are masked by MSR[EE] in non-hypervisor modes and use
SRR0/1 not HSRR0/1.
* There is no virtual RMA (VRMA) mode; the guest must use an RMO
(real mode offset) area.
* The TLB entries are not tagged with the LPID, so it is necessary to
flush the whole TLB on partition switch. Furthermore, when switching
partitions we have to ensure that no other CPU is executing the tlbie
or tlbsync instructions in either the old or the new partition,
otherwise undefined behaviour can occur.
* The PMU has 8 counters (PMC registers) rather than 6.
* The DSCR, PURR, SPURR, AMR, AMOR, UAMOR registers don't exist.
* The SLB has 64 entries rather than 32.
* There is no mediated external interrupt facility, so if we switch to
a guest that has a virtual external interrupt pending but the guest
has MSR[EE] = 0, we have to arrange to have an interrupt pending for
it so that we can get control back once it re-enables interrupts. We
do that by sending ourselves an IPI with smp_send_reschedule after
hard-disabling interrupts.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:40:08 +08:00
|
|
|
* We don't have to lock against tlbies but we do
|
|
|
|
|
* have to coordinate the hardware threads.
|
2018-03-07 19:17:20 +08:00
|
|
|
* Here STACK_SLOT_TRAP(r1) contains the trap number.
|
KVM: PPC: book3s_hv: Add support for PPC970-family processors
This adds support for running KVM guests in supervisor mode on those
PPC970 processors that have a usable hypervisor mode. Unfortunately,
Apple G5 machines have supervisor mode disabled (MSR[HV] is forced to
1), but the YDL PowerStation does have a usable hypervisor mode.
There are several differences between the PPC970 and POWER7 in how
guests are managed. These differences are accommodated using the
CPU_FTR_ARCH_201 (PPC970) and CPU_FTR_ARCH_206 (POWER7) CPU feature
bits. Notably, on PPC970:
* The LPCR, LPID or RMOR registers don't exist, and the functions of
those registers are provided by bits in HID4 and one bit in HID0.
* External interrupts can be directed to the hypervisor, but unlike
POWER7 they are masked by MSR[EE] in non-hypervisor modes and use
SRR0/1 not HSRR0/1.
* There is no virtual RMA (VRMA) mode; the guest must use an RMO
(real mode offset) area.
* The TLB entries are not tagged with the LPID, so it is necessary to
flush the whole TLB on partition switch. Furthermore, when switching
partitions we have to ensure that no other CPU is executing the tlbie
or tlbsync instructions in either the old or the new partition,
otherwise undefined behaviour can occur.
* The PMU has 8 counters (PMC registers) rather than 6.
* The DSCR, PURR, SPURR, AMR, AMOR, UAMOR registers don't exist.
* The SLB has 64 entries rather than 32.
* There is no mediated external interrupt facility, so if we switch to
a guest that has a virtual external interrupt pending but the guest
has MSR[EE] = 0, we have to arrange to have an interrupt pending for
it so that we can get control back once it re-enables interrupts. We
do that by sending ourselves an IPI with smp_send_reschedule after
hard-disabling interrupts.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:40:08 +08:00
|
|
|
*/
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
kvmhv_switch_to_host:
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
/* Secondary threads wait for primary to do partition switch */
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
ld r5,HSTATE_KVM_VCORE(r13)
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
|
|
|
|
|
lbz r3,HSTATE_PTID(r13)
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
cmpwi r3,0
|
|
|
|
|
beq 15f
|
|
|
|
|
HMT_LOW
|
|
|
|
|
13: lbz r3,VCORE_IN_GUEST(r5)
|
|
|
|
|
cmpwi r3,0
|
|
|
|
|
bne 13b
|
|
|
|
|
HMT_MEDIUM
|
|
|
|
|
b 16f
|
|
|
|
|
|
|
|
|
|
/* Primary thread waits for all the secondaries to exit guest */
|
|
|
|
|
15: lwz r3,VCORE_ENTRY_EXIT(r5)
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
rlwinm r0,r3,32-8,0xff
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
clrldi r3,r3,56
|
|
|
|
|
cmpw r3,r0
|
|
|
|
|
bne 15b
|
|
|
|
|
isync
|
|
|
|
|
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
/* Did we actually switch to the guest at all? */
|
|
|
|
|
lbz r6, VCORE_IN_GUEST(r5)
|
|
|
|
|
cmpwi r6, 0
|
|
|
|
|
beq 19f
|
|
|
|
|
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
/* Primary thread switches back to host partition */
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
lwz r7,KVM_HOST_LPID(r4)
|
2016-11-16 19:25:20 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r6,KVM_HOST_SDR1(r4)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
li r8,LPID_RSVD /* switch to reserved LPID */
|
|
|
|
|
mtspr SPRN_LPID,r8
|
|
|
|
|
ptesync
|
2016-11-16 19:25:20 +08:00
|
|
|
mtspr SPRN_SDR1,r6 /* switch to host page table */
|
|
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
mtspr SPRN_LPID,r7
|
|
|
|
|
isync
|
KVM: PPC: Book3S HV: Implement timebase offset for guests
This allows guests to have a different timebase origin from the host.
This is needed for migration, where a guest can migrate from one host
to another and the two hosts might have a different timebase origin.
However, the timebase seen by the guest must not go backwards, and
should go forwards only by a small amount corresponding to the time
taken for the migration.
Therefore this provides a new per-vcpu value accessed via the one_reg
interface using the new KVM_REG_PPC_TB_OFFSET identifier. This value
defaults to 0 and is not modified by KVM. On entering the guest, this
value is added onto the timebase, and on exiting the guest, it is
subtracted from the timebase.
This is only supported for recent POWER hardware which has the TBU40
(timebase upper 40 bits) register. Writing to the TBU40 register only
alters the upper 40 bits of the timebase, leaving the lower 24 bits
unchanged. This provides a way to modify the timebase for guest
migration without disturbing the synchronization of the timebase
registers across CPU cores. The kernel rounds up the value given
to a multiple of 2^24.
Timebase values stored in KVM structures (struct kvm_vcpu, struct
kvmppc_vcore, etc.) are stored as host timebase values. The timebase
values in the dispatch trace log need to be guest timebase values,
however, since that is read directly by the guest. This moves the
setting of vcpu->arch.dec_expires on guest exit to a point after we
have restored the host timebase so that vcpu->arch.dec_expires is a
host timebase value.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-09-06 11:17:46 +08:00
|
|
|
|
2014-01-08 18:25:21 +08:00
|
|
|
BEGIN_FTR_SECTION
|
2016-09-15 11:42:52 +08:00
|
|
|
/* DPDES and VTB are shared between threads */
|
2014-01-08 18:25:21 +08:00
|
|
|
mfspr r7, SPRN_DPDES
|
2016-09-15 11:42:52 +08:00
|
|
|
mfspr r8, SPRN_VTB
|
2014-01-08 18:25:21 +08:00
|
|
|
std r7, VCORE_DPDES(r5)
|
2016-09-15 11:42:52 +08:00
|
|
|
std r8, VCORE_VTB(r5)
|
2014-01-08 18:25:21 +08:00
|
|
|
/* clear DPDES so we don't get guest doorbells in the host */
|
|
|
|
|
li r8, 0
|
|
|
|
|
mtspr SPRN_DPDES, r8
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Implement timebase offset for guests
This allows guests to have a different timebase origin from the host.
This is needed for migration, where a guest can migrate from one host
to another and the two hosts might have a different timebase origin.
However, the timebase seen by the guest must not go backwards, and
should go forwards only by a small amount corresponding to the time
taken for the migration.
Therefore this provides a new per-vcpu value accessed via the one_reg
interface using the new KVM_REG_PPC_TB_OFFSET identifier. This value
defaults to 0 and is not modified by KVM. On entering the guest, this
value is added onto the timebase, and on exiting the guest, it is
subtracted from the timebase.
This is only supported for recent POWER hardware which has the TBU40
(timebase upper 40 bits) register. Writing to the TBU40 register only
alters the upper 40 bits of the timebase, leaving the lower 24 bits
unchanged. This provides a way to modify the timebase for guest
migration without disturbing the synchronization of the timebase
registers across CPU cores. The kernel rounds up the value given
to a multiple of 2^24.
Timebase values stored in KVM structures (struct kvm_vcpu, struct
kvmppc_vcore, etc.) are stored as host timebase values. The timebase
values in the dispatch trace log need to be guest timebase values,
however, since that is read directly by the guest. This moves the
setting of vcpu->arch.dec_expires on guest exit to a point after we
have restored the host timebase so that vcpu->arch.dec_expires is a
host timebase value.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-09-06 11:17:46 +08:00
|
|
|
/* Subtract timebase offset from timebase */
|
KVM: PPC: Book3S HV: Snapshot timebase offset on guest entry
Currently, the HV KVM guest entry/exit code adds the timebase offset
from the vcore struct to the timebase on guest entry, and subtracts
it on guest exit. Which is fine, except that it is possible for
userspace to change the offset using the SET_ONE_REG interface while
the vcore is running, as there is only one timebase offset per vcore
but potentially multiple VCPUs in the vcore. If that were to happen,
KVM would subtract a different offset on guest exit from that which
it had added on guest entry, leading to the timebase being out of sync
between cores in the host, which then leads to bad things happening
such as hangs and spurious watchdog timeouts.
To fix this, we add a new field 'tb_offset_applied' to the vcore struct
which stores the offset that is currently applied to the timebase.
This value is set from the vcore tb_offset field on guest entry, and
is what is subtracted from the timebase on guest exit. Since it is
zero when the timebase offset is not applied, we can simplify the
logic in kvmhv_start_timing and kvmhv_accumulate_time.
In addition, we had secondary threads reading the timebase while
running concurrently with code on the primary thread which would
eventually add or subtract the timebase offset from the timebase.
This occurred while saving or restoring the DEC register value on
the secondary threads. Although no specific incorrect behaviour has
been observed, this is a race which should be fixed. To fix it, we
move the DEC saving code to just before we call kvmhv_commence_exit,
and the DEC restoring code to after the point where we have waited
for the primary thread to switch the MMU context and add the timebase
offset. That way we are sure that the timebase contains the guest
timebase value in both cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-04-20 20:51:11 +08:00
|
|
|
ld r8, VCORE_TB_OFFSET_APPL(r5)
|
KVM: PPC: Book3S HV: Implement timebase offset for guests
This allows guests to have a different timebase origin from the host.
This is needed for migration, where a guest can migrate from one host
to another and the two hosts might have a different timebase origin.
However, the timebase seen by the guest must not go backwards, and
should go forwards only by a small amount corresponding to the time
taken for the migration.
Therefore this provides a new per-vcpu value accessed via the one_reg
interface using the new KVM_REG_PPC_TB_OFFSET identifier. This value
defaults to 0 and is not modified by KVM. On entering the guest, this
value is added onto the timebase, and on exiting the guest, it is
subtracted from the timebase.
This is only supported for recent POWER hardware which has the TBU40
(timebase upper 40 bits) register. Writing to the TBU40 register only
alters the upper 40 bits of the timebase, leaving the lower 24 bits
unchanged. This provides a way to modify the timebase for guest
migration without disturbing the synchronization of the timebase
registers across CPU cores. The kernel rounds up the value given
to a multiple of 2^24.
Timebase values stored in KVM structures (struct kvm_vcpu, struct
kvmppc_vcore, etc.) are stored as host timebase values. The timebase
values in the dispatch trace log need to be guest timebase values,
however, since that is read directly by the guest. This moves the
setting of vcpu->arch.dec_expires on guest exit to a point after we
have restored the host timebase so that vcpu->arch.dec_expires is a
host timebase value.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-09-06 11:17:46 +08:00
|
|
|
cmpdi r8,0
|
|
|
|
|
beq 17f
|
KVM: PPC: Book3S HV: Snapshot timebase offset on guest entry
Currently, the HV KVM guest entry/exit code adds the timebase offset
from the vcore struct to the timebase on guest entry, and subtracts
it on guest exit. Which is fine, except that it is possible for
userspace to change the offset using the SET_ONE_REG interface while
the vcore is running, as there is only one timebase offset per vcore
but potentially multiple VCPUs in the vcore. If that were to happen,
KVM would subtract a different offset on guest exit from that which
it had added on guest entry, leading to the timebase being out of sync
between cores in the host, which then leads to bad things happening
such as hangs and spurious watchdog timeouts.
To fix this, we add a new field 'tb_offset_applied' to the vcore struct
which stores the offset that is currently applied to the timebase.
This value is set from the vcore tb_offset field on guest entry, and
is what is subtracted from the timebase on guest exit. Since it is
zero when the timebase offset is not applied, we can simplify the
logic in kvmhv_start_timing and kvmhv_accumulate_time.
In addition, we had secondary threads reading the timebase while
running concurrently with code on the primary thread which would
eventually add or subtract the timebase offset from the timebase.
This occurred while saving or restoring the DEC register value on
the secondary threads. Although no specific incorrect behaviour has
been observed, this is a race which should be fixed. To fix it, we
move the DEC saving code to just before we call kvmhv_commence_exit,
and the DEC restoring code to after the point where we have waited
for the primary thread to switch the MMU context and add the timebase
offset. That way we are sure that the timebase contains the guest
timebase value in both cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-04-20 20:51:11 +08:00
|
|
|
li r0, 0
|
|
|
|
|
std r0, VCORE_TB_OFFSET_APPL(r5)
|
2014-03-25 07:47:07 +08:00
|
|
|
mftb r6 /* current guest timebase */
|
KVM: PPC: Book3S HV: Implement timebase offset for guests
This allows guests to have a different timebase origin from the host.
This is needed for migration, where a guest can migrate from one host
to another and the two hosts might have a different timebase origin.
However, the timebase seen by the guest must not go backwards, and
should go forwards only by a small amount corresponding to the time
taken for the migration.
Therefore this provides a new per-vcpu value accessed via the one_reg
interface using the new KVM_REG_PPC_TB_OFFSET identifier. This value
defaults to 0 and is not modified by KVM. On entering the guest, this
value is added onto the timebase, and on exiting the guest, it is
subtracted from the timebase.
This is only supported for recent POWER hardware which has the TBU40
(timebase upper 40 bits) register. Writing to the TBU40 register only
alters the upper 40 bits of the timebase, leaving the lower 24 bits
unchanged. This provides a way to modify the timebase for guest
migration without disturbing the synchronization of the timebase
registers across CPU cores. The kernel rounds up the value given
to a multiple of 2^24.
Timebase values stored in KVM structures (struct kvm_vcpu, struct
kvmppc_vcore, etc.) are stored as host timebase values. The timebase
values in the dispatch trace log need to be guest timebase values,
however, since that is read directly by the guest. This moves the
setting of vcpu->arch.dec_expires on guest exit to a point after we
have restored the host timebase so that vcpu->arch.dec_expires is a
host timebase value.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-09-06 11:17:46 +08:00
|
|
|
subf r8,r8,r6
|
|
|
|
|
mtspr SPRN_TBU40,r8 /* update upper 40 bits */
|
|
|
|
|
mftb r7 /* check if lower 24 bits overflowed */
|
|
|
|
|
clrldi r6,r6,40
|
|
|
|
|
clrldi r7,r7,40
|
|
|
|
|
cmpld r7,r6
|
|
|
|
|
bge 17f
|
|
|
|
|
addis r8,r8,0x100 /* if so, increment upper 40 bits */
|
|
|
|
|
mtspr SPRN_TBU40,r8
|
|
|
|
|
|
2018-10-08 13:30:52 +08:00
|
|
|
17:
|
|
|
|
|
/*
|
|
|
|
|
* If this is an HMI, we called kvmppc_realmode_hmi_handler
|
|
|
|
|
* above, which may or may not have already called
|
|
|
|
|
* kvmppc_subcore_exit_guest. Fortunately, all that
|
|
|
|
|
* kvmppc_subcore_exit_guest does is clear a flag, so calling
|
|
|
|
|
* it again here is benign even if kvmppc_realmode_hmi_handler
|
|
|
|
|
* has already called it.
|
|
|
|
|
*/
|
|
|
|
|
bl kvmppc_subcore_exit_guest
|
KVM: PPC: Book3S HV: Fix TB corruption in guest exit path on HMI interrupt
When a guest is assigned to a core it converts the host Timebase (TB)
into guest TB by adding guest timebase offset before entering into
guest. During guest exit it restores the guest TB to host TB. This means
under certain conditions (Guest migration) host TB and guest TB can differ.
When we get an HMI for TB related issues the opal HMI handler would
try fixing errors and restore the correct host TB value. With no guest
running, we don't have any issues. But with guest running on the core
we run into TB corruption issues.
If we get an HMI while in the guest, the current HMI handler invokes opal
hmi handler before forcing guest to exit. The guest exit path subtracts
the guest TB offset from the current TB value which may have already
been restored with host value by opal hmi handler. This leads to incorrect
host and guest TB values.
With split-core, things become more complex. With split-core, TB also gets
split and each subcore gets its own TB register. When a hmi handler fixes
a TB error and restores the TB value, it affects all the TB values of
sibling subcores on the same core. On TB errors all the thread in the core
gets HMI. With existing code, the individual threads call opal hmi handle
independently which can easily throw TB out of sync if we have guest
running on subcores. Hence we will need to co-ordinate with all the
threads before making opal hmi handler call followed by TB resync.
This patch introduces a sibling subcore state structure (shared by all
threads in the core) in paca which holds information about whether sibling
subcores are in Guest mode or host mode. An array in_guest[] of size
MAX_SUBCORE_PER_CORE=4 is used to maintain the state of each subcore.
The subcore id is used as index into in_guest[] array. Only primary
thread entering/exiting the guest is responsible to set/unset its
designated array element.
On TB error, we get HMI interrupt on every thread on the core. Upon HMI,
this patch will now force guest to vacate the core/subcore. Primary
thread from each subcore will then turn off its respective bit
from the above bitmap during the guest exit path just after the
guest->host partition switch is complete.
All other threads that have just exited the guest OR were already in host
will wait until all other subcores clears their respective bit.
Once all the subcores turn off their respective bit, all threads will
will make call to opal hmi handler.
It is not necessary that opal hmi handler would resync the TB value for
every HMI interrupts. It would do so only for the HMI caused due to
TB errors. For rest, it would not touch TB value. Hence to make things
simpler, primary thread would call TB resync explicitly once for each
core immediately after opal hmi handler instead of subtracting guest
offset from TB. TB resync call will restore the TB with host value.
Thus we can be sure about the TB state.
One of the primary threads exiting the guest will take up the
responsibility of calling TB resync. It will use one of the top bits
(bit 63) from subcore state flags bitmap to make the decision. The first
primary thread (among the subcores) that is able to set the bit will
have to call the TB resync. Rest all other threads will wait until TB
resync is complete. Once TB resync is complete all threads will then
proceed.
Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-05-15 12:14:26 +08:00
|
|
|
nop
|
|
|
|
|
30: ld r5,HSTATE_KVM_VCORE(r13)
|
|
|
|
|
ld r4,VCORE_KVM(r5) /* pointer to struct kvm */
|
|
|
|
|
|
2013-09-21 12:35:02 +08:00
|
|
|
/* Reset PCR */
|
KVM: PPC: Book3S HV: Fix TB corruption in guest exit path on HMI interrupt
When a guest is assigned to a core it converts the host Timebase (TB)
into guest TB by adding guest timebase offset before entering into
guest. During guest exit it restores the guest TB to host TB. This means
under certain conditions (Guest migration) host TB and guest TB can differ.
When we get an HMI for TB related issues the opal HMI handler would
try fixing errors and restore the correct host TB value. With no guest
running, we don't have any issues. But with guest running on the core
we run into TB corruption issues.
If we get an HMI while in the guest, the current HMI handler invokes opal
hmi handler before forcing guest to exit. The guest exit path subtracts
the guest TB offset from the current TB value which may have already
been restored with host value by opal hmi handler. This leads to incorrect
host and guest TB values.
With split-core, things become more complex. With split-core, TB also gets
split and each subcore gets its own TB register. When a hmi handler fixes
a TB error and restores the TB value, it affects all the TB values of
sibling subcores on the same core. On TB errors all the thread in the core
gets HMI. With existing code, the individual threads call opal hmi handle
independently which can easily throw TB out of sync if we have guest
running on subcores. Hence we will need to co-ordinate with all the
threads before making opal hmi handler call followed by TB resync.
This patch introduces a sibling subcore state structure (shared by all
threads in the core) in paca which holds information about whether sibling
subcores are in Guest mode or host mode. An array in_guest[] of size
MAX_SUBCORE_PER_CORE=4 is used to maintain the state of each subcore.
The subcore id is used as index into in_guest[] array. Only primary
thread entering/exiting the guest is responsible to set/unset its
designated array element.
On TB error, we get HMI interrupt on every thread on the core. Upon HMI,
this patch will now force guest to vacate the core/subcore. Primary
thread from each subcore will then turn off its respective bit
from the above bitmap during the guest exit path just after the
guest->host partition switch is complete.
All other threads that have just exited the guest OR were already in host
will wait until all other subcores clears their respective bit.
Once all the subcores turn off their respective bit, all threads will
will make call to opal hmi handler.
It is not necessary that opal hmi handler would resync the TB value for
every HMI interrupts. It would do so only for the HMI caused due to
TB errors. For rest, it would not touch TB value. Hence to make things
simpler, primary thread would call TB resync explicitly once for each
core immediately after opal hmi handler instead of subtracting guest
offset from TB. TB resync call will restore the TB with host value.
Thus we can be sure about the TB state.
One of the primary threads exiting the guest will take up the
responsibility of calling TB resync. It will use one of the top bits
(bit 63) from subcore state flags bitmap to make the decision. The first
primary thread (among the subcores) that is able to set the bit will
have to call the TB resync. Rest all other threads will wait until TB
resync is complete. Once TB resync is complete all threads will then
proceed.
Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-05-15 12:14:26 +08:00
|
|
|
ld r0, VCORE_PCR(r5)
|
2013-09-21 12:35:02 +08:00
|
|
|
cmpdi r0, 0
|
|
|
|
|
beq 18f
|
|
|
|
|
li r0, 0
|
|
|
|
|
mtspr SPRN_PCR, r0
|
|
|
|
|
18:
|
KVM: PPC: Book3S HV: Implement timebase offset for guests
This allows guests to have a different timebase origin from the host.
This is needed for migration, where a guest can migrate from one host
to another and the two hosts might have a different timebase origin.
However, the timebase seen by the guest must not go backwards, and
should go forwards only by a small amount corresponding to the time
taken for the migration.
Therefore this provides a new per-vcpu value accessed via the one_reg
interface using the new KVM_REG_PPC_TB_OFFSET identifier. This value
defaults to 0 and is not modified by KVM. On entering the guest, this
value is added onto the timebase, and on exiting the guest, it is
subtracted from the timebase.
This is only supported for recent POWER hardware which has the TBU40
(timebase upper 40 bits) register. Writing to the TBU40 register only
alters the upper 40 bits of the timebase, leaving the lower 24 bits
unchanged. This provides a way to modify the timebase for guest
migration without disturbing the synchronization of the timebase
registers across CPU cores. The kernel rounds up the value given
to a multiple of 2^24.
Timebase values stored in KVM structures (struct kvm_vcpu, struct
kvmppc_vcore, etc.) are stored as host timebase values. The timebase
values in the dispatch trace log need to be guest timebase values,
however, since that is read directly by the guest. This moves the
setting of vcpu->arch.dec_expires on guest exit to a point after we
have restored the host timebase so that vcpu->arch.dec_expires is a
host timebase value.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-09-06 11:17:46 +08:00
|
|
|
/* Signal secondary CPUs to continue */
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
stb r0,VCORE_IN_GUEST(r5)
|
KVM: PPC: Book3S HV: Implement dynamic micro-threading on POWER8
This builds on the ability to run more than one vcore on a physical
core by using the micro-threading (split-core) modes of the POWER8
chip. Previously, only vcores from the same VM could be run together,
and (on POWER8) only if they had just one thread per core. With the
ability to split the core on guest entry and unsplit it on guest exit,
we can run up to 8 vcpu threads from up to 4 different VMs, and we can
run multiple vcores with 2 or 4 vcpus per vcore.
Dynamic micro-threading is only available if the static configuration
of the cores is whole-core mode (unsplit), and only on POWER8.
To manage this, we introduce a new kvm_split_mode struct which is
shared across all of the subcores in the core, with a pointer in the
paca on each thread. In addition we extend the core_info struct to
have information on each subcore. When deciding whether to add a
vcore to the set already on the core, we now have two possibilities:
(a) piggyback the vcore onto an existing subcore, or (b) start a new
subcore.
Currently, when any vcpu needs to exit the guest and switch to host
virtual mode, we interrupt all the threads in all subcores and switch
the core back to whole-core mode. It may be possible in future to
allow some of the subcores to keep executing in the guest while
subcore 0 switches to the host, but that is not implemented in this
patch.
This adds a module parameter called dynamic_mt_modes which controls
which micro-threading (split-core) modes the code will consider, as a
bitmap. In other words, if it is 0, no micro-threading mode is
considered; if it is 2, only 2-way micro-threading is considered; if
it is 4, only 4-way, and if it is 6, both 2-way and 4-way
micro-threading mode will be considered. The default is 6.
With this, we now have secondary threads which are the primary thread
for their subcore and therefore need to do the MMU switch. These
threads will need to be started even if they have no vcpu to run, so
we use the vcore pointer in the PACA rather than the vcpu pointer to
trigger them.
It is now possible for thread 0 to find that an exit has been
requested before it gets to switch the subcore state to the guest. In
that case we haven't added the guest's timebase offset to the
timebase, so we need to be careful not to subtract the offset in the
guest exit path. In fact we just skip the whole path that switches
back to host context, since we haven't switched to the guest context.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-07-02 18:38:16 +08:00
|
|
|
19: lis r8,0x7fff /* MAX_INT@h */
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
mtspr SPRN_HDEC,r8
|
|
|
|
|
|
KVM: PPC: Book3S HV: Run HPT guests on POWER9 radix hosts
This patch removes the restriction that a radix host can only run
radix guests, allowing us to run HPT (hashed page table) guests as
well. This is useful because it provides a way to run old guest
kernels that know about POWER8 but not POWER9.
Unfortunately, POWER9 currently has a restriction that all threads
in a given code must either all be in HPT mode, or all in radix mode.
This means that when entering a HPT guest, we have to obtain control
of all 4 threads in the core and get them to switch their LPIDR and
LPCR registers, even if they are not going to run a guest. On guest
exit we also have to get all threads to switch LPIDR and LPCR back
to host values.
To make this feasible, we require that KVM not be in the "independent
threads" mode, and that the CPU cores be in single-threaded mode from
the host kernel's perspective (only thread 0 online; threads 1, 2 and
3 offline). That allows us to use the same code as on POWER8 for
obtaining control of the secondary threads.
To manage the LPCR/LPIDR changes required, we extend the kvm_split_info
struct to contain the information needed by the secondary threads.
All threads perform a barrier synchronization (where all threads wait
for every other thread to reach the synchronization point) on guest
entry, both before and after loading LPCR and LPIDR. On guest exit,
they all once again perform a barrier synchronization both before
and after loading host values into LPCR and LPIDR.
Finally, it is also currently necessary to flush the entire TLB every
time we enter a HPT guest on a radix host. We do this on thread 0
with a loop of tlbiel instructions.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-19 11:11:23 +08:00
|
|
|
16:
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
/* On POWER9 with HPT-on-radix we need to wait for all other threads */
|
|
|
|
|
ld r3, HSTATE_SPLIT_MODE(r13)
|
|
|
|
|
cmpdi r3, 0
|
|
|
|
|
beq 47f
|
|
|
|
|
lwz r8, KVM_SPLIT_DO_RESTORE(r3)
|
|
|
|
|
cmpwi r8, 0
|
|
|
|
|
beq 47f
|
|
|
|
|
bl kvmhv_p9_restore_lpcr
|
|
|
|
|
nop
|
|
|
|
|
b 48f
|
|
|
|
|
47:
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
|
|
|
|
ld r8,KVM_HOST_LPCR(r4)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
mtspr SPRN_LPCR,r8
|
|
|
|
|
isync
|
KVM: PPC: Book3S HV: Run HPT guests on POWER9 radix hosts
This patch removes the restriction that a radix host can only run
radix guests, allowing us to run HPT (hashed page table) guests as
well. This is useful because it provides a way to run old guest
kernels that know about POWER8 but not POWER9.
Unfortunately, POWER9 currently has a restriction that all threads
in a given code must either all be in HPT mode, or all in radix mode.
This means that when entering a HPT guest, we have to obtain control
of all 4 threads in the core and get them to switch their LPIDR and
LPCR registers, even if they are not going to run a guest. On guest
exit we also have to get all threads to switch LPIDR and LPCR back
to host values.
To make this feasible, we require that KVM not be in the "independent
threads" mode, and that the CPU cores be in single-threaded mode from
the host kernel's perspective (only thread 0 online; threads 1, 2 and
3 offline). That allows us to use the same code as on POWER8 for
obtaining control of the secondary threads.
To manage the LPCR/LPIDR changes required, we extend the kvm_split_info
struct to contain the information needed by the secondary threads.
All threads perform a barrier synchronization (where all threads wait
for every other thread to reach the synchronization point) on guest
entry, both before and after loading LPCR and LPIDR. On guest exit,
they all once again perform a barrier synchronization both before
and after loading host values into LPCR and LPIDR.
Finally, it is also currently necessary to flush the entire TLB every
time we enter a HPT guest on a radix host. We do this on thread 0
with a loop of tlbiel instructions.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-19 11:11:23 +08:00
|
|
|
48:
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
|
|
|
|
/* Finish timing, if we have a vcpu */
|
|
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
cmpdi r4, 0
|
|
|
|
|
li r3, 0
|
|
|
|
|
beq 2f
|
|
|
|
|
bl kvmhv_accumulate_time
|
|
|
|
|
2:
|
|
|
|
|
#endif
|
KVM: PPC: Book3S HV: Better handling of exceptions that happen in real mode
When an interrupt or exception happens in the guest that comes to the
host, the CPU goes to hypervisor real mode (MMU off) to handle the
exception but doesn't change the MMU context. After saving a few
registers, we then clear the "in guest" flag. If, for any reason,
we get an exception in the real-mode code, that then gets handled
by the normal kernel exception handlers, which turn the MMU on. This
is disastrous if the MMU is still set to the guest context, since we
end up executing instructions from random places in the guest kernel
with hypervisor privilege.
In order to catch this situation, we define a new value for the "in guest"
flag, KVM_GUEST_MODE_HOST_HV, to indicate that we are in hypervisor real
mode with guest MMU context. If the "in guest" flag is set to this value,
we branch off to an emergency handler. For the moment, this just does
a branch to self to stop the CPU from doing anything further.
While we're here, we define another new flag value to indicate that we
are in a HV guest, as distinct from a PR guest. This will be useful
when we have a kernel that can support both PR and HV guests concurrently.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-10-04 19:45:04 +08:00
|
|
|
/* Unset guest mode */
|
|
|
|
|
li r0, KVM_GUEST_MODE_NONE
|
|
|
|
|
stb r0, HSTATE_IN_GUEST(r13)
|
|
|
|
|
|
2018-03-07 19:17:20 +08:00
|
|
|
lwz r12, STACK_SLOT_TRAP(r1) /* return trap # in r12 */
|
2017-06-16 09:53:19 +08:00
|
|
|
ld r0, SFS+PPC_LR_STKOFF(r1)
|
|
|
|
|
addi r1, r1, SFS
|
2013-09-06 11:23:44 +08:00
|
|
|
mtlr r0
|
|
|
|
|
blr
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
|
2018-10-08 13:30:52 +08:00
|
|
|
kvmppc_guest_external:
|
|
|
|
|
/* External interrupt, first check for host_ipi. If this is
|
|
|
|
|
* set, we know the host wants us out so let's do it now
|
|
|
|
|
*/
|
|
|
|
|
bl kvmppc_read_intr
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Restore the active volatile registers after returning from
|
|
|
|
|
* a C function.
|
|
|
|
|
*/
|
|
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
li r12, BOOK3S_INTERRUPT_EXTERNAL
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* kvmppc_read_intr return codes:
|
|
|
|
|
*
|
|
|
|
|
* Exit to host (r3 > 0)
|
|
|
|
|
* 1 An interrupt is pending that needs to be handled by the host
|
|
|
|
|
* Exit guest and return to host by branching to guest_exit_cont
|
|
|
|
|
*
|
|
|
|
|
* 2 Passthrough that needs completion in the host
|
|
|
|
|
* Exit guest and return to host by branching to guest_exit_cont
|
|
|
|
|
* However, we also set r12 to BOOK3S_INTERRUPT_HV_RM_HARD
|
|
|
|
|
* to indicate to the host to complete handling the interrupt
|
|
|
|
|
*
|
|
|
|
|
* Before returning to guest, we check if any CPU is heading out
|
|
|
|
|
* to the host and if so, we head out also. If no CPUs are heading
|
|
|
|
|
* check return values <= 0.
|
|
|
|
|
*
|
|
|
|
|
* Return to guest (r3 <= 0)
|
|
|
|
|
* 0 No external interrupt is pending
|
|
|
|
|
* -1 A guest wakeup IPI (which has now been cleared)
|
|
|
|
|
* In either case, we return to guest to deliver any pending
|
|
|
|
|
* guest interrupts.
|
|
|
|
|
*
|
|
|
|
|
* -2 A PCI passthrough external interrupt was handled
|
|
|
|
|
* (interrupt was delivered directly to guest)
|
|
|
|
|
* Return to guest to deliver any pending guest interrupts.
|
|
|
|
|
*/
|
|
|
|
|
|
|
|
|
|
cmpdi r3, 1
|
|
|
|
|
ble 1f
|
|
|
|
|
|
|
|
|
|
/* Return code = 2 */
|
|
|
|
|
li r12, BOOK3S_INTERRUPT_HV_RM_HARD
|
|
|
|
|
stw r12, VCPU_TRAP(r9)
|
|
|
|
|
b guest_exit_cont
|
|
|
|
|
|
|
|
|
|
1: /* Return code <= 1 */
|
|
|
|
|
cmpdi r3, 0
|
|
|
|
|
bgt guest_exit_cont
|
|
|
|
|
|
|
|
|
|
/* Return code <= 0 */
|
|
|
|
|
maybe_reenter_guest:
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
lwz r0, VCORE_ENTRY_EXIT(r5)
|
|
|
|
|
cmpwi r0, 0x100
|
|
|
|
|
mr r4, r9
|
|
|
|
|
blt deliver_guest_interrupt
|
|
|
|
|
b guest_exit_cont
|
|
|
|
|
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
|
|
|
|
|
/*
|
|
|
|
|
* Softpatch interrupt for transactional memory emulation cases
|
|
|
|
|
* on POWER9 DD2.2. This is early in the guest exit path - we
|
|
|
|
|
* haven't saved registers or done a treclaim yet.
|
|
|
|
|
*/
|
|
|
|
|
kvmppc_tm_emul:
|
|
|
|
|
/* Save instruction image in HEIR */
|
|
|
|
|
mfspr r3, SPRN_HEIR
|
|
|
|
|
stw r3, VCPU_HEIR(r9)
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The cases we want to handle here are those where the guest
|
|
|
|
|
* is in real suspend mode and is trying to transition to
|
|
|
|
|
* transactional mode.
|
|
|
|
|
*/
|
|
|
|
|
lbz r0, HSTATE_FAKE_SUSPEND(r13)
|
|
|
|
|
cmpwi r0, 0 /* keep exiting guest if in fake suspend */
|
|
|
|
|
bne guest_exit_cont
|
|
|
|
|
rldicl r3, r11, 64 - MSR_TS_S_LG, 62
|
|
|
|
|
cmpwi r3, 1 /* or if not in suspend state */
|
|
|
|
|
bne guest_exit_cont
|
|
|
|
|
|
|
|
|
|
/* Call C code to do the emulation */
|
|
|
|
|
mr r3, r9
|
|
|
|
|
bl kvmhv_p9_tm_emulation_early
|
|
|
|
|
nop
|
|
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
li r12, BOOK3S_INTERRUPT_HV_SOFTPATCH
|
|
|
|
|
cmpwi r3, 0
|
|
|
|
|
beq guest_exit_cont /* continue exiting if not handled */
|
|
|
|
|
ld r10, VCPU_PC(r9)
|
|
|
|
|
ld r11, VCPU_MSR(r9)
|
|
|
|
|
b fast_interrupt_c_return /* go back to guest if handled */
|
|
|
|
|
#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
|
|
|
|
|
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
/*
|
|
|
|
|
* Check whether an HDSI is an HPTE not found fault or something else.
|
|
|
|
|
* If it is an HPTE not found fault that is due to the guest accessing
|
|
|
|
|
* a page that they have mapped but which we have paged out, then
|
|
|
|
|
* we continue on with the guest exit path. In all other cases,
|
|
|
|
|
* reflect the HDSI to the guest as a DSI.
|
|
|
|
|
*/
|
|
|
|
|
kvmppc_hdsi:
|
2017-01-30 18:21:45 +08:00
|
|
|
ld r3, VCPU_KVM(r9)
|
|
|
|
|
lbz r0, KVM_RADIX(r3)
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
mfspr r4, SPRN_HDAR
|
|
|
|
|
mfspr r6, SPRN_HDSISR
|
2017-09-15 13:26:14 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
/* Look for DSISR canary. If we find it, retry instruction */
|
|
|
|
|
cmpdi r6, 0x7fff
|
|
|
|
|
beq 6f
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
|
|
|
|
cmpwi r0, 0
|
2017-01-30 18:21:45 +08:00
|
|
|
bne .Lradix_hdsi /* on radix, just save DAR/DSISR/ASDR */
|
2011-12-12 20:38:51 +08:00
|
|
|
/* HPTE not found fault or protection fault? */
|
|
|
|
|
andis. r0, r6, (DSISR_NOHPTE | DSISR_PROTFAULT)@h
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
beq 1f /* if not, send it to the guest */
|
2017-02-28 08:05:47 +08:00
|
|
|
andi. r0, r11, MSR_DR /* data relocation enabled? */
|
|
|
|
|
beq 3f
|
2017-01-30 18:21:43 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
mfspr r5, SPRN_ASDR /* on POWER9, use ASDR to get VSID */
|
|
|
|
|
b 4f
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
clrrdi r0, r4, 28
|
2012-06-25 21:33:10 +08:00
|
|
|
PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
|
2015-10-27 13:10:20 +08:00
|
|
|
li r0, BOOK3S_INTERRUPT_DATA_SEGMENT
|
|
|
|
|
bne 7f /* if no SLB entry found */
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
4: std r4, VCPU_FAULT_DAR(r9)
|
|
|
|
|
stw r6, VCPU_FAULT_DSISR(r9)
|
|
|
|
|
|
|
|
|
|
/* Search the hash table. */
|
|
|
|
|
mr r3, r9 /* vcpu pointer */
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
li r7, 1 /* data fault */
|
2014-02-04 13:04:35 +08:00
|
|
|
bl kvmppc_hpte_hv_fault
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
ld r10, VCPU_PC(r9)
|
|
|
|
|
ld r11, VCPU_MSR(r9)
|
|
|
|
|
li r12, BOOK3S_INTERRUPT_H_DATA_STORAGE
|
|
|
|
|
cmpdi r3, 0 /* retry the instruction */
|
|
|
|
|
beq 6f
|
|
|
|
|
cmpdi r3, -1 /* handle in kernel mode */
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
beq guest_exit_cont
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
cmpdi r3, -2 /* MMIO emulation; need instr word */
|
|
|
|
|
beq 2f
|
|
|
|
|
|
2015-10-27 13:10:20 +08:00
|
|
|
/* Synthesize a DSI (or DSegI) for the guest */
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
ld r4, VCPU_FAULT_DAR(r9)
|
|
|
|
|
mr r6, r3
|
2015-10-27 13:10:20 +08:00
|
|
|
1: li r0, BOOK3S_INTERRUPT_DATA_STORAGE
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
mtspr SPRN_DSISR, r6
|
2015-10-27 13:10:20 +08:00
|
|
|
7: mtspr SPRN_DAR, r4
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
mtspr SPRN_SRR0, r10
|
|
|
|
|
mtspr SPRN_SRR1, r11
|
2015-10-27 13:10:20 +08:00
|
|
|
mr r10, r0
|
2014-03-25 07:47:02 +08:00
|
|
|
bl kvmppc_msr_interrupt
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
fast_interrupt_c_return:
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
6: ld r7, VCPU_CTR(r9)
|
2015-05-27 07:56:57 +08:00
|
|
|
ld r8, VCPU_XER(r9)
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
mtctr r7
|
|
|
|
|
mtxer r8
|
|
|
|
|
mr r4, r9
|
|
|
|
|
b fast_guest_return
|
|
|
|
|
|
|
|
|
|
3: ld r5, VCPU_KVM(r9) /* not relocated, use VRMA */
|
|
|
|
|
ld r5, KVM_VRMA_SLB_V(r5)
|
|
|
|
|
b 4b
|
|
|
|
|
|
|
|
|
|
/* If this is for emulated MMIO, load the instruction word */
|
|
|
|
|
2: li r8, KVM_INST_FETCH_FAILED /* In case lwz faults */
|
|
|
|
|
|
|
|
|
|
/* Set guest mode to 'jump over instruction' so if lwz faults
|
|
|
|
|
* we'll just continue at the next IP. */
|
|
|
|
|
li r0, KVM_GUEST_MODE_SKIP
|
|
|
|
|
stb r0, HSTATE_IN_GUEST(r13)
|
|
|
|
|
|
|
|
|
|
/* Do the access with MSR:DR enabled */
|
|
|
|
|
mfmsr r3
|
|
|
|
|
ori r4, r3, MSR_DR /* Enable paging for data */
|
|
|
|
|
mtmsrd r4
|
|
|
|
|
lwz r8, 0(r10)
|
|
|
|
|
mtmsrd r3
|
|
|
|
|
|
|
|
|
|
/* Store the result */
|
|
|
|
|
stw r8, VCPU_LAST_INST(r9)
|
|
|
|
|
|
|
|
|
|
/* Unset guest mode. */
|
KVM: PPC: Book3S HV: Better handling of exceptions that happen in real mode
When an interrupt or exception happens in the guest that comes to the
host, the CPU goes to hypervisor real mode (MMU off) to handle the
exception but doesn't change the MMU context. After saving a few
registers, we then clear the "in guest" flag. If, for any reason,
we get an exception in the real-mode code, that then gets handled
by the normal kernel exception handlers, which turn the MMU on. This
is disastrous if the MMU is still set to the guest context, since we
end up executing instructions from random places in the guest kernel
with hypervisor privilege.
In order to catch this situation, we define a new value for the "in guest"
flag, KVM_GUEST_MODE_HOST_HV, to indicate that we are in hypervisor real
mode with guest MMU context. If the "in guest" flag is set to this value,
we branch off to an emergency handler. For the moment, this just does
a branch to self to stop the CPU from doing anything further.
While we're here, we define another new flag value to indicate that we
are in a HV guest, as distinct from a PR guest. This will be useful
when we have a kernel that can support both PR and HV guests concurrently.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-10-04 19:45:04 +08:00
|
|
|
li r0, KVM_GUEST_MODE_HOST_HV
|
KVM: PPC: Implement MMIO emulation support for Book3S HV guests
This provides the low-level support for MMIO emulation in Book3S HV
guests. When the guest tries to map a page which is not covered by
any memslot, that page is taken to be an MMIO emulation page. Instead
of inserting a valid HPTE, we insert an HPTE that has the valid bit
clear but another hypervisor software-use bit set, which we call
HPTE_V_ABSENT, to indicate that this is an absent page. An
absent page is treated much like a valid page as far as guest hcalls
(H_ENTER, H_REMOVE, H_READ etc.) are concerned, except of course that
an absent HPTE doesn't need to be invalidated with tlbie since it
was never valid as far as the hardware is concerned.
When the guest accesses a page for which there is an absent HPTE, it
will take a hypervisor data storage interrupt (HDSI) since we now set
the VPM1 bit in the LPCR. Our HDSI handler for HPTE-not-present faults
looks up the hash table and if it finds an absent HPTE mapping the
requested virtual address, will switch to kernel mode and handle the
fault in kvmppc_book3s_hv_page_fault(), which at present just calls
kvmppc_hv_emulate_mmio() to set up the MMIO emulation.
This is based on an earlier patch by Benjamin Herrenschmidt, but since
heavily reworked.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:36:37 +08:00
|
|
|
stb r0, HSTATE_IN_GUEST(r13)
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
b guest_exit_cont
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
|
2017-01-30 18:21:45 +08:00
|
|
|
.Lradix_hdsi:
|
|
|
|
|
std r4, VCPU_FAULT_DAR(r9)
|
|
|
|
|
stw r6, VCPU_FAULT_DSISR(r9)
|
|
|
|
|
.Lradix_hisi:
|
|
|
|
|
mfspr r5, SPRN_ASDR
|
|
|
|
|
std r5, VCPU_FAULT_GPA(r9)
|
|
|
|
|
b guest_exit_cont
|
|
|
|
|
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
/*
|
|
|
|
|
* Similarly for an HISI, reflect it to the guest as an ISI unless
|
|
|
|
|
* it is an HPTE not found fault for a page that we have paged out.
|
|
|
|
|
*/
|
|
|
|
|
kvmppc_hisi:
|
2017-01-30 18:21:45 +08:00
|
|
|
ld r3, VCPU_KVM(r9)
|
|
|
|
|
lbz r0, KVM_RADIX(r3)
|
|
|
|
|
cmpwi r0, 0
|
|
|
|
|
bne .Lradix_hisi /* for radix, just save ASDR */
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
andis. r0, r11, SRR1_ISI_NOPT@h
|
|
|
|
|
beq 1f
|
2017-02-28 08:05:47 +08:00
|
|
|
andi. r0, r11, MSR_IR /* instruction relocation enabled? */
|
|
|
|
|
beq 3f
|
2017-01-30 18:21:43 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
mfspr r5, SPRN_ASDR /* on POWER9, use ASDR to get VSID */
|
|
|
|
|
b 4f
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
clrrdi r0, r10, 28
|
2012-06-25 21:33:10 +08:00
|
|
|
PPC_SLBFEE_DOT(R5, R0) /* if so, look up SLB */
|
2015-10-27 13:10:20 +08:00
|
|
|
li r0, BOOK3S_INTERRUPT_INST_SEGMENT
|
|
|
|
|
bne 7f /* if no SLB entry found */
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
4:
|
|
|
|
|
/* Search the hash table. */
|
|
|
|
|
mr r3, r9 /* vcpu pointer */
|
|
|
|
|
mr r4, r10
|
|
|
|
|
mr r6, r11
|
|
|
|
|
li r7, 0 /* instruction fault */
|
2014-02-04 13:04:35 +08:00
|
|
|
bl kvmppc_hpte_hv_fault
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
ld r10, VCPU_PC(r9)
|
|
|
|
|
ld r11, VCPU_MSR(r9)
|
|
|
|
|
li r12, BOOK3S_INTERRUPT_H_INST_STORAGE
|
|
|
|
|
cmpdi r3, 0 /* retry the instruction */
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
beq fast_interrupt_c_return
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
cmpdi r3, -1 /* handle in kernel mode */
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
beq guest_exit_cont
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
|
2015-10-27 13:10:20 +08:00
|
|
|
/* Synthesize an ISI (or ISegI) for the guest */
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
mr r11, r3
|
2015-10-27 13:10:20 +08:00
|
|
|
1: li r0, BOOK3S_INTERRUPT_INST_STORAGE
|
|
|
|
|
7: mtspr SPRN_SRR0, r10
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
mtspr SPRN_SRR1, r11
|
2015-10-27 13:10:20 +08:00
|
|
|
mr r10, r0
|
2014-03-25 07:47:02 +08:00
|
|
|
bl kvmppc_msr_interrupt
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
b fast_interrupt_c_return
|
KVM: PPC: Implement MMU notifiers for Book3S HV guests
This adds the infrastructure to enable us to page out pages underneath
a Book3S HV guest, on processors that support virtualized partition
memory, that is, POWER7. Instead of pinning all the guest's pages,
we now look in the host userspace Linux page tables to find the
mapping for a given guest page. Then, if the userspace Linux PTE
gets invalidated, kvm_unmap_hva() gets called for that address, and
we replace all the guest HPTEs that refer to that page with absent
HPTEs, i.e. ones with the valid bit clear and the HPTE_V_ABSENT bit
set, which will cause an HDSI when the guest tries to access them.
Finally, the page fault handler is extended to reinstantiate the
guest HPTE when the guest tries to access a page which has been paged
out.
Since we can't intercept the guest DSI and ISI interrupts on PPC970,
we still have to pin all the guest pages on PPC970. We have a new flag,
kvm->arch.using_mmu_notifiers, that indicates whether we can page
guest pages out. If it is not set, the MMU notifier callbacks do
nothing and everything operates as before.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
2011-12-12 20:38:05 +08:00
|
|
|
|
|
|
|
|
3: ld r6, VCPU_KVM(r9) /* not relocated, use VRMA */
|
|
|
|
|
ld r5, KVM_VRMA_SLB_V(r6)
|
|
|
|
|
b 4b
|
|
|
|
|
|
2011-06-29 08:22:05 +08:00
|
|
|
/*
|
|
|
|
|
* Try to handle an hcall in real mode.
|
|
|
|
|
* Returns to the guest if we handle it, or continues on up to
|
|
|
|
|
* the kernel if we can't (i.e. if we don't have a handler for
|
|
|
|
|
* it, or if the handler returns H_TOO_HARD).
|
2015-03-28 11:21:04 +08:00
|
|
|
*
|
|
|
|
|
* r5 - r8 contain hcall args,
|
|
|
|
|
* r9 = vcpu, r10 = pc, r11 = msr, r12 = trap, r13 = paca
|
2011-06-29 08:22:05 +08:00
|
|
|
*/
|
|
|
|
|
hcall_try_real_mode:
|
2012-06-25 21:33:10 +08:00
|
|
|
ld r3,VCPU_GPR(R3)(r9)
|
2011-06-29 08:22:05 +08:00
|
|
|
andi. r0,r11,MSR_PR
|
2013-11-19 14:12:48 +08:00
|
|
|
/* sc 1 from userspace - reflect to guest syscall */
|
|
|
|
|
bne sc_1_fast_return
|
2018-10-08 13:31:04 +08:00
|
|
|
/* sc 1 from nested guest - give it to L1 to handle */
|
|
|
|
|
ld r0, VCPU_NESTED(r9)
|
|
|
|
|
cmpdi r0, 0
|
|
|
|
|
bne guest_exit_cont
|
2011-06-29 08:22:05 +08:00
|
|
|
clrrdi r3,r3,2
|
|
|
|
|
cmpldi r3,hcall_real_table_end - hcall_real_table
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
bge guest_exit_cont
|
2014-06-02 09:02:59 +08:00
|
|
|
/* See if this hcall is enabled for in-kernel handling */
|
|
|
|
|
ld r4, VCPU_KVM(r9)
|
|
|
|
|
srdi r0, r3, 8 /* r0 = (r3 / 4) >> 6 */
|
|
|
|
|
sldi r0, r0, 3 /* index into kvm->arch.enabled_hcalls[] */
|
|
|
|
|
add r4, r4, r0
|
|
|
|
|
ld r0, KVM_ENABLED_HCALLS(r4)
|
|
|
|
|
rlwinm r4, r3, 32-2, 0x3f /* r4 = (r3 / 4) & 0x3f */
|
|
|
|
|
srd r0, r0, r4
|
|
|
|
|
andi. r0, r0, 1
|
|
|
|
|
beq guest_exit_cont
|
|
|
|
|
/* Get pointer to handler, if any, and call it */
|
2011-06-29 08:22:05 +08:00
|
|
|
LOAD_REG_ADDR(r4, hcall_real_table)
|
2013-07-08 18:09:53 +08:00
|
|
|
lwax r3,r3,r4
|
2011-06-29 08:22:05 +08:00
|
|
|
cmpwi r3,0
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
beq guest_exit_cont
|
2014-06-12 16:16:10 +08:00
|
|
|
add r12,r3,r4
|
|
|
|
|
mtctr r12
|
2011-06-29 08:22:05 +08:00
|
|
|
mr r3,r9 /* get vcpu pointer */
|
2012-06-25 21:33:10 +08:00
|
|
|
ld r4,VCPU_GPR(R4)(r9)
|
2011-06-29 08:22:05 +08:00
|
|
|
bctrl
|
|
|
|
|
cmpdi r3,H_TOO_HARD
|
|
|
|
|
beq hcall_real_fallback
|
|
|
|
|
ld r4,HSTATE_KVM_VCPU(r13)
|
2012-06-25 21:33:10 +08:00
|
|
|
std r3,VCPU_GPR(R3)(r4)
|
2011-06-29 08:22:05 +08:00
|
|
|
ld r10,VCPU_PC(r4)
|
|
|
|
|
ld r11,VCPU_MSR(r4)
|
|
|
|
|
b fast_guest_return
|
|
|
|
|
|
2013-11-19 14:12:48 +08:00
|
|
|
sc_1_fast_return:
|
|
|
|
|
mtspr SPRN_SRR0,r10
|
|
|
|
|
mtspr SPRN_SRR1,r11
|
|
|
|
|
li r10, BOOK3S_INTERRUPT_SYSCALL
|
2014-03-25 07:47:02 +08:00
|
|
|
bl kvmppc_msr_interrupt
|
2013-11-19 14:12:48 +08:00
|
|
|
mr r4,r9
|
|
|
|
|
b fast_guest_return
|
|
|
|
|
|
2011-06-29 08:22:05 +08:00
|
|
|
/* We've attempted a real mode hcall, but it's punted it back
|
|
|
|
|
* to userspace. We need to restore some clobbered volatiles
|
|
|
|
|
* before resuming the pass-it-to-qemu path */
|
|
|
|
|
hcall_real_fallback:
|
|
|
|
|
li r12,BOOK3S_INTERRUPT_SYSCALL
|
|
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
b guest_exit_cont
|
2011-06-29 08:22:05 +08:00
|
|
|
|
|
|
|
|
.globl hcall_real_table
|
|
|
|
|
hcall_real_table:
|
|
|
|
|
.long 0 /* 0 - unused */
|
2014-02-04 13:07:01 +08:00
|
|
|
.long DOTSYM(kvmppc_h_remove) - hcall_real_table
|
|
|
|
|
.long DOTSYM(kvmppc_h_enter) - hcall_real_table
|
|
|
|
|
.long DOTSYM(kvmppc_h_read) - hcall_real_table
|
2015-06-24 19:18:07 +08:00
|
|
|
.long DOTSYM(kvmppc_h_clear_mod) - hcall_real_table
|
|
|
|
|
.long DOTSYM(kvmppc_h_clear_ref) - hcall_real_table
|
2014-02-04 13:07:01 +08:00
|
|
|
.long DOTSYM(kvmppc_h_protect) - hcall_real_table
|
2019-02-21 11:28:48 +08:00
|
|
|
#ifdef CONFIG_SPAPR_TCE_IOMMU
|
2014-02-04 13:07:01 +08:00
|
|
|
.long DOTSYM(kvmppc_h_get_tce) - hcall_real_table
|
2016-03-18 10:50:42 +08:00
|
|
|
.long DOTSYM(kvmppc_rm_h_put_tce) - hcall_real_table
|
2019-02-21 11:28:48 +08:00
|
|
|
#else
|
|
|
|
|
.long 0 /* 0x1c */
|
|
|
|
|
.long 0 /* 0x20 */
|
|
|
|
|
#endif
|
2011-06-29 08:22:05 +08:00
|
|
|
.long 0 /* 0x24 - H_SET_SPRG0 */
|
2014-02-04 13:07:01 +08:00
|
|
|
.long DOTSYM(kvmppc_h_set_dabr) - hcall_real_table
|
2019-03-22 14:05:45 +08:00
|
|
|
.long DOTSYM(kvmppc_rm_h_page_init) - hcall_real_table
|
2011-06-29 08:22:05 +08:00
|
|
|
.long 0 /* 0x30 */
|
|
|
|
|
.long 0 /* 0x34 */
|
|
|
|
|
.long 0 /* 0x38 */
|
|
|
|
|
.long 0 /* 0x3c */
|
|
|
|
|
.long 0 /* 0x40 */
|
|
|
|
|
.long 0 /* 0x44 */
|
|
|
|
|
.long 0 /* 0x48 */
|
|
|
|
|
.long 0 /* 0x4c */
|
|
|
|
|
.long 0 /* 0x50 */
|
|
|
|
|
.long 0 /* 0x54 */
|
|
|
|
|
.long 0 /* 0x58 */
|
|
|
|
|
.long 0 /* 0x5c */
|
|
|
|
|
.long 0 /* 0x60 */
|
2013-04-18 04:31:15 +08:00
|
|
|
#ifdef CONFIG_KVM_XICS
|
2014-02-04 13:07:01 +08:00
|
|
|
.long DOTSYM(kvmppc_rm_h_eoi) - hcall_real_table
|
|
|
|
|
.long DOTSYM(kvmppc_rm_h_cppr) - hcall_real_table
|
|
|
|
|
.long DOTSYM(kvmppc_rm_h_ipi) - hcall_real_table
|
2017-04-05 15:54:56 +08:00
|
|
|
.long DOTSYM(kvmppc_rm_h_ipoll) - hcall_real_table
|
2014-02-04 13:07:01 +08:00
|
|
|
.long DOTSYM(kvmppc_rm_h_xirr) - hcall_real_table
|
2013-04-18 04:31:15 +08:00
|
|
|
#else
|
|
|
|
|
.long 0 /* 0x64 - H_EOI */
|
|
|
|
|
.long 0 /* 0x68 - H_CPPR */
|
|
|
|
|
.long 0 /* 0x6c - H_IPI */
|
|
|
|
|
.long 0 /* 0x70 - H_IPOLL */
|
|
|
|
|
.long 0 /* 0x74 - H_XIRR */
|
|
|
|
|
#endif
|
2011-06-29 08:22:05 +08:00
|
|
|
.long 0 /* 0x78 */
|
|
|
|
|
.long 0 /* 0x7c */
|
|
|
|
|
.long 0 /* 0x80 */
|
|
|
|
|
.long 0 /* 0x84 */
|
|
|
|
|
.long 0 /* 0x88 */
|
|
|
|
|
.long 0 /* 0x8c */
|
|
|
|
|
.long 0 /* 0x90 */
|
|
|
|
|
.long 0 /* 0x94 */
|
|
|
|
|
.long 0 /* 0x98 */
|
|
|
|
|
.long 0 /* 0x9c */
|
|
|
|
|
.long 0 /* 0xa0 */
|
|
|
|
|
.long 0 /* 0xa4 */
|
|
|
|
|
.long 0 /* 0xa8 */
|
|
|
|
|
.long 0 /* 0xac */
|
|
|
|
|
.long 0 /* 0xb0 */
|
|
|
|
|
.long 0 /* 0xb4 */
|
|
|
|
|
.long 0 /* 0xb8 */
|
|
|
|
|
.long 0 /* 0xbc */
|
|
|
|
|
.long 0 /* 0xc0 */
|
|
|
|
|
.long 0 /* 0xc4 */
|
|
|
|
|
.long 0 /* 0xc8 */
|
|
|
|
|
.long 0 /* 0xcc */
|
|
|
|
|
.long 0 /* 0xd0 */
|
|
|
|
|
.long 0 /* 0xd4 */
|
|
|
|
|
.long 0 /* 0xd8 */
|
|
|
|
|
.long 0 /* 0xdc */
|
2014-02-04 13:07:01 +08:00
|
|
|
.long DOTSYM(kvmppc_h_cede) - hcall_real_table
|
2014-12-03 10:30:40 +08:00
|
|
|
.long DOTSYM(kvmppc_rm_h_confer) - hcall_real_table
|
2011-06-29 08:22:05 +08:00
|
|
|
.long 0 /* 0xe8 */
|
|
|
|
|
.long 0 /* 0xec */
|
|
|
|
|
.long 0 /* 0xf0 */
|
|
|
|
|
.long 0 /* 0xf4 */
|
|
|
|
|
.long 0 /* 0xf8 */
|
|
|
|
|
.long 0 /* 0xfc */
|
|
|
|
|
.long 0 /* 0x100 */
|
|
|
|
|
.long 0 /* 0x104 */
|
|
|
|
|
.long 0 /* 0x108 */
|
|
|
|
|
.long 0 /* 0x10c */
|
|
|
|
|
.long 0 /* 0x110 */
|
|
|
|
|
.long 0 /* 0x114 */
|
|
|
|
|
.long 0 /* 0x118 */
|
|
|
|
|
.long 0 /* 0x11c */
|
|
|
|
|
.long 0 /* 0x120 */
|
2014-02-04 13:07:01 +08:00
|
|
|
.long DOTSYM(kvmppc_h_bulk_remove) - hcall_real_table
|
KVM: PPC: Book3S HV: Add support for DABRX register on POWER7
The DABRX (DABR extension) register on POWER7 processors provides finer
control over which accesses cause a data breakpoint interrupt. It
contains 3 bits which indicate whether to enable accesses in user,
kernel and hypervisor modes respectively to cause data breakpoint
interrupts, plus one bit that enables both real mode and virtual mode
accesses to cause interrupts. Currently, KVM sets DABRX to allow
both kernel and user accesses to cause interrupts while in the guest.
This adds support for the guest to specify other values for DABRX.
PAPR defines a H_SET_XDABR hcall to allow the guest to set both DABR
and DABRX with one call. This adds a real-mode implementation of
H_SET_XDABR, which shares most of its code with the existing H_SET_DABR
implementation. To support this, we add a per-vcpu field to store the
DABRX value plus code to get and set it via the ONE_REG interface.
For Linux guests to use this new hcall, userspace needs to add
"hcall-xdabr" to the set of strings in the /chosen/hypertas-functions
property in the device tree. If userspace does this and then migrates
the guest to a host where the kernel doesn't include this patch, then
userspace will need to implement H_SET_XDABR by writing the specified
DABR value to the DABR using the ONE_REG interface. In that case, the
old kernel will set DABRX to DABRX_USER | DABRX_KERNEL. That should
still work correctly, at least for Linux guests, since Linux guests
cope with getting data breakpoint interrupts in modes that weren't
requested by just ignoring the interrupt, and Linux guests never set
DABRX_BTI.
The other thing this does is to make H_SET_DABR and H_SET_XDABR work
on POWER8, which has the DAWR and DAWRX instead of DABR/X. Guests that
know about POWER8 should use H_SET_MODE rather than H_SET_[X]DABR, but
guests running in POWER7 compatibility mode will still use H_SET_[X]DABR.
For them, this adds the logic to convert DABR/X values into DAWR/X values
on POWER8.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:29 +08:00
|
|
|
.long 0 /* 0x128 */
|
|
|
|
|
.long 0 /* 0x12c */
|
|
|
|
|
.long 0 /* 0x130 */
|
2014-02-04 13:07:01 +08:00
|
|
|
.long DOTSYM(kvmppc_h_set_xdabr) - hcall_real_table
|
2019-02-21 11:28:48 +08:00
|
|
|
#ifdef CONFIG_SPAPR_TCE_IOMMU
|
2016-03-18 10:50:42 +08:00
|
|
|
.long DOTSYM(kvmppc_rm_h_stuff_tce) - hcall_real_table
|
2016-02-15 09:55:09 +08:00
|
|
|
.long DOTSYM(kvmppc_rm_h_put_tce_indirect) - hcall_real_table
|
2019-02-21 11:28:48 +08:00
|
|
|
#else
|
|
|
|
|
.long 0 /* 0x138 */
|
|
|
|
|
.long 0 /* 0x13c */
|
|
|
|
|
#endif
|
2015-03-20 17:39:41 +08:00
|
|
|
.long 0 /* 0x140 */
|
|
|
|
|
.long 0 /* 0x144 */
|
|
|
|
|
.long 0 /* 0x148 */
|
|
|
|
|
.long 0 /* 0x14c */
|
|
|
|
|
.long 0 /* 0x150 */
|
|
|
|
|
.long 0 /* 0x154 */
|
|
|
|
|
.long 0 /* 0x158 */
|
|
|
|
|
.long 0 /* 0x15c */
|
|
|
|
|
.long 0 /* 0x160 */
|
|
|
|
|
.long 0 /* 0x164 */
|
|
|
|
|
.long 0 /* 0x168 */
|
|
|
|
|
.long 0 /* 0x16c */
|
|
|
|
|
.long 0 /* 0x170 */
|
|
|
|
|
.long 0 /* 0x174 */
|
|
|
|
|
.long 0 /* 0x178 */
|
|
|
|
|
.long 0 /* 0x17c */
|
|
|
|
|
.long 0 /* 0x180 */
|
|
|
|
|
.long 0 /* 0x184 */
|
|
|
|
|
.long 0 /* 0x188 */
|
|
|
|
|
.long 0 /* 0x18c */
|
|
|
|
|
.long 0 /* 0x190 */
|
|
|
|
|
.long 0 /* 0x194 */
|
|
|
|
|
.long 0 /* 0x198 */
|
|
|
|
|
.long 0 /* 0x19c */
|
|
|
|
|
.long 0 /* 0x1a0 */
|
|
|
|
|
.long 0 /* 0x1a4 */
|
|
|
|
|
.long 0 /* 0x1a8 */
|
|
|
|
|
.long 0 /* 0x1ac */
|
|
|
|
|
.long 0 /* 0x1b0 */
|
|
|
|
|
.long 0 /* 0x1b4 */
|
|
|
|
|
.long 0 /* 0x1b8 */
|
|
|
|
|
.long 0 /* 0x1bc */
|
|
|
|
|
.long 0 /* 0x1c0 */
|
|
|
|
|
.long 0 /* 0x1c4 */
|
|
|
|
|
.long 0 /* 0x1c8 */
|
|
|
|
|
.long 0 /* 0x1cc */
|
|
|
|
|
.long 0 /* 0x1d0 */
|
|
|
|
|
.long 0 /* 0x1d4 */
|
|
|
|
|
.long 0 /* 0x1d8 */
|
|
|
|
|
.long 0 /* 0x1dc */
|
|
|
|
|
.long 0 /* 0x1e0 */
|
|
|
|
|
.long 0 /* 0x1e4 */
|
|
|
|
|
.long 0 /* 0x1e8 */
|
|
|
|
|
.long 0 /* 0x1ec */
|
|
|
|
|
.long 0 /* 0x1f0 */
|
|
|
|
|
.long 0 /* 0x1f4 */
|
|
|
|
|
.long 0 /* 0x1f8 */
|
|
|
|
|
.long 0 /* 0x1fc */
|
|
|
|
|
.long 0 /* 0x200 */
|
|
|
|
|
.long 0 /* 0x204 */
|
|
|
|
|
.long 0 /* 0x208 */
|
|
|
|
|
.long 0 /* 0x20c */
|
|
|
|
|
.long 0 /* 0x210 */
|
|
|
|
|
.long 0 /* 0x214 */
|
|
|
|
|
.long 0 /* 0x218 */
|
|
|
|
|
.long 0 /* 0x21c */
|
|
|
|
|
.long 0 /* 0x220 */
|
|
|
|
|
.long 0 /* 0x224 */
|
|
|
|
|
.long 0 /* 0x228 */
|
|
|
|
|
.long 0 /* 0x22c */
|
|
|
|
|
.long 0 /* 0x230 */
|
|
|
|
|
.long 0 /* 0x234 */
|
|
|
|
|
.long 0 /* 0x238 */
|
|
|
|
|
.long 0 /* 0x23c */
|
|
|
|
|
.long 0 /* 0x240 */
|
|
|
|
|
.long 0 /* 0x244 */
|
|
|
|
|
.long 0 /* 0x248 */
|
|
|
|
|
.long 0 /* 0x24c */
|
|
|
|
|
.long 0 /* 0x250 */
|
|
|
|
|
.long 0 /* 0x254 */
|
|
|
|
|
.long 0 /* 0x258 */
|
|
|
|
|
.long 0 /* 0x25c */
|
|
|
|
|
.long 0 /* 0x260 */
|
|
|
|
|
.long 0 /* 0x264 */
|
|
|
|
|
.long 0 /* 0x268 */
|
|
|
|
|
.long 0 /* 0x26c */
|
|
|
|
|
.long 0 /* 0x270 */
|
|
|
|
|
.long 0 /* 0x274 */
|
|
|
|
|
.long 0 /* 0x278 */
|
|
|
|
|
.long 0 /* 0x27c */
|
|
|
|
|
.long 0 /* 0x280 */
|
|
|
|
|
.long 0 /* 0x284 */
|
|
|
|
|
.long 0 /* 0x288 */
|
|
|
|
|
.long 0 /* 0x28c */
|
|
|
|
|
.long 0 /* 0x290 */
|
|
|
|
|
.long 0 /* 0x294 */
|
|
|
|
|
.long 0 /* 0x298 */
|
|
|
|
|
.long 0 /* 0x29c */
|
|
|
|
|
.long 0 /* 0x2a0 */
|
|
|
|
|
.long 0 /* 0x2a4 */
|
|
|
|
|
.long 0 /* 0x2a8 */
|
|
|
|
|
.long 0 /* 0x2ac */
|
|
|
|
|
.long 0 /* 0x2b0 */
|
|
|
|
|
.long 0 /* 0x2b4 */
|
|
|
|
|
.long 0 /* 0x2b8 */
|
|
|
|
|
.long 0 /* 0x2bc */
|
|
|
|
|
.long 0 /* 0x2c0 */
|
|
|
|
|
.long 0 /* 0x2c4 */
|
|
|
|
|
.long 0 /* 0x2c8 */
|
|
|
|
|
.long 0 /* 0x2cc */
|
|
|
|
|
.long 0 /* 0x2d0 */
|
|
|
|
|
.long 0 /* 0x2d4 */
|
|
|
|
|
.long 0 /* 0x2d8 */
|
|
|
|
|
.long 0 /* 0x2dc */
|
|
|
|
|
.long 0 /* 0x2e0 */
|
|
|
|
|
.long 0 /* 0x2e4 */
|
|
|
|
|
.long 0 /* 0x2e8 */
|
|
|
|
|
.long 0 /* 0x2ec */
|
|
|
|
|
.long 0 /* 0x2f0 */
|
|
|
|
|
.long 0 /* 0x2f4 */
|
|
|
|
|
.long 0 /* 0x2f8 */
|
2017-04-05 15:54:56 +08:00
|
|
|
#ifdef CONFIG_KVM_XICS
|
|
|
|
|
.long DOTSYM(kvmppc_rm_h_xirr_x) - hcall_real_table
|
|
|
|
|
#else
|
|
|
|
|
.long 0 /* 0x2fc - H_XIRR_X*/
|
|
|
|
|
#endif
|
2015-03-20 17:39:41 +08:00
|
|
|
.long DOTSYM(kvmppc_h_random) - hcall_real_table
|
2014-06-02 09:03:00 +08:00
|
|
|
.globl hcall_real_table_end
|
2011-06-29 08:22:05 +08:00
|
|
|
hcall_real_table_end:
|
|
|
|
|
|
KVM: PPC: Book3S HV: Add support for DABRX register on POWER7
The DABRX (DABR extension) register on POWER7 processors provides finer
control over which accesses cause a data breakpoint interrupt. It
contains 3 bits which indicate whether to enable accesses in user,
kernel and hypervisor modes respectively to cause data breakpoint
interrupts, plus one bit that enables both real mode and virtual mode
accesses to cause interrupts. Currently, KVM sets DABRX to allow
both kernel and user accesses to cause interrupts while in the guest.
This adds support for the guest to specify other values for DABRX.
PAPR defines a H_SET_XDABR hcall to allow the guest to set both DABR
and DABRX with one call. This adds a real-mode implementation of
H_SET_XDABR, which shares most of its code with the existing H_SET_DABR
implementation. To support this, we add a per-vcpu field to store the
DABRX value plus code to get and set it via the ONE_REG interface.
For Linux guests to use this new hcall, userspace needs to add
"hcall-xdabr" to the set of strings in the /chosen/hypertas-functions
property in the device tree. If userspace does this and then migrates
the guest to a host where the kernel doesn't include this patch, then
userspace will need to implement H_SET_XDABR by writing the specified
DABR value to the DABR using the ONE_REG interface. In that case, the
old kernel will set DABRX to DABRX_USER | DABRX_KERNEL. That should
still work correctly, at least for Linux guests, since Linux guests
cope with getting data breakpoint interrupts in modes that weren't
requested by just ignoring the interrupt, and Linux guests never set
DABRX_BTI.
The other thing this does is to make H_SET_DABR and H_SET_XDABR work
on POWER8, which has the DAWR and DAWRX instead of DABR/X. Guests that
know about POWER8 should use H_SET_MODE rather than H_SET_[X]DABR, but
guests running in POWER7 compatibility mode will still use H_SET_[X]DABR.
For them, this adds the logic to convert DABR/X values into DAWR/X values
on POWER8.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:29 +08:00
|
|
|
_GLOBAL(kvmppc_h_set_xdabr)
|
2018-10-08 13:31:06 +08:00
|
|
|
EXPORT_SYMBOL_GPL(kvmppc_h_set_xdabr)
|
KVM: PPC: Book3S HV: Add support for DABRX register on POWER7
The DABRX (DABR extension) register on POWER7 processors provides finer
control over which accesses cause a data breakpoint interrupt. It
contains 3 bits which indicate whether to enable accesses in user,
kernel and hypervisor modes respectively to cause data breakpoint
interrupts, plus one bit that enables both real mode and virtual mode
accesses to cause interrupts. Currently, KVM sets DABRX to allow
both kernel and user accesses to cause interrupts while in the guest.
This adds support for the guest to specify other values for DABRX.
PAPR defines a H_SET_XDABR hcall to allow the guest to set both DABR
and DABRX with one call. This adds a real-mode implementation of
H_SET_XDABR, which shares most of its code with the existing H_SET_DABR
implementation. To support this, we add a per-vcpu field to store the
DABRX value plus code to get and set it via the ONE_REG interface.
For Linux guests to use this new hcall, userspace needs to add
"hcall-xdabr" to the set of strings in the /chosen/hypertas-functions
property in the device tree. If userspace does this and then migrates
the guest to a host where the kernel doesn't include this patch, then
userspace will need to implement H_SET_XDABR by writing the specified
DABR value to the DABR using the ONE_REG interface. In that case, the
old kernel will set DABRX to DABRX_USER | DABRX_KERNEL. That should
still work correctly, at least for Linux guests, since Linux guests
cope with getting data breakpoint interrupts in modes that weren't
requested by just ignoring the interrupt, and Linux guests never set
DABRX_BTI.
The other thing this does is to make H_SET_DABR and H_SET_XDABR work
on POWER8, which has the DAWR and DAWRX instead of DABR/X. Guests that
know about POWER8 should use H_SET_MODE rather than H_SET_[X]DABR, but
guests running in POWER7 compatibility mode will still use H_SET_[X]DABR.
For them, this adds the logic to convert DABR/X values into DAWR/X values
on POWER8.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:29 +08:00
|
|
|
andi. r0, r5, DABRX_USER | DABRX_KERNEL
|
|
|
|
|
beq 6f
|
|
|
|
|
li r0, DABRX_USER | DABRX_KERNEL | DABRX_BTI
|
|
|
|
|
andc. r0, r5, r0
|
|
|
|
|
beq 3f
|
|
|
|
|
6: li r3, H_PARAMETER
|
|
|
|
|
blr
|
|
|
|
|
|
2011-06-29 08:22:05 +08:00
|
|
|
_GLOBAL(kvmppc_h_set_dabr)
|
2018-10-08 13:31:06 +08:00
|
|
|
EXPORT_SYMBOL_GPL(kvmppc_h_set_dabr)
|
KVM: PPC: Book3S HV: Add support for DABRX register on POWER7
The DABRX (DABR extension) register on POWER7 processors provides finer
control over which accesses cause a data breakpoint interrupt. It
contains 3 bits which indicate whether to enable accesses in user,
kernel and hypervisor modes respectively to cause data breakpoint
interrupts, plus one bit that enables both real mode and virtual mode
accesses to cause interrupts. Currently, KVM sets DABRX to allow
both kernel and user accesses to cause interrupts while in the guest.
This adds support for the guest to specify other values for DABRX.
PAPR defines a H_SET_XDABR hcall to allow the guest to set both DABR
and DABRX with one call. This adds a real-mode implementation of
H_SET_XDABR, which shares most of its code with the existing H_SET_DABR
implementation. To support this, we add a per-vcpu field to store the
DABRX value plus code to get and set it via the ONE_REG interface.
For Linux guests to use this new hcall, userspace needs to add
"hcall-xdabr" to the set of strings in the /chosen/hypertas-functions
property in the device tree. If userspace does this and then migrates
the guest to a host where the kernel doesn't include this patch, then
userspace will need to implement H_SET_XDABR by writing the specified
DABR value to the DABR using the ONE_REG interface. In that case, the
old kernel will set DABRX to DABRX_USER | DABRX_KERNEL. That should
still work correctly, at least for Linux guests, since Linux guests
cope with getting data breakpoint interrupts in modes that weren't
requested by just ignoring the interrupt, and Linux guests never set
DABRX_BTI.
The other thing this does is to make H_SET_DABR and H_SET_XDABR work
on POWER8, which has the DAWR and DAWRX instead of DABR/X. Guests that
know about POWER8 should use H_SET_MODE rather than H_SET_[X]DABR, but
guests running in POWER7 compatibility mode will still use H_SET_[X]DABR.
For them, this adds the logic to convert DABR/X values into DAWR/X values
on POWER8.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:29 +08:00
|
|
|
li r5, DABRX_USER | DABRX_KERNEL
|
|
|
|
|
3:
|
2014-01-08 18:25:19 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
b 2f
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
2011-06-29 08:22:05 +08:00
|
|
|
std r4,VCPU_DABR(r3)
|
KVM: PPC: Book3S HV: Add support for DABRX register on POWER7
The DABRX (DABR extension) register on POWER7 processors provides finer
control over which accesses cause a data breakpoint interrupt. It
contains 3 bits which indicate whether to enable accesses in user,
kernel and hypervisor modes respectively to cause data breakpoint
interrupts, plus one bit that enables both real mode and virtual mode
accesses to cause interrupts. Currently, KVM sets DABRX to allow
both kernel and user accesses to cause interrupts while in the guest.
This adds support for the guest to specify other values for DABRX.
PAPR defines a H_SET_XDABR hcall to allow the guest to set both DABR
and DABRX with one call. This adds a real-mode implementation of
H_SET_XDABR, which shares most of its code with the existing H_SET_DABR
implementation. To support this, we add a per-vcpu field to store the
DABRX value plus code to get and set it via the ONE_REG interface.
For Linux guests to use this new hcall, userspace needs to add
"hcall-xdabr" to the set of strings in the /chosen/hypertas-functions
property in the device tree. If userspace does this and then migrates
the guest to a host where the kernel doesn't include this patch, then
userspace will need to implement H_SET_XDABR by writing the specified
DABR value to the DABR using the ONE_REG interface. In that case, the
old kernel will set DABRX to DABRX_USER | DABRX_KERNEL. That should
still work correctly, at least for Linux guests, since Linux guests
cope with getting data breakpoint interrupts in modes that weren't
requested by just ignoring the interrupt, and Linux guests never set
DABRX_BTI.
The other thing this does is to make H_SET_DABR and H_SET_XDABR work
on POWER8, which has the DAWR and DAWRX instead of DABR/X. Guests that
know about POWER8 should use H_SET_MODE rather than H_SET_[X]DABR, but
guests running in POWER7 compatibility mode will still use H_SET_[X]DABR.
For them, this adds the logic to convert DABR/X values into DAWR/X values
on POWER8.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:29 +08:00
|
|
|
stw r5, VCPU_DABRX(r3)
|
|
|
|
|
mtspr SPRN_DABRX, r5
|
2012-03-02 09:38:23 +08:00
|
|
|
/* Work around P7 bug where DABR can get corrupted on mtspr */
|
|
|
|
|
1: mtspr SPRN_DABR,r4
|
|
|
|
|
mfspr r5, SPRN_DABR
|
|
|
|
|
cmpd r4, r5
|
|
|
|
|
bne 1b
|
|
|
|
|
isync
|
2011-06-29 08:22:05 +08:00
|
|
|
li r3,0
|
|
|
|
|
blr
|
|
|
|
|
|
2018-03-27 12:37:21 +08:00
|
|
|
2:
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
/* POWER9 with disabled DAWR */
|
2018-03-30 19:57:24 +08:00
|
|
|
li r3, H_HARDWARE
|
2018-03-27 12:37:21 +08:00
|
|
|
blr
|
|
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_DAWR)
|
KVM: PPC: Book3S HV: Add support for DABRX register on POWER7
The DABRX (DABR extension) register on POWER7 processors provides finer
control over which accesses cause a data breakpoint interrupt. It
contains 3 bits which indicate whether to enable accesses in user,
kernel and hypervisor modes respectively to cause data breakpoint
interrupts, plus one bit that enables both real mode and virtual mode
accesses to cause interrupts. Currently, KVM sets DABRX to allow
both kernel and user accesses to cause interrupts while in the guest.
This adds support for the guest to specify other values for DABRX.
PAPR defines a H_SET_XDABR hcall to allow the guest to set both DABR
and DABRX with one call. This adds a real-mode implementation of
H_SET_XDABR, which shares most of its code with the existing H_SET_DABR
implementation. To support this, we add a per-vcpu field to store the
DABRX value plus code to get and set it via the ONE_REG interface.
For Linux guests to use this new hcall, userspace needs to add
"hcall-xdabr" to the set of strings in the /chosen/hypertas-functions
property in the device tree. If userspace does this and then migrates
the guest to a host where the kernel doesn't include this patch, then
userspace will need to implement H_SET_XDABR by writing the specified
DABR value to the DABR using the ONE_REG interface. In that case, the
old kernel will set DABRX to DABRX_USER | DABRX_KERNEL. That should
still work correctly, at least for Linux guests, since Linux guests
cope with getting data breakpoint interrupts in modes that weren't
requested by just ignoring the interrupt, and Linux guests never set
DABRX_BTI.
The other thing this does is to make H_SET_DABR and H_SET_XDABR work
on POWER8, which has the DAWR and DAWRX instead of DABR/X. Guests that
know about POWER8 should use H_SET_MODE rather than H_SET_[X]DABR, but
guests running in POWER7 compatibility mode will still use H_SET_[X]DABR.
For them, this adds the logic to convert DABR/X values into DAWR/X values
on POWER8.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:29 +08:00
|
|
|
/* Emulate H_SET_DABR/X on P8 for the sake of compat mode guests */
|
2018-03-27 12:37:21 +08:00
|
|
|
rlwimi r5, r4, 5, DAWRX_DR | DAWRX_DW
|
2015-11-20 16:11:45 +08:00
|
|
|
rlwimi r5, r4, 2, DAWRX_WT
|
KVM: PPC: Book3S HV: Add support for DABRX register on POWER7
The DABRX (DABR extension) register on POWER7 processors provides finer
control over which accesses cause a data breakpoint interrupt. It
contains 3 bits which indicate whether to enable accesses in user,
kernel and hypervisor modes respectively to cause data breakpoint
interrupts, plus one bit that enables both real mode and virtual mode
accesses to cause interrupts. Currently, KVM sets DABRX to allow
both kernel and user accesses to cause interrupts while in the guest.
This adds support for the guest to specify other values for DABRX.
PAPR defines a H_SET_XDABR hcall to allow the guest to set both DABR
and DABRX with one call. This adds a real-mode implementation of
H_SET_XDABR, which shares most of its code with the existing H_SET_DABR
implementation. To support this, we add a per-vcpu field to store the
DABRX value plus code to get and set it via the ONE_REG interface.
For Linux guests to use this new hcall, userspace needs to add
"hcall-xdabr" to the set of strings in the /chosen/hypertas-functions
property in the device tree. If userspace does this and then migrates
the guest to a host where the kernel doesn't include this patch, then
userspace will need to implement H_SET_XDABR by writing the specified
DABR value to the DABR using the ONE_REG interface. In that case, the
old kernel will set DABRX to DABRX_USER | DABRX_KERNEL. That should
still work correctly, at least for Linux guests, since Linux guests
cope with getting data breakpoint interrupts in modes that weren't
requested by just ignoring the interrupt, and Linux guests never set
DABRX_BTI.
The other thing this does is to make H_SET_DABR and H_SET_XDABR work
on POWER8, which has the DAWR and DAWRX instead of DABR/X. Guests that
know about POWER8 should use H_SET_MODE rather than H_SET_[X]DABR, but
guests running in POWER7 compatibility mode will still use H_SET_[X]DABR.
For them, this adds the logic to convert DABR/X values into DAWR/X values
on POWER8.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:29 +08:00
|
|
|
clrrdi r4, r4, 3
|
|
|
|
|
std r4, VCPU_DAWR(r3)
|
|
|
|
|
std r5, VCPU_DAWRX(r3)
|
|
|
|
|
mtspr SPRN_DAWR, r4
|
|
|
|
|
mtspr SPRN_DAWRX, r5
|
|
|
|
|
li r3, 0
|
2011-06-29 08:22:05 +08:00
|
|
|
blr
|
|
|
|
|
|
2015-03-28 11:21:04 +08:00
|
|
|
_GLOBAL(kvmppc_h_cede) /* r3 = vcpu pointer, r11 = msr, r13 = paca */
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
ori r11,r11,MSR_EE
|
|
|
|
|
std r11,VCPU_MSR(r3)
|
|
|
|
|
li r0,1
|
|
|
|
|
stb r0,VCPU_CEDED(r3)
|
|
|
|
|
sync /* order setting ceded vs. testing prodded */
|
|
|
|
|
lbz r5,VCPU_PRODDED(r3)
|
|
|
|
|
cmpwi r5,0
|
2012-08-06 08:03:28 +08:00
|
|
|
bne kvm_cede_prodded
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
li r12,0 /* set trap to 0 to say hcall is handled */
|
|
|
|
|
stw r12,VCPU_TRAP(r3)
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
li r0,H_SUCCESS
|
2012-06-25 21:33:10 +08:00
|
|
|
std r0,VCPU_GPR(R3)(r3)
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Set our bit in the bitmask of napping threads unless all the
|
|
|
|
|
* other threads are already napping, in which case we send this
|
|
|
|
|
* up to the host.
|
|
|
|
|
*/
|
|
|
|
|
ld r5,HSTATE_KVM_VCORE(r13)
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
lbz r6,HSTATE_PTID(r13)
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
lwz r8,VCORE_ENTRY_EXIT(r5)
|
|
|
|
|
clrldi r8,r8,56
|
|
|
|
|
li r0,1
|
|
|
|
|
sld r0,r0,r6
|
|
|
|
|
addi r6,r5,VCORE_NAPPING_THREADS
|
|
|
|
|
31: lwarx r4,0,r6
|
|
|
|
|
or r4,r4,r0
|
2015-03-28 11:21:09 +08:00
|
|
|
cmpw r4,r8
|
|
|
|
|
beq kvm_cede_exit
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
stwcx. r4,0,r6
|
|
|
|
|
bne 31b
|
2015-03-28 11:21:09 +08:00
|
|
|
/* order napping_threads update vs testing entry_exit_map */
|
2013-11-16 14:46:03 +08:00
|
|
|
isync
|
KVM: PPC: Book3S HV: Align physical and virtual CPU thread numbers
On a threaded processor such as POWER7, we group VCPUs into virtual
cores and arrange that the VCPUs in a virtual core run on the same
physical core. Currently we don't enforce any correspondence between
virtual thread numbers within a virtual core and physical thread
numbers. Physical threads are allocated starting at 0 on a first-come
first-served basis to runnable virtual threads (VCPUs).
POWER8 implements a new "msgsndp" instruction which guest kernels can
use to interrupt other threads in the same core or sub-core. Since
the instruction takes the destination physical thread ID as a parameter,
it becomes necessary to align the physical thread IDs with the virtual
thread IDs, that is, to make sure virtual thread N within a virtual
core always runs on physical thread N.
This means that it's possible that thread 0, which is where we call
__kvmppc_vcore_entry, may end up running some other vcpu than the
one whose task called kvmppc_run_core(), or it may end up running
no vcpu at all, if for example thread 0 of the virtual core is
currently executing in userspace. However, we do need thread 0
to be responsible for switching the MMU -- a previous version of
this patch that had other threads switching the MMU was found to
be responsible for occasional memory corruption and machine check
interrupts in the guest on POWER7 machines.
To accommodate this, we no longer pass the vcpu pointer to
__kvmppc_vcore_entry, but instead let the assembly code load it from
the PACA. Since the assembly code will need to know the kvm pointer
and the thread ID for threads which don't have a vcpu, we move the
thread ID into the PACA and we add a kvm pointer to the virtual core
structure.
In the case where thread 0 has no vcpu to run, it still calls into
kvmppc_hv_entry in order to do the MMU switch, and then naps until
either its vcpu is ready to run in the guest, or some other thread
needs to exit the guest. In the latter case, thread 0 jumps to the
code that switches the MMU back to the host. This control flow means
that now we switch the MMU before loading any guest vcpu state.
Similarly, on guest exit we now save all the guest vcpu state before
switching the MMU back to the host. This has required substantial
code movement, making the diff rather large.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:20 +08:00
|
|
|
li r0,NAPPING_CEDE
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
stb r0,HSTATE_NAPPING(r13)
|
|
|
|
|
lwz r7,VCORE_ENTRY_EXIT(r5)
|
|
|
|
|
cmpwi r7,0x100
|
|
|
|
|
bge 33f /* another thread already exiting */
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Although not specifically required by the architecture, POWER7
|
|
|
|
|
* preserves the following registers in nap mode, even if an SMT mode
|
|
|
|
|
* switch occurs: SLB entries, PURR, SPURR, AMOR, UAMOR, AMR, SPRG0-3,
|
|
|
|
|
* DAR, DSISR, DABR, DABRX, DSCR, PMCx, MMCRx, SIAR, SDAR.
|
|
|
|
|
*/
|
|
|
|
|
/* Save non-volatile GPRs */
|
2012-06-25 21:33:10 +08:00
|
|
|
std r14, VCPU_GPR(R14)(r3)
|
|
|
|
|
std r15, VCPU_GPR(R15)(r3)
|
|
|
|
|
std r16, VCPU_GPR(R16)(r3)
|
|
|
|
|
std r17, VCPU_GPR(R17)(r3)
|
|
|
|
|
std r18, VCPU_GPR(R18)(r3)
|
|
|
|
|
std r19, VCPU_GPR(R19)(r3)
|
|
|
|
|
std r20, VCPU_GPR(R20)(r3)
|
|
|
|
|
std r21, VCPU_GPR(R21)(r3)
|
|
|
|
|
std r22, VCPU_GPR(R22)(r3)
|
|
|
|
|
std r23, VCPU_GPR(R23)(r3)
|
|
|
|
|
std r24, VCPU_GPR(R24)(r3)
|
|
|
|
|
std r25, VCPU_GPR(R25)(r3)
|
|
|
|
|
std r26, VCPU_GPR(R26)(r3)
|
|
|
|
|
std r27, VCPU_GPR(R27)(r3)
|
|
|
|
|
std r28, VCPU_GPR(R28)(r3)
|
|
|
|
|
std r29, VCPU_GPR(R29)(r3)
|
|
|
|
|
std r30, VCPU_GPR(R30)(r3)
|
|
|
|
|
std r31, VCPU_GPR(R31)(r3)
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
|
|
|
|
|
/* save FP state */
|
2013-10-15 17:43:04 +08:00
|
|
|
bl kvmppc_save_fp
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Save/restore TM state in H_CEDE
It turns out that if the guest does a H_CEDE while the CPU is in
a transactional state, and the H_CEDE does a nap, and the nap
loses the architected state of the CPU (which is is allowed to do),
then we lose the checkpointed state of the virtual CPU. In addition,
the transactional-memory state recorded in the MSR gets reset back
to non-transactional, and when we try to return to the guest, we take
a TM bad thing type of program interrupt because we are trying to
transition from non-transactional to transactional with a hrfid
instruction, which is not permitted.
The result of the program interrupt occurring at that point is that
the host CPU will hang in an infinite loop with interrupts disabled.
Thus this is a denial of service vulnerability in the host which can
be triggered by any guest (and depending on the guest kernel, it can
potentially triggered by unprivileged userspace in the guest).
This vulnerability has been assigned the ID CVE-2016-5412.
To fix this, we save the TM state before napping and restore it
on exit from the nap, when handling a H_CEDE in real mode. The
case where H_CEDE exits to host virtual mode is already OK (as are
other hcalls which exit to host virtual mode) because the exit
path saves the TM state.
Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-06-22 13:52:55 +08:00
|
|
|
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
/*
|
|
|
|
|
* Branch around the call if both CPU_FTR_TM and
|
|
|
|
|
* CPU_FTR_P9_TM_HV_ASSIST are off.
|
|
|
|
|
*/
|
KVM: PPC: Book3S HV: Save/restore TM state in H_CEDE
It turns out that if the guest does a H_CEDE while the CPU is in
a transactional state, and the H_CEDE does a nap, and the nap
loses the architected state of the CPU (which is is allowed to do),
then we lose the checkpointed state of the virtual CPU. In addition,
the transactional-memory state recorded in the MSR gets reset back
to non-transactional, and when we try to return to the guest, we take
a TM bad thing type of program interrupt because we are trying to
transition from non-transactional to transactional with a hrfid
instruction, which is not permitted.
The result of the program interrupt occurring at that point is that
the host CPU will hang in an infinite loop with interrupts disabled.
Thus this is a denial of service vulnerability in the host which can
be triggered by any guest (and depending on the guest kernel, it can
potentially triggered by unprivileged userspace in the guest).
This vulnerability has been assigned the ID CVE-2016-5412.
To fix this, we save the TM state before napping and restore it
on exit from the nap, when handling a H_CEDE in real mode. The
case where H_CEDE exits to host virtual mode is already OK (as are
other hcalls which exit to host virtual mode) because the exit
path saves the TM state.
Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-06-22 13:52:55 +08:00
|
|
|
BEGIN_FTR_SECTION
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
b 91f
|
|
|
|
|
END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
|
KVM: PPC: Book3S HV: Fix bug causing host SLB to be restored incorrectly
Aneesh Kumar reported seeing host crashes when running recent kernels
on POWER8. The symptom was an oops like this:
Unable to handle kernel paging request for data at address 0xf00000000786c620
Faulting instruction address: 0xc00000000030e1e4
Oops: Kernel access of bad area, sig: 11 [#1]
LE SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in: powernv_op_panel
CPU: 24 PID: 6663 Comm: qemu-system-ppc Tainted: G W 4.13.0-rc7-43932-gfc36c59 #2
task: c000000fdeadfe80 task.stack: c000000fdeb68000
NIP: c00000000030e1e4 LR: c00000000030de6c CTR: c000000000103620
REGS: c000000fdeb6b450 TRAP: 0300 Tainted: G W (4.13.0-rc7-43932-gfc36c59)
MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE> CR: 24044428 XER: 20000000
CFAR: c00000000030e134 DAR: f00000000786c620 DSISR: 40000000 SOFTE: 0
GPR00: 0000000000000000 c000000fdeb6b6d0 c0000000010bd000 000000000000e1b0
GPR04: c00000000115e168 c000001fffa6e4b0 c00000000115d000 c000001e1b180386
GPR08: f000000000000000 c000000f9a8913e0 f00000000786c600 00007fff587d0000
GPR12: c000000fdeb68000 c00000000fb0f000 0000000000000001 00007fff587cffff
GPR16: 0000000000000000 c000000000000000 00000000003fffff c000000fdebfe1f8
GPR20: 0000000000000004 c000000fdeb6b8a8 0000000000000001 0008000000000040
GPR24: 07000000000000c0 00007fff587cffff c000000fdec20bf8 00007fff587d0000
GPR28: c000000fdeca9ac0 00007fff587d0000 00007fff587c0000 00007fff587d0000
NIP [c00000000030e1e4] __get_user_pages_fast+0x434/0x1070
LR [c00000000030de6c] __get_user_pages_fast+0xbc/0x1070
Call Trace:
[c000000fdeb6b6d0] [c00000000139dab8] lock_classes+0x0/0x35fe50 (unreliable)
[c000000fdeb6b7e0] [c00000000030ef38] get_user_pages_fast+0xf8/0x120
[c000000fdeb6b830] [c000000000112318] kvmppc_book3s_hv_page_fault+0x308/0xf30
[c000000fdeb6b960] [c00000000010e10c] kvmppc_vcpu_run_hv+0xfdc/0x1f00
[c000000fdeb6bb20] [c0000000000e915c] kvmppc_vcpu_run+0x2c/0x40
[c000000fdeb6bb40] [c0000000000e5650] kvm_arch_vcpu_ioctl_run+0x110/0x300
[c000000fdeb6bbe0] [c0000000000d6468] kvm_vcpu_ioctl+0x528/0x900
[c000000fdeb6bd40] [c0000000003bc04c] do_vfs_ioctl+0xcc/0x950
[c000000fdeb6bde0] [c0000000003bc930] SyS_ioctl+0x60/0x100
[c000000fdeb6be30] [c00000000000b96c] system_call+0x58/0x6c
Instruction dump:
7ca81a14 2fa50000 41de0010 7cc8182a 68c60002 78c6ffe2 0b060000 3cc2000a
794a3664 390610d8 e9080000 7d485214 <e90a0020> 7d435378 790507e1 408202f0
---[ end trace fad4a342d0414aa2 ]---
It turns out that what has happened is that the SLB entry for the
vmmemap region hasn't been reloaded on exit from a guest, and it has
the wrong page size. Then, when the host next accesses the vmemmap
region, it gets a page fault.
Commit a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with
KVM", 2017-07-24) modified the guest exit code so that it now only clears
out the SLB for hash guest. The code tests the radix flag and puts the
result in a non-volatile CR field, CR2, and later branches based on CR2.
Unfortunately, the kvmppc_save_tm function, which gets called between
those two points, modifies all the user-visible registers in the case
where the guest was in transactional or suspended state, except for a
few which it restores (namely r1, r2, r9 and r13). Thus the hash/radix indication in CR2 gets corrupted.
This fixes the problem by re-doing the comparison just before the
result is needed. For good measure, this also adds comments next to
the call sites of kvmppc_save_tm and kvmppc_restore_tm pointing out
that non-volatile register state will be lost.
Cc: stable@vger.kernel.org # v4.13
Fixes: a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with KVM")
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-12 11:47:23 +08:00
|
|
|
/*
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
* NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
|
KVM: PPC: Book3S HV: Fix bug causing host SLB to be restored incorrectly
Aneesh Kumar reported seeing host crashes when running recent kernels
on POWER8. The symptom was an oops like this:
Unable to handle kernel paging request for data at address 0xf00000000786c620
Faulting instruction address: 0xc00000000030e1e4
Oops: Kernel access of bad area, sig: 11 [#1]
LE SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in: powernv_op_panel
CPU: 24 PID: 6663 Comm: qemu-system-ppc Tainted: G W 4.13.0-rc7-43932-gfc36c59 #2
task: c000000fdeadfe80 task.stack: c000000fdeb68000
NIP: c00000000030e1e4 LR: c00000000030de6c CTR: c000000000103620
REGS: c000000fdeb6b450 TRAP: 0300 Tainted: G W (4.13.0-rc7-43932-gfc36c59)
MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE> CR: 24044428 XER: 20000000
CFAR: c00000000030e134 DAR: f00000000786c620 DSISR: 40000000 SOFTE: 0
GPR00: 0000000000000000 c000000fdeb6b6d0 c0000000010bd000 000000000000e1b0
GPR04: c00000000115e168 c000001fffa6e4b0 c00000000115d000 c000001e1b180386
GPR08: f000000000000000 c000000f9a8913e0 f00000000786c600 00007fff587d0000
GPR12: c000000fdeb68000 c00000000fb0f000 0000000000000001 00007fff587cffff
GPR16: 0000000000000000 c000000000000000 00000000003fffff c000000fdebfe1f8
GPR20: 0000000000000004 c000000fdeb6b8a8 0000000000000001 0008000000000040
GPR24: 07000000000000c0 00007fff587cffff c000000fdec20bf8 00007fff587d0000
GPR28: c000000fdeca9ac0 00007fff587d0000 00007fff587c0000 00007fff587d0000
NIP [c00000000030e1e4] __get_user_pages_fast+0x434/0x1070
LR [c00000000030de6c] __get_user_pages_fast+0xbc/0x1070
Call Trace:
[c000000fdeb6b6d0] [c00000000139dab8] lock_classes+0x0/0x35fe50 (unreliable)
[c000000fdeb6b7e0] [c00000000030ef38] get_user_pages_fast+0xf8/0x120
[c000000fdeb6b830] [c000000000112318] kvmppc_book3s_hv_page_fault+0x308/0xf30
[c000000fdeb6b960] [c00000000010e10c] kvmppc_vcpu_run_hv+0xfdc/0x1f00
[c000000fdeb6bb20] [c0000000000e915c] kvmppc_vcpu_run+0x2c/0x40
[c000000fdeb6bb40] [c0000000000e5650] kvm_arch_vcpu_ioctl_run+0x110/0x300
[c000000fdeb6bbe0] [c0000000000d6468] kvm_vcpu_ioctl+0x528/0x900
[c000000fdeb6bd40] [c0000000003bc04c] do_vfs_ioctl+0xcc/0x950
[c000000fdeb6bde0] [c0000000003bc930] SyS_ioctl+0x60/0x100
[c000000fdeb6be30] [c00000000000b96c] system_call+0x58/0x6c
Instruction dump:
7ca81a14 2fa50000 41de0010 7cc8182a 68c60002 78c6ffe2 0b060000 3cc2000a
794a3664 390610d8 e9080000 7d485214 <e90a0020> 7d435378 790507e1 408202f0
---[ end trace fad4a342d0414aa2 ]---
It turns out that what has happened is that the SLB entry for the
vmmemap region hasn't been reloaded on exit from a guest, and it has
the wrong page size. Then, when the host next accesses the vmemmap
region, it gets a page fault.
Commit a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with
KVM", 2017-07-24) modified the guest exit code so that it now only clears
out the SLB for hash guest. The code tests the radix flag and puts the
result in a non-volatile CR field, CR2, and later branches based on CR2.
Unfortunately, the kvmppc_save_tm function, which gets called between
those two points, modifies all the user-visible registers in the case
where the guest was in transactional or suspended state, except for a
few which it restores (namely r1, r2, r9 and r13). Thus the hash/radix indication in CR2 gets corrupted.
This fixes the problem by re-doing the comparison just before the
result is needed. For good measure, this also adds comments next to
the call sites of kvmppc_save_tm and kvmppc_restore_tm pointing out
that non-volatile register state will be lost.
Cc: stable@vger.kernel.org # v4.13
Fixes: a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with KVM")
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-12 11:47:23 +08:00
|
|
|
*/
|
2018-05-23 15:01:48 +08:00
|
|
|
ld r3, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
ld r4, VCPU_MSR(r3)
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
li r5, 0 /* don't preserve non-vol regs */
|
2018-05-30 18:07:52 +08:00
|
|
|
bl kvmppc_save_tm_hv
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
nop
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
91:
|
KVM: PPC: Book3S HV: Save/restore TM state in H_CEDE
It turns out that if the guest does a H_CEDE while the CPU is in
a transactional state, and the H_CEDE does a nap, and the nap
loses the architected state of the CPU (which is is allowed to do),
then we lose the checkpointed state of the virtual CPU. In addition,
the transactional-memory state recorded in the MSR gets reset back
to non-transactional, and when we try to return to the guest, we take
a TM bad thing type of program interrupt because we are trying to
transition from non-transactional to transactional with a hrfid
instruction, which is not permitted.
The result of the program interrupt occurring at that point is that
the host CPU will hang in an infinite loop with interrupts disabled.
Thus this is a denial of service vulnerability in the host which can
be triggered by any guest (and depending on the guest kernel, it can
potentially triggered by unprivileged userspace in the guest).
This vulnerability has been assigned the ID CVE-2016-5412.
To fix this, we save the TM state before napping and restore it
on exit from the nap, when handling a H_CEDE in real mode. The
case where H_CEDE exits to host virtual mode is already OK (as are
other hcalls which exit to host virtual mode) because the exit
path saves the TM state.
Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-06-22 13:52:55 +08:00
|
|
|
#endif
|
|
|
|
|
|
2015-03-28 11:21:08 +08:00
|
|
|
/*
|
|
|
|
|
* Set DEC to the smaller of DEC and HDEC, so that we wake
|
|
|
|
|
* no later than the end of our timeslice (HDEC interrupts
|
|
|
|
|
* don't wake us from nap).
|
|
|
|
|
*/
|
|
|
|
|
mfspr r3, SPRN_DEC
|
|
|
|
|
mfspr r4, SPRN_HDEC
|
|
|
|
|
mftb r5
|
2017-05-22 14:55:16 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
/* On P9 check whether the guest has large decrementer mode enabled */
|
|
|
|
|
ld r6, HSTATE_KVM_VCORE(r13)
|
|
|
|
|
ld r6, VCORE_LPCR(r6)
|
|
|
|
|
andis. r6, r6, LPCR_LD@h
|
|
|
|
|
bne 68f
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
2017-05-22 14:25:14 +08:00
|
|
|
extsw r3, r3
|
2017-05-22 14:55:16 +08:00
|
|
|
68: EXTEND_HDEC(r4)
|
2017-05-22 14:25:14 +08:00
|
|
|
cmpd r3, r4
|
2015-03-28 11:21:08 +08:00
|
|
|
ble 67f
|
|
|
|
|
mtspr SPRN_DEC, r4
|
|
|
|
|
67:
|
|
|
|
|
/* save expiry time of guest decrementer */
|
|
|
|
|
add r3, r3, r5
|
|
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
KVM: PPC: Book3S HV: Snapshot timebase offset on guest entry
Currently, the HV KVM guest entry/exit code adds the timebase offset
from the vcore struct to the timebase on guest entry, and subtracts
it on guest exit. Which is fine, except that it is possible for
userspace to change the offset using the SET_ONE_REG interface while
the vcore is running, as there is only one timebase offset per vcore
but potentially multiple VCPUs in the vcore. If that were to happen,
KVM would subtract a different offset on guest exit from that which
it had added on guest entry, leading to the timebase being out of sync
between cores in the host, which then leads to bad things happening
such as hangs and spurious watchdog timeouts.
To fix this, we add a new field 'tb_offset_applied' to the vcore struct
which stores the offset that is currently applied to the timebase.
This value is set from the vcore tb_offset field on guest entry, and
is what is subtracted from the timebase on guest exit. Since it is
zero when the timebase offset is not applied, we can simplify the
logic in kvmhv_start_timing and kvmhv_accumulate_time.
In addition, we had secondary threads reading the timebase while
running concurrently with code on the primary thread which would
eventually add or subtract the timebase offset from the timebase.
This occurred while saving or restoring the DEC register value on
the secondary threads. Although no specific incorrect behaviour has
been observed, this is a race which should be fixed. To fix it, we
move the DEC saving code to just before we call kvmhv_commence_exit,
and the DEC restoring code to after the point where we have waited
for the primary thread to switch the MMU context and add the timebase
offset. That way we are sure that the timebase contains the guest
timebase value in both cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-04-20 20:51:11 +08:00
|
|
|
ld r6, VCORE_TB_OFFSET_APPL(r5)
|
2015-03-28 11:21:08 +08:00
|
|
|
subf r3, r6, r3 /* convert to host TB value */
|
|
|
|
|
std r3, VCPU_DEC_EXPIRES(r4)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
|
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
addi r3, r4, VCPU_TB_CEDE
|
|
|
|
|
bl kvmhv_accumulate_time
|
|
|
|
|
#endif
|
|
|
|
|
|
2015-03-28 11:21:07 +08:00
|
|
|
lis r3, LPCR_PECEDP@h /* Do wake on privileged doorbell */
|
|
|
|
|
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
/*
|
2014-01-08 18:25:26 +08:00
|
|
|
* Take a nap until a decrementer or external or doobell interrupt
|
2015-03-28 11:21:07 +08:00
|
|
|
* occurs, with PECE1 and PECE0 set in LPCR.
|
KVM: PPC: Book3S HV: Use msgsnd for signalling threads on POWER8
This uses msgsnd where possible for signalling other threads within
the same core on POWER8 systems, rather than IPIs through the XICS
interrupt controller. This includes waking secondary threads to run
the guest, the interrupts generated by the virtual XICS, and the
interrupts to bring the other threads out of the guest when exiting.
Aggregated statistics from debugfs across vcpus for a guest with 32
vcpus, 8 threads/vcore, running on a POWER8, show this before the
change:
rm_entry: 3387.6ns (228 - 86600, 1008969 samples)
rm_exit: 4561.5ns (12 - 3477452, 1009402 samples)
rm_intr: 1660.0ns (12 - 553050, 3600051 samples)
and this after the change:
rm_entry: 3060.1ns (212 - 65138, 953873 samples)
rm_exit: 4244.1ns (12 - 9693408, 954331 samples)
rm_intr: 1342.3ns (12 - 1104718, 3405326 samples)
for a test of booting Fedora 20 big-endian to the login prompt.
The time taken for a H_PROD hcall (which is handled in the host
kernel) went down from about 35 microseconds to about 16 microseconds
with this change.
The noinline added to kvmppc_run_core turned out to be necessary for
good performance, at least with gcc 4.9.2 as packaged with Fedora 21
and a little-endian POWER8 host.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:12 +08:00
|
|
|
* On POWER8, set PECEDH, and if we are ceding, also set PECEDP.
|
2015-03-28 11:21:07 +08:00
|
|
|
* Also clear the runlatch bit before napping.
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
*/
|
powerpc/powernv: Return to cpu offline loop when finished in KVM guest
When a secondary hardware thread has finished running a KVM guest, we
currently put that thread into nap mode using a nap instruction in
the KVM code. This changes the code so that instead of doing a nap
instruction directly, we instead cause the call to power7_nap() that
put the thread into nap mode to return. The reason for doing this is
to avoid having the KVM code having to know what low-power mode to
put the thread into.
In the case of a secondary thread used to run a KVM guest, the thread
will be offline from the point of view of the host kernel, and the
relevant power7_nap() call is the one in pnv_smp_cpu_disable().
In this case we don't want to clear pending IPIs in the offline loop
in that function, since that might cause us to miss the wakeup for
the next time the thread needs to run a guest. To tell whether or
not to clear the interrupt, we use the SRR1 value returned from
power7_nap(), and check if it indicates an external interrupt. We
arrange that the return from power7_nap() when we have finished running
a guest returns 0, so pending interrupts don't get flushed in that
case.
Note that it is important a secondary thread that has finished
executing in the guest, or that didn't have a guest to run, should
not return to power7_nap's caller while the kvm_hstate.hwthread_req
flag in the PACA is non-zero, because the return from power7_nap
will reenable the MMU, and the MMU might still be in guest context.
In this situation we spin at low priority in real mode waiting for
hwthread_req to become zero.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-12-03 11:48:40 +08:00
|
|
|
kvm_do_nap:
|
2015-03-28 11:21:04 +08:00
|
|
|
mfspr r0, SPRN_CTRLF
|
|
|
|
|
clrrdi r0, r0, 1
|
|
|
|
|
mtspr SPRN_CTRLT, r0
|
2014-04-11 18:32:08 +08:00
|
|
|
|
2012-02-03 08:54:17 +08:00
|
|
|
li r0,1
|
|
|
|
|
stb r0,HSTATE_HWTHREAD_REQ(r13)
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
mfspr r5,SPRN_LPCR
|
|
|
|
|
ori r5,r5,LPCR_PECE0 | LPCR_PECE1
|
2014-01-08 18:25:26 +08:00
|
|
|
BEGIN_FTR_SECTION
|
KVM: PPC: Book3S HV: Use msgsnd for signalling threads on POWER8
This uses msgsnd where possible for signalling other threads within
the same core on POWER8 systems, rather than IPIs through the XICS
interrupt controller. This includes waking secondary threads to run
the guest, the interrupts generated by the virtual XICS, and the
interrupts to bring the other threads out of the guest when exiting.
Aggregated statistics from debugfs across vcpus for a guest with 32
vcpus, 8 threads/vcore, running on a POWER8, show this before the
change:
rm_entry: 3387.6ns (228 - 86600, 1008969 samples)
rm_exit: 4561.5ns (12 - 3477452, 1009402 samples)
rm_intr: 1660.0ns (12 - 553050, 3600051 samples)
and this after the change:
rm_entry: 3060.1ns (212 - 65138, 953873 samples)
rm_exit: 4244.1ns (12 - 9693408, 954331 samples)
rm_intr: 1342.3ns (12 - 1104718, 3405326 samples)
for a test of booting Fedora 20 big-endian to the login prompt.
The time taken for a H_PROD hcall (which is handled in the host
kernel) went down from about 35 microseconds to about 16 microseconds
with this change.
The noinline added to kvmppc_run_core turned out to be necessary for
good performance, at least with gcc 4.9.2 as packaged with Fedora 21
and a little-endian POWER8 host.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:12 +08:00
|
|
|
ori r5, r5, LPCR_PECEDH
|
2015-03-28 11:21:07 +08:00
|
|
|
rlwimi r5, r3, 0, LPCR_PECEDP
|
2014-01-08 18:25:26 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
2016-11-18 11:34:07 +08:00
|
|
|
|
|
|
|
|
kvm_nap_sequence: /* desired LPCR value in r5 */
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
/*
|
|
|
|
|
* PSSCR bits: exit criterion = 1 (wakeup based on LPCR at sreset)
|
|
|
|
|
* enable state loss = 1 (allow SMT mode switch)
|
|
|
|
|
* requested level = 0 (just stop dispatching)
|
|
|
|
|
*/
|
|
|
|
|
lis r3, (PSSCR_EC | PSSCR_ESL)@h
|
|
|
|
|
mtspr SPRN_PSSCR, r3
|
|
|
|
|
/* Set LPCR_PECE_HVEE bit to enable wakeup by HV interrupts */
|
|
|
|
|
li r4, LPCR_PECE_HVEE@higher
|
|
|
|
|
sldi r4, r4, 32
|
|
|
|
|
or r5, r5, r4
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
mtspr SPRN_LPCR,r5
|
|
|
|
|
isync
|
|
|
|
|
li r0, 0
|
|
|
|
|
std r0, HSTATE_SCRATCH0(r13)
|
|
|
|
|
ptesync
|
|
|
|
|
ld r0, HSTATE_SCRATCH0(r13)
|
|
|
|
|
1: cmpd r0, r0
|
|
|
|
|
bne 1b
|
2016-11-18 11:34:07 +08:00
|
|
|
BEGIN_FTR_SECTION
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
nap
|
2016-11-18 11:34:07 +08:00
|
|
|
FTR_SECTION_ELSE
|
|
|
|
|
PPC_STOP
|
|
|
|
|
ALT_FTR_SECTION_END_IFCLR(CPU_FTR_ARCH_300)
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
b .
|
|
|
|
|
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
33: mr r4, r3
|
|
|
|
|
li r3, 0
|
|
|
|
|
li r12, 0
|
|
|
|
|
b 34f
|
|
|
|
|
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
kvm_end_cede:
|
2013-04-18 04:31:41 +08:00
|
|
|
/* get vcpu pointer */
|
|
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
/* Woken by external or decrementer interrupt */
|
|
|
|
|
ld r1, HSTATE_HOST_R1(r13)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
|
|
|
|
addi r3, r4, VCPU_TB_RMINTR
|
|
|
|
|
bl kvmhv_accumulate_time
|
|
|
|
|
#endif
|
|
|
|
|
|
KVM: PPC: Book3S HV: Save/restore TM state in H_CEDE
It turns out that if the guest does a H_CEDE while the CPU is in
a transactional state, and the H_CEDE does a nap, and the nap
loses the architected state of the CPU (which is is allowed to do),
then we lose the checkpointed state of the virtual CPU. In addition,
the transactional-memory state recorded in the MSR gets reset back
to non-transactional, and when we try to return to the guest, we take
a TM bad thing type of program interrupt because we are trying to
transition from non-transactional to transactional with a hrfid
instruction, which is not permitted.
The result of the program interrupt occurring at that point is that
the host CPU will hang in an infinite loop with interrupts disabled.
Thus this is a denial of service vulnerability in the host which can
be triggered by any guest (and depending on the guest kernel, it can
potentially triggered by unprivileged userspace in the guest).
This vulnerability has been assigned the ID CVE-2016-5412.
To fix this, we save the TM state before napping and restore it
on exit from the nap, when handling a H_CEDE in real mode. The
case where H_CEDE exits to host virtual mode is already OK (as are
other hcalls which exit to host virtual mode) because the exit
path saves the TM state.
Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-06-22 13:52:55 +08:00
|
|
|
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
/*
|
|
|
|
|
* Branch around the call if both CPU_FTR_TM and
|
|
|
|
|
* CPU_FTR_P9_TM_HV_ASSIST are off.
|
|
|
|
|
*/
|
KVM: PPC: Book3S HV: Save/restore TM state in H_CEDE
It turns out that if the guest does a H_CEDE while the CPU is in
a transactional state, and the H_CEDE does a nap, and the nap
loses the architected state of the CPU (which is is allowed to do),
then we lose the checkpointed state of the virtual CPU. In addition,
the transactional-memory state recorded in the MSR gets reset back
to non-transactional, and when we try to return to the guest, we take
a TM bad thing type of program interrupt because we are trying to
transition from non-transactional to transactional with a hrfid
instruction, which is not permitted.
The result of the program interrupt occurring at that point is that
the host CPU will hang in an infinite loop with interrupts disabled.
Thus this is a denial of service vulnerability in the host which can
be triggered by any guest (and depending on the guest kernel, it can
potentially triggered by unprivileged userspace in the guest).
This vulnerability has been assigned the ID CVE-2016-5412.
To fix this, we save the TM state before napping and restore it
on exit from the nap, when handling a H_CEDE in real mode. The
case where H_CEDE exits to host virtual mode is already OK (as are
other hcalls which exit to host virtual mode) because the exit
path saves the TM state.
Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-06-22 13:52:55 +08:00
|
|
|
BEGIN_FTR_SECTION
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
b 91f
|
|
|
|
|
END_FTR_SECTION(CPU_FTR_TM | CPU_FTR_P9_TM_HV_ASSIST, 0)
|
KVM: PPC: Book3S HV: Fix bug causing host SLB to be restored incorrectly
Aneesh Kumar reported seeing host crashes when running recent kernels
on POWER8. The symptom was an oops like this:
Unable to handle kernel paging request for data at address 0xf00000000786c620
Faulting instruction address: 0xc00000000030e1e4
Oops: Kernel access of bad area, sig: 11 [#1]
LE SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in: powernv_op_panel
CPU: 24 PID: 6663 Comm: qemu-system-ppc Tainted: G W 4.13.0-rc7-43932-gfc36c59 #2
task: c000000fdeadfe80 task.stack: c000000fdeb68000
NIP: c00000000030e1e4 LR: c00000000030de6c CTR: c000000000103620
REGS: c000000fdeb6b450 TRAP: 0300 Tainted: G W (4.13.0-rc7-43932-gfc36c59)
MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE> CR: 24044428 XER: 20000000
CFAR: c00000000030e134 DAR: f00000000786c620 DSISR: 40000000 SOFTE: 0
GPR00: 0000000000000000 c000000fdeb6b6d0 c0000000010bd000 000000000000e1b0
GPR04: c00000000115e168 c000001fffa6e4b0 c00000000115d000 c000001e1b180386
GPR08: f000000000000000 c000000f9a8913e0 f00000000786c600 00007fff587d0000
GPR12: c000000fdeb68000 c00000000fb0f000 0000000000000001 00007fff587cffff
GPR16: 0000000000000000 c000000000000000 00000000003fffff c000000fdebfe1f8
GPR20: 0000000000000004 c000000fdeb6b8a8 0000000000000001 0008000000000040
GPR24: 07000000000000c0 00007fff587cffff c000000fdec20bf8 00007fff587d0000
GPR28: c000000fdeca9ac0 00007fff587d0000 00007fff587c0000 00007fff587d0000
NIP [c00000000030e1e4] __get_user_pages_fast+0x434/0x1070
LR [c00000000030de6c] __get_user_pages_fast+0xbc/0x1070
Call Trace:
[c000000fdeb6b6d0] [c00000000139dab8] lock_classes+0x0/0x35fe50 (unreliable)
[c000000fdeb6b7e0] [c00000000030ef38] get_user_pages_fast+0xf8/0x120
[c000000fdeb6b830] [c000000000112318] kvmppc_book3s_hv_page_fault+0x308/0xf30
[c000000fdeb6b960] [c00000000010e10c] kvmppc_vcpu_run_hv+0xfdc/0x1f00
[c000000fdeb6bb20] [c0000000000e915c] kvmppc_vcpu_run+0x2c/0x40
[c000000fdeb6bb40] [c0000000000e5650] kvm_arch_vcpu_ioctl_run+0x110/0x300
[c000000fdeb6bbe0] [c0000000000d6468] kvm_vcpu_ioctl+0x528/0x900
[c000000fdeb6bd40] [c0000000003bc04c] do_vfs_ioctl+0xcc/0x950
[c000000fdeb6bde0] [c0000000003bc930] SyS_ioctl+0x60/0x100
[c000000fdeb6be30] [c00000000000b96c] system_call+0x58/0x6c
Instruction dump:
7ca81a14 2fa50000 41de0010 7cc8182a 68c60002 78c6ffe2 0b060000 3cc2000a
794a3664 390610d8 e9080000 7d485214 <e90a0020> 7d435378 790507e1 408202f0
---[ end trace fad4a342d0414aa2 ]---
It turns out that what has happened is that the SLB entry for the
vmmemap region hasn't been reloaded on exit from a guest, and it has
the wrong page size. Then, when the host next accesses the vmemmap
region, it gets a page fault.
Commit a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with
KVM", 2017-07-24) modified the guest exit code so that it now only clears
out the SLB for hash guest. The code tests the radix flag and puts the
result in a non-volatile CR field, CR2, and later branches based on CR2.
Unfortunately, the kvmppc_save_tm function, which gets called between
those two points, modifies all the user-visible registers in the case
where the guest was in transactional or suspended state, except for a
few which it restores (namely r1, r2, r9 and r13). Thus the hash/radix indication in CR2 gets corrupted.
This fixes the problem by re-doing the comparison just before the
result is needed. For good measure, this also adds comments next to
the call sites of kvmppc_save_tm and kvmppc_restore_tm pointing out
that non-volatile register state will be lost.
Cc: stable@vger.kernel.org # v4.13
Fixes: a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with KVM")
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-12 11:47:23 +08:00
|
|
|
/*
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
* NOTE THAT THIS TRASHES ALL NON-VOLATILE REGISTERS (but not CR)
|
KVM: PPC: Book3S HV: Fix bug causing host SLB to be restored incorrectly
Aneesh Kumar reported seeing host crashes when running recent kernels
on POWER8. The symptom was an oops like this:
Unable to handle kernel paging request for data at address 0xf00000000786c620
Faulting instruction address: 0xc00000000030e1e4
Oops: Kernel access of bad area, sig: 11 [#1]
LE SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in: powernv_op_panel
CPU: 24 PID: 6663 Comm: qemu-system-ppc Tainted: G W 4.13.0-rc7-43932-gfc36c59 #2
task: c000000fdeadfe80 task.stack: c000000fdeb68000
NIP: c00000000030e1e4 LR: c00000000030de6c CTR: c000000000103620
REGS: c000000fdeb6b450 TRAP: 0300 Tainted: G W (4.13.0-rc7-43932-gfc36c59)
MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE> CR: 24044428 XER: 20000000
CFAR: c00000000030e134 DAR: f00000000786c620 DSISR: 40000000 SOFTE: 0
GPR00: 0000000000000000 c000000fdeb6b6d0 c0000000010bd000 000000000000e1b0
GPR04: c00000000115e168 c000001fffa6e4b0 c00000000115d000 c000001e1b180386
GPR08: f000000000000000 c000000f9a8913e0 f00000000786c600 00007fff587d0000
GPR12: c000000fdeb68000 c00000000fb0f000 0000000000000001 00007fff587cffff
GPR16: 0000000000000000 c000000000000000 00000000003fffff c000000fdebfe1f8
GPR20: 0000000000000004 c000000fdeb6b8a8 0000000000000001 0008000000000040
GPR24: 07000000000000c0 00007fff587cffff c000000fdec20bf8 00007fff587d0000
GPR28: c000000fdeca9ac0 00007fff587d0000 00007fff587c0000 00007fff587d0000
NIP [c00000000030e1e4] __get_user_pages_fast+0x434/0x1070
LR [c00000000030de6c] __get_user_pages_fast+0xbc/0x1070
Call Trace:
[c000000fdeb6b6d0] [c00000000139dab8] lock_classes+0x0/0x35fe50 (unreliable)
[c000000fdeb6b7e0] [c00000000030ef38] get_user_pages_fast+0xf8/0x120
[c000000fdeb6b830] [c000000000112318] kvmppc_book3s_hv_page_fault+0x308/0xf30
[c000000fdeb6b960] [c00000000010e10c] kvmppc_vcpu_run_hv+0xfdc/0x1f00
[c000000fdeb6bb20] [c0000000000e915c] kvmppc_vcpu_run+0x2c/0x40
[c000000fdeb6bb40] [c0000000000e5650] kvm_arch_vcpu_ioctl_run+0x110/0x300
[c000000fdeb6bbe0] [c0000000000d6468] kvm_vcpu_ioctl+0x528/0x900
[c000000fdeb6bd40] [c0000000003bc04c] do_vfs_ioctl+0xcc/0x950
[c000000fdeb6bde0] [c0000000003bc930] SyS_ioctl+0x60/0x100
[c000000fdeb6be30] [c00000000000b96c] system_call+0x58/0x6c
Instruction dump:
7ca81a14 2fa50000 41de0010 7cc8182a 68c60002 78c6ffe2 0b060000 3cc2000a
794a3664 390610d8 e9080000 7d485214 <e90a0020> 7d435378 790507e1 408202f0
---[ end trace fad4a342d0414aa2 ]---
It turns out that what has happened is that the SLB entry for the
vmmemap region hasn't been reloaded on exit from a guest, and it has
the wrong page size. Then, when the host next accesses the vmemmap
region, it gets a page fault.
Commit a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with
KVM", 2017-07-24) modified the guest exit code so that it now only clears
out the SLB for hash guest. The code tests the radix flag and puts the
result in a non-volatile CR field, CR2, and later branches based on CR2.
Unfortunately, the kvmppc_save_tm function, which gets called between
those two points, modifies all the user-visible registers in the case
where the guest was in transactional or suspended state, except for a
few which it restores (namely r1, r2, r9 and r13). Thus the hash/radix indication in CR2 gets corrupted.
This fixes the problem by re-doing the comparison just before the
result is needed. For good measure, this also adds comments next to
the call sites of kvmppc_save_tm and kvmppc_restore_tm pointing out
that non-volatile register state will be lost.
Cc: stable@vger.kernel.org # v4.13
Fixes: a25bd72badfa ("powerpc/mm/radix: Workaround prefetch issue with KVM")
Tested-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-12 11:47:23 +08:00
|
|
|
*/
|
2018-05-23 15:01:48 +08:00
|
|
|
mr r3, r4
|
|
|
|
|
ld r4, VCPU_MSR(r3)
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
li r5, 0 /* don't preserve non-vol regs */
|
2018-05-30 18:07:52 +08:00
|
|
|
bl kvmppc_restore_tm_hv
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
nop
|
2018-05-23 15:01:48 +08:00
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
91:
|
KVM: PPC: Book3S HV: Save/restore TM state in H_CEDE
It turns out that if the guest does a H_CEDE while the CPU is in
a transactional state, and the H_CEDE does a nap, and the nap
loses the architected state of the CPU (which is is allowed to do),
then we lose the checkpointed state of the virtual CPU. In addition,
the transactional-memory state recorded in the MSR gets reset back
to non-transactional, and when we try to return to the guest, we take
a TM bad thing type of program interrupt because we are trying to
transition from non-transactional to transactional with a hrfid
instruction, which is not permitted.
The result of the program interrupt occurring at that point is that
the host CPU will hang in an infinite loop with interrupts disabled.
Thus this is a denial of service vulnerability in the host which can
be triggered by any guest (and depending on the guest kernel, it can
potentially triggered by unprivileged userspace in the guest).
This vulnerability has been assigned the ID CVE-2016-5412.
To fix this, we save the TM state before napping and restore it
on exit from the nap, when handling a H_CEDE in real mode. The
case where H_CEDE exits to host virtual mode is already OK (as are
other hcalls which exit to host virtual mode) because the exit
path saves the TM state.
Cc: stable@vger.kernel.org # v3.15+
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-06-22 13:52:55 +08:00
|
|
|
#endif
|
|
|
|
|
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
/* load up FP state */
|
|
|
|
|
bl kvmppc_load_fp
|
|
|
|
|
|
2015-03-28 11:21:08 +08:00
|
|
|
/* Restore guest decrementer */
|
|
|
|
|
ld r3, VCPU_DEC_EXPIRES(r4)
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
KVM: PPC: Book3S HV: Snapshot timebase offset on guest entry
Currently, the HV KVM guest entry/exit code adds the timebase offset
from the vcore struct to the timebase on guest entry, and subtracts
it on guest exit. Which is fine, except that it is possible for
userspace to change the offset using the SET_ONE_REG interface while
the vcore is running, as there is only one timebase offset per vcore
but potentially multiple VCPUs in the vcore. If that were to happen,
KVM would subtract a different offset on guest exit from that which
it had added on guest entry, leading to the timebase being out of sync
between cores in the host, which then leads to bad things happening
such as hangs and spurious watchdog timeouts.
To fix this, we add a new field 'tb_offset_applied' to the vcore struct
which stores the offset that is currently applied to the timebase.
This value is set from the vcore tb_offset field on guest entry, and
is what is subtracted from the timebase on guest exit. Since it is
zero when the timebase offset is not applied, we can simplify the
logic in kvmhv_start_timing and kvmhv_accumulate_time.
In addition, we had secondary threads reading the timebase while
running concurrently with code on the primary thread which would
eventually add or subtract the timebase offset from the timebase.
This occurred while saving or restoring the DEC register value on
the secondary threads. Although no specific incorrect behaviour has
been observed, this is a race which should be fixed. To fix it, we
move the DEC saving code to just before we call kvmhv_commence_exit,
and the DEC restoring code to after the point where we have waited
for the primary thread to switch the MMU context and add the timebase
offset. That way we are sure that the timebase contains the guest
timebase value in both cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-04-20 20:51:11 +08:00
|
|
|
ld r6, VCORE_TB_OFFSET_APPL(r5)
|
2015-03-28 11:21:08 +08:00
|
|
|
add r3, r3, r6 /* convert host TB to guest TB value */
|
|
|
|
|
mftb r7
|
|
|
|
|
subf r3, r7, r3
|
|
|
|
|
mtspr SPRN_DEC, r3
|
|
|
|
|
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
/* Load NV GPRS */
|
2012-06-25 21:33:10 +08:00
|
|
|
ld r14, VCPU_GPR(R14)(r4)
|
|
|
|
|
ld r15, VCPU_GPR(R15)(r4)
|
|
|
|
|
ld r16, VCPU_GPR(R16)(r4)
|
|
|
|
|
ld r17, VCPU_GPR(R17)(r4)
|
|
|
|
|
ld r18, VCPU_GPR(R18)(r4)
|
|
|
|
|
ld r19, VCPU_GPR(R19)(r4)
|
|
|
|
|
ld r20, VCPU_GPR(R20)(r4)
|
|
|
|
|
ld r21, VCPU_GPR(R21)(r4)
|
|
|
|
|
ld r22, VCPU_GPR(R22)(r4)
|
|
|
|
|
ld r23, VCPU_GPR(R23)(r4)
|
|
|
|
|
ld r24, VCPU_GPR(R24)(r4)
|
|
|
|
|
ld r25, VCPU_GPR(R25)(r4)
|
|
|
|
|
ld r26, VCPU_GPR(R26)(r4)
|
|
|
|
|
ld r27, VCPU_GPR(R27)(r4)
|
|
|
|
|
ld r28, VCPU_GPR(R28)(r4)
|
|
|
|
|
ld r29, VCPU_GPR(R29)(r4)
|
|
|
|
|
ld r30, VCPU_GPR(R30)(r4)
|
|
|
|
|
ld r31, VCPU_GPR(R31)(r4)
|
2016-08-19 13:35:46 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
/* Check the wake reason in SRR1 to see why we got here */
|
|
|
|
|
bl kvmppc_check_wake_reason
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
|
2016-08-19 13:35:46 +08:00
|
|
|
/*
|
|
|
|
|
* Restore volatile registers since we could have called a
|
|
|
|
|
* C routine in kvmppc_check_wake_reason
|
|
|
|
|
* r4 = VCPU
|
|
|
|
|
* r3 tells us whether we need to return to host or not
|
|
|
|
|
* WARNING: it gets checked further down:
|
|
|
|
|
* should not modify r3 until this check is done.
|
|
|
|
|
*/
|
|
|
|
|
ld r4, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
/* clear our bit in vcore->napping_threads */
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
34: ld r5,HSTATE_KVM_VCORE(r13)
|
|
|
|
|
lbz r7,HSTATE_PTID(r13)
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
li r0,1
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
sld r0,r0,r7
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
addi r6,r5,VCORE_NAPPING_THREADS
|
|
|
|
|
32: lwarx r7,0,r6
|
|
|
|
|
andc r7,r7,r0
|
|
|
|
|
stwcx. r7,0,r6
|
|
|
|
|
bne 32b
|
|
|
|
|
li r0,0
|
|
|
|
|
stb r0,HSTATE_NAPPING(r13)
|
|
|
|
|
|
2016-08-19 13:35:46 +08:00
|
|
|
/* See if the wake reason saved in r3 means we need to exit */
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
stw r12, VCPU_TRAP(r4)
|
2013-04-18 04:31:41 +08:00
|
|
|
mr r9, r4
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
cmpdi r3, 0
|
|
|
|
|
bgt guest_exit_cont
|
2018-10-08 13:30:52 +08:00
|
|
|
b maybe_reenter_guest
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
|
|
|
|
|
/* cede when already previously prodded case */
|
2012-08-06 08:03:28 +08:00
|
|
|
kvm_cede_prodded:
|
|
|
|
|
li r0,0
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
stb r0,VCPU_PRODDED(r3)
|
|
|
|
|
sync /* order testing prodded vs. clearing ceded */
|
|
|
|
|
stb r0,VCPU_CEDED(r3)
|
|
|
|
|
li r3,H_SUCCESS
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
/* we've ceded but we want to give control to the host */
|
2012-08-06 08:03:28 +08:00
|
|
|
kvm_cede_exit:
|
KVM: PPC: Book3S HV: Streamline guest entry and exit
On entry to the guest, secondary threads now wait for the primary to
switch the MMU after loading up most of their state, rather than before.
This means that the secondary threads get into the guest sooner, in the
common case where the secondary threads get to kvmppc_hv_entry before
the primary thread.
On exit, the first thread out increments the exit count and interrupts
the other threads (to get them out of the guest) before saving most
of its state, rather than after. That means that the other threads
exit sooner and means that the first thread doesn't spend so much
time waiting for the other threads at the point where the MMU gets
switched back to the host.
This pulls out the code that increments the exit count and interrupts
other threads into a separate function, kvmhv_commence_exit().
This also makes sure that r12 and vcpu->arch.trap are set correctly
in some corner cases.
Statistics from /sys/kernel/debug/kvm/vm*/vcpu*/timings show the
improvement. Aggregating across vcpus for a guest with 32 vcpus,
8 threads/vcore, running on a POWER8, gives this before the change:
rm_entry: avg 4537.3ns (222 - 48444, 1068878 samples)
rm_exit: avg 4787.6ns (152 - 165490, 1010717 samples)
rm_intr: avg 1673.6ns (12 - 341304, 3818691 samples)
and this after the change:
rm_entry: avg 3427.7ns (232 - 68150, 1118921 samples)
rm_exit: avg 4716.0ns (12 - 150720, 1119477 samples)
rm_intr: avg 1614.8ns (12 - 522436, 3850432 samples)
showing a substantial reduction in the time spent per guest entry in
the real-mode guest entry code, and smaller reductions in the real
mode guest exit and interrupt handling times. (The test was to start
the guest and boot Fedora 20 big-endian to the login prompt.)
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:10 +08:00
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
2018-01-12 10:37:16 +08:00
|
|
|
#ifdef CONFIG_KVM_XICS
|
|
|
|
|
/* Abort if we still have a pending escalation */
|
|
|
|
|
lbz r5, VCPU_XIVE_ESC_ON(r9)
|
|
|
|
|
cmpwi r5, 0
|
|
|
|
|
beq 1f
|
|
|
|
|
li r0, 0
|
|
|
|
|
stb r0, VCPU_CEDED(r9)
|
|
|
|
|
1: /* Enable XIVE escalation */
|
|
|
|
|
li r5, XIVE_ESB_SET_PQ_00
|
|
|
|
|
mfmsr r0
|
|
|
|
|
andi. r0, r0, MSR_DR /* in real mode? */
|
|
|
|
|
beq 1f
|
|
|
|
|
ld r10, VCPU_XIVE_ESC_VADDR(r9)
|
|
|
|
|
cmpdi r10, 0
|
|
|
|
|
beq 3f
|
|
|
|
|
ldx r0, r10, r5
|
|
|
|
|
b 2f
|
|
|
|
|
1: ld r10, VCPU_XIVE_ESC_RADDR(r9)
|
|
|
|
|
cmpdi r10, 0
|
|
|
|
|
beq 3f
|
|
|
|
|
ldcix r0, r10, r5
|
|
|
|
|
2: sync
|
|
|
|
|
li r0, 1
|
|
|
|
|
stb r0, VCPU_XIVE_ESC_ON(r9)
|
|
|
|
|
#endif /* CONFIG_KVM_XICS */
|
|
|
|
|
3: b guest_exit_cont
|
KVM: PPC: Implement H_CEDE hcall for book3s_hv in real-mode code
With a KVM guest operating in SMT4 mode (i.e. 4 hardware threads per
core), whenever a CPU goes idle, we have to pull all the other
hardware threads in the core out of the guest, because the H_CEDE
hcall is handled in the kernel. This is inefficient.
This adds code to book3s_hv_rmhandlers.S to handle the H_CEDE hcall
in real mode. When a guest vcpu does an H_CEDE hcall, we now only
exit to the kernel if all the other vcpus in the same core are also
idle. Otherwise we mark this vcpu as napping, save state that could
be lost in nap mode (mainly GPRs and FPRs), and execute the nap
instruction. When the thread wakes up, because of a decrementer or
external interrupt, we come back in at kvm_start_guest (from the
system reset interrupt vector), find the `napping' flag set in the
paca, and go to the resume path.
This has some other ramifications. First, when starting a core, we
now start all the threads, both those that are immediately runnable and
those that are idle. This is so that we don't have to pull all the
threads out of the guest when an idle thread gets a decrementer interrupt
and wants to start running. In fact the idle threads will all start
with the H_CEDE hcall returning; being idle they will just do another
H_CEDE immediately and go to nap mode.
This required some changes to kvmppc_run_core() and kvmppc_run_vcpu().
These functions have been restructured to make them simpler and clearer.
We introduce a level of indirection in the wait queue that gets woken
when external and decrementer interrupts get generated for a vcpu, so
that we can have the 4 vcpus in a vcore using the same wait queue.
We need this because the 4 vcpus are being handled by one thread.
Secondly, when we need to exit from the guest to the kernel, we now
have to generate an IPI for any napping threads, because an HDEC
interrupt doesn't wake up a napping thread.
Thirdly, we now need to be able to handle virtual external interrupts
and decrementer interrupts becoming pending while a thread is napping,
and deliver those interrupts to the guest when the thread wakes.
This is done in kvmppc_cede_reentry, just before fast_guest_return.
Finally, since we are not using the generic kvm_vcpu_block for book3s_hv,
and hence not calling kvm_arch_vcpu_runnable, we can remove the #ifdef
from kvm_arch_vcpu_runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-07-23 15:42:46 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Simplify machine check handling
This makes the handling of machine check interrupts that occur inside
a guest simpler and more robust, with less done in assembler code and
in real mode.
Now, when a machine check occurs inside a guest, we always get the
machine check event struct and put a copy in the vcpu struct for the
vcpu where the machine check occurred. We no longer call
machine_check_queue_event() from kvmppc_realmode_mc_power7(), because
on POWER8, when a vcpu is running on an offline secondary thread and
we call machine_check_queue_event(), that calls irq_work_queue(),
which doesn't work because the CPU is offline, but instead triggers
the WARN_ON(lazy_irq_pending()) in pnv_smp_cpu_kill_self() (which
fires again and again because nothing clears the condition).
All that machine_check_queue_event() actually does is to cause the
event to be printed to the console. For a machine check occurring in
the guest, we now print the event in kvmppc_handle_exit_hv()
instead.
The assembly code at label machine_check_realmode now just calls C
code and then continues exiting the guest. We no longer either
synthesize a machine check for the guest in assembly code or return
to the guest without a machine check.
The code in kvmppc_handle_exit_hv() is extended to handle the case
where the guest is not FWNMI-capable. In that case we now always
synthesize a machine check interrupt for the guest. Previously, if
the host thinks it has recovered the machine check fully, it would
return to the guest without any notification that the machine check
had occurred. If the machine check was caused by some action of the
guest (such as creating duplicate SLB entries), it is much better to
tell the guest that it has caused a problem. Therefore we now always
generate a machine check interrupt for guests that are not
FWNMI-capable.
Reviewed-by: Aravinda Prasad <aravinda@linux.vnet.ibm.com>
Reviewed-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2019-02-21 10:38:49 +08:00
|
|
|
/* Try to do machine check recovery in real mode */
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
machine_check_realmode:
|
|
|
|
|
mr r3, r9 /* get vcpu pointer */
|
2014-02-04 13:04:35 +08:00
|
|
|
bl kvmppc_realmode_machine_check
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
nop
|
KVM: PPC: Book3S HV: Simplify machine check handling
This makes the handling of machine check interrupts that occur inside
a guest simpler and more robust, with less done in assembler code and
in real mode.
Now, when a machine check occurs inside a guest, we always get the
machine check event struct and put a copy in the vcpu struct for the
vcpu where the machine check occurred. We no longer call
machine_check_queue_event() from kvmppc_realmode_mc_power7(), because
on POWER8, when a vcpu is running on an offline secondary thread and
we call machine_check_queue_event(), that calls irq_work_queue(),
which doesn't work because the CPU is offline, but instead triggers
the WARN_ON(lazy_irq_pending()) in pnv_smp_cpu_kill_self() (which
fires again and again because nothing clears the condition).
All that machine_check_queue_event() actually does is to cause the
event to be printed to the console. For a machine check occurring in
the guest, we now print the event in kvmppc_handle_exit_hv()
instead.
The assembly code at label machine_check_realmode now just calls C
code and then continues exiting the guest. We no longer either
synthesize a machine check for the guest in assembly code or return
to the guest without a machine check.
The code in kvmppc_handle_exit_hv() is extended to handle the case
where the guest is not FWNMI-capable. In that case we now always
synthesize a machine check interrupt for the guest. Previously, if
the host thinks it has recovered the machine check fully, it would
return to the guest without any notification that the machine check
had occurred. If the machine check was caused by some action of the
guest (such as creating duplicate SLB entries), it is much better to
tell the guest that it has caused a problem. Therefore we now always
generate a machine check interrupt for guests that are not
FWNMI-capable.
Reviewed-by: Aravinda Prasad <aravinda@linux.vnet.ibm.com>
Reviewed-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2019-02-21 10:38:49 +08:00
|
|
|
/* all machine checks go to virtual mode for further handling */
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
li r12, BOOK3S_INTERRUPT_MACHINE_CHECK
|
KVM: PPC: Book3S HV: Simplify machine check handling
This makes the handling of machine check interrupts that occur inside
a guest simpler and more robust, with less done in assembler code and
in real mode.
Now, when a machine check occurs inside a guest, we always get the
machine check event struct and put a copy in the vcpu struct for the
vcpu where the machine check occurred. We no longer call
machine_check_queue_event() from kvmppc_realmode_mc_power7(), because
on POWER8, when a vcpu is running on an offline secondary thread and
we call machine_check_queue_event(), that calls irq_work_queue(),
which doesn't work because the CPU is offline, but instead triggers
the WARN_ON(lazy_irq_pending()) in pnv_smp_cpu_kill_self() (which
fires again and again because nothing clears the condition).
All that machine_check_queue_event() actually does is to cause the
event to be printed to the console. For a machine check occurring in
the guest, we now print the event in kvmppc_handle_exit_hv()
instead.
The assembly code at label machine_check_realmode now just calls C
code and then continues exiting the guest. We no longer either
synthesize a machine check for the guest in assembly code or return
to the guest without a machine check.
The code in kvmppc_handle_exit_hv() is extended to handle the case
where the guest is not FWNMI-capable. In that case we now always
synthesize a machine check interrupt for the guest. Previously, if
the host thinks it has recovered the machine check fully, it would
return to the guest without any notification that the machine check
had occurred. If the machine check was caused by some action of the
guest (such as creating duplicate SLB entries), it is much better to
tell the guest that it has caused a problem. Therefore we now always
generate a machine check interrupt for guests that are not
FWNMI-capable.
Reviewed-by: Aravinda Prasad <aravinda@linux.vnet.ibm.com>
Reviewed-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2019-02-21 10:38:49 +08:00
|
|
|
b guest_exit_cont
|
KVM: PPC: Book3S HV: Handle guest-caused machine checks on POWER7 without panicking
Currently, if a machine check interrupt happens while we are in the
guest, we exit the guest and call the host's machine check handler,
which tends to cause the host to panic. Some machine checks can be
triggered by the guest; for example, if the guest creates two entries
in the SLB that map the same effective address, and then accesses that
effective address, the CPU will take a machine check interrupt.
To handle this better, when a machine check happens inside the guest,
we call a new function, kvmppc_realmode_machine_check(), while still in
real mode before exiting the guest. On POWER7, it handles the cases
that the guest can trigger, either by flushing and reloading the SLB,
or by flushing the TLB, and then it delivers the machine check interrupt
directly to the guest without going back to the host. On POWER7, the
OPAL firmware patches the machine check interrupt vector so that it
gets control first, and it leaves behind its analysis of the situation
in a structure pointed to by the opal_mc_evt field of the paca. The
kvmppc_realmode_machine_check() function looks at this, and if OPAL
reports that there was no error, or that it has handled the error, we
also go straight back to the guest with a machine check. We have to
deliver a machine check to the guest since the machine check interrupt
might have trashed valid values in SRR0/1.
If the machine check is one we can't handle in real mode, and one that
OPAL hasn't already handled, or on PPC970, we exit the guest and call
the host's machine check handler. We do this by jumping to the
machine_check_fwnmi label, rather than absolute address 0x200, because
we don't want to re-execute OPAL's handler on POWER7. On PPC970, the
two are equivalent because address 0x200 just contains a branch.
Then, if the host machine check handler decides that the system can
continue executing, kvmppc_handle_exit() delivers a machine check
interrupt to the guest -- once again to let the guest know that SRR0/1
have been modified.
Signed-off-by: Paul Mackerras <paulus@samba.org>
[agraf: fix checkpatch warnings]
Signed-off-by: Alexander Graf <agraf@suse.de>
2012-11-24 06:37:50 +08:00
|
|
|
|
2018-10-08 13:30:52 +08:00
|
|
|
/*
|
|
|
|
|
* Call C code to handle a HMI in real mode.
|
|
|
|
|
* Only the primary thread does the call, secondary threads are handled
|
|
|
|
|
* by calling hmi_exception_realmode() after kvmppc_hv_entry returns.
|
|
|
|
|
* r9 points to the vcpu on entry
|
|
|
|
|
*/
|
|
|
|
|
hmi_realmode:
|
|
|
|
|
lbz r0, HSTATE_PTID(r13)
|
|
|
|
|
cmpwi r0, 0
|
|
|
|
|
bne guest_exit_cont
|
|
|
|
|
bl kvmppc_realmode_hmi_handler
|
|
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
li r12, BOOK3S_INTERRUPT_HMI
|
|
|
|
|
b guest_exit_cont
|
|
|
|
|
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
/*
|
|
|
|
|
* Check the reason we woke from nap, and take appropriate action.
|
2015-03-28 11:21:04 +08:00
|
|
|
* Returns (in r3):
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
* 0 if nothing needs to be done
|
|
|
|
|
* 1 if something happened that needs to be handled by the host
|
KVM: PPC: Book3S HV: Use msgsnd for signalling threads on POWER8
This uses msgsnd where possible for signalling other threads within
the same core on POWER8 systems, rather than IPIs through the XICS
interrupt controller. This includes waking secondary threads to run
the guest, the interrupts generated by the virtual XICS, and the
interrupts to bring the other threads out of the guest when exiting.
Aggregated statistics from debugfs across vcpus for a guest with 32
vcpus, 8 threads/vcore, running on a POWER8, show this before the
change:
rm_entry: 3387.6ns (228 - 86600, 1008969 samples)
rm_exit: 4561.5ns (12 - 3477452, 1009402 samples)
rm_intr: 1660.0ns (12 - 553050, 3600051 samples)
and this after the change:
rm_entry: 3060.1ns (212 - 65138, 953873 samples)
rm_exit: 4244.1ns (12 - 9693408, 954331 samples)
rm_intr: 1342.3ns (12 - 1104718, 3405326 samples)
for a test of booting Fedora 20 big-endian to the login prompt.
The time taken for a H_PROD hcall (which is handled in the host
kernel) went down from about 35 microseconds to about 16 microseconds
with this change.
The noinline added to kvmppc_run_core turned out to be necessary for
good performance, at least with gcc 4.9.2 as packaged with Fedora 21
and a little-endian POWER8 host.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:12 +08:00
|
|
|
* -1 if there was a guest wakeup (IPI or msgsnd)
|
KVM: PPC: Book3S HV: Handle passthrough interrupts in guest
Currently, KVM switches back to the host to handle any external
interrupt (when the interrupt is received while running in the
guest). This patch updates real-mode KVM to check if an interrupt
is generated by a passthrough adapter that is owned by this guest.
If so, the real mode KVM will directly inject the corresponding
virtual interrupt to the guest VCPU's ICS and also EOI the interrupt
in hardware. In short, the interrupt is handled entirely in real
mode in the guest context without switching back to the host.
In some rare cases, the interrupt cannot be completely handled in
real mode, for instance, a VCPU that is sleeping needs to be woken
up. In this case, KVM simply switches back to the host with trap
reason set to 0x500. This works, but it is clearly not very efficient.
A following patch will distinguish this case and handle it
correctly in the host. Note that we can use the existing
check_too_hard() routine even though we are not in a hypercall to
determine if there is unfinished business that needs to be
completed in host virtual mode.
The patch assumes that the mapping between hardware interrupt IRQ
and virtual IRQ to be injected to the guest already exists for the
PCI passthrough interrupts that need to be handled in real mode.
If the mapping does not exist, KVM falls back to the default
existing behavior.
The KVM real mode code reads mappings from the mapped array in the
passthrough IRQ map without taking any lock. We carefully order the
loads and stores of the fields in the kvmppc_irq_map data structure
using memory barriers to avoid an inconsistent mapping being seen by
the reader. Thus, although it is possible to miss a map entry, it is
not possible to read a stale value.
[paulus@ozlabs.org - get irq_chip from irq_map rather than pimap,
pulled out powernv eoi change into a separate patch, made
kvmppc_read_intr get the vcpu from the paca rather than being
passed in, rewrote the logic at the end of kvmppc_read_intr to
avoid deep indentation, simplified logic in book3s_hv_rmhandlers.S
since we were always restoring SRR0/1 anyway, get rid of the cached
array (just use the mapped array), removed the kick_all_cpus_sync()
call, clear saved_xirr PACA field when we handle the interrupt in
real mode, fix compilation with CONFIG_KVM_XICS=n.]
Signed-off-by: Suresh Warrier <warrier@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-08-19 13:35:51 +08:00
|
|
|
* -2 if we handled a PCI passthrough interrupt (returned by
|
|
|
|
|
* kvmppc_read_intr only)
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
*
|
|
|
|
|
* Also sets r12 to the interrupt vector for any interrupt that needs
|
|
|
|
|
* to be handled now by the host (0x500 for external interrupt), or zero.
|
2016-08-19 13:35:46 +08:00
|
|
|
* Modifies all volatile registers (since it may call a C function).
|
|
|
|
|
* This routine calls kvmppc_read_intr, a C function, if an external
|
|
|
|
|
* interrupt is pending.
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
*/
|
|
|
|
|
kvmppc_check_wake_reason:
|
|
|
|
|
mfspr r6, SPRN_SRR1
|
2014-01-08 18:25:26 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
rlwinm r6, r6, 45-31, 0xf /* extract wake reason field (P8) */
|
|
|
|
|
FTR_SECTION_ELSE
|
|
|
|
|
rlwinm r6, r6, 45-31, 0xe /* P7 wake reason field is 3 bits */
|
|
|
|
|
ALT_FTR_SECTION_END_IFSET(CPU_FTR_ARCH_207S)
|
|
|
|
|
cmpwi r6, 8 /* was it an external interrupt? */
|
2016-08-19 13:35:46 +08:00
|
|
|
beq 7f /* if so, see what it was */
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
li r3, 0
|
|
|
|
|
li r12, 0
|
|
|
|
|
cmpwi r6, 6 /* was it the decrementer? */
|
|
|
|
|
beq 0f
|
2014-01-08 18:25:26 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
cmpwi r6, 5 /* privileged doorbell? */
|
|
|
|
|
beq 0f
|
2014-01-08 18:25:28 +08:00
|
|
|
cmpwi r6, 3 /* hypervisor doorbell? */
|
|
|
|
|
beq 3f
|
2014-01-08 18:25:26 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
KVM: PPC: Book3S HV: Fix TB corruption in guest exit path on HMI interrupt
When a guest is assigned to a core it converts the host Timebase (TB)
into guest TB by adding guest timebase offset before entering into
guest. During guest exit it restores the guest TB to host TB. This means
under certain conditions (Guest migration) host TB and guest TB can differ.
When we get an HMI for TB related issues the opal HMI handler would
try fixing errors and restore the correct host TB value. With no guest
running, we don't have any issues. But with guest running on the core
we run into TB corruption issues.
If we get an HMI while in the guest, the current HMI handler invokes opal
hmi handler before forcing guest to exit. The guest exit path subtracts
the guest TB offset from the current TB value which may have already
been restored with host value by opal hmi handler. This leads to incorrect
host and guest TB values.
With split-core, things become more complex. With split-core, TB also gets
split and each subcore gets its own TB register. When a hmi handler fixes
a TB error and restores the TB value, it affects all the TB values of
sibling subcores on the same core. On TB errors all the thread in the core
gets HMI. With existing code, the individual threads call opal hmi handle
independently which can easily throw TB out of sync if we have guest
running on subcores. Hence we will need to co-ordinate with all the
threads before making opal hmi handler call followed by TB resync.
This patch introduces a sibling subcore state structure (shared by all
threads in the core) in paca which holds information about whether sibling
subcores are in Guest mode or host mode. An array in_guest[] of size
MAX_SUBCORE_PER_CORE=4 is used to maintain the state of each subcore.
The subcore id is used as index into in_guest[] array. Only primary
thread entering/exiting the guest is responsible to set/unset its
designated array element.
On TB error, we get HMI interrupt on every thread on the core. Upon HMI,
this patch will now force guest to vacate the core/subcore. Primary
thread from each subcore will then turn off its respective bit
from the above bitmap during the guest exit path just after the
guest->host partition switch is complete.
All other threads that have just exited the guest OR were already in host
will wait until all other subcores clears their respective bit.
Once all the subcores turn off their respective bit, all threads will
will make call to opal hmi handler.
It is not necessary that opal hmi handler would resync the TB value for
every HMI interrupts. It would do so only for the HMI caused due to
TB errors. For rest, it would not touch TB value. Hence to make things
simpler, primary thread would call TB resync explicitly once for each
core immediately after opal hmi handler instead of subtracting guest
offset from TB. TB resync call will restore the TB with host value.
Thus we can be sure about the TB state.
One of the primary threads exiting the guest will take up the
responsibility of calling TB resync. It will use one of the top bits
(bit 63) from subcore state flags bitmap to make the decision. The first
primary thread (among the subcores) that is able to set the bit will
have to call the TB resync. Rest all other threads will wait until TB
resync is complete. Once TB resync is complete all threads will then
proceed.
Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-05-15 12:14:26 +08:00
|
|
|
cmpwi r6, 0xa /* Hypervisor maintenance ? */
|
|
|
|
|
beq 4f
|
KVM: PPC: Book3S HV: Consolidate code that checks reason for wake from nap
Currently in book3s_hv_rmhandlers.S we have three places where we
have woken up from nap mode and we check the reason field in SRR1
to see what event woke us up. This consolidates them into a new
function, kvmppc_check_wake_reason. It looks at the wake reason
field in SRR1, and if it indicates that an external interrupt caused
the wakeup, calls kvmppc_read_intr to check what sort of interrupt
it was.
This also consolidates the two places where we synthesize an external
interrupt (0x500 vector) for the guest. Now, if the guest exit code
finds that there was an external interrupt which has been handled
(i.e. it was an IPI indicating that there is now an interrupt pending
for the guest), it jumps to deliver_guest_interrupt, which is in the
last part of the guest entry code, where we synthesize guest external
and decrementer interrupts. That code has been streamlined a little
and now clears LPCR[MER] when appropriate as well as setting it.
The extra clearing of any pending IPI on a secondary, offline CPU
thread before going back to nap mode has been removed. It is no longer
necessary now that we have code to read and acknowledge IPIs in the
guest exit path.
This fixes a minor bug in the H_CEDE real-mode handling - previously,
if we found that other threads were already exiting the guest when we
were about to go to nap mode, we would branch to the cede wakeup path
and end up looking in SRR1 for a wakeup reason. Now we branch to a
point after we have checked the wakeup reason.
This also fixes a minor bug in kvmppc_read_intr - previously it could
return 0xff rather than 1, in the case where we find that a host IPI
is pending after we have cleared the IPI. Now it returns 1.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2014-01-08 18:25:25 +08:00
|
|
|
li r3, 1 /* anything else, return 1 */
|
|
|
|
|
0: blr
|
|
|
|
|
|
2014-01-08 18:25:28 +08:00
|
|
|
/* hypervisor doorbell */
|
|
|
|
|
3: li r12, BOOK3S_INTERRUPT_H_DOORBELL
|
2015-10-15 13:59:58 +08:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Clear the doorbell as we will invoke the handler
|
|
|
|
|
* explicitly in the guest exit path.
|
|
|
|
|
*/
|
|
|
|
|
lis r6, (PPC_DBELL_SERVER << (63-36))@h
|
|
|
|
|
PPC_MSGCLR(6)
|
KVM: PPC: Book3S HV: Use msgsnd for signalling threads on POWER8
This uses msgsnd where possible for signalling other threads within
the same core on POWER8 systems, rather than IPIs through the XICS
interrupt controller. This includes waking secondary threads to run
the guest, the interrupts generated by the virtual XICS, and the
interrupts to bring the other threads out of the guest when exiting.
Aggregated statistics from debugfs across vcpus for a guest with 32
vcpus, 8 threads/vcore, running on a POWER8, show this before the
change:
rm_entry: 3387.6ns (228 - 86600, 1008969 samples)
rm_exit: 4561.5ns (12 - 3477452, 1009402 samples)
rm_intr: 1660.0ns (12 - 553050, 3600051 samples)
and this after the change:
rm_entry: 3060.1ns (212 - 65138, 953873 samples)
rm_exit: 4244.1ns (12 - 9693408, 954331 samples)
rm_intr: 1342.3ns (12 - 1104718, 3405326 samples)
for a test of booting Fedora 20 big-endian to the login prompt.
The time taken for a H_PROD hcall (which is handled in the host
kernel) went down from about 35 microseconds to about 16 microseconds
with this change.
The noinline added to kvmppc_run_core turned out to be necessary for
good performance, at least with gcc 4.9.2 as packaged with Fedora 21
and a little-endian POWER8 host.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:12 +08:00
|
|
|
/* see if it's a host IPI */
|
2014-01-08 18:25:28 +08:00
|
|
|
li r3, 1
|
2017-10-10 18:18:28 +08:00
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
PPC_MSGSYNC
|
|
|
|
|
lwsync
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_300)
|
KVM: PPC: Book3S HV: Use msgsnd for signalling threads on POWER8
This uses msgsnd where possible for signalling other threads within
the same core on POWER8 systems, rather than IPIs through the XICS
interrupt controller. This includes waking secondary threads to run
the guest, the interrupts generated by the virtual XICS, and the
interrupts to bring the other threads out of the guest when exiting.
Aggregated statistics from debugfs across vcpus for a guest with 32
vcpus, 8 threads/vcore, running on a POWER8, show this before the
change:
rm_entry: 3387.6ns (228 - 86600, 1008969 samples)
rm_exit: 4561.5ns (12 - 3477452, 1009402 samples)
rm_intr: 1660.0ns (12 - 553050, 3600051 samples)
and this after the change:
rm_entry: 3060.1ns (212 - 65138, 953873 samples)
rm_exit: 4244.1ns (12 - 9693408, 954331 samples)
rm_intr: 1342.3ns (12 - 1104718, 3405326 samples)
for a test of booting Fedora 20 big-endian to the login prompt.
The time taken for a H_PROD hcall (which is handled in the host
kernel) went down from about 35 microseconds to about 16 microseconds
with this change.
The noinline added to kvmppc_run_core turned out to be necessary for
good performance, at least with gcc 4.9.2 as packaged with Fedora 21
and a little-endian POWER8 host.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:12 +08:00
|
|
|
lbz r0, HSTATE_HOST_IPI(r13)
|
|
|
|
|
cmpwi r0, 0
|
|
|
|
|
bnelr
|
2015-10-15 13:59:58 +08:00
|
|
|
/* if not, return -1 */
|
KVM: PPC: Book3S HV: Use msgsnd for signalling threads on POWER8
This uses msgsnd where possible for signalling other threads within
the same core on POWER8 systems, rather than IPIs through the XICS
interrupt controller. This includes waking secondary threads to run
the guest, the interrupts generated by the virtual XICS, and the
interrupts to bring the other threads out of the guest when exiting.
Aggregated statistics from debugfs across vcpus for a guest with 32
vcpus, 8 threads/vcore, running on a POWER8, show this before the
change:
rm_entry: 3387.6ns (228 - 86600, 1008969 samples)
rm_exit: 4561.5ns (12 - 3477452, 1009402 samples)
rm_intr: 1660.0ns (12 - 553050, 3600051 samples)
and this after the change:
rm_entry: 3060.1ns (212 - 65138, 953873 samples)
rm_exit: 4244.1ns (12 - 9693408, 954331 samples)
rm_intr: 1342.3ns (12 - 1104718, 3405326 samples)
for a test of booting Fedora 20 big-endian to the login prompt.
The time taken for a H_PROD hcall (which is handled in the host
kernel) went down from about 35 microseconds to about 16 microseconds
with this change.
The noinline added to kvmppc_run_core turned out to be necessary for
good performance, at least with gcc 4.9.2 as packaged with Fedora 21
and a little-endian POWER8 host.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:12 +08:00
|
|
|
li r3, -1
|
2014-01-08 18:25:28 +08:00
|
|
|
blr
|
|
|
|
|
|
KVM: PPC: Book3S HV: Fix TB corruption in guest exit path on HMI interrupt
When a guest is assigned to a core it converts the host Timebase (TB)
into guest TB by adding guest timebase offset before entering into
guest. During guest exit it restores the guest TB to host TB. This means
under certain conditions (Guest migration) host TB and guest TB can differ.
When we get an HMI for TB related issues the opal HMI handler would
try fixing errors and restore the correct host TB value. With no guest
running, we don't have any issues. But with guest running on the core
we run into TB corruption issues.
If we get an HMI while in the guest, the current HMI handler invokes opal
hmi handler before forcing guest to exit. The guest exit path subtracts
the guest TB offset from the current TB value which may have already
been restored with host value by opal hmi handler. This leads to incorrect
host and guest TB values.
With split-core, things become more complex. With split-core, TB also gets
split and each subcore gets its own TB register. When a hmi handler fixes
a TB error and restores the TB value, it affects all the TB values of
sibling subcores on the same core. On TB errors all the thread in the core
gets HMI. With existing code, the individual threads call opal hmi handle
independently which can easily throw TB out of sync if we have guest
running on subcores. Hence we will need to co-ordinate with all the
threads before making opal hmi handler call followed by TB resync.
This patch introduces a sibling subcore state structure (shared by all
threads in the core) in paca which holds information about whether sibling
subcores are in Guest mode or host mode. An array in_guest[] of size
MAX_SUBCORE_PER_CORE=4 is used to maintain the state of each subcore.
The subcore id is used as index into in_guest[] array. Only primary
thread entering/exiting the guest is responsible to set/unset its
designated array element.
On TB error, we get HMI interrupt on every thread on the core. Upon HMI,
this patch will now force guest to vacate the core/subcore. Primary
thread from each subcore will then turn off its respective bit
from the above bitmap during the guest exit path just after the
guest->host partition switch is complete.
All other threads that have just exited the guest OR were already in host
will wait until all other subcores clears their respective bit.
Once all the subcores turn off their respective bit, all threads will
will make call to opal hmi handler.
It is not necessary that opal hmi handler would resync the TB value for
every HMI interrupts. It would do so only for the HMI caused due to
TB errors. For rest, it would not touch TB value. Hence to make things
simpler, primary thread would call TB resync explicitly once for each
core immediately after opal hmi handler instead of subtracting guest
offset from TB. TB resync call will restore the TB with host value.
Thus we can be sure about the TB state.
One of the primary threads exiting the guest will take up the
responsibility of calling TB resync. It will use one of the top bits
(bit 63) from subcore state flags bitmap to make the decision. The first
primary thread (among the subcores) that is able to set the bit will
have to call the TB resync. Rest all other threads will wait until TB
resync is complete. Once TB resync is complete all threads will then
proceed.
Signed-off-by: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2016-05-15 12:14:26 +08:00
|
|
|
/* Woken up due to Hypervisor maintenance interrupt */
|
|
|
|
|
4: li r12, BOOK3S_INTERRUPT_HMI
|
|
|
|
|
li r3, 1
|
|
|
|
|
blr
|
|
|
|
|
|
2016-08-19 13:35:46 +08:00
|
|
|
/* external interrupt - create a stack frame so we can call C */
|
|
|
|
|
7: mflr r0
|
|
|
|
|
std r0, PPC_LR_STKOFF(r1)
|
|
|
|
|
stdu r1, -PPC_MIN_STKFRM(r1)
|
|
|
|
|
bl kvmppc_read_intr
|
|
|
|
|
nop
|
|
|
|
|
li r12, BOOK3S_INTERRUPT_EXTERNAL
|
2016-08-19 13:35:52 +08:00
|
|
|
cmpdi r3, 1
|
|
|
|
|
ble 1f
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Return code of 2 means PCI passthrough interrupt, but
|
|
|
|
|
* we need to return back to host to complete handling the
|
|
|
|
|
* interrupt. Trap reason is expected in r12 by guest
|
|
|
|
|
* exit code.
|
|
|
|
|
*/
|
|
|
|
|
li r12, BOOK3S_INTERRUPT_HV_RM_HARD
|
|
|
|
|
1:
|
2016-08-19 13:35:46 +08:00
|
|
|
ld r0, PPC_MIN_STKFRM+PPC_LR_STKOFF(r1)
|
|
|
|
|
addi r1, r1, PPC_MIN_STKFRM
|
|
|
|
|
mtlr r0
|
|
|
|
|
blr
|
KVM: PPC: Allow book3s_hv guests to use SMT processor modes
This lifts the restriction that book3s_hv guests can only run one
hardware thread per core, and allows them to use up to 4 threads
per core on POWER7. The host still has to run single-threaded.
This capability is advertised to qemu through a new KVM_CAP_PPC_SMT
capability. The return value of the ioctl querying this capability
is the number of vcpus per virtual CPU core (vcore), currently 4.
To use this, the host kernel should be booted with all threads
active, and then all the secondary threads should be offlined.
This will put the secondary threads into nap mode. KVM will then
wake them from nap mode and use them for running guest code (while
they are still offline). To wake the secondary threads, we send
them an IPI using a new xics_wake_cpu() function, implemented in
arch/powerpc/sysdev/xics/icp-native.c. In other words, at this stage
we assume that the platform has a XICS interrupt controller and
we are using icp-native.c to drive it. Since the woken thread will
need to acknowledge and clear the IPI, we also export the base
physical address of the XICS registers using kvmppc_set_xics_phys()
for use in the low-level KVM book3s code.
When a vcpu is created, it is assigned to a virtual CPU core.
The vcore number is obtained by dividing the vcpu number by the
number of threads per core in the host. This number is exported
to userspace via the KVM_CAP_PPC_SMT capability. If qemu wishes
to run the guest in single-threaded mode, it should make all vcpu
numbers be multiples of the number of threads per core.
We distinguish three states of a vcpu: runnable (i.e., ready to execute
the guest), blocked (that is, idle), and busy in host. We currently
implement a policy that the vcore can run only when all its threads
are runnable or blocked. This way, if a vcpu needs to execute elsewhere
in the kernel or in qemu, it can do so without being starved of CPU
by the other vcpus.
When a vcore starts to run, it executes in the context of one of the
vcpu threads. The other vcpu threads all go to sleep and stay asleep
until something happens requiring the vcpu thread to return to qemu,
or to wake up to run the vcore (this can happen when another vcpu
thread goes from busy in host state to blocked).
It can happen that a vcpu goes from blocked to runnable state (e.g.
because of an interrupt), and the vcore it belongs to is already
running. In that case it can start to run immediately as long as
the none of the vcpus in the vcore have started to exit the guest.
We send the next free thread in the vcore an IPI to get it to start
to execute the guest. It synchronizes with the other threads via
the vcore->entry_exit_count field to make sure that it doesn't go
into the guest if the other vcpus are exiting by the time that it
is ready to actually enter the guest.
Note that there is no fixed relationship between the hardware thread
number and the vcpu number. Hardware threads are assigned to vcpus
as they become runnable, so we will always use the lower-numbered
hardware threads in preference to higher-numbered threads if not all
the vcpus in the vcore are runnable, regardless of which vcpus are
runnable.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:23:08 +08:00
|
|
|
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
/*
|
|
|
|
|
* Save away FP, VMX and VSX registers.
|
|
|
|
|
* r3 = vcpu pointer
|
2013-10-15 17:43:04 +08:00
|
|
|
* N.B. r30 and r31 are volatile across this function,
|
|
|
|
|
* thus it is not callable from C.
|
2011-06-29 08:22:05 +08:00
|
|
|
*/
|
2013-10-15 17:43:04 +08:00
|
|
|
kvmppc_save_fp:
|
|
|
|
|
mflr r30
|
|
|
|
|
mr r31,r3
|
2012-03-02 09:38:23 +08:00
|
|
|
mfmsr r5
|
|
|
|
|
ori r8,r5,MSR_FP
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
#ifdef CONFIG_ALTIVEC
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
oris r8,r8,MSR_VEC@h
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
|
|
|
|
|
#endif
|
|
|
|
|
#ifdef CONFIG_VSX
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
oris r8,r8,MSR_VSX@h
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_VSX)
|
|
|
|
|
#endif
|
|
|
|
|
mtmsrd r8
|
2013-10-15 17:43:04 +08:00
|
|
|
addi r3,r3,VCPU_FPRS
|
2014-06-16 20:41:15 +08:00
|
|
|
bl store_fp_state
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
#ifdef CONFIG_ALTIVEC
|
|
|
|
|
BEGIN_FTR_SECTION
|
2013-10-15 17:43:04 +08:00
|
|
|
addi r3,r31,VCPU_VRS
|
2014-06-16 20:41:15 +08:00
|
|
|
bl store_vr_state
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
|
|
|
|
|
#endif
|
|
|
|
|
mfspr r6,SPRN_VRSAVE
|
2014-03-13 17:02:48 +08:00
|
|
|
stw r6,VCPU_VRSAVE(r31)
|
2013-10-15 17:43:04 +08:00
|
|
|
mtlr r30
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Load up FP, VMX and VSX registers
|
|
|
|
|
* r4 = vcpu pointer
|
2013-10-15 17:43:04 +08:00
|
|
|
* N.B. r30 and r31 are volatile across this function,
|
|
|
|
|
* thus it is not callable from C.
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
*/
|
|
|
|
|
kvmppc_load_fp:
|
2013-10-15 17:43:04 +08:00
|
|
|
mflr r30
|
|
|
|
|
mr r31,r4
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
mfmsr r9
|
|
|
|
|
ori r8,r9,MSR_FP
|
|
|
|
|
#ifdef CONFIG_ALTIVEC
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
oris r8,r8,MSR_VEC@h
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
|
|
|
|
|
#endif
|
|
|
|
|
#ifdef CONFIG_VSX
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
oris r8,r8,MSR_VSX@h
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_VSX)
|
|
|
|
|
#endif
|
|
|
|
|
mtmsrd r8
|
2013-10-15 17:43:04 +08:00
|
|
|
addi r3,r4,VCPU_FPRS
|
2014-06-16 20:41:15 +08:00
|
|
|
bl load_fp_state
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
#ifdef CONFIG_ALTIVEC
|
|
|
|
|
BEGIN_FTR_SECTION
|
2013-10-15 17:43:04 +08:00
|
|
|
addi r3,r31,VCPU_VRS
|
2014-06-16 20:41:15 +08:00
|
|
|
bl load_vr_state
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
|
|
|
|
|
#endif
|
2014-03-13 17:02:48 +08:00
|
|
|
lwz r7,VCPU_VRSAVE(r31)
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
mtspr SPRN_VRSAVE,r7
|
2013-10-15 17:43:04 +08:00
|
|
|
mtlr r30
|
|
|
|
|
mr r4,r31
|
KVM: PPC: Add support for Book3S processors in hypervisor mode
This adds support for KVM running on 64-bit Book 3S processors,
specifically POWER7, in hypervisor mode. Using hypervisor mode means
that the guest can use the processor's supervisor mode. That means
that the guest can execute privileged instructions and access privileged
registers itself without trapping to the host. This gives excellent
performance, but does mean that KVM cannot emulate a processor
architecture other than the one that the hardware implements.
This code assumes that the guest is running paravirtualized using the
PAPR (Power Architecture Platform Requirements) interface, which is the
interface that IBM's PowerVM hypervisor uses. That means that existing
Linux distributions that run on IBM pSeries machines will also run
under KVM without modification. In order to communicate the PAPR
hypercalls to qemu, this adds a new KVM_EXIT_PAPR_HCALL exit code
to include/linux/kvm.h.
Currently the choice between book3s_hv support and book3s_pr support
(i.e. the existing code, which runs the guest in user mode) has to be
made at kernel configuration time, so a given kernel binary can only
do one or the other.
This new book3s_hv code doesn't support MMIO emulation at present.
Since we are running paravirtualized guests, this isn't a serious
restriction.
With the guest running in supervisor mode, most exceptions go straight
to the guest. We will never get data or instruction storage or segment
interrupts, alignment interrupts, decrementer interrupts, program
interrupts, single-step interrupts, etc., coming to the hypervisor from
the guest. Therefore this introduces a new KVMTEST_NONHV macro for the
exception entry path so that we don't have to do the KVM test on entry
to those exception handlers.
We do however get hypervisor decrementer, hypervisor data storage,
hypervisor instruction storage, and hypervisor emulation assist
interrupts, so we have to handle those.
In hypervisor mode, real-mode accesses can access all of RAM, not just
a limited amount. Therefore we put all the guest state in the vcpu.arch
and use the shadow_vcpu in the PACA only for temporary scratch space.
We allocate the vcpu with kzalloc rather than vzalloc, and we don't use
anything in the kvmppc_vcpu_book3s struct, so we don't allocate it.
We don't have a shared page with the guest, but we still need a
kvm_vcpu_arch_shared struct to store the values of various registers,
so we include one in the vcpu_arch struct.
The POWER7 processor has a restriction that all threads in a core have
to be in the same partition. MMU-on kernel code counts as a partition
(partition 0), so we have to do a partition switch on every entry to and
exit from the guest. At present we require the host and guest to run
in single-thread mode because of this hardware restriction.
This code allocates a hashed page table for the guest and initializes
it with HPTEs for the guest's Virtual Real Memory Area (VRMA). We
require that the guest memory is allocated using 16MB huge pages, in
order to simplify the low-level memory management. This also means that
we can get away without tracking paging activity in the host for now,
since huge pages can't be paged or swapped.
This also adds a few new exports needed by the book3s_hv code.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2011-06-29 08:21:34 +08:00
|
|
|
blr
|
KVM: PPC: Book3S HV: Better handling of exceptions that happen in real mode
When an interrupt or exception happens in the guest that comes to the
host, the CPU goes to hypervisor real mode (MMU off) to handle the
exception but doesn't change the MMU context. After saving a few
registers, we then clear the "in guest" flag. If, for any reason,
we get an exception in the real-mode code, that then gets handled
by the normal kernel exception handlers, which turn the MMU on. This
is disastrous if the MMU is still set to the guest context, since we
end up executing instructions from random places in the guest kernel
with hypervisor privilege.
In order to catch this situation, we define a new value for the "in guest"
flag, KVM_GUEST_MODE_HOST_HV, to indicate that we are in hypervisor real
mode with guest MMU context. If the "in guest" flag is set to this value,
we branch off to an emergency handler. For the moment, this just does
a branch to self to stop the CPU from doing anything further.
While we're here, we define another new flag value to indicate that we
are in a HV guest, as distinct from a PR guest. This will be useful
when we have a kernel that can support both PR and HV guests concurrently.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-10-04 19:45:04 +08:00
|
|
|
|
2016-06-22 12:21:59 +08:00
|
|
|
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
|
|
|
|
|
/*
|
|
|
|
|
* Save transactional state and TM-related registers.
|
2018-05-23 15:01:48 +08:00
|
|
|
* Called with r3 pointing to the vcpu struct and r4 containing
|
|
|
|
|
* the guest MSR value.
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
* r5 is non-zero iff non-volatile register state needs to be maintained.
|
|
|
|
|
* If r5 == 0, this can modify all checkpointed registers, but
|
2018-05-23 15:01:48 +08:00
|
|
|
* restores r1 and r2 before exit.
|
2016-06-22 12:21:59 +08:00
|
|
|
*/
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
_GLOBAL_TOC(kvmppc_save_tm_hv)
|
|
|
|
|
EXPORT_SYMBOL_GPL(kvmppc_save_tm_hv)
|
2018-05-30 18:07:52 +08:00
|
|
|
/* See if we need to handle fake suspend mode */
|
|
|
|
|
BEGIN_FTR_SECTION
|
2018-05-23 15:01:50 +08:00
|
|
|
b __kvmppc_save_tm
|
2018-05-30 18:07:52 +08:00
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
|
|
|
|
|
|
|
|
|
|
lbz r0, HSTATE_FAKE_SUSPEND(r13) /* Were we fake suspended? */
|
|
|
|
|
cmpwi r0, 0
|
2018-05-23 15:01:50 +08:00
|
|
|
beq __kvmppc_save_tm
|
2018-05-30 18:07:52 +08:00
|
|
|
|
|
|
|
|
/* The following code handles the fake_suspend = 1 case */
|
2016-06-22 12:21:59 +08:00
|
|
|
mflr r0
|
|
|
|
|
std r0, PPC_LR_STKOFF(r1)
|
2018-03-21 18:32:02 +08:00
|
|
|
stdu r1, -PPC_MIN_STKFRM(r1)
|
2016-06-22 12:21:59 +08:00
|
|
|
|
|
|
|
|
/* Turn on TM. */
|
|
|
|
|
mfmsr r8
|
|
|
|
|
li r0, 1
|
|
|
|
|
rldimi r8, r0, MSR_TM_LG, 63-MSR_TM_LG
|
|
|
|
|
mtmsrd r8
|
|
|
|
|
|
2018-03-21 18:32:02 +08:00
|
|
|
rldicl. r8, r8, 64 - MSR_TS_S_LG, 62 /* Did we actually hrfid? */
|
|
|
|
|
beq 4f
|
2018-05-30 18:07:52 +08:00
|
|
|
BEGIN_FTR_SECTION
|
2018-03-21 18:32:02 +08:00
|
|
|
bl pnv_power9_force_smt4_catch
|
2018-05-30 18:07:52 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
|
2018-03-21 18:32:02 +08:00
|
|
|
nop
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
|
2018-05-30 18:07:52 +08:00
|
|
|
/* We have to treclaim here because that's the only way to do S->N */
|
|
|
|
|
li r3, TM_CAUSE_KVM_RESCHED
|
2016-06-22 12:21:59 +08:00
|
|
|
TRECLAIM(R3)
|
|
|
|
|
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
/*
|
|
|
|
|
* We were in fake suspend, so we are not going to save the
|
|
|
|
|
* register state as the guest checkpointed state (since
|
|
|
|
|
* we already have it), therefore we can now use any volatile GPR.
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
* In fact treclaim in fake suspend state doesn't modify
|
|
|
|
|
* any registers.
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
*/
|
2018-05-30 18:07:52 +08:00
|
|
|
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
BEGIN_FTR_SECTION
|
2018-03-21 18:32:02 +08:00
|
|
|
bl pnv_power9_force_smt4_release
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_P9_TM_XER_SO_BUG)
|
2018-03-21 18:32:02 +08:00
|
|
|
nop
|
|
|
|
|
|
|
|
|
|
4:
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
mfspr r3, SPRN_PSSCR
|
|
|
|
|
/* PSSCR_FAKE_SUSPEND is a write-only bit, but clear it anyway */
|
|
|
|
|
li r0, PSSCR_FAKE_SUSPEND
|
|
|
|
|
andc r3, r3, r0
|
|
|
|
|
mtspr SPRN_PSSCR, r3
|
|
|
|
|
|
KVM: PPC: Book3S HV: Work around TEXASR bug in fake suspend state
This works around a hardware bug in "Nimbus" POWER9 DD2.2 processors,
where the contents of the TEXASR can get corrupted while a thread is
in fake suspend state. The workaround is for the instruction emulation
code to use the value saved at the most recent guest exit in real
suspend mode. We achieve this by simply not saving the TEXASR into
the vcpu struct on an exit in fake suspend state. We also have to
take care to set the orig_texasr field only on guest exit in real
suspend state.
This also means that on guest entry in fake suspend state, TEXASR
will be restored to the value it had on the last exit in real suspend
state, effectively counteracting any hardware-caused corruption. This
works because TEXASR may not be written in suspend state.
With this, the guest might see the wrong values in TEXASR if it reads
it while in suspend state, but will see the correct value in
non-transactional state (e.g. after a treclaim), and treclaim will
work correctly.
With this workaround, the code will actually run slightly faster, and
will operate correctly on systems without the TEXASR bug (since TEXASR
may not be written in suspend state, and is only changed by failure
recording, which will have already been done before we get into fake
suspend state). Therefore these changes are not made subject to a CPU
feature bit.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:03 +08:00
|
|
|
/* Don't save TEXASR, use value from last exit in real suspend state */
|
2016-06-22 12:21:59 +08:00
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
mfspr r5, SPRN_TFHAR
|
|
|
|
|
mfspr r6, SPRN_TFIAR
|
|
|
|
|
std r5, VCPU_TFHAR(r9)
|
|
|
|
|
std r6, VCPU_TFIAR(r9)
|
|
|
|
|
|
2018-03-21 18:32:02 +08:00
|
|
|
addi r1, r1, PPC_MIN_STKFRM
|
2016-06-22 12:21:59 +08:00
|
|
|
ld r0, PPC_LR_STKOFF(r1)
|
|
|
|
|
mtlr r0
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Restore transactional state and TM-related registers.
|
2018-05-23 15:01:48 +08:00
|
|
|
* Called with r3 pointing to the vcpu struct
|
|
|
|
|
* and r4 containing the guest MSR value.
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
* r5 is non-zero iff non-volatile register state needs to be maintained.
|
2016-06-22 12:21:59 +08:00
|
|
|
* This potentially modifies all checkpointed registers.
|
2018-05-23 15:01:48 +08:00
|
|
|
* It restores r1 and r2 from the PACA.
|
2016-06-22 12:21:59 +08:00
|
|
|
*/
|
KVM: PPC: Book3S: Rework TM save/restore code and make it C-callable
This adds a parameter to __kvmppc_save_tm and __kvmppc_restore_tm
which allows the caller to indicate whether it wants the nonvolatile
register state to be preserved across the call, as required by the C
calling conventions. This parameter being non-zero also causes the
MSR bits that enable TM, FP, VMX and VSX to be preserved. The
condition register and DSCR are now always preserved.
With this, kvmppc_save_tm_hv and kvmppc_restore_tm_hv can be called
from C code provided the 3rd parameter is non-zero. So that these
functions can be called from modules, they now include code to set
the TOC pointer (r2) on entry, as they can call other built-in C
functions which will assume the TOC to have been set.
Also, the fake suspend code in kvmppc_save_tm_hv is modified here to
assume that treclaim in fake-suspend state does not modify any registers,
which is the case on POWER9. This enables the code to be simplified
quite a bit.
_kvmppc_save_tm_pr and _kvmppc_restore_tm_pr become much simpler with
this change, since they now only need to save and restore TAR and pass
1 for the 3rd argument to __kvmppc_{save,restore}_tm.
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-10-08 13:30:53 +08:00
|
|
|
_GLOBAL_TOC(kvmppc_restore_tm_hv)
|
|
|
|
|
EXPORT_SYMBOL_GPL(kvmppc_restore_tm_hv)
|
2018-05-30 18:07:52 +08:00
|
|
|
/*
|
|
|
|
|
* If we are doing TM emulation for the guest on a POWER9 DD2,
|
|
|
|
|
* then we don't actually do a trechkpt -- we either set up
|
|
|
|
|
* fake-suspend mode, or emulate a TM rollback.
|
|
|
|
|
*/
|
|
|
|
|
BEGIN_FTR_SECTION
|
2018-05-23 15:01:50 +08:00
|
|
|
b __kvmppc_restore_tm
|
2018-05-30 18:07:52 +08:00
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_P9_TM_HV_ASSIST)
|
2016-06-22 12:21:59 +08:00
|
|
|
mflr r0
|
|
|
|
|
std r0, PPC_LR_STKOFF(r1)
|
|
|
|
|
|
2018-05-30 18:07:52 +08:00
|
|
|
li r0, 0
|
|
|
|
|
stb r0, HSTATE_FAKE_SUSPEND(r13)
|
|
|
|
|
|
|
|
|
|
/* Turn on TM so we can restore TM SPRs */
|
2016-06-22 12:21:59 +08:00
|
|
|
mfmsr r5
|
2018-05-30 18:07:52 +08:00
|
|
|
li r0, 1
|
|
|
|
|
rldimi r5, r0, MSR_TM_LG, 63-MSR_TM_LG
|
2016-06-22 12:21:59 +08:00
|
|
|
mtmsrd r5
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* The user may change these outside of a transaction, so they must
|
|
|
|
|
* always be context switched.
|
|
|
|
|
*/
|
2018-05-23 15:01:48 +08:00
|
|
|
ld r5, VCPU_TFHAR(r3)
|
|
|
|
|
ld r6, VCPU_TFIAR(r3)
|
|
|
|
|
ld r7, VCPU_TEXASR(r3)
|
2016-06-22 12:21:59 +08:00
|
|
|
mtspr SPRN_TFHAR, r5
|
|
|
|
|
mtspr SPRN_TFIAR, r6
|
|
|
|
|
mtspr SPRN_TEXASR, r7
|
|
|
|
|
|
2018-05-23 15:01:48 +08:00
|
|
|
rldicl. r5, r4, 64 - MSR_TS_S_LG, 62
|
2016-06-22 12:21:59 +08:00
|
|
|
beqlr /* TM not active in guest */
|
|
|
|
|
|
2018-05-30 18:07:52 +08:00
|
|
|
/* Make sure the failure summary is set */
|
2016-06-22 12:21:59 +08:00
|
|
|
oris r7, r7, (TEXASR_FS)@h
|
|
|
|
|
mtspr SPRN_TEXASR, r7
|
|
|
|
|
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
cmpwi r5, 1 /* check for suspended state */
|
|
|
|
|
bgt 10f
|
|
|
|
|
stb r5, HSTATE_FAKE_SUSPEND(r13)
|
2018-05-30 18:07:52 +08:00
|
|
|
b 9f /* and return */
|
KVM: PPC: Book3S HV: Work around transactional memory bugs in POWER9
POWER9 has hardware bugs relating to transactional memory and thread
reconfiguration (changes to hardware SMT mode). Specifically, the core
does not have enough storage to store a complete checkpoint of all the
architected state for all four threads. The DD2.2 version of POWER9
includes hardware modifications designed to allow hypervisor software
to implement workarounds for these problems. This patch implements
those workarounds in KVM code so that KVM guests see a full, working
transactional memory implementation.
The problems center around the use of TM suspended state, where the
CPU has a checkpointed state but execution is not transactional. The
workaround is to implement a "fake suspend" state, which looks to the
guest like suspended state but the CPU does not store a checkpoint.
In this state, any instruction that would cause a transition to
transactional state (rfid, rfebb, mtmsrd, tresume) or would use the
checkpointed state (treclaim) causes a "soft patch" interrupt (vector
0x1500) to the hypervisor so that it can be emulated. The trechkpt
instruction also causes a soft patch interrupt.
On POWER9 DD2.2, we avoid returning to the guest in any state which
would require a checkpoint to be present. The trechkpt in the guest
entry path which would normally create that checkpoint is replaced by
either a transition to fake suspend state, if the guest is in suspend
state, or a rollback to the pre-transactional state if the guest is in
transactional state. Fake suspend state is indicated by a flag in the
PACA plus a new bit in the PSSCR. The new PSSCR bit is write-only and
reads back as 0.
On exit from the guest, if the guest is in fake suspend state, we still
do the treclaim instruction as we would in real suspend state, in order
to get into non-transactional state, but we do not save the resulting
register state since there was no checkpoint.
Emulation of the instructions that cause a softpatch interrupt is
handled in two paths. If the guest is in real suspend mode, we call
kvmhv_p9_tm_emulation_early() to handle the cases where the guest is
transitioning to transactional state. This is called before we do the
treclaim in the guest exit path; because we haven't done treclaim, we
can get back to the guest with the transaction still active. If the
instruction is a case that kvmhv_p9_tm_emulation_early() doesn't
handle, or if the guest is in fake suspend state, then we proceed to
do the complete guest exit path and subsequently call
kvmhv_p9_tm_emulation() in host context with the MMU on. This handles
all the cases including the cases that generate program interrupts
(illegal instruction or TM Bad Thing) and facility unavailable
interrupts.
The emulation is reasonably straightforward and is mostly concerned
with checking for exception conditions and updating the state of
registers such as MSR and CR0. The treclaim emulation takes care to
ensure that the TEXASR register gets updated as if it were the guest
treclaim instruction that had done failure recording, not the treclaim
done in hypervisor state in the guest exit path.
With this, the KVM_CAP_PPC_HTM capability returns true (1) even if
transactional memory is not available to host userspace.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-03-21 18:32:01 +08:00
|
|
|
10: stdu r1, -PPC_MIN_STKFRM(r1)
|
|
|
|
|
/* guest is in transactional state, so simulate rollback */
|
|
|
|
|
bl kvmhv_emulate_tm_rollback
|
|
|
|
|
nop
|
|
|
|
|
addi r1, r1, PPC_MIN_STKFRM
|
2018-05-30 18:07:52 +08:00
|
|
|
9: ld r0, PPC_LR_STKOFF(r1)
|
|
|
|
|
mtlr r0
|
|
|
|
|
blr
|
|
|
|
|
#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
|
2016-06-22 12:21:59 +08:00
|
|
|
|
KVM: PPC: Book3S HV: Better handling of exceptions that happen in real mode
When an interrupt or exception happens in the guest that comes to the
host, the CPU goes to hypervisor real mode (MMU off) to handle the
exception but doesn't change the MMU context. After saving a few
registers, we then clear the "in guest" flag. If, for any reason,
we get an exception in the real-mode code, that then gets handled
by the normal kernel exception handlers, which turn the MMU on. This
is disastrous if the MMU is still set to the guest context, since we
end up executing instructions from random places in the guest kernel
with hypervisor privilege.
In order to catch this situation, we define a new value for the "in guest"
flag, KVM_GUEST_MODE_HOST_HV, to indicate that we are in hypervisor real
mode with guest MMU context. If the "in guest" flag is set to this value,
we branch off to an emergency handler. For the moment, this just does
a branch to self to stop the CPU from doing anything further.
While we're here, we define another new flag value to indicate that we
are in a HV guest, as distinct from a PR guest. This will be useful
when we have a kernel that can support both PR and HV guests concurrently.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-10-04 19:45:04 +08:00
|
|
|
/*
|
|
|
|
|
* We come here if we get any exception or interrupt while we are
|
|
|
|
|
* executing host real mode code while in guest MMU context.
|
KVM: PPC: Book3S HV: Handle unexpected interrupts better
At present, if an interrupt (i.e. an exception or trap) occurs in the
code where KVM is switching the MMU to or from guest context, we jump
to kvmppc_bad_host_intr, where we simply spin with interrupts disabled.
In this situation, it is hard to debug what happened because we get no
indication as to which interrupt occurred or where. Typically we get
a cascade of stall and soft lockup warnings from other CPUs.
In order to get more information for debugging, this adds code to
create a stack frame on the emergency stack and save register values
to it. We start half-way down the emergency stack in order to give
ourselves some chance of being able to do a stack trace on secondary
threads that are already on the emergency stack.
On POWER7 or POWER8, we then just spin, as before, because we don't
know what state the MMU context is in or what other threads are doing,
and we can't switch back to host context without coordinating with
other threads. On POWER9 we can do better; there we load up the host
MMU context and jump to C code, which prints an oops message to the
console and panics.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-01 14:17:27 +08:00
|
|
|
* r12 is (CR << 32) | vector
|
|
|
|
|
* r13 points to our PACA
|
|
|
|
|
* r12 is saved in HSTATE_SCRATCH0(r13)
|
|
|
|
|
* ctr is saved in HSTATE_SCRATCH1(r13) if RELOCATABLE
|
|
|
|
|
* r9 is saved in HSTATE_SCRATCH2(r13)
|
|
|
|
|
* r13 is saved in HSPRG1
|
|
|
|
|
* cfar is saved in HSTATE_CFAR(r13)
|
|
|
|
|
* ppr is saved in HSTATE_PPR(r13)
|
KVM: PPC: Book3S HV: Better handling of exceptions that happen in real mode
When an interrupt or exception happens in the guest that comes to the
host, the CPU goes to hypervisor real mode (MMU off) to handle the
exception but doesn't change the MMU context. After saving a few
registers, we then clear the "in guest" flag. If, for any reason,
we get an exception in the real-mode code, that then gets handled
by the normal kernel exception handlers, which turn the MMU on. This
is disastrous if the MMU is still set to the guest context, since we
end up executing instructions from random places in the guest kernel
with hypervisor privilege.
In order to catch this situation, we define a new value for the "in guest"
flag, KVM_GUEST_MODE_HOST_HV, to indicate that we are in hypervisor real
mode with guest MMU context. If the "in guest" flag is set to this value,
we branch off to an emergency handler. For the moment, this just does
a branch to self to stop the CPU from doing anything further.
While we're here, we define another new flag value to indicate that we
are in a HV guest, as distinct from a PR guest. This will be useful
when we have a kernel that can support both PR and HV guests concurrently.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-10-04 19:45:04 +08:00
|
|
|
*/
|
|
|
|
|
kvmppc_bad_host_intr:
|
KVM: PPC: Book3S HV: Handle unexpected interrupts better
At present, if an interrupt (i.e. an exception or trap) occurs in the
code where KVM is switching the MMU to or from guest context, we jump
to kvmppc_bad_host_intr, where we simply spin with interrupts disabled.
In this situation, it is hard to debug what happened because we get no
indication as to which interrupt occurred or where. Typically we get
a cascade of stall and soft lockup warnings from other CPUs.
In order to get more information for debugging, this adds code to
create a stack frame on the emergency stack and save register values
to it. We start half-way down the emergency stack in order to give
ourselves some chance of being able to do a stack trace on secondary
threads that are already on the emergency stack.
On POWER7 or POWER8, we then just spin, as before, because we don't
know what state the MMU context is in or what other threads are doing,
and we can't switch back to host context without coordinating with
other threads. On POWER9 we can do better; there we load up the host
MMU context and jump to C code, which prints an oops message to the
console and panics.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-01 14:17:27 +08:00
|
|
|
/*
|
|
|
|
|
* Switch to the emergency stack, but start half-way down in
|
|
|
|
|
* case we were already on it.
|
|
|
|
|
*/
|
|
|
|
|
mr r9, r1
|
|
|
|
|
std r1, PACAR1(r13)
|
|
|
|
|
ld r1, PACAEMERGSP(r13)
|
|
|
|
|
subi r1, r1, THREAD_SIZE/2 + INT_FRAME_SIZE
|
|
|
|
|
std r9, 0(r1)
|
|
|
|
|
std r0, GPR0(r1)
|
|
|
|
|
std r9, GPR1(r1)
|
|
|
|
|
std r2, GPR2(r1)
|
|
|
|
|
SAVE_4GPRS(3, r1)
|
|
|
|
|
SAVE_2GPRS(7, r1)
|
|
|
|
|
srdi r0, r12, 32
|
|
|
|
|
clrldi r12, r12, 32
|
|
|
|
|
std r0, _CCR(r1)
|
|
|
|
|
std r12, _TRAP(r1)
|
|
|
|
|
andi. r0, r12, 2
|
|
|
|
|
beq 1f
|
|
|
|
|
mfspr r3, SPRN_HSRR0
|
|
|
|
|
mfspr r4, SPRN_HSRR1
|
|
|
|
|
mfspr r5, SPRN_HDAR
|
|
|
|
|
mfspr r6, SPRN_HDSISR
|
|
|
|
|
b 2f
|
|
|
|
|
1: mfspr r3, SPRN_SRR0
|
|
|
|
|
mfspr r4, SPRN_SRR1
|
|
|
|
|
mfspr r5, SPRN_DAR
|
|
|
|
|
mfspr r6, SPRN_DSISR
|
|
|
|
|
2: std r3, _NIP(r1)
|
|
|
|
|
std r4, _MSR(r1)
|
|
|
|
|
std r5, _DAR(r1)
|
|
|
|
|
std r6, _DSISR(r1)
|
|
|
|
|
ld r9, HSTATE_SCRATCH2(r13)
|
|
|
|
|
ld r12, HSTATE_SCRATCH0(r13)
|
|
|
|
|
GET_SCRATCH0(r0)
|
|
|
|
|
SAVE_4GPRS(9, r1)
|
|
|
|
|
std r0, GPR13(r1)
|
|
|
|
|
SAVE_NVGPRS(r1)
|
|
|
|
|
ld r5, HSTATE_CFAR(r13)
|
|
|
|
|
std r5, ORIG_GPR3(r1)
|
|
|
|
|
mflr r3
|
|
|
|
|
#ifdef CONFIG_RELOCATABLE
|
|
|
|
|
ld r4, HSTATE_SCRATCH1(r13)
|
|
|
|
|
#else
|
|
|
|
|
mfctr r4
|
|
|
|
|
#endif
|
|
|
|
|
mfxer r5
|
2017-12-20 11:55:50 +08:00
|
|
|
lbz r6, PACAIRQSOFTMASK(r13)
|
KVM: PPC: Book3S HV: Handle unexpected interrupts better
At present, if an interrupt (i.e. an exception or trap) occurs in the
code where KVM is switching the MMU to or from guest context, we jump
to kvmppc_bad_host_intr, where we simply spin with interrupts disabled.
In this situation, it is hard to debug what happened because we get no
indication as to which interrupt occurred or where. Typically we get
a cascade of stall and soft lockup warnings from other CPUs.
In order to get more information for debugging, this adds code to
create a stack frame on the emergency stack and save register values
to it. We start half-way down the emergency stack in order to give
ourselves some chance of being able to do a stack trace on secondary
threads that are already on the emergency stack.
On POWER7 or POWER8, we then just spin, as before, because we don't
know what state the MMU context is in or what other threads are doing,
and we can't switch back to host context without coordinating with
other threads. On POWER9 we can do better; there we load up the host
MMU context and jump to C code, which prints an oops message to the
console and panics.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-01 14:17:27 +08:00
|
|
|
std r3, _LINK(r1)
|
|
|
|
|
std r4, _CTR(r1)
|
|
|
|
|
std r5, _XER(r1)
|
|
|
|
|
std r6, SOFTE(r1)
|
|
|
|
|
ld r2, PACATOC(r13)
|
|
|
|
|
LOAD_REG_IMMEDIATE(3, 0x7265677368657265)
|
|
|
|
|
std r3, STACK_FRAME_OVERHEAD-16(r1)
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* On POWER9 do a minimal restore of the MMU and call C code,
|
|
|
|
|
* which will print a message and panic.
|
|
|
|
|
* XXX On POWER7 and POWER8, we just spin here since we don't
|
|
|
|
|
* know what the other threads are doing (and we don't want to
|
|
|
|
|
* coordinate with them) - but at least we now have register state
|
|
|
|
|
* in memory that we might be able to look at from another CPU.
|
|
|
|
|
*/
|
|
|
|
|
BEGIN_FTR_SECTION
|
KVM: PPC: Book3S HV: Better handling of exceptions that happen in real mode
When an interrupt or exception happens in the guest that comes to the
host, the CPU goes to hypervisor real mode (MMU off) to handle the
exception but doesn't change the MMU context. After saving a few
registers, we then clear the "in guest" flag. If, for any reason,
we get an exception in the real-mode code, that then gets handled
by the normal kernel exception handlers, which turn the MMU on. This
is disastrous if the MMU is still set to the guest context, since we
end up executing instructions from random places in the guest kernel
with hypervisor privilege.
In order to catch this situation, we define a new value for the "in guest"
flag, KVM_GUEST_MODE_HOST_HV, to indicate that we are in hypervisor real
mode with guest MMU context. If the "in guest" flag is set to this value,
we branch off to an emergency handler. For the moment, this just does
a branch to self to stop the CPU from doing anything further.
While we're here, we define another new flag value to indicate that we
are in a HV guest, as distinct from a PR guest. This will be useful
when we have a kernel that can support both PR and HV guests concurrently.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2013-10-04 19:45:04 +08:00
|
|
|
b .
|
KVM: PPC: Book3S HV: Handle unexpected interrupts better
At present, if an interrupt (i.e. an exception or trap) occurs in the
code where KVM is switching the MMU to or from guest context, we jump
to kvmppc_bad_host_intr, where we simply spin with interrupts disabled.
In this situation, it is hard to debug what happened because we get no
indication as to which interrupt occurred or where. Typically we get
a cascade of stall and soft lockup warnings from other CPUs.
In order to get more information for debugging, this adds code to
create a stack frame on the emergency stack and save register values
to it. We start half-way down the emergency stack in order to give
ourselves some chance of being able to do a stack trace on secondary
threads that are already on the emergency stack.
On POWER7 or POWER8, we then just spin, as before, because we don't
know what state the MMU context is in or what other threads are doing,
and we can't switch back to host context without coordinating with
other threads. On POWER9 we can do better; there we load up the host
MMU context and jump to C code, which prints an oops message to the
console and panics.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-01 14:17:27 +08:00
|
|
|
END_FTR_SECTION_IFCLR(CPU_FTR_ARCH_300)
|
|
|
|
|
ld r9, HSTATE_KVM_VCPU(r13)
|
|
|
|
|
ld r10, VCPU_KVM(r9)
|
|
|
|
|
|
|
|
|
|
li r0, 0
|
|
|
|
|
mtspr SPRN_AMR, r0
|
|
|
|
|
mtspr SPRN_IAMR, r0
|
|
|
|
|
mtspr SPRN_CIABR, r0
|
|
|
|
|
mtspr SPRN_DAWRX, r0
|
|
|
|
|
|
|
|
|
|
BEGIN_MMU_FTR_SECTION
|
|
|
|
|
b 4f
|
|
|
|
|
END_MMU_FTR_SECTION_IFSET(MMU_FTR_TYPE_RADIX)
|
|
|
|
|
|
|
|
|
|
slbmte r0, r0
|
|
|
|
|
slbia
|
|
|
|
|
ptesync
|
|
|
|
|
ld r8, PACA_SLBSHADOWPTR(r13)
|
|
|
|
|
.rept SLB_NUM_BOLTED
|
|
|
|
|
li r3, SLBSHADOW_SAVEAREA
|
|
|
|
|
LDX_BE r5, r8, r3
|
|
|
|
|
addi r3, r3, 8
|
|
|
|
|
LDX_BE r6, r8, r3
|
|
|
|
|
andis. r7, r5, SLB_ESID_V@h
|
|
|
|
|
beq 3f
|
|
|
|
|
slbmte r6, r5
|
|
|
|
|
3: addi r8, r8, 16
|
|
|
|
|
.endr
|
|
|
|
|
|
|
|
|
|
4: lwz r7, KVM_HOST_LPID(r10)
|
|
|
|
|
mtspr SPRN_LPID, r7
|
|
|
|
|
mtspr SPRN_PID, r0
|
|
|
|
|
ld r8, KVM_HOST_LPCR(r10)
|
|
|
|
|
mtspr SPRN_LPCR, r8
|
|
|
|
|
isync
|
|
|
|
|
li r0, KVM_GUEST_MODE_NONE
|
|
|
|
|
stb r0, HSTATE_IN_GUEST(r13)
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Turn on the MMU and jump to C code
|
|
|
|
|
*/
|
|
|
|
|
bcl 20, 31, .+4
|
|
|
|
|
5: mflr r3
|
|
|
|
|
addi r3, r3, 9f - 5b
|
2018-05-18 01:49:43 +08:00
|
|
|
li r4, -1
|
|
|
|
|
rldimi r3, r4, 62, 0 /* ensure 0xc000000000000000 bits are set */
|
KVM: PPC: Book3S HV: Handle unexpected interrupts better
At present, if an interrupt (i.e. an exception or trap) occurs in the
code where KVM is switching the MMU to or from guest context, we jump
to kvmppc_bad_host_intr, where we simply spin with interrupts disabled.
In this situation, it is hard to debug what happened because we get no
indication as to which interrupt occurred or where. Typically we get
a cascade of stall and soft lockup warnings from other CPUs.
In order to get more information for debugging, this adds code to
create a stack frame on the emergency stack and save register values
to it. We start half-way down the emergency stack in order to give
ourselves some chance of being able to do a stack trace on secondary
threads that are already on the emergency stack.
On POWER7 or POWER8, we then just spin, as before, because we don't
know what state the MMU context is in or what other threads are doing,
and we can't switch back to host context without coordinating with
other threads. On POWER9 we can do better; there we load up the host
MMU context and jump to C code, which prints an oops message to the
console and panics.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-01 14:17:27 +08:00
|
|
|
ld r4, PACAKMSR(r13)
|
|
|
|
|
mtspr SPRN_SRR0, r3
|
|
|
|
|
mtspr SPRN_SRR1, r4
|
2018-01-10 00:07:15 +08:00
|
|
|
RFI_TO_KERNEL
|
KVM: PPC: Book3S HV: Handle unexpected interrupts better
At present, if an interrupt (i.e. an exception or trap) occurs in the
code where KVM is switching the MMU to or from guest context, we jump
to kvmppc_bad_host_intr, where we simply spin with interrupts disabled.
In this situation, it is hard to debug what happened because we get no
indication as to which interrupt occurred or where. Typically we get
a cascade of stall and soft lockup warnings from other CPUs.
In order to get more information for debugging, this adds code to
create a stack frame on the emergency stack and save register values
to it. We start half-way down the emergency stack in order to give
ourselves some chance of being able to do a stack trace on secondary
threads that are already on the emergency stack.
On POWER7 or POWER8, we then just spin, as before, because we don't
know what state the MMU context is in or what other threads are doing,
and we can't switch back to host context without coordinating with
other threads. On POWER9 we can do better; there we load up the host
MMU context and jump to C code, which prints an oops message to the
console and panics.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-09-01 14:17:27 +08:00
|
|
|
9: addi r3, r1, STACK_FRAME_OVERHEAD
|
|
|
|
|
bl kvmppc_bad_interrupt
|
|
|
|
|
b 9b
|
2014-03-25 07:47:02 +08:00
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* This mimics the MSR transition on IRQ delivery. The new guest MSR is taken
|
|
|
|
|
* from VCPU_INTR_MSR and is modified based on the required TM state changes.
|
|
|
|
|
* r11 has the guest MSR value (in/out)
|
|
|
|
|
* r9 has a vcpu pointer (in)
|
|
|
|
|
* r0 is used as a scratch register
|
|
|
|
|
*/
|
|
|
|
|
kvmppc_msr_interrupt:
|
|
|
|
|
rldicl r0, r11, 64 - MSR_TS_S_LG, 62
|
|
|
|
|
cmpwi r0, 2 /* Check if we are in transactional state.. */
|
|
|
|
|
ld r11, VCPU_INTR_MSR(r9)
|
|
|
|
|
bne 1f
|
|
|
|
|
/* ... if transactional, change to suspended */
|
|
|
|
|
li r0, 1
|
|
|
|
|
1: rldimi r11, r0, MSR_TS_S_LG, 63 - MSR_TS_T_LG
|
|
|
|
|
blr
|
2014-05-26 17:48:40 +08:00
|
|
|
|
2018-10-08 13:30:51 +08:00
|
|
|
/*
|
|
|
|
|
* Load up guest PMU state. R3 points to the vcpu struct.
|
|
|
|
|
*/
|
|
|
|
|
_GLOBAL(kvmhv_load_guest_pmu)
|
|
|
|
|
EXPORT_SYMBOL_GPL(kvmhv_load_guest_pmu)
|
|
|
|
|
mr r4, r3
|
|
|
|
|
mflr r0
|
|
|
|
|
li r3, 1
|
|
|
|
|
sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
|
|
|
|
|
mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
|
|
|
|
|
isync
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r3, VCPU_MMCR(r4)
|
|
|
|
|
andi. r5, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
|
|
|
|
|
cmpwi r5, MMCR0_PMAO
|
|
|
|
|
beql kvmppc_fix_pmao
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
|
|
|
|
|
lwz r3, VCPU_PMC(r4) /* always load up guest PMU registers */
|
|
|
|
|
lwz r5, VCPU_PMC + 4(r4) /* to prevent information leak */
|
|
|
|
|
lwz r6, VCPU_PMC + 8(r4)
|
|
|
|
|
lwz r7, VCPU_PMC + 12(r4)
|
|
|
|
|
lwz r8, VCPU_PMC + 16(r4)
|
|
|
|
|
lwz r9, VCPU_PMC + 20(r4)
|
|
|
|
|
mtspr SPRN_PMC1, r3
|
|
|
|
|
mtspr SPRN_PMC2, r5
|
|
|
|
|
mtspr SPRN_PMC3, r6
|
|
|
|
|
mtspr SPRN_PMC4, r7
|
|
|
|
|
mtspr SPRN_PMC5, r8
|
|
|
|
|
mtspr SPRN_PMC6, r9
|
|
|
|
|
ld r3, VCPU_MMCR(r4)
|
|
|
|
|
ld r5, VCPU_MMCR + 8(r4)
|
|
|
|
|
ld r6, VCPU_MMCR + 16(r4)
|
|
|
|
|
ld r7, VCPU_SIAR(r4)
|
|
|
|
|
ld r8, VCPU_SDAR(r4)
|
|
|
|
|
mtspr SPRN_MMCR1, r5
|
|
|
|
|
mtspr SPRN_MMCRA, r6
|
|
|
|
|
mtspr SPRN_SIAR, r7
|
|
|
|
|
mtspr SPRN_SDAR, r8
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r5, VCPU_MMCR + 24(r4)
|
|
|
|
|
ld r6, VCPU_SIER(r4)
|
|
|
|
|
mtspr SPRN_MMCR2, r5
|
|
|
|
|
mtspr SPRN_SIER, r6
|
|
|
|
|
BEGIN_FTR_SECTION_NESTED(96)
|
|
|
|
|
lwz r7, VCPU_PMC + 24(r4)
|
|
|
|
|
lwz r8, VCPU_PMC + 28(r4)
|
|
|
|
|
ld r9, VCPU_MMCR + 32(r4)
|
|
|
|
|
mtspr SPRN_SPMC1, r7
|
|
|
|
|
mtspr SPRN_SPMC2, r8
|
|
|
|
|
mtspr SPRN_MMCRS, r9
|
|
|
|
|
END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
|
|
|
|
mtspr SPRN_MMCR0, r3
|
|
|
|
|
isync
|
|
|
|
|
mtlr r0
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Reload host PMU state saved in the PACA by kvmhv_save_host_pmu.
|
|
|
|
|
*/
|
|
|
|
|
_GLOBAL(kvmhv_load_host_pmu)
|
|
|
|
|
EXPORT_SYMBOL_GPL(kvmhv_load_host_pmu)
|
|
|
|
|
mflr r0
|
|
|
|
|
lbz r4, PACA_PMCINUSE(r13) /* is the host using the PMU? */
|
|
|
|
|
cmpwi r4, 0
|
|
|
|
|
beq 23f /* skip if not */
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r3, HSTATE_MMCR0(r13)
|
|
|
|
|
andi. r4, r3, MMCR0_PMAO_SYNC | MMCR0_PMAO
|
|
|
|
|
cmpwi r4, MMCR0_PMAO
|
|
|
|
|
beql kvmppc_fix_pmao
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_PMAO_BUG)
|
|
|
|
|
lwz r3, HSTATE_PMC1(r13)
|
|
|
|
|
lwz r4, HSTATE_PMC2(r13)
|
|
|
|
|
lwz r5, HSTATE_PMC3(r13)
|
|
|
|
|
lwz r6, HSTATE_PMC4(r13)
|
|
|
|
|
lwz r8, HSTATE_PMC5(r13)
|
|
|
|
|
lwz r9, HSTATE_PMC6(r13)
|
|
|
|
|
mtspr SPRN_PMC1, r3
|
|
|
|
|
mtspr SPRN_PMC2, r4
|
|
|
|
|
mtspr SPRN_PMC3, r5
|
|
|
|
|
mtspr SPRN_PMC4, r6
|
|
|
|
|
mtspr SPRN_PMC5, r8
|
|
|
|
|
mtspr SPRN_PMC6, r9
|
|
|
|
|
ld r3, HSTATE_MMCR0(r13)
|
|
|
|
|
ld r4, HSTATE_MMCR1(r13)
|
|
|
|
|
ld r5, HSTATE_MMCRA(r13)
|
|
|
|
|
ld r6, HSTATE_SIAR(r13)
|
|
|
|
|
ld r7, HSTATE_SDAR(r13)
|
|
|
|
|
mtspr SPRN_MMCR1, r4
|
|
|
|
|
mtspr SPRN_MMCRA, r5
|
|
|
|
|
mtspr SPRN_SIAR, r6
|
|
|
|
|
mtspr SPRN_SDAR, r7
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
ld r8, HSTATE_MMCR2(r13)
|
|
|
|
|
ld r9, HSTATE_SIER(r13)
|
|
|
|
|
mtspr SPRN_MMCR2, r8
|
|
|
|
|
mtspr SPRN_SIER, r9
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
|
|
|
|
mtspr SPRN_MMCR0, r3
|
|
|
|
|
isync
|
|
|
|
|
mtlr r0
|
|
|
|
|
23: blr
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Save guest PMU state into the vcpu struct.
|
|
|
|
|
* r3 = vcpu, r4 = full save flag (PMU in use flag set in VPA)
|
|
|
|
|
*/
|
|
|
|
|
_GLOBAL(kvmhv_save_guest_pmu)
|
|
|
|
|
EXPORT_SYMBOL_GPL(kvmhv_save_guest_pmu)
|
|
|
|
|
mr r9, r3
|
|
|
|
|
mr r8, r4
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
/*
|
|
|
|
|
* POWER8 seems to have a hardware bug where setting
|
|
|
|
|
* MMCR0[PMAE] along with MMCR0[PMC1CE] and/or MMCR0[PMCjCE]
|
|
|
|
|
* when some counters are already negative doesn't seem
|
|
|
|
|
* to cause a performance monitor alert (and hence interrupt).
|
|
|
|
|
* The effect of this is that when saving the PMU state,
|
|
|
|
|
* if there is no PMU alert pending when we read MMCR0
|
|
|
|
|
* before freezing the counters, but one becomes pending
|
|
|
|
|
* before we read the counters, we lose it.
|
|
|
|
|
* To work around this, we need a way to freeze the counters
|
|
|
|
|
* before reading MMCR0. Normally, freezing the counters
|
|
|
|
|
* is done by writing MMCR0 (to set MMCR0[FC]) which
|
|
|
|
|
* unavoidably writes MMCR0[PMA0] as well. On POWER8,
|
|
|
|
|
* we can also freeze the counters using MMCR2, by writing
|
|
|
|
|
* 1s to all the counter freeze condition bits (there are
|
|
|
|
|
* 9 bits each for 6 counters).
|
|
|
|
|
*/
|
|
|
|
|
li r3, -1 /* set all freeze bits */
|
|
|
|
|
clrrdi r3, r3, 10
|
|
|
|
|
mfspr r10, SPRN_MMCR2
|
|
|
|
|
mtspr SPRN_MMCR2, r3
|
|
|
|
|
isync
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
|
|
|
|
li r3, 1
|
|
|
|
|
sldi r3, r3, 31 /* MMCR0_FC (freeze counters) bit */
|
|
|
|
|
mfspr r4, SPRN_MMCR0 /* save MMCR0 */
|
|
|
|
|
mtspr SPRN_MMCR0, r3 /* freeze all counters, disable ints */
|
|
|
|
|
mfspr r6, SPRN_MMCRA
|
|
|
|
|
/* Clear MMCRA in order to disable SDAR updates */
|
|
|
|
|
li r7, 0
|
|
|
|
|
mtspr SPRN_MMCRA, r7
|
|
|
|
|
isync
|
|
|
|
|
cmpwi r8, 0 /* did they ask for PMU stuff to be saved? */
|
|
|
|
|
bne 21f
|
|
|
|
|
std r3, VCPU_MMCR(r9) /* if not, set saved MMCR0 to FC */
|
|
|
|
|
b 22f
|
|
|
|
|
21: mfspr r5, SPRN_MMCR1
|
|
|
|
|
mfspr r7, SPRN_SIAR
|
|
|
|
|
mfspr r8, SPRN_SDAR
|
|
|
|
|
std r4, VCPU_MMCR(r9)
|
|
|
|
|
std r5, VCPU_MMCR + 8(r9)
|
|
|
|
|
std r6, VCPU_MMCR + 16(r9)
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
std r10, VCPU_MMCR + 24(r9)
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
|
|
|
|
std r7, VCPU_SIAR(r9)
|
|
|
|
|
std r8, VCPU_SDAR(r9)
|
|
|
|
|
mfspr r3, SPRN_PMC1
|
|
|
|
|
mfspr r4, SPRN_PMC2
|
|
|
|
|
mfspr r5, SPRN_PMC3
|
|
|
|
|
mfspr r6, SPRN_PMC4
|
|
|
|
|
mfspr r7, SPRN_PMC5
|
|
|
|
|
mfspr r8, SPRN_PMC6
|
|
|
|
|
stw r3, VCPU_PMC(r9)
|
|
|
|
|
stw r4, VCPU_PMC + 4(r9)
|
|
|
|
|
stw r5, VCPU_PMC + 8(r9)
|
|
|
|
|
stw r6, VCPU_PMC + 12(r9)
|
|
|
|
|
stw r7, VCPU_PMC + 16(r9)
|
|
|
|
|
stw r8, VCPU_PMC + 20(r9)
|
|
|
|
|
BEGIN_FTR_SECTION
|
|
|
|
|
mfspr r5, SPRN_SIER
|
|
|
|
|
std r5, VCPU_SIER(r9)
|
|
|
|
|
BEGIN_FTR_SECTION_NESTED(96)
|
|
|
|
|
mfspr r6, SPRN_SPMC1
|
|
|
|
|
mfspr r7, SPRN_SPMC2
|
|
|
|
|
mfspr r8, SPRN_MMCRS
|
|
|
|
|
stw r6, VCPU_PMC + 24(r9)
|
|
|
|
|
stw r7, VCPU_PMC + 28(r9)
|
|
|
|
|
std r8, VCPU_MMCR + 32(r9)
|
|
|
|
|
lis r4, 0x8000
|
|
|
|
|
mtspr SPRN_MMCRS, r4
|
|
|
|
|
END_FTR_SECTION_NESTED(CPU_FTR_ARCH_300, 0, 96)
|
|
|
|
|
END_FTR_SECTION_IFSET(CPU_FTR_ARCH_207S)
|
|
|
|
|
22: blr
|
|
|
|
|
|
2014-05-26 17:48:40 +08:00
|
|
|
/*
|
|
|
|
|
* This works around a hardware bug on POWER8E processors, where
|
|
|
|
|
* writing a 1 to the MMCR0[PMAO] bit doesn't generate a
|
|
|
|
|
* performance monitor interrupt. Instead, when we need to have
|
|
|
|
|
* an interrupt pending, we have to arrange for a counter to overflow.
|
|
|
|
|
*/
|
|
|
|
|
kvmppc_fix_pmao:
|
|
|
|
|
li r3, 0
|
|
|
|
|
mtspr SPRN_MMCR2, r3
|
|
|
|
|
lis r3, (MMCR0_PMXE | MMCR0_FCECE)@h
|
|
|
|
|
ori r3, r3, MMCR0_PMCjCE | MMCR0_C56RUN
|
|
|
|
|
mtspr SPRN_MMCR0, r3
|
|
|
|
|
lis r3, 0x7fff
|
|
|
|
|
ori r3, r3, 0xffff
|
|
|
|
|
mtspr SPRN_PMC6, r3
|
|
|
|
|
isync
|
|
|
|
|
blr
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
|
|
|
|
|
#ifdef CONFIG_KVM_BOOK3S_HV_EXIT_TIMING
|
|
|
|
|
/*
|
|
|
|
|
* Start timing an activity
|
|
|
|
|
* r3 = pointer to time accumulation struct, r4 = vcpu
|
|
|
|
|
*/
|
|
|
|
|
kvmhv_start_timing:
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
KVM: PPC: Book3S HV: Snapshot timebase offset on guest entry
Currently, the HV KVM guest entry/exit code adds the timebase offset
from the vcore struct to the timebase on guest entry, and subtracts
it on guest exit. Which is fine, except that it is possible for
userspace to change the offset using the SET_ONE_REG interface while
the vcore is running, as there is only one timebase offset per vcore
but potentially multiple VCPUs in the vcore. If that were to happen,
KVM would subtract a different offset on guest exit from that which
it had added on guest entry, leading to the timebase being out of sync
between cores in the host, which then leads to bad things happening
such as hangs and spurious watchdog timeouts.
To fix this, we add a new field 'tb_offset_applied' to the vcore struct
which stores the offset that is currently applied to the timebase.
This value is set from the vcore tb_offset field on guest entry, and
is what is subtracted from the timebase on guest exit. Since it is
zero when the timebase offset is not applied, we can simplify the
logic in kvmhv_start_timing and kvmhv_accumulate_time.
In addition, we had secondary threads reading the timebase while
running concurrently with code on the primary thread which would
eventually add or subtract the timebase offset from the timebase.
This occurred while saving or restoring the DEC register value on
the secondary threads. Although no specific incorrect behaviour has
been observed, this is a race which should be fixed. To fix it, we
move the DEC saving code to just before we call kvmhv_commence_exit,
and the DEC restoring code to after the point where we have waited
for the primary thread to switch the MMU context and add the timebase
offset. That way we are sure that the timebase contains the guest
timebase value in both cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-04-20 20:51:11 +08:00
|
|
|
ld r6, VCORE_TB_OFFSET_APPL(r5)
|
|
|
|
|
mftb r5
|
|
|
|
|
subf r5, r6, r5 /* subtract current timebase offset */
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
std r3, VCPU_CUR_ACTIVITY(r4)
|
|
|
|
|
std r5, VCPU_ACTIVITY_START(r4)
|
|
|
|
|
blr
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Accumulate time to one activity and start another.
|
|
|
|
|
* r3 = pointer to new time accumulation struct, r4 = vcpu
|
|
|
|
|
*/
|
|
|
|
|
kvmhv_accumulate_time:
|
|
|
|
|
ld r5, HSTATE_KVM_VCORE(r13)
|
KVM: PPC: Book3S HV: Snapshot timebase offset on guest entry
Currently, the HV KVM guest entry/exit code adds the timebase offset
from the vcore struct to the timebase on guest entry, and subtracts
it on guest exit. Which is fine, except that it is possible for
userspace to change the offset using the SET_ONE_REG interface while
the vcore is running, as there is only one timebase offset per vcore
but potentially multiple VCPUs in the vcore. If that were to happen,
KVM would subtract a different offset on guest exit from that which
it had added on guest entry, leading to the timebase being out of sync
between cores in the host, which then leads to bad things happening
such as hangs and spurious watchdog timeouts.
To fix this, we add a new field 'tb_offset_applied' to the vcore struct
which stores the offset that is currently applied to the timebase.
This value is set from the vcore tb_offset field on guest entry, and
is what is subtracted from the timebase on guest exit. Since it is
zero when the timebase offset is not applied, we can simplify the
logic in kvmhv_start_timing and kvmhv_accumulate_time.
In addition, we had secondary threads reading the timebase while
running concurrently with code on the primary thread which would
eventually add or subtract the timebase offset from the timebase.
This occurred while saving or restoring the DEC register value on
the secondary threads. Although no specific incorrect behaviour has
been observed, this is a race which should be fixed. To fix it, we
move the DEC saving code to just before we call kvmhv_commence_exit,
and the DEC restoring code to after the point where we have waited
for the primary thread to switch the MMU context and add the timebase
offset. That way we are sure that the timebase contains the guest
timebase value in both cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-04-20 20:51:11 +08:00
|
|
|
ld r8, VCORE_TB_OFFSET_APPL(r5)
|
|
|
|
|
ld r5, VCPU_CUR_ACTIVITY(r4)
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
ld r6, VCPU_ACTIVITY_START(r4)
|
|
|
|
|
std r3, VCPU_CUR_ACTIVITY(r4)
|
|
|
|
|
mftb r7
|
KVM: PPC: Book3S HV: Snapshot timebase offset on guest entry
Currently, the HV KVM guest entry/exit code adds the timebase offset
from the vcore struct to the timebase on guest entry, and subtracts
it on guest exit. Which is fine, except that it is possible for
userspace to change the offset using the SET_ONE_REG interface while
the vcore is running, as there is only one timebase offset per vcore
but potentially multiple VCPUs in the vcore. If that were to happen,
KVM would subtract a different offset on guest exit from that which
it had added on guest entry, leading to the timebase being out of sync
between cores in the host, which then leads to bad things happening
such as hangs and spurious watchdog timeouts.
To fix this, we add a new field 'tb_offset_applied' to the vcore struct
which stores the offset that is currently applied to the timebase.
This value is set from the vcore tb_offset field on guest entry, and
is what is subtracted from the timebase on guest exit. Since it is
zero when the timebase offset is not applied, we can simplify the
logic in kvmhv_start_timing and kvmhv_accumulate_time.
In addition, we had secondary threads reading the timebase while
running concurrently with code on the primary thread which would
eventually add or subtract the timebase offset from the timebase.
This occurred while saving or restoring the DEC register value on
the secondary threads. Although no specific incorrect behaviour has
been observed, this is a race which should be fixed. To fix it, we
move the DEC saving code to just before we call kvmhv_commence_exit,
and the DEC restoring code to after the point where we have waited
for the primary thread to switch the MMU context and add the timebase
offset. That way we are sure that the timebase contains the guest
timebase value in both cases.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-04-20 20:51:11 +08:00
|
|
|
subf r7, r8, r7 /* subtract current timebase offset */
|
KVM: PPC: Book3S HV: Accumulate timing information for real-mode code
This reads the timebase at various points in the real-mode guest
entry/exit code and uses that to accumulate total, minimum and
maximum time spent in those parts of the code. Currently these
times are accumulated per vcpu in 5 parts of the code:
* rm_entry - time taken from the start of kvmppc_hv_entry() until
just before entering the guest.
* rm_intr - time from when we take a hypervisor interrupt in the
guest until we either re-enter the guest or decide to exit to the
host. This includes time spent handling hcalls in real mode.
* rm_exit - time from when we decide to exit the guest until the
return from kvmppc_hv_entry().
* guest - time spend in the guest
* cede - time spent napping in real mode due to an H_CEDE hcall
while other threads in the same vcore are active.
These times are exposed in debugfs in a directory per vcpu that
contains a file called "timings". This file contains one line for
each of the 5 timings above, with the name followed by a colon and
4 numbers, which are the count (number of times the code has been
executed), the total time, the minimum time, and the maximum time,
all in nanoseconds.
The overhead of the extra code amounts to about 30ns for an hcall that
is handled in real mode (e.g. H_SET_DABR), which is about 25%. Since
production environments may not wish to incur this overhead, the new
code is conditional on a new config symbol,
CONFIG_KVM_BOOK3S_HV_EXIT_TIMING.
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
2015-03-28 11:21:02 +08:00
|
|
|
std r7, VCPU_ACTIVITY_START(r4)
|
|
|
|
|
cmpdi r5, 0
|
|
|
|
|
beqlr
|
|
|
|
|
subf r3, r6, r7
|
|
|
|
|
ld r8, TAS_SEQCOUNT(r5)
|
|
|
|
|
cmpdi r8, 0
|
|
|
|
|
addi r8, r8, 1
|
|
|
|
|
std r8, TAS_SEQCOUNT(r5)
|
|
|
|
|
lwsync
|
|
|
|
|
ld r7, TAS_TOTAL(r5)
|
|
|
|
|
add r7, r7, r3
|
|
|
|
|
std r7, TAS_TOTAL(r5)
|
|
|
|
|
ld r6, TAS_MIN(r5)
|
|
|
|
|
ld r7, TAS_MAX(r5)
|
|
|
|
|
beq 3f
|
|
|
|
|
cmpd r3, r6
|
|
|
|
|
bge 1f
|
|
|
|
|
3: std r3, TAS_MIN(r5)
|
|
|
|
|
1: cmpd r3, r7
|
|
|
|
|
ble 2f
|
|
|
|
|
std r3, TAS_MAX(r5)
|
|
|
|
|
2: lwsync
|
|
|
|
|
addi r8, r8, 1
|
|
|
|
|
std r8, TAS_SEQCOUNT(r5)
|
|
|
|
|
blr
|
|
|
|
|
#endif
|