OpenCloudOS-Kernel/arch/powerpc/kernel/process.c

2300 lines
58 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-or-later
/*
* Derived from "arch/i386/kernel/process.c"
* Copyright (C) 1995 Linus Torvalds
*
* Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
* Paul Mackerras (paulus@cs.anu.edu.au)
*
* PowerPC version
* Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
*/
#include <linux/errno.h>
#include <linux/sched.h>
#include <linux/sched/debug.h>
#include <linux/sched/task.h>
#include <linux/sched/task_stack.h>
#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/smp.h>
#include <linux/stddef.h>
#include <linux/unistd.h>
#include <linux/ptrace.h>
#include <linux/slab.h>
#include <linux/user.h>
#include <linux/elf.h>
#include <linux/prctl.h>
#include <linux/init_task.h>
#include <linux/export.h>
#include <linux/kallsyms.h>
#include <linux/mqueue.h>
#include <linux/hardirq.h>
#include <linux/utsname.h>
#include <linux/ftrace.h>
2008-12-31 22:11:38 +08:00
#include <linux/kernel_stat.h>
#include <linux/personality.h>
#include <linux/random.h>
#include <linux/hw_breakpoint.h>
#include <linux/uaccess.h>
#include <linux/elf-randomize.h>
#include <linux/pkeys.h>
#include <linux/seq_buf.h>
#include <asm/interrupt.h>
#include <asm/io.h>
#include <asm/processor.h>
#include <asm/mmu.h>
#include <asm/prom.h>
#include <asm/machdep.h>
powerpc: Implement accurate task and CPU time accounting This implements accurate task and cpu time accounting for 64-bit powerpc kernels. Instead of accounting a whole jiffy of time to a task on a timer interrupt because that task happened to be running at the time, we now account time in units of timebase ticks according to the actual time spent by the task in user mode and kernel mode. We also count the time spent processing hardware and software interrupts accurately. This is conditional on CONFIG_VIRT_CPU_ACCOUNTING. If that is not set, we do tick-based approximate accounting as before. To get this accurate information, we read either the PURR (processor utilization of resources register) on POWER5 machines, or the timebase on other machines on * each entry to the kernel from usermode * each exit to usermode * transitions between process context, hard irq context and soft irq context in kernel mode * context switches. On POWER5 systems with shared-processor logical partitioning we also read both the PURR and the timebase at each timer interrupt and context switch in order to determine how much time has been taken by the hypervisor to run other partitions ("steal" time). Unfortunately, since we need values of the PURR on both threads at the same time to accurately calculate the steal time, and since we can only calculate steal time on a per-core basis, the apportioning of the steal time between idle time (time which we ceded to the hypervisor in the idle loop) and actual stolen time is somewhat approximate at the moment. This is all based quite heavily on what s390 does, and it uses the generic interfaces that were added by the s390 developers, i.e. account_system_time(), account_user_time(), etc. This patch doesn't add any new interfaces between the kernel and userspace, and doesn't change the units in which time is reported to userspace by things such as /proc/stat, /proc/<pid>/stat, getrusage(), times(), etc. Internally the various task and cpu times are stored in timebase units, but they are converted to USER_HZ units (1/100th of a second) when reported to userspace. Some precision is therefore lost but there should not be any accumulating error, since the internal accumulation is at full precision. Signed-off-by: Paul Mackerras <paulus@samba.org>
2006-02-24 07:06:59 +08:00
#include <asm/time.h>
#include <asm/runlatch.h>
#include <asm/syscalls.h>
#include <asm/switch_to.h>
#include <asm/tm.h>
#include <asm/debug.h>
#ifdef CONFIG_PPC64
#include <asm/firmware.h>
#include <asm/hw_irq.h>
#endif
#include <asm/code-patching.h>
#include <asm/exec.h>
#include <asm/livepatch.h>
#include <asm/cpu_has_feature.h>
#include <asm/asm-prototypes.h>
#include <asm/stacktrace.h>
powerpc: Add force enable of DAWR on P9 option This adds a flag so that the DAWR can be enabled on P9 via: echo Y > /sys/kernel/debug/powerpc/dawr_enable_dangerous The DAWR was previously force disabled on POWER9 in: 9654153158 powerpc: Disable DAWR in the base POWER9 CPU features Also see Documentation/powerpc/DAWR-POWER9.txt This is a dangerous setting, USE AT YOUR OWN RISK. Some users may not care about a bad user crashing their box (ie. single user/desktop systems) and really want the DAWR. This allows them to force enable DAWR. This flag can also be used to disable DAWR access. Once this is cleared, all DAWR access should be cleared immediately and your machine once again safe from crashing. Userspace may get confused by toggling this. If DAWR is force enabled/disabled between getting the number of breakpoints (via PTRACE_GETHWDBGINFO) and setting the breakpoint, userspace will get an inconsistent view of what's available. Similarly for guests. For the DAWR to be enabled in a KVM guest, the DAWR needs to be force enabled in the host AND the guest. For this reason, this won't work on POWERVM as it doesn't allow the HCALL to work. Writes of 'Y' to the dawr_enable_dangerous file will fail if the hypervisor doesn't support writing the DAWR. To double check the DAWR is working, run this kernel selftest: tools/testing/selftests/powerpc/ptrace/ptrace-hwbreak.c Any errors/failures/skips mean something is wrong. Signed-off-by: Michael Neuling <mikey@neuling.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2019-04-01 14:03:12 +08:00
#include <asm/hw_breakpoint.h>
#include <linux/kprobes.h>
#include <linux/kdebug.h>
/* Transactional Memory debug */
#ifdef TM_DEBUG_SW
#define TM_DEBUG(x...) printk(KERN_INFO x)
#else
#define TM_DEBUG(x...) do { } while(0)
#endif
extern unsigned long _get_SP(void);
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
/*
* Are we running in "Suspend disabled" mode? If so we have to block any
* sigreturn that would get us into suspended state, and we also warn in some
* other paths that we should never reach with suspend disabled.
*/
bool tm_suspend_disabled __ro_after_init = false;
static void check_if_tm_restore_required(struct task_struct *tsk)
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
{
/*
* If we are saving the current thread's registers, and the
* thread is in a transactional state, set the TIF_RESTORE_TM
* bit so that we know to restore the registers before
* returning to userspace.
*/
if (tsk == current && tsk->thread.regs &&
MSR_TM_ACTIVE(tsk->thread.regs->msr) &&
!test_thread_flag(TIF_RESTORE_TM)) {
tsk->thread.ckpt_regs.msr = tsk->thread.regs->msr;
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
set_thread_flag(TIF_RESTORE_TM);
}
}
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
#else
static inline void check_if_tm_restore_required(struct task_struct *tsk) { }
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
bool strict_msr_control;
EXPORT_SYMBOL(strict_msr_control);
static int __init enable_strict_msr_control(char *str)
{
strict_msr_control = true;
pr_info("Enabling strict facility control\n");
return 0;
}
early_param("ppc_strict_facility_enable", enable_strict_msr_control);
/* notrace because it's called by restore_math */
unsigned long notrace msr_check_and_set(unsigned long bits)
{
unsigned long oldmsr = mfmsr();
unsigned long newmsr;
newmsr = oldmsr | bits;
if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
newmsr |= MSR_VSX;
if (oldmsr != newmsr)
mtmsr_isync(newmsr);
return newmsr;
}
EXPORT_SYMBOL_GPL(msr_check_and_set);
/* notrace because it's called by restore_math */
void notrace __msr_check_and_clear(unsigned long bits)
{
unsigned long oldmsr = mfmsr();
unsigned long newmsr;
newmsr = oldmsr & ~bits;
if (cpu_has_feature(CPU_FTR_VSX) && (bits & MSR_FP))
newmsr &= ~MSR_VSX;
if (oldmsr != newmsr)
mtmsr_isync(newmsr);
}
EXPORT_SYMBOL(__msr_check_and_clear);
#ifdef CONFIG_PPC_FPU
static void __giveup_fpu(struct task_struct *tsk)
{
unsigned long msr;
save_fpu(tsk);
msr = tsk->thread.regs->msr;
msr &= ~(MSR_FP|MSR_FE0|MSR_FE1);
if (cpu_has_feature(CPU_FTR_VSX))
msr &= ~MSR_VSX;
tsk->thread.regs->msr = msr;
}
void giveup_fpu(struct task_struct *tsk)
{
check_if_tm_restore_required(tsk);
msr_check_and_set(MSR_FP);
__giveup_fpu(tsk);
msr_check_and_clear(MSR_FP);
}
EXPORT_SYMBOL(giveup_fpu);
/*
* Make sure the floating-point register state in the
* the thread_struct is up to date for task tsk.
*/
void flush_fp_to_thread(struct task_struct *tsk)
{
if (tsk->thread.regs) {
/*
* We need to disable preemption here because if we didn't,
* another process could get scheduled after the regs->msr
* test but before we have finished saving the FP registers
* to the thread_struct. That process could take over the
* FPU, and then when we get scheduled again we would store
* bogus values for the remaining FP registers.
*/
preempt_disable();
if (tsk->thread.regs->msr & MSR_FP) {
/*
* This should only ever be called for current or
* for a stopped child process. Since we save away
* the FP register state on context switch,
* there is something wrong if a stopped child appears
* to still have its FP state in the CPU registers.
*/
BUG_ON(tsk != current);
giveup_fpu(tsk);
}
preempt_enable();
}
}
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
EXPORT_SYMBOL_GPL(flush_fp_to_thread);
void enable_kernel_fp(void)
{
unsigned long cpumsr;
WARN_ON(preemptible());
cpumsr = msr_check_and_set(MSR_FP);
if (current->thread.regs && (current->thread.regs->msr & MSR_FP)) {
check_if_tm_restore_required(current);
/*
* If a thread has already been reclaimed then the
* checkpointed registers are on the CPU but have definitely
* been saved by the reclaim code. Don't need to and *cannot*
* giveup as this would save to the 'live' structure not the
* checkpointed structure.
*/
if (!MSR_TM_ACTIVE(cpumsr) &&
MSR_TM_ACTIVE(current->thread.regs->msr))
return;
__giveup_fpu(current);
}
}
EXPORT_SYMBOL(enable_kernel_fp);
#else
static inline void __giveup_fpu(struct task_struct *tsk) { }
#endif /* CONFIG_PPC_FPU */
#ifdef CONFIG_ALTIVEC
static void __giveup_altivec(struct task_struct *tsk)
{
unsigned long msr;
save_altivec(tsk);
msr = tsk->thread.regs->msr;
msr &= ~MSR_VEC;
if (cpu_has_feature(CPU_FTR_VSX))
msr &= ~MSR_VSX;
tsk->thread.regs->msr = msr;
}
void giveup_altivec(struct task_struct *tsk)
{
check_if_tm_restore_required(tsk);
msr_check_and_set(MSR_VEC);
__giveup_altivec(tsk);
msr_check_and_clear(MSR_VEC);
}
EXPORT_SYMBOL(giveup_altivec);
void enable_kernel_altivec(void)
{
unsigned long cpumsr;
WARN_ON(preemptible());
cpumsr = msr_check_and_set(MSR_VEC);
if (current->thread.regs && (current->thread.regs->msr & MSR_VEC)) {
check_if_tm_restore_required(current);
/*
* If a thread has already been reclaimed then the
* checkpointed registers are on the CPU but have definitely
* been saved by the reclaim code. Don't need to and *cannot*
* giveup as this would save to the 'live' structure not the
* checkpointed structure.
*/
if (!MSR_TM_ACTIVE(cpumsr) &&
MSR_TM_ACTIVE(current->thread.regs->msr))
return;
__giveup_altivec(current);
}
}
EXPORT_SYMBOL(enable_kernel_altivec);
/*
* Make sure the VMX/Altivec register state in the
* the thread_struct is up to date for task tsk.
*/
void flush_altivec_to_thread(struct task_struct *tsk)
{
if (tsk->thread.regs) {
preempt_disable();
if (tsk->thread.regs->msr & MSR_VEC) {
BUG_ON(tsk != current);
giveup_altivec(tsk);
}
preempt_enable();
}
}
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
EXPORT_SYMBOL_GPL(flush_altivec_to_thread);
#endif /* CONFIG_ALTIVEC */
#ifdef CONFIG_VSX
static void __giveup_vsx(struct task_struct *tsk)
{
unsigned long msr = tsk->thread.regs->msr;
/*
* We should never be ssetting MSR_VSX without also setting
* MSR_FP and MSR_VEC
*/
WARN_ON((msr & MSR_VSX) && !((msr & MSR_FP) && (msr & MSR_VEC)));
/* __giveup_fpu will clear MSR_VSX */
if (msr & MSR_FP)
__giveup_fpu(tsk);
if (msr & MSR_VEC)
__giveup_altivec(tsk);
}
static void giveup_vsx(struct task_struct *tsk)
{
check_if_tm_restore_required(tsk);
msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
__giveup_vsx(tsk);
msr_check_and_clear(MSR_FP|MSR_VEC|MSR_VSX);
}
void enable_kernel_vsx(void)
{
unsigned long cpumsr;
WARN_ON(preemptible());
cpumsr = msr_check_and_set(MSR_FP|MSR_VEC|MSR_VSX);
if (current->thread.regs &&
(current->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP))) {
check_if_tm_restore_required(current);
/*
* If a thread has already been reclaimed then the
* checkpointed registers are on the CPU but have definitely
* been saved by the reclaim code. Don't need to and *cannot*
* giveup as this would save to the 'live' structure not the
* checkpointed structure.
*/
if (!MSR_TM_ACTIVE(cpumsr) &&
MSR_TM_ACTIVE(current->thread.regs->msr))
return;
__giveup_vsx(current);
}
}
EXPORT_SYMBOL(enable_kernel_vsx);
void flush_vsx_to_thread(struct task_struct *tsk)
{
if (tsk->thread.regs) {
preempt_disable();
if (tsk->thread.regs->msr & (MSR_VSX|MSR_VEC|MSR_FP)) {
BUG_ON(tsk != current);
giveup_vsx(tsk);
}
preempt_enable();
}
}
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
EXPORT_SYMBOL_GPL(flush_vsx_to_thread);
#endif /* CONFIG_VSX */
#ifdef CONFIG_SPE
void giveup_spe(struct task_struct *tsk)
{
check_if_tm_restore_required(tsk);
msr_check_and_set(MSR_SPE);
__giveup_spe(tsk);
msr_check_and_clear(MSR_SPE);
}
EXPORT_SYMBOL(giveup_spe);
void enable_kernel_spe(void)
{
WARN_ON(preemptible());
msr_check_and_set(MSR_SPE);
if (current->thread.regs && (current->thread.regs->msr & MSR_SPE)) {
check_if_tm_restore_required(current);
__giveup_spe(current);
}
}
EXPORT_SYMBOL(enable_kernel_spe);
void flush_spe_to_thread(struct task_struct *tsk)
{
if (tsk->thread.regs) {
preempt_disable();
if (tsk->thread.regs->msr & MSR_SPE) {
BUG_ON(tsk != current);
tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
giveup_spe(tsk);
}
preempt_enable();
}
}
#endif /* CONFIG_SPE */
static unsigned long msr_all_available;
static int __init init_msr_all_available(void)
{
if (IS_ENABLED(CONFIG_PPC_FPU))
msr_all_available |= MSR_FP;
if (cpu_has_feature(CPU_FTR_ALTIVEC))
msr_all_available |= MSR_VEC;
if (cpu_has_feature(CPU_FTR_VSX))
msr_all_available |= MSR_VSX;
if (cpu_has_feature(CPU_FTR_SPE))
msr_all_available |= MSR_SPE;
return 0;
}
early_initcall(init_msr_all_available);
void giveup_all(struct task_struct *tsk)
{
unsigned long usermsr;
if (!tsk->thread.regs)
return;
powerpc/tm: Fix FP/VMX unavailable exceptions inside a transaction When we take an FP unavailable exception in a transaction we have to account for the hardware FP TM checkpointed registers being incorrect. In this case for this process we know the current and checkpointed FP registers must be the same (since FP wasn't used inside the transaction) hence in the thread_struct we copy the current FP registers to the checkpointed ones. This copy is done in tm_reclaim_thread(). We use thread->ckpt_regs.msr to determine if FP was on when in userspace. thread->ckpt_regs.msr represents the state of the MSR when exiting userspace. This is setup by check_if_tm_restore_required(). Unfortunatley there is an optimisation in giveup_all() which returns early if tsk->thread.regs->msr (via local variable `usermsr`) has FP=VEC=VSX=SPE=0. This optimisation means that check_if_tm_restore_required() is not called and hence thread->ckpt_regs.msr is not updated and will contain an old value. This can happen if due to load_fp=255 we start a userspace process with MSR FP=1 and then we are context switched out. In this case thread->ckpt_regs.msr will contain FP=1. If that same process is then context switched in and load_fp overflows, MSR will have FP=0. If that process now enters a transaction and does an FP instruction, the FP unavailable will not update thread->ckpt_regs.msr (the bug) and MSR FP=1 will be retained in thread->ckpt_regs.msr. tm_reclaim_thread() will then not perform the required memcpy and the checkpointed FP regs in the thread struct will contain the wrong values. The code path for this happening is: Userspace: Kernel Start userspace with MSR FP/VEC/VSX/SPE=0 TM=1 < ----- ... tbegin bne fp instruction FP unavailable ---- > fp_unavailable_tm() tm_reclaim_current() tm_reclaim_thread() giveup_all() return early since FP/VMX/VSX=0 /* ckpt MSR not updated (Incorrect) */ tm_reclaim() /* thread_struct ckpt FP regs contain junk (OK) */ /* Sees ckpt MSR FP=1 (Incorrect) */ no memcpy() performed /* thread_struct ckpt FP regs not fixed (Incorrect) */ tm_recheckpoint() /* Put junk in hardware checkpoint FP regs */ .... < ----- Return to userspace with MSR TM=1 FP=1 with junk in the FP TM checkpoint TM rollback reads FP junk This is a data integrity problem for the current process as the FP registers are corrupted. It's also a security problem as the FP registers from one process may be leaked to another. This patch moves up check_if_tm_restore_required() in giveup_all() to ensure thread->ckpt_regs.msr is updated correctly. A simple testcase to replicate this will be posted to tools/testing/selftests/powerpc/tm/tm-poison.c Similarly for VMX. This fixes CVE-2019-15030. Fixes: f48e91e87e67 ("powerpc/tm: Fix FP and VMX register corruption") Cc: stable@vger.kernel.org # 4.12+ Signed-off-by: Gustavo Romero <gromero@linux.vnet.ibm.com> Signed-off-by: Michael Neuling <mikey@neuling.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20190904045529.23002-1-gromero@linux.vnet.ibm.com
2019-09-04 12:55:27 +08:00
check_if_tm_restore_required(tsk);
usermsr = tsk->thread.regs->msr;
if ((usermsr & msr_all_available) == 0)
return;
msr_check_and_set(msr_all_available);
WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
if (usermsr & MSR_FP)
__giveup_fpu(tsk);
if (usermsr & MSR_VEC)
__giveup_altivec(tsk);
if (usermsr & MSR_SPE)
__giveup_spe(tsk);
msr_check_and_clear(msr_all_available);
}
EXPORT_SYMBOL(giveup_all);
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
#ifdef CONFIG_PPC_BOOK3S_64
#ifdef CONFIG_PPC_FPU
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
static bool should_restore_fp(void)
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
{
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
if (current->thread.load_fp) {
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
current->thread.load_fp++;
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
return true;
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
}
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
return false;
}
static void do_restore_fp(void)
{
load_fp_state(&current->thread.fp_state);
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
}
#else
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
static bool should_restore_fp(void) { return false; }
static void do_restore_fp(void) { }
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
#endif /* CONFIG_PPC_FPU */
#ifdef CONFIG_ALTIVEC
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
static bool should_restore_altivec(void)
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
{
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
if (cpu_has_feature(CPU_FTR_ALTIVEC) && (current->thread.load_vec)) {
current->thread.load_vec++;
return true;
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
}
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
return false;
}
static void do_restore_altivec(void)
{
load_vr_state(&current->thread.vr_state);
current->thread.used_vr = 1;
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
}
#else
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
static bool should_restore_altivec(void) { return false; }
static void do_restore_altivec(void) { }
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
#endif /* CONFIG_ALTIVEC */
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
static bool should_restore_vsx(void)
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
{
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
if (cpu_has_feature(CPU_FTR_VSX))
return true;
return false;
}
#ifdef CONFIG_VSX
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
static void do_restore_vsx(void)
{
current->thread.used_vsr = 1;
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
}
#else
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
static void do_restore_vsx(void) { }
powerpc/64s: Implement interrupt exit logic in C Implement the bulk of interrupt return logic in C. The asm return code must handle a few cases: restoring full GPRs, and emulating stack store. The stack store emulation is significantly simplfied, rather than creating a new return frame and switching to that before performing the store, it uses the PACA to keep a scratch register around to perform the store. The asm return code is moved into 64e for now. The new logic has made allowance for 64e, but I don't have a full environment that works well to test it, and even booting in emulated qemu is not great for stress testing. 64e shouldn't be too far off working with this, given a bit more testing and auditing of the logic. This is slightly faster on a POWER9 (page fault speed increases about 1.1%), probably due to reduced mtmsrd. mpe: Includes fixes from Nick for _TIF_EMULATE_STACK_STORE handling (including the fast_interrupt_return path), to remove trace_hardirqs_on(), and fixes the interrupt-return part of the MSR_VSX restore bug caught by tm-unavailable selftest. mpe: Incorporate fix from Nick: The return-to-kernel path has to replay any soft-pending interrupts if it is returning to a context that had interrupts soft-enabled. It has to do this carefully and avoid plain enabling interrupts if this is an irq context, which can cause multiple nesting of interrupts on the stack, and other unexpected issues. The code which avoided this case got the soft-mask state wrong, and marked interrupts as enabled before going around again to retry. This seems to be mostly harmless except when PREEMPT=y, this calls preempt_schedule_irq with irqs apparently enabled and runs into a BUG in kernel/sched/core.c Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michal Suchanek <msuchanek@suse.de> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200225173541.1549955-29-npiggin@gmail.com
2020-02-26 01:35:37 +08:00
#endif /* CONFIG_VSX */
/*
* The exception exit path calls restore_math() with interrupts hard disabled
* but the soft irq state not "reconciled". ftrace code that calls
* local_irq_save/restore causes warnings.
*
* Rather than complicate the exit path, just don't trace restore_math. This
* could be done by having ftrace entry code check for this un-reconciled
* condition where MSR[EE]=0 and PACA_IRQ_HARD_DIS is not set, and
* temporarily fix it up for the duration of the ftrace call.
*/
void notrace restore_math(struct pt_regs *regs)
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
{
unsigned long msr;
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
unsigned long new_msr = 0;
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
msr = regs->msr;
/*
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
* new_msr tracks the facilities that are to be restored. Only reload
* if the bit is not set in the user MSR (if it is set, the registers
* are live for the user thread).
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
*/
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
if ((!(msr & MSR_FP)) && should_restore_fp())
powerpc/64s: Fix crash in load_fp_state() due to fpexc_mode The recent commit 01eb01877f33 ("powerpc/64s: Fix restore_math unnecessarily changing MSR") changed some of the handling of floating point/vector restore. In particular it caused current->thread.fpexc_mode to be copied into the current MSR (via msr_check_and_set()), rather than just into regs->msr (which is moved into MSR on return to userspace). This can lead to a crash in the kernel if we take a floating point exception when restoring FPSCR: Oops: Exception in kernel mode, sig: 8 [#1] LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV Modules linked in: CPU: 3 PID: 101213 Comm: ld64.so.2 Not tainted 5.9.0-rc1-00098-g18445bf405cb-dirty #9 NIP: c00000000000fbb4 LR: c00000000001a7ac CTR: c000000000183570 REGS: c0000016b7cfb3b0 TRAP: 0700 Not tainted (5.9.0-rc1-00098-g18445bf405cb-dirty) MSR: 900000000290b933 <SF,HV,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 44002444 XER: 00000000 CFAR: c00000000001a7a8 IRQMASK: 1 GPR00: c00000000001ae40 c0000016b7cfb640 c0000000011b7f00 c000001542a0f740 GPR04: c000001542a0f720 c000001542a0eb00 0000000000000900 c000001542a0eb00 GPR08: 000000000000000a 0000000000002000 9000000000009033 0000000000000000 GPR12: 0000000000004000 c0000017ffffd900 0000000000000001 c000000000df5a58 GPR16: c000000000e19c18 c0000000010e1123 0000000000000001 c000000000e1a638 GPR20: 0000000000000000 c0000000044b1d00 0000000000000000 c000001542a0f2a0 GPR24: 00000016c7fe0000 c000001542a0f720 c000000001c93da0 c000000000fe5f28 GPR28: c000001542a0f720 0000000000800000 c0000016b7cfbe90 0000000002802900 NIP load_fp_state+0x4/0x214 LR restore_math+0x17c/0x1f0 Call Trace: 0xc0000016b7cfb680 (unreliable) __switch_to+0x330/0x460 __schedule+0x318/0x920 schedule+0x74/0x140 schedule_timeout+0x318/0x3f0 wait_for_completion+0xc8/0x210 call_usermodehelper_exec+0x234/0x280 do_coredump+0xedc/0x13c0 get_signal+0x1d4/0xbe0 do_notify_resume+0x1a0/0x490 interrupt_exit_user_prepare+0x1c4/0x230 interrupt_return+0x14/0x1c0 Instruction dump: ebe10168 e88101a0 7c8ff120 382101e0 e8010010 7c0803a6 4e800020 790605c4 782905c4 7c0008a8 7c0008a8 c8030200 <fffe058e> 48000088 c8030000 c8230010 Fix it by only loading the fpexc_mode value into regs->msr. Also add a comment to explain that although VSX is subject to the value of fpexc_mode, we don't have to handle that separately because we only allow VSX to be enabled if FP is also enabled. Fixes: 01eb01877f33 ("powerpc/64s: Fix restore_math unnecessarily changing MSR") Reported-by: Milton Miller <miltonm@us.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Reviewed-by: Nicholas Piggin <npiggin@gmail.com> Link: https://lore.kernel.org/r/20200825093424.3967813-1-mpe@ellerman.id.au
2020-08-25 17:34:24 +08:00
new_msr |= MSR_FP;
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
if ((!(msr & MSR_VEC)) && should_restore_altivec())
new_msr |= MSR_VEC;
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
if ((!(msr & MSR_VSX)) && should_restore_vsx()) {
if (((msr | new_msr) & (MSR_FP | MSR_VEC)) == (MSR_FP | MSR_VEC))
new_msr |= MSR_VSX;
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
}
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
if (new_msr) {
powerpc/64s: Fix crash in load_fp_state() due to fpexc_mode The recent commit 01eb01877f33 ("powerpc/64s: Fix restore_math unnecessarily changing MSR") changed some of the handling of floating point/vector restore. In particular it caused current->thread.fpexc_mode to be copied into the current MSR (via msr_check_and_set()), rather than just into regs->msr (which is moved into MSR on return to userspace). This can lead to a crash in the kernel if we take a floating point exception when restoring FPSCR: Oops: Exception in kernel mode, sig: 8 [#1] LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV Modules linked in: CPU: 3 PID: 101213 Comm: ld64.so.2 Not tainted 5.9.0-rc1-00098-g18445bf405cb-dirty #9 NIP: c00000000000fbb4 LR: c00000000001a7ac CTR: c000000000183570 REGS: c0000016b7cfb3b0 TRAP: 0700 Not tainted (5.9.0-rc1-00098-g18445bf405cb-dirty) MSR: 900000000290b933 <SF,HV,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 44002444 XER: 00000000 CFAR: c00000000001a7a8 IRQMASK: 1 GPR00: c00000000001ae40 c0000016b7cfb640 c0000000011b7f00 c000001542a0f740 GPR04: c000001542a0f720 c000001542a0eb00 0000000000000900 c000001542a0eb00 GPR08: 000000000000000a 0000000000002000 9000000000009033 0000000000000000 GPR12: 0000000000004000 c0000017ffffd900 0000000000000001 c000000000df5a58 GPR16: c000000000e19c18 c0000000010e1123 0000000000000001 c000000000e1a638 GPR20: 0000000000000000 c0000000044b1d00 0000000000000000 c000001542a0f2a0 GPR24: 00000016c7fe0000 c000001542a0f720 c000000001c93da0 c000000000fe5f28 GPR28: c000001542a0f720 0000000000800000 c0000016b7cfbe90 0000000002802900 NIP load_fp_state+0x4/0x214 LR restore_math+0x17c/0x1f0 Call Trace: 0xc0000016b7cfb680 (unreliable) __switch_to+0x330/0x460 __schedule+0x318/0x920 schedule+0x74/0x140 schedule_timeout+0x318/0x3f0 wait_for_completion+0xc8/0x210 call_usermodehelper_exec+0x234/0x280 do_coredump+0xedc/0x13c0 get_signal+0x1d4/0xbe0 do_notify_resume+0x1a0/0x490 interrupt_exit_user_prepare+0x1c4/0x230 interrupt_return+0x14/0x1c0 Instruction dump: ebe10168 e88101a0 7c8ff120 382101e0 e8010010 7c0803a6 4e800020 790605c4 782905c4 7c0008a8 7c0008a8 c8030200 <fffe058e> 48000088 c8030000 c8230010 Fix it by only loading the fpexc_mode value into regs->msr. Also add a comment to explain that although VSX is subject to the value of fpexc_mode, we don't have to handle that separately because we only allow VSX to be enabled if FP is also enabled. Fixes: 01eb01877f33 ("powerpc/64s: Fix restore_math unnecessarily changing MSR") Reported-by: Milton Miller <miltonm@us.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Reviewed-by: Nicholas Piggin <npiggin@gmail.com> Link: https://lore.kernel.org/r/20200825093424.3967813-1-mpe@ellerman.id.au
2020-08-25 17:34:24 +08:00
unsigned long fpexc_mode = 0;
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
msr_check_and_set(new_msr);
powerpc/64s: Fix crash in load_fp_state() due to fpexc_mode The recent commit 01eb01877f33 ("powerpc/64s: Fix restore_math unnecessarily changing MSR") changed some of the handling of floating point/vector restore. In particular it caused current->thread.fpexc_mode to be copied into the current MSR (via msr_check_and_set()), rather than just into regs->msr (which is moved into MSR on return to userspace). This can lead to a crash in the kernel if we take a floating point exception when restoring FPSCR: Oops: Exception in kernel mode, sig: 8 [#1] LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV Modules linked in: CPU: 3 PID: 101213 Comm: ld64.so.2 Not tainted 5.9.0-rc1-00098-g18445bf405cb-dirty #9 NIP: c00000000000fbb4 LR: c00000000001a7ac CTR: c000000000183570 REGS: c0000016b7cfb3b0 TRAP: 0700 Not tainted (5.9.0-rc1-00098-g18445bf405cb-dirty) MSR: 900000000290b933 <SF,HV,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 44002444 XER: 00000000 CFAR: c00000000001a7a8 IRQMASK: 1 GPR00: c00000000001ae40 c0000016b7cfb640 c0000000011b7f00 c000001542a0f740 GPR04: c000001542a0f720 c000001542a0eb00 0000000000000900 c000001542a0eb00 GPR08: 000000000000000a 0000000000002000 9000000000009033 0000000000000000 GPR12: 0000000000004000 c0000017ffffd900 0000000000000001 c000000000df5a58 GPR16: c000000000e19c18 c0000000010e1123 0000000000000001 c000000000e1a638 GPR20: 0000000000000000 c0000000044b1d00 0000000000000000 c000001542a0f2a0 GPR24: 00000016c7fe0000 c000001542a0f720 c000000001c93da0 c000000000fe5f28 GPR28: c000001542a0f720 0000000000800000 c0000016b7cfbe90 0000000002802900 NIP load_fp_state+0x4/0x214 LR restore_math+0x17c/0x1f0 Call Trace: 0xc0000016b7cfb680 (unreliable) __switch_to+0x330/0x460 __schedule+0x318/0x920 schedule+0x74/0x140 schedule_timeout+0x318/0x3f0 wait_for_completion+0xc8/0x210 call_usermodehelper_exec+0x234/0x280 do_coredump+0xedc/0x13c0 get_signal+0x1d4/0xbe0 do_notify_resume+0x1a0/0x490 interrupt_exit_user_prepare+0x1c4/0x230 interrupt_return+0x14/0x1c0 Instruction dump: ebe10168 e88101a0 7c8ff120 382101e0 e8010010 7c0803a6 4e800020 790605c4 782905c4 7c0008a8 7c0008a8 c8030200 <fffe058e> 48000088 c8030000 c8230010 Fix it by only loading the fpexc_mode value into regs->msr. Also add a comment to explain that although VSX is subject to the value of fpexc_mode, we don't have to handle that separately because we only allow VSX to be enabled if FP is also enabled. Fixes: 01eb01877f33 ("powerpc/64s: Fix restore_math unnecessarily changing MSR") Reported-by: Milton Miller <miltonm@us.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Reviewed-by: Nicholas Piggin <npiggin@gmail.com> Link: https://lore.kernel.org/r/20200825093424.3967813-1-mpe@ellerman.id.au
2020-08-25 17:34:24 +08:00
if (new_msr & MSR_FP) {
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
do_restore_fp();
powerpc/64s: Fix crash in load_fp_state() due to fpexc_mode The recent commit 01eb01877f33 ("powerpc/64s: Fix restore_math unnecessarily changing MSR") changed some of the handling of floating point/vector restore. In particular it caused current->thread.fpexc_mode to be copied into the current MSR (via msr_check_and_set()), rather than just into regs->msr (which is moved into MSR on return to userspace). This can lead to a crash in the kernel if we take a floating point exception when restoring FPSCR: Oops: Exception in kernel mode, sig: 8 [#1] LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV Modules linked in: CPU: 3 PID: 101213 Comm: ld64.so.2 Not tainted 5.9.0-rc1-00098-g18445bf405cb-dirty #9 NIP: c00000000000fbb4 LR: c00000000001a7ac CTR: c000000000183570 REGS: c0000016b7cfb3b0 TRAP: 0700 Not tainted (5.9.0-rc1-00098-g18445bf405cb-dirty) MSR: 900000000290b933 <SF,HV,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 44002444 XER: 00000000 CFAR: c00000000001a7a8 IRQMASK: 1 GPR00: c00000000001ae40 c0000016b7cfb640 c0000000011b7f00 c000001542a0f740 GPR04: c000001542a0f720 c000001542a0eb00 0000000000000900 c000001542a0eb00 GPR08: 000000000000000a 0000000000002000 9000000000009033 0000000000000000 GPR12: 0000000000004000 c0000017ffffd900 0000000000000001 c000000000df5a58 GPR16: c000000000e19c18 c0000000010e1123 0000000000000001 c000000000e1a638 GPR20: 0000000000000000 c0000000044b1d00 0000000000000000 c000001542a0f2a0 GPR24: 00000016c7fe0000 c000001542a0f720 c000000001c93da0 c000000000fe5f28 GPR28: c000001542a0f720 0000000000800000 c0000016b7cfbe90 0000000002802900 NIP load_fp_state+0x4/0x214 LR restore_math+0x17c/0x1f0 Call Trace: 0xc0000016b7cfb680 (unreliable) __switch_to+0x330/0x460 __schedule+0x318/0x920 schedule+0x74/0x140 schedule_timeout+0x318/0x3f0 wait_for_completion+0xc8/0x210 call_usermodehelper_exec+0x234/0x280 do_coredump+0xedc/0x13c0 get_signal+0x1d4/0xbe0 do_notify_resume+0x1a0/0x490 interrupt_exit_user_prepare+0x1c4/0x230 interrupt_return+0x14/0x1c0 Instruction dump: ebe10168 e88101a0 7c8ff120 382101e0 e8010010 7c0803a6 4e800020 790605c4 782905c4 7c0008a8 7c0008a8 c8030200 <fffe058e> 48000088 c8030000 c8230010 Fix it by only loading the fpexc_mode value into regs->msr. Also add a comment to explain that although VSX is subject to the value of fpexc_mode, we don't have to handle that separately because we only allow VSX to be enabled if FP is also enabled. Fixes: 01eb01877f33 ("powerpc/64s: Fix restore_math unnecessarily changing MSR") Reported-by: Milton Miller <miltonm@us.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Reviewed-by: Nicholas Piggin <npiggin@gmail.com> Link: https://lore.kernel.org/r/20200825093424.3967813-1-mpe@ellerman.id.au
2020-08-25 17:34:24 +08:00
// This also covers VSX, because VSX implies FP
fpexc_mode = current->thread.fpexc_mode;
}
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
if (new_msr & MSR_VEC)
do_restore_altivec();
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
if (new_msr & MSR_VSX)
do_restore_vsx();
msr_check_and_clear(new_msr);
powerpc/64s: Fix crash in load_fp_state() due to fpexc_mode The recent commit 01eb01877f33 ("powerpc/64s: Fix restore_math unnecessarily changing MSR") changed some of the handling of floating point/vector restore. In particular it caused current->thread.fpexc_mode to be copied into the current MSR (via msr_check_and_set()), rather than just into regs->msr (which is moved into MSR on return to userspace). This can lead to a crash in the kernel if we take a floating point exception when restoring FPSCR: Oops: Exception in kernel mode, sig: 8 [#1] LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV Modules linked in: CPU: 3 PID: 101213 Comm: ld64.so.2 Not tainted 5.9.0-rc1-00098-g18445bf405cb-dirty #9 NIP: c00000000000fbb4 LR: c00000000001a7ac CTR: c000000000183570 REGS: c0000016b7cfb3b0 TRAP: 0700 Not tainted (5.9.0-rc1-00098-g18445bf405cb-dirty) MSR: 900000000290b933 <SF,HV,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 44002444 XER: 00000000 CFAR: c00000000001a7a8 IRQMASK: 1 GPR00: c00000000001ae40 c0000016b7cfb640 c0000000011b7f00 c000001542a0f740 GPR04: c000001542a0f720 c000001542a0eb00 0000000000000900 c000001542a0eb00 GPR08: 000000000000000a 0000000000002000 9000000000009033 0000000000000000 GPR12: 0000000000004000 c0000017ffffd900 0000000000000001 c000000000df5a58 GPR16: c000000000e19c18 c0000000010e1123 0000000000000001 c000000000e1a638 GPR20: 0000000000000000 c0000000044b1d00 0000000000000000 c000001542a0f2a0 GPR24: 00000016c7fe0000 c000001542a0f720 c000000001c93da0 c000000000fe5f28 GPR28: c000001542a0f720 0000000000800000 c0000016b7cfbe90 0000000002802900 NIP load_fp_state+0x4/0x214 LR restore_math+0x17c/0x1f0 Call Trace: 0xc0000016b7cfb680 (unreliable) __switch_to+0x330/0x460 __schedule+0x318/0x920 schedule+0x74/0x140 schedule_timeout+0x318/0x3f0 wait_for_completion+0xc8/0x210 call_usermodehelper_exec+0x234/0x280 do_coredump+0xedc/0x13c0 get_signal+0x1d4/0xbe0 do_notify_resume+0x1a0/0x490 interrupt_exit_user_prepare+0x1c4/0x230 interrupt_return+0x14/0x1c0 Instruction dump: ebe10168 e88101a0 7c8ff120 382101e0 e8010010 7c0803a6 4e800020 790605c4 782905c4 7c0008a8 7c0008a8 c8030200 <fffe058e> 48000088 c8030000 c8230010 Fix it by only loading the fpexc_mode value into regs->msr. Also add a comment to explain that although VSX is subject to the value of fpexc_mode, we don't have to handle that separately because we only allow VSX to be enabled if FP is also enabled. Fixes: 01eb01877f33 ("powerpc/64s: Fix restore_math unnecessarily changing MSR") Reported-by: Milton Miller <miltonm@us.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Reviewed-by: Nicholas Piggin <npiggin@gmail.com> Link: https://lore.kernel.org/r/20200825093424.3967813-1-mpe@ellerman.id.au
2020-08-25 17:34:24 +08:00
regs->msr |= new_msr | fpexc_mode;
powerpc/64s: Fix restore_math unnecessarily changing MSR Before returning to user, if there are missing FP/VEC/VSX bits from the user MSR then those registers had been saved and must be restored again before use. restore_math will decide whether to restore immediately, or skip the restore and let fp/vec/vsx unavailable faults demand load the registers. Each time restore_math restores one of the FP/VSX or VEC register sets is loaded, an 8-bit counter is incremented (load_fp and load_vec). When these wrap to zero, restore_math no longer restores that register set until after they are next demand faulted. It's quite usual for those counters to have different values, so if one wraps to zero and restore_math no longer restores its registers or user MSR bit but the other is not zero yet does not need to be restored (because the kernel is not frequently using the FPU), then restore_math will be called and it will also not return in the early exit check. This causes msr_check_and_set to test and set the MSR at every kernel exit despite having no work to do. This can cause workloads (e.g., a NULL syscall microbenchmark) to run fast for a time while both counters are non-zero, then slow down when one of the counters reaches zero, then speed up again after the second counter reaches zero. The cost is significant, about 10% slowdown on a NULL syscall benchmark, and the jittery behaviour is very undesirable. Fix this by having restore_math test all conditions first, and only update MSR if we will be loading registers. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200623234139.2262227-2-npiggin@gmail.com
2020-06-24 07:41:38 +08:00
}
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
}
#endif /* CONFIG_PPC_BOOK3S_64 */
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
static void save_all(struct task_struct *tsk)
{
unsigned long usermsr;
if (!tsk->thread.regs)
return;
usermsr = tsk->thread.regs->msr;
if ((usermsr & msr_all_available) == 0)
return;
msr_check_and_set(msr_all_available);
WARN_ON((usermsr & MSR_VSX) && !((usermsr & MSR_FP) && (usermsr & MSR_VEC)));
if (usermsr & MSR_FP)
save_fpu(tsk);
if (usermsr & MSR_VEC)
save_altivec(tsk);
if (usermsr & MSR_SPE)
__giveup_spe(tsk);
msr_check_and_clear(msr_all_available);
}
void flush_all_to_thread(struct task_struct *tsk)
{
if (tsk->thread.regs) {
preempt_disable();
BUG_ON(tsk != current);
#ifdef CONFIG_SPE
if (tsk->thread.regs->msr & MSR_SPE)
tsk->thread.spefscr = mfspr(SPRN_SPEFSCR);
#endif
save_all(tsk);
preempt_enable();
}
}
EXPORT_SYMBOL(flush_all_to_thread);
#ifdef CONFIG_PPC_ADV_DEBUG_REGS
void do_send_trap(struct pt_regs *regs, unsigned long address,
unsigned long error_code, int breakpt)
{
current->thread.trap_nr = TRAP_HWBKPT;
if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
11, SIGSEGV) == NOTIFY_STOP)
return;
/* Deliver the signal to userspace */
force_sig_ptrace_errno_trap(breakpt, /* breakpoint or watchpoint id */
(void __user *)address);
}
#else /* !CONFIG_PPC_ADV_DEBUG_REGS */
static void do_break_handler(struct pt_regs *regs)
{
struct arch_hw_breakpoint null_brk = {0};
struct arch_hw_breakpoint *info;
struct ppc_inst instr = ppc_inst(0);
int type = 0;
int size = 0;
unsigned long ea;
int i;
/*
* If underneath hw supports only one watchpoint, we know it
* caused exception. 8xx also falls into this category.
*/
if (nr_wp_slots() == 1) {
__set_breakpoint(0, &null_brk);
current->thread.hw_brk[0] = null_brk;
current->thread.hw_brk[0].flags |= HW_BRK_FLAG_DISABLED;
return;
}
/* Otherwise findout which DAWR caused exception and disable it. */
wp_get_instr_detail(regs, &instr, &type, &size, &ea);
for (i = 0; i < nr_wp_slots(); i++) {
info = &current->thread.hw_brk[i];
if (!info->address)
continue;
if (wp_check_constraints(regs, instr, ea, type, size, info)) {
__set_breakpoint(i, &null_brk);
current->thread.hw_brk[i] = null_brk;
current->thread.hw_brk[i].flags |= HW_BRK_FLAG_DISABLED;
}
}
}
DEFINE_INTERRUPT_HANDLER(do_break)
{
current->thread.trap_nr = TRAP_HWBKPT;
if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, regs->dsisr,
11, SIGSEGV) == NOTIFY_STOP)
return;
if (debugger_break_match(regs))
return;
/*
* We reach here only when watchpoint exception is generated by ptrace
* event (or hw is buggy!). Now if CONFIG_HAVE_HW_BREAKPOINT is set,
* watchpoint is already handled by hw_breakpoint_handler() so we don't
* have to do anything. But when CONFIG_HAVE_HW_BREAKPOINT is not set,
* we need to manually handle the watchpoint here.
*/
if (!IS_ENABLED(CONFIG_HAVE_HW_BREAKPOINT))
do_break_handler(regs);
/* Deliver the signal to userspace */
force_sig_fault(SIGTRAP, TRAP_HWBKPT, (void __user *)regs->dar);
}
#endif /* CONFIG_PPC_ADV_DEBUG_REGS */
static DEFINE_PER_CPU(struct arch_hw_breakpoint, current_brk[HBP_NUM_MAX]);
#ifdef CONFIG_PPC_ADV_DEBUG_REGS
/*
* Set the debug registers back to their default "safe" values.
*/
static void set_debug_reg_defaults(struct thread_struct *thread)
{
thread->debug.iac1 = thread->debug.iac2 = 0;
#if CONFIG_PPC_ADV_DEBUG_IACS > 2
thread->debug.iac3 = thread->debug.iac4 = 0;
#endif
thread->debug.dac1 = thread->debug.dac2 = 0;
#if CONFIG_PPC_ADV_DEBUG_DVCS > 0
thread->debug.dvc1 = thread->debug.dvc2 = 0;
#endif
thread->debug.dbcr0 = 0;
#ifdef CONFIG_BOOKE
/*
* Force User/Supervisor bits to b11 (user-only MSR[PR]=1)
*/
thread->debug.dbcr1 = DBCR1_IAC1US | DBCR1_IAC2US |
DBCR1_IAC3US | DBCR1_IAC4US;
/*
* Force Data Address Compare User/Supervisor bits to be User-only
* (0b11 MSR[PR]=1) and set all other bits in DBCR2 register to be 0.
*/
thread->debug.dbcr2 = DBCR2_DAC1US | DBCR2_DAC2US;
#else
thread->debug.dbcr1 = 0;
#endif
}
static void prime_debug_regs(struct debug_reg *debug)
{
/*
* We could have inherited MSR_DE from userspace, since
* it doesn't get cleared on exception entry. Make sure
* MSR_DE is clear before we enable any debug events.
*/
mtmsr(mfmsr() & ~MSR_DE);
mtspr(SPRN_IAC1, debug->iac1);
mtspr(SPRN_IAC2, debug->iac2);
#if CONFIG_PPC_ADV_DEBUG_IACS > 2
mtspr(SPRN_IAC3, debug->iac3);
mtspr(SPRN_IAC4, debug->iac4);
#endif
mtspr(SPRN_DAC1, debug->dac1);
mtspr(SPRN_DAC2, debug->dac2);
#if CONFIG_PPC_ADV_DEBUG_DVCS > 0
mtspr(SPRN_DVC1, debug->dvc1);
mtspr(SPRN_DVC2, debug->dvc2);
#endif
mtspr(SPRN_DBCR0, debug->dbcr0);
mtspr(SPRN_DBCR1, debug->dbcr1);
#ifdef CONFIG_BOOKE
mtspr(SPRN_DBCR2, debug->dbcr2);
#endif
}
/*
* Unless neither the old or new thread are making use of the
* debug registers, set the debug registers from the values
* stored in the new thread.
*/
void switch_booke_debug_regs(struct debug_reg *new_debug)
{
if ((current->thread.debug.dbcr0 & DBCR0_IDM)
|| (new_debug->dbcr0 & DBCR0_IDM))
prime_debug_regs(new_debug);
}
EXPORT_SYMBOL_GPL(switch_booke_debug_regs);
#else /* !CONFIG_PPC_ADV_DEBUG_REGS */
#ifndef CONFIG_HAVE_HW_BREAKPOINT
static void set_breakpoint(int i, struct arch_hw_breakpoint *brk)
{
preempt_disable();
__set_breakpoint(i, brk);
preempt_enable();
}
static void set_debug_reg_defaults(struct thread_struct *thread)
{
int i;
struct arch_hw_breakpoint null_brk = {0};
for (i = 0; i < nr_wp_slots(); i++) {
thread->hw_brk[i] = null_brk;
if (ppc_breakpoint_available())
set_breakpoint(i, &thread->hw_brk[i]);
}
}
static inline bool hw_brk_match(struct arch_hw_breakpoint *a,
struct arch_hw_breakpoint *b)
{
if (a->address != b->address)
return false;
if (a->type != b->type)
return false;
if (a->len != b->len)
return false;
/* no need to check hw_len. it's calculated from address and len */
return true;
}
static void switch_hw_breakpoint(struct task_struct *new)
{
int i;
for (i = 0; i < nr_wp_slots(); i++) {
if (likely(hw_brk_match(this_cpu_ptr(&current_brk[i]),
&new->thread.hw_brk[i])))
continue;
__set_breakpoint(i, &new->thread.hw_brk[i]);
}
}
#endif /* !CONFIG_HAVE_HW_BREAKPOINT */
#endif /* CONFIG_PPC_ADV_DEBUG_REGS */
static inline int set_dabr(struct arch_hw_breakpoint *brk)
{
unsigned long dabr, dabrx;
dabr = brk->address | (brk->type & HW_BRK_TYPE_DABR);
dabrx = ((brk->type >> 3) & 0x7);
if (ppc_md.set_dabr)
return ppc_md.set_dabr(dabr, dabrx);
if (IS_ENABLED(CONFIG_PPC_ADV_DEBUG_REGS)) {
mtspr(SPRN_DAC1, dabr);
if (IS_ENABLED(CONFIG_PPC_47x))
isync();
return 0;
} else if (IS_ENABLED(CONFIG_PPC_BOOK3S)) {
mtspr(SPRN_DABR, dabr);
if (cpu_has_feature(CPU_FTR_DABRX))
mtspr(SPRN_DABRX, dabrx);
return 0;
} else {
return -EINVAL;
}
}
static inline int set_breakpoint_8xx(struct arch_hw_breakpoint *brk)
{
unsigned long lctrl1 = LCTRL1_CTE_GT | LCTRL1_CTF_LT | LCTRL1_CRWE_RW |
LCTRL1_CRWF_RW;
unsigned long lctrl2 = LCTRL2_LW0EN | LCTRL2_LW0LADC | LCTRL2_SLW0EN;
unsigned long start_addr = ALIGN_DOWN(brk->address, HW_BREAKPOINT_SIZE);
unsigned long end_addr = ALIGN(brk->address + brk->len, HW_BREAKPOINT_SIZE);
if (start_addr == 0)
lctrl2 |= LCTRL2_LW0LA_F;
else if (end_addr == 0)
lctrl2 |= LCTRL2_LW0LA_E;
else
lctrl2 |= LCTRL2_LW0LA_EandF;
mtspr(SPRN_LCTRL2, 0);
if ((brk->type & HW_BRK_TYPE_RDWR) == 0)
return 0;
if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_READ)
lctrl1 |= LCTRL1_CRWE_RO | LCTRL1_CRWF_RO;
if ((brk->type & HW_BRK_TYPE_RDWR) == HW_BRK_TYPE_WRITE)
lctrl1 |= LCTRL1_CRWE_WO | LCTRL1_CRWF_WO;
mtspr(SPRN_CMPE, start_addr - 1);
mtspr(SPRN_CMPF, end_addr);
mtspr(SPRN_LCTRL1, lctrl1);
mtspr(SPRN_LCTRL2, lctrl2);
return 0;
}
void __set_breakpoint(int nr, struct arch_hw_breakpoint *brk)
{
memcpy(this_cpu_ptr(&current_brk[nr]), brk, sizeof(*brk));
powerpc: Add force enable of DAWR on P9 option This adds a flag so that the DAWR can be enabled on P9 via: echo Y > /sys/kernel/debug/powerpc/dawr_enable_dangerous The DAWR was previously force disabled on POWER9 in: 9654153158 powerpc: Disable DAWR in the base POWER9 CPU features Also see Documentation/powerpc/DAWR-POWER9.txt This is a dangerous setting, USE AT YOUR OWN RISK. Some users may not care about a bad user crashing their box (ie. single user/desktop systems) and really want the DAWR. This allows them to force enable DAWR. This flag can also be used to disable DAWR access. Once this is cleared, all DAWR access should be cleared immediately and your machine once again safe from crashing. Userspace may get confused by toggling this. If DAWR is force enabled/disabled between getting the number of breakpoints (via PTRACE_GETHWDBGINFO) and setting the breakpoint, userspace will get an inconsistent view of what's available. Similarly for guests. For the DAWR to be enabled in a KVM guest, the DAWR needs to be force enabled in the host AND the guest. For this reason, this won't work on POWERVM as it doesn't allow the HCALL to work. Writes of 'Y' to the dawr_enable_dangerous file will fail if the hypervisor doesn't support writing the DAWR. To double check the DAWR is working, run this kernel selftest: tools/testing/selftests/powerpc/ptrace/ptrace-hwbreak.c Any errors/failures/skips mean something is wrong. Signed-off-by: Michael Neuling <mikey@neuling.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2019-04-01 14:03:12 +08:00
if (dawr_enabled())
// Power8 or later
set_dawr(nr, brk);
else if (IS_ENABLED(CONFIG_PPC_8xx))
set_breakpoint_8xx(brk);
else if (!cpu_has_feature(CPU_FTR_ARCH_207S))
// Power7 or earlier
set_dabr(brk);
else
// Shouldn't happen due to higher level checks
WARN_ON_ONCE(1);
}
/* Check if we have DAWR or DABR hardware */
bool ppc_breakpoint_available(void)
{
powerpc: Add force enable of DAWR on P9 option This adds a flag so that the DAWR can be enabled on P9 via: echo Y > /sys/kernel/debug/powerpc/dawr_enable_dangerous The DAWR was previously force disabled on POWER9 in: 9654153158 powerpc: Disable DAWR in the base POWER9 CPU features Also see Documentation/powerpc/DAWR-POWER9.txt This is a dangerous setting, USE AT YOUR OWN RISK. Some users may not care about a bad user crashing their box (ie. single user/desktop systems) and really want the DAWR. This allows them to force enable DAWR. This flag can also be used to disable DAWR access. Once this is cleared, all DAWR access should be cleared immediately and your machine once again safe from crashing. Userspace may get confused by toggling this. If DAWR is force enabled/disabled between getting the number of breakpoints (via PTRACE_GETHWDBGINFO) and setting the breakpoint, userspace will get an inconsistent view of what's available. Similarly for guests. For the DAWR to be enabled in a KVM guest, the DAWR needs to be force enabled in the host AND the guest. For this reason, this won't work on POWERVM as it doesn't allow the HCALL to work. Writes of 'Y' to the dawr_enable_dangerous file will fail if the hypervisor doesn't support writing the DAWR. To double check the DAWR is working, run this kernel selftest: tools/testing/selftests/powerpc/ptrace/ptrace-hwbreak.c Any errors/failures/skips mean something is wrong. Signed-off-by: Michael Neuling <mikey@neuling.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2019-04-01 14:03:12 +08:00
if (dawr_enabled())
return true; /* POWER8 DAWR or POWER9 forced DAWR */
if (cpu_has_feature(CPU_FTR_ARCH_207S))
return false; /* POWER9 with DAWR disabled */
/* DABR: Everything but POWER8 and POWER9 */
return true;
}
EXPORT_SYMBOL_GPL(ppc_breakpoint_available);
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
static inline bool tm_enabled(struct task_struct *tsk)
{
return tsk && tsk->thread.regs && (tsk->thread.regs->msr & MSR_TM);
}
static void tm_reclaim_thread(struct thread_struct *thr, uint8_t cause)
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
{
powerpc/tm: Check for already reclaimed tasks Currently we can hit a scenario where we'll tm_reclaim() twice. This results in a TM bad thing exception because the second reclaim occurs when not in suspend mode. The scenario in which this can happen is the following. We attempt to deliver a signal to userspace. To do this we need obtain the stack pointer to write the signal context. To get this stack pointer we must tm_reclaim() in case we need to use the checkpointed stack pointer (see get_tm_stackpointer()). Normally we'd then return directly to userspace to deliver the signal without going through __switch_to(). Unfortunatley, if at this point we get an error (such as a bad userspace stack pointer), we need to exit the process. The exit will result in a __switch_to(). __switch_to() will attempt to save the process state which results in another tm_reclaim(). This tm_reclaim() now causes a TM Bad Thing exception as this state has already been saved and the processor is no longer in TM suspend mode. Whee! This patch checks the state of the MSR to ensure we are TM suspended before we attempt the tm_reclaim(). If we've already saved the state away, we should no longer be in TM suspend mode. This has the additional advantage of checking for a potential TM Bad Thing exception. Found using syscall fuzzer. Fixes: fb09692e71f1 ("powerpc: Add reclaim and recheckpoint functions for context switching transactional memory processes") Cc: stable@vger.kernel.org # v3.9+ Signed-off-by: Michael Neuling <mikey@neuling.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2015-11-19 12:44:45 +08:00
/*
* Use the current MSR TM suspended bit to track if we have
* checkpointed state outstanding.
* On signal delivery, we'd normally reclaim the checkpointed
* state to obtain stack pointer (see:get_tm_stackpointer()).
* This will then directly return to userspace without going
* through __switch_to(). However, if the stack frame is bad,
* we need to exit this thread which calls __switch_to() which
* will again attempt to reclaim the already saved tm state.
* Hence we need to check that we've not already reclaimed
* this state.
* We do this using the current MSR, rather tracking it in
* some specific thread_struct bit, as it has the additional
* benefit of checking for a potential TM bad thing exception.
powerpc/tm: Check for already reclaimed tasks Currently we can hit a scenario where we'll tm_reclaim() twice. This results in a TM bad thing exception because the second reclaim occurs when not in suspend mode. The scenario in which this can happen is the following. We attempt to deliver a signal to userspace. To do this we need obtain the stack pointer to write the signal context. To get this stack pointer we must tm_reclaim() in case we need to use the checkpointed stack pointer (see get_tm_stackpointer()). Normally we'd then return directly to userspace to deliver the signal without going through __switch_to(). Unfortunatley, if at this point we get an error (such as a bad userspace stack pointer), we need to exit the process. The exit will result in a __switch_to(). __switch_to() will attempt to save the process state which results in another tm_reclaim(). This tm_reclaim() now causes a TM Bad Thing exception as this state has already been saved and the processor is no longer in TM suspend mode. Whee! This patch checks the state of the MSR to ensure we are TM suspended before we attempt the tm_reclaim(). If we've already saved the state away, we should no longer be in TM suspend mode. This has the additional advantage of checking for a potential TM Bad Thing exception. Found using syscall fuzzer. Fixes: fb09692e71f1 ("powerpc: Add reclaim and recheckpoint functions for context switching transactional memory processes") Cc: stable@vger.kernel.org # v3.9+ Signed-off-by: Michael Neuling <mikey@neuling.org> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2015-11-19 12:44:45 +08:00
*/
if (!MSR_TM_SUSPENDED(mfmsr()))
return;
powerpc: Force reload for recheckpoint during tm {fp, vec, vsx} unavailable exception Lazy save and restore of FP/Altivec means that a userspace process can be sent to userspace with FP or Altivec disabled and loaded only as required (by way of an FP/Altivec unavailable exception). Transactional Memory complicates this situation as a transaction could be started without FP/Altivec being loaded up. This causes the hardware to checkpoint incorrect registers. Handling FP/Altivec unavailable exceptions while a thread is transactional requires a reclaim and recheckpoint to ensure the CPU has correct state for both sets of registers. tm_reclaim() has optimisations to not always save the FP/Altivec registers to the checkpointed save area. This was originally done because the caller might have information that the checkpointed registers aren't valid due to lazy save and restore. We've also been a little vague as to how tm_reclaim() leaves the FP/Altivec state since it doesn't necessarily always save it to the thread struct. This has lead to an (incorrect) assumption that it leaves the checkpointed state on the CPU. tm_recheckpoint() has similar optimisations in reverse. It may not always reload the checkpointed FP/Altivec registers from the thread struct before the trecheckpoint. It is therefore quite unclear where it expects to get the state from. This didn't help with the assumption made about tm_reclaim(). This patch is a minimal fix for ease of backporting. A more correct fix which removes the msr parameter to tm_reclaim() and tm_recheckpoint() altogether has been upstreamed to apply on top of this patch. Fixes: dc3106690b20 ("powerpc: tm: Always use fp_state and vr_state to store live registers") Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-11-02 11:09:04 +08:00
giveup_all(container_of(thr, struct task_struct, thread));
powerpc: Always save/restore checkpointed regs during treclaim/trecheckpoint Lazy save and restore of FP/Altivec means that a userspace process can be sent to userspace with FP or Altivec disabled and loaded only as required (by way of an FP/Altivec unavailable exception). Transactional Memory complicates this situation as a transaction could be started without FP/Altivec being loaded up. This causes the hardware to checkpoint incorrect registers. Handling FP/Altivec unavailable exceptions while a thread is transactional requires a reclaim and recheckpoint to ensure the CPU has correct state for both sets of registers. tm_reclaim() has optimisations to not always save the FP/Altivec registers to the checkpointed save area. This was originally done because the caller might have information that the checkpointed registers aren't valid due to lazy save and restore. We've also been a little vague as to how tm_reclaim() leaves the FP/Altivec state since it doesn't necessarily always save it to the thread struct. This has lead to an (incorrect) assumption that it leaves the checkpointed state on the CPU. tm_recheckpoint() has similar optimisations in reverse. It may not always reload the checkpointed FP/Altivec registers from the thread struct before the trecheckpoint. It is therefore quite unclear where it expects to get the state from. This didn't help with the assumption made about tm_reclaim(). These optimisations sit in what is by definition a slow path. If a process has to go through a reclaim/recheckpoint then its transaction will be doomed on returning to userspace. This mean that the process will be unable to complete its transaction and be forced to its failure handler. This is already an out if line case for userspace. Furthermore, the cost of copying 64 times 128 bits from registers isn't very long[0] (at all) on modern processors. As such it appears these optimisations have only served to increase code complexity and are unlikely to have had a measurable performance impact. Our transactional memory handling has been riddled with bugs. A cause of this has been difficulty in following the code flow, code complexity has not been our friend here. It makes sense to remove these optimisations in favour of a (hopefully) more stable implementation. This patch does mean that some times the assembly will needlessly save 'junk' registers which will subsequently get overwritten with the correct value by the C code which calls the assembly function. This small inefficiency is far outweighed by the reduction in complexity for general TM code, context switching paths, and transactional facility unavailable exception handler. 0: I tried to measure it once for other work and found that it was hiding in the noise of everything else I was working with. I find it exceedingly likely this will be the case here. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-11-02 11:09:05 +08:00
tm_reclaim(thr, cause);
/*
* If we are in a transaction and FP is off then we can't have
* used FP inside that transaction. Hence the checkpointed
* state is the same as the live state. We need to copy the
* live state to the checkpointed state so that when the
* transaction is restored, the checkpointed state is correct
* and the aborted transaction sees the correct state. We use
* ckpt_regs.msr here as that's what tm_reclaim will use to
* determine if it's going to write the checkpointed state or
* not. So either this will write the checkpointed registers,
* or reclaim will. Similarly for VMX.
*/
if ((thr->ckpt_regs.msr & MSR_FP) == 0)
memcpy(&thr->ckfp_state, &thr->fp_state,
sizeof(struct thread_fp_state));
if ((thr->ckpt_regs.msr & MSR_VEC) == 0)
memcpy(&thr->ckvr_state, &thr->vr_state,
sizeof(struct thread_vr_state));
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
}
void tm_reclaim_current(uint8_t cause)
{
tm_enable();
tm_reclaim_thread(&current->thread, cause);
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
}
static inline void tm_reclaim_task(struct task_struct *tsk)
{
/* We have to work out if we're switching from/to a task that's in the
* middle of a transaction.
*
* In switching we need to maintain a 2nd register state as
* oldtask->thread.ckpt_regs. We tm_reclaim(oldproc); this saves the
* checkpointed (tbegin) state in ckpt_regs, ckfp_state and
* ckvr_state
*
* We also context switch (save) TFHAR/TEXASR/TFIAR in here.
*/
struct thread_struct *thr = &tsk->thread;
if (!thr->regs)
return;
if (!MSR_TM_ACTIVE(thr->regs->msr))
goto out_and_saveregs;
WARN_ON(tm_suspend_disabled);
TM_DEBUG("--- tm_reclaim on pid %d (NIP=%lx, "
"ccr=%lx, msr=%lx, trap=%lx)\n",
tsk->pid, thr->regs->nip,
thr->regs->ccr, thr->regs->msr,
thr->regs->trap);
tm_reclaim_thread(thr, TM_CAUSE_RESCHED);
TM_DEBUG("--- tm_reclaim on pid %d complete\n",
tsk->pid);
out_and_saveregs:
/* Always save the regs here, even if a transaction's not active.
* This context-switches a thread's TM info SPRs. We do it here to
* be consistent with the restore path (in recheckpoint) which
* cannot happen later in _switch().
*/
tm_save_sprs(thr);
}
powerpc: Always save/restore checkpointed regs during treclaim/trecheckpoint Lazy save and restore of FP/Altivec means that a userspace process can be sent to userspace with FP or Altivec disabled and loaded only as required (by way of an FP/Altivec unavailable exception). Transactional Memory complicates this situation as a transaction could be started without FP/Altivec being loaded up. This causes the hardware to checkpoint incorrect registers. Handling FP/Altivec unavailable exceptions while a thread is transactional requires a reclaim and recheckpoint to ensure the CPU has correct state for both sets of registers. tm_reclaim() has optimisations to not always save the FP/Altivec registers to the checkpointed save area. This was originally done because the caller might have information that the checkpointed registers aren't valid due to lazy save and restore. We've also been a little vague as to how tm_reclaim() leaves the FP/Altivec state since it doesn't necessarily always save it to the thread struct. This has lead to an (incorrect) assumption that it leaves the checkpointed state on the CPU. tm_recheckpoint() has similar optimisations in reverse. It may not always reload the checkpointed FP/Altivec registers from the thread struct before the trecheckpoint. It is therefore quite unclear where it expects to get the state from. This didn't help with the assumption made about tm_reclaim(). These optimisations sit in what is by definition a slow path. If a process has to go through a reclaim/recheckpoint then its transaction will be doomed on returning to userspace. This mean that the process will be unable to complete its transaction and be forced to its failure handler. This is already an out if line case for userspace. Furthermore, the cost of copying 64 times 128 bits from registers isn't very long[0] (at all) on modern processors. As such it appears these optimisations have only served to increase code complexity and are unlikely to have had a measurable performance impact. Our transactional memory handling has been riddled with bugs. A cause of this has been difficulty in following the code flow, code complexity has not been our friend here. It makes sense to remove these optimisations in favour of a (hopefully) more stable implementation. This patch does mean that some times the assembly will needlessly save 'junk' registers which will subsequently get overwritten with the correct value by the C code which calls the assembly function. This small inefficiency is far outweighed by the reduction in complexity for general TM code, context switching paths, and transactional facility unavailable exception handler. 0: I tried to measure it once for other work and found that it was hiding in the noise of everything else I was working with. I find it exceedingly likely this will be the case here. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-11-02 11:09:05 +08:00
extern void __tm_recheckpoint(struct thread_struct *thread);
powerpc/tm: Disable IRQ in tm_recheckpoint We can't take an IRQ when we're about to do a trechkpt as our GPR state is set to user GPR values. We've hit this when running some IBM Java stress tests in the lab resulting in the following dump: cpu 0x3f: Vector: 700 (Program Check) at [c000000007eb3d40] pc: c000000000050074: restore_gprs+0xc0/0x148 lr: 00000000b52a8184 sp: ac57d360 msr: 8000000100201030 current = 0xc00000002c500000 paca = 0xc000000007dbfc00 softe: 0 irq_happened: 0x00 pid = 34535, comm = Pooled Thread # R00 = 00000000b52a8184 R16 = 00000000b3e48fda R01 = 00000000ac57d360 R17 = 00000000ade79bd8 R02 = 00000000ac586930 R18 = 000000000fac9bcc R03 = 00000000ade60000 R19 = 00000000ac57f930 R04 = 00000000f6624918 R20 = 00000000ade79be8 R05 = 00000000f663f238 R21 = 00000000ac218a54 R06 = 0000000000000002 R22 = 000000000f956280 R07 = 0000000000000008 R23 = 000000000000007e R08 = 000000000000000a R24 = 000000000000000c R09 = 00000000b6e69160 R25 = 00000000b424cf00 R10 = 0000000000000181 R26 = 00000000f66256d4 R11 = 000000000f365ec0 R27 = 00000000b6fdcdd0 R12 = 00000000f66400f0 R28 = 0000000000000001 R13 = 00000000ada71900 R29 = 00000000ade5a300 R14 = 00000000ac2185a8 R30 = 00000000f663f238 R15 = 0000000000000004 R31 = 00000000f6624918 pc = c000000000050074 restore_gprs+0xc0/0x148 cfar= c00000000004fe28 dont_restore_vec+0x1c/0x1a4 lr = 00000000b52a8184 msr = 8000000100201030 cr = 24804888 ctr = 0000000000000000 xer = 0000000000000000 trap = 700 This moves tm_recheckpoint to a C function and moves the tm_restore_sprs into that function. It then adds IRQ disabling over the trechkpt critical section. It also sets the TEXASR FS in the signals code to ensure this is never set now that we explictly write the TM sprs in tm_recheckpoint. Signed-off-by: Michael Neuling <mikey@neuling.org> cc: stable@vger.kernel.org Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-04-04 17:19:48 +08:00
powerpc: Always save/restore checkpointed regs during treclaim/trecheckpoint Lazy save and restore of FP/Altivec means that a userspace process can be sent to userspace with FP or Altivec disabled and loaded only as required (by way of an FP/Altivec unavailable exception). Transactional Memory complicates this situation as a transaction could be started without FP/Altivec being loaded up. This causes the hardware to checkpoint incorrect registers. Handling FP/Altivec unavailable exceptions while a thread is transactional requires a reclaim and recheckpoint to ensure the CPU has correct state for both sets of registers. tm_reclaim() has optimisations to not always save the FP/Altivec registers to the checkpointed save area. This was originally done because the caller might have information that the checkpointed registers aren't valid due to lazy save and restore. We've also been a little vague as to how tm_reclaim() leaves the FP/Altivec state since it doesn't necessarily always save it to the thread struct. This has lead to an (incorrect) assumption that it leaves the checkpointed state on the CPU. tm_recheckpoint() has similar optimisations in reverse. It may not always reload the checkpointed FP/Altivec registers from the thread struct before the trecheckpoint. It is therefore quite unclear where it expects to get the state from. This didn't help with the assumption made about tm_reclaim(). These optimisations sit in what is by definition a slow path. If a process has to go through a reclaim/recheckpoint then its transaction will be doomed on returning to userspace. This mean that the process will be unable to complete its transaction and be forced to its failure handler. This is already an out if line case for userspace. Furthermore, the cost of copying 64 times 128 bits from registers isn't very long[0] (at all) on modern processors. As such it appears these optimisations have only served to increase code complexity and are unlikely to have had a measurable performance impact. Our transactional memory handling has been riddled with bugs. A cause of this has been difficulty in following the code flow, code complexity has not been our friend here. It makes sense to remove these optimisations in favour of a (hopefully) more stable implementation. This patch does mean that some times the assembly will needlessly save 'junk' registers which will subsequently get overwritten with the correct value by the C code which calls the assembly function. This small inefficiency is far outweighed by the reduction in complexity for general TM code, context switching paths, and transactional facility unavailable exception handler. 0: I tried to measure it once for other work and found that it was hiding in the noise of everything else I was working with. I find it exceedingly likely this will be the case here. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-11-02 11:09:05 +08:00
void tm_recheckpoint(struct thread_struct *thread)
powerpc/tm: Disable IRQ in tm_recheckpoint We can't take an IRQ when we're about to do a trechkpt as our GPR state is set to user GPR values. We've hit this when running some IBM Java stress tests in the lab resulting in the following dump: cpu 0x3f: Vector: 700 (Program Check) at [c000000007eb3d40] pc: c000000000050074: restore_gprs+0xc0/0x148 lr: 00000000b52a8184 sp: ac57d360 msr: 8000000100201030 current = 0xc00000002c500000 paca = 0xc000000007dbfc00 softe: 0 irq_happened: 0x00 pid = 34535, comm = Pooled Thread # R00 = 00000000b52a8184 R16 = 00000000b3e48fda R01 = 00000000ac57d360 R17 = 00000000ade79bd8 R02 = 00000000ac586930 R18 = 000000000fac9bcc R03 = 00000000ade60000 R19 = 00000000ac57f930 R04 = 00000000f6624918 R20 = 00000000ade79be8 R05 = 00000000f663f238 R21 = 00000000ac218a54 R06 = 0000000000000002 R22 = 000000000f956280 R07 = 0000000000000008 R23 = 000000000000007e R08 = 000000000000000a R24 = 000000000000000c R09 = 00000000b6e69160 R25 = 00000000b424cf00 R10 = 0000000000000181 R26 = 00000000f66256d4 R11 = 000000000f365ec0 R27 = 00000000b6fdcdd0 R12 = 00000000f66400f0 R28 = 0000000000000001 R13 = 00000000ada71900 R29 = 00000000ade5a300 R14 = 00000000ac2185a8 R30 = 00000000f663f238 R15 = 0000000000000004 R31 = 00000000f6624918 pc = c000000000050074 restore_gprs+0xc0/0x148 cfar= c00000000004fe28 dont_restore_vec+0x1c/0x1a4 lr = 00000000b52a8184 msr = 8000000100201030 cr = 24804888 ctr = 0000000000000000 xer = 0000000000000000 trap = 700 This moves tm_recheckpoint to a C function and moves the tm_restore_sprs into that function. It then adds IRQ disabling over the trechkpt critical section. It also sets the TEXASR FS in the signals code to ensure this is never set now that we explictly write the TM sprs in tm_recheckpoint. Signed-off-by: Michael Neuling <mikey@neuling.org> cc: stable@vger.kernel.org Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-04-04 17:19:48 +08:00
{
unsigned long flags;
if (!(thread->regs->msr & MSR_TM))
return;
powerpc/tm: Disable IRQ in tm_recheckpoint We can't take an IRQ when we're about to do a trechkpt as our GPR state is set to user GPR values. We've hit this when running some IBM Java stress tests in the lab resulting in the following dump: cpu 0x3f: Vector: 700 (Program Check) at [c000000007eb3d40] pc: c000000000050074: restore_gprs+0xc0/0x148 lr: 00000000b52a8184 sp: ac57d360 msr: 8000000100201030 current = 0xc00000002c500000 paca = 0xc000000007dbfc00 softe: 0 irq_happened: 0x00 pid = 34535, comm = Pooled Thread # R00 = 00000000b52a8184 R16 = 00000000b3e48fda R01 = 00000000ac57d360 R17 = 00000000ade79bd8 R02 = 00000000ac586930 R18 = 000000000fac9bcc R03 = 00000000ade60000 R19 = 00000000ac57f930 R04 = 00000000f6624918 R20 = 00000000ade79be8 R05 = 00000000f663f238 R21 = 00000000ac218a54 R06 = 0000000000000002 R22 = 000000000f956280 R07 = 0000000000000008 R23 = 000000000000007e R08 = 000000000000000a R24 = 000000000000000c R09 = 00000000b6e69160 R25 = 00000000b424cf00 R10 = 0000000000000181 R26 = 00000000f66256d4 R11 = 000000000f365ec0 R27 = 00000000b6fdcdd0 R12 = 00000000f66400f0 R28 = 0000000000000001 R13 = 00000000ada71900 R29 = 00000000ade5a300 R14 = 00000000ac2185a8 R30 = 00000000f663f238 R15 = 0000000000000004 R31 = 00000000f6624918 pc = c000000000050074 restore_gprs+0xc0/0x148 cfar= c00000000004fe28 dont_restore_vec+0x1c/0x1a4 lr = 00000000b52a8184 msr = 8000000100201030 cr = 24804888 ctr = 0000000000000000 xer = 0000000000000000 trap = 700 This moves tm_recheckpoint to a C function and moves the tm_restore_sprs into that function. It then adds IRQ disabling over the trechkpt critical section. It also sets the TEXASR FS in the signals code to ensure this is never set now that we explictly write the TM sprs in tm_recheckpoint. Signed-off-by: Michael Neuling <mikey@neuling.org> cc: stable@vger.kernel.org Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-04-04 17:19:48 +08:00
/* We really can't be interrupted here as the TEXASR registers can't
* change and later in the trecheckpoint code, we have a userspace R1.
* So let's hard disable over this region.
*/
local_irq_save(flags);
hard_irq_disable();
/* The TM SPRs are restored here, so that TEXASR.FS can be set
* before the trecheckpoint and no explosion occurs.
*/
tm_restore_sprs(thread);
powerpc: Always save/restore checkpointed regs during treclaim/trecheckpoint Lazy save and restore of FP/Altivec means that a userspace process can be sent to userspace with FP or Altivec disabled and loaded only as required (by way of an FP/Altivec unavailable exception). Transactional Memory complicates this situation as a transaction could be started without FP/Altivec being loaded up. This causes the hardware to checkpoint incorrect registers. Handling FP/Altivec unavailable exceptions while a thread is transactional requires a reclaim and recheckpoint to ensure the CPU has correct state for both sets of registers. tm_reclaim() has optimisations to not always save the FP/Altivec registers to the checkpointed save area. This was originally done because the caller might have information that the checkpointed registers aren't valid due to lazy save and restore. We've also been a little vague as to how tm_reclaim() leaves the FP/Altivec state since it doesn't necessarily always save it to the thread struct. This has lead to an (incorrect) assumption that it leaves the checkpointed state on the CPU. tm_recheckpoint() has similar optimisations in reverse. It may not always reload the checkpointed FP/Altivec registers from the thread struct before the trecheckpoint. It is therefore quite unclear where it expects to get the state from. This didn't help with the assumption made about tm_reclaim(). These optimisations sit in what is by definition a slow path. If a process has to go through a reclaim/recheckpoint then its transaction will be doomed on returning to userspace. This mean that the process will be unable to complete its transaction and be forced to its failure handler. This is already an out if line case for userspace. Furthermore, the cost of copying 64 times 128 bits from registers isn't very long[0] (at all) on modern processors. As such it appears these optimisations have only served to increase code complexity and are unlikely to have had a measurable performance impact. Our transactional memory handling has been riddled with bugs. A cause of this has been difficulty in following the code flow, code complexity has not been our friend here. It makes sense to remove these optimisations in favour of a (hopefully) more stable implementation. This patch does mean that some times the assembly will needlessly save 'junk' registers which will subsequently get overwritten with the correct value by the C code which calls the assembly function. This small inefficiency is far outweighed by the reduction in complexity for general TM code, context switching paths, and transactional facility unavailable exception handler. 0: I tried to measure it once for other work and found that it was hiding in the noise of everything else I was working with. I find it exceedingly likely this will be the case here. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-11-02 11:09:05 +08:00
__tm_recheckpoint(thread);
powerpc/tm: Disable IRQ in tm_recheckpoint We can't take an IRQ when we're about to do a trechkpt as our GPR state is set to user GPR values. We've hit this when running some IBM Java stress tests in the lab resulting in the following dump: cpu 0x3f: Vector: 700 (Program Check) at [c000000007eb3d40] pc: c000000000050074: restore_gprs+0xc0/0x148 lr: 00000000b52a8184 sp: ac57d360 msr: 8000000100201030 current = 0xc00000002c500000 paca = 0xc000000007dbfc00 softe: 0 irq_happened: 0x00 pid = 34535, comm = Pooled Thread # R00 = 00000000b52a8184 R16 = 00000000b3e48fda R01 = 00000000ac57d360 R17 = 00000000ade79bd8 R02 = 00000000ac586930 R18 = 000000000fac9bcc R03 = 00000000ade60000 R19 = 00000000ac57f930 R04 = 00000000f6624918 R20 = 00000000ade79be8 R05 = 00000000f663f238 R21 = 00000000ac218a54 R06 = 0000000000000002 R22 = 000000000f956280 R07 = 0000000000000008 R23 = 000000000000007e R08 = 000000000000000a R24 = 000000000000000c R09 = 00000000b6e69160 R25 = 00000000b424cf00 R10 = 0000000000000181 R26 = 00000000f66256d4 R11 = 000000000f365ec0 R27 = 00000000b6fdcdd0 R12 = 00000000f66400f0 R28 = 0000000000000001 R13 = 00000000ada71900 R29 = 00000000ade5a300 R14 = 00000000ac2185a8 R30 = 00000000f663f238 R15 = 0000000000000004 R31 = 00000000f6624918 pc = c000000000050074 restore_gprs+0xc0/0x148 cfar= c00000000004fe28 dont_restore_vec+0x1c/0x1a4 lr = 00000000b52a8184 msr = 8000000100201030 cr = 24804888 ctr = 0000000000000000 xer = 0000000000000000 trap = 700 This moves tm_recheckpoint to a C function and moves the tm_restore_sprs into that function. It then adds IRQ disabling over the trechkpt critical section. It also sets the TEXASR FS in the signals code to ensure this is never set now that we explictly write the TM sprs in tm_recheckpoint. Signed-off-by: Michael Neuling <mikey@neuling.org> cc: stable@vger.kernel.org Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-04-04 17:19:48 +08:00
local_irq_restore(flags);
}
static inline void tm_recheckpoint_new_task(struct task_struct *new)
{
if (!cpu_has_feature(CPU_FTR_TM))
return;
/* Recheckpoint the registers of the thread we're about to switch to.
*
* If the task was using FP, we non-lazily reload both the original and
* the speculative FP register states. This is because the kernel
* doesn't see if/when a TM rollback occurs, so if we take an FP
powerpc: tm: Always use fp_state and vr_state to store live registers There is currently an inconsistency as to how the entire CPU register state is saved and restored when a thread uses transactional memory (TM). Using transactional memory results in the CPU having duplicated (almost) all of its register state. This duplication results in a set of registers which can be considered 'live', those being currently modified by the instructions being executed and another set that is frozen at a point in time. On context switch, both sets of state have to be saved and (later) restored. These two states are often called a variety of different things. Common terms for the state which only exists after the CPU has entered a transaction (performed a TBEGIN instruction) in hardware are 'transactional' or 'speculative'. Between a TBEGIN and a TEND or TABORT (or an event that causes the hardware to abort), regardless of the use of TSUSPEND the transactional state can be referred to as the live state. The second state is often to referred to as the 'checkpointed' state and is a duplication of the live state when the TBEGIN instruction is executed. This state is kept in the hardware and will be rolled back to on transaction failure. Currently all the registers stored in pt_regs are ALWAYS the live registers, that is, when a thread has transactional registers their values are stored in pt_regs and the checkpointed state is in ckpt_regs. A strange opposite is true for fp_state/vr_state. When a thread is non transactional fp_state/vr_state holds the live registers. When a thread has initiated a transaction fp_state/vr_state holds the checkpointed state and transact_fp/transact_vr become the structure which holds the live state (at this point it is a transactional state). This method creates confusion as to where the live state is, in some circumstances it requires extra work to determine where to put the live state and prevents the use of common functions designed (probably before TM) to save the live state. With this patch pt_regs, fp_state and vr_state all represent the same thing and the other structures [pending rename] are for checkpointed state. Acked-by: Simon Guo <wei.guo.simon@gmail.com> Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-09-23 14:18:24 +08:00
* unavailable later, we are unable to determine which set of FP regs
* need to be restored.
*/
if (!tm_enabled(new))
return;
powerpc/tm: Disable IRQ in tm_recheckpoint We can't take an IRQ when we're about to do a trechkpt as our GPR state is set to user GPR values. We've hit this when running some IBM Java stress tests in the lab resulting in the following dump: cpu 0x3f: Vector: 700 (Program Check) at [c000000007eb3d40] pc: c000000000050074: restore_gprs+0xc0/0x148 lr: 00000000b52a8184 sp: ac57d360 msr: 8000000100201030 current = 0xc00000002c500000 paca = 0xc000000007dbfc00 softe: 0 irq_happened: 0x00 pid = 34535, comm = Pooled Thread # R00 = 00000000b52a8184 R16 = 00000000b3e48fda R01 = 00000000ac57d360 R17 = 00000000ade79bd8 R02 = 00000000ac586930 R18 = 000000000fac9bcc R03 = 00000000ade60000 R19 = 00000000ac57f930 R04 = 00000000f6624918 R20 = 00000000ade79be8 R05 = 00000000f663f238 R21 = 00000000ac218a54 R06 = 0000000000000002 R22 = 000000000f956280 R07 = 0000000000000008 R23 = 000000000000007e R08 = 000000000000000a R24 = 000000000000000c R09 = 00000000b6e69160 R25 = 00000000b424cf00 R10 = 0000000000000181 R26 = 00000000f66256d4 R11 = 000000000f365ec0 R27 = 00000000b6fdcdd0 R12 = 00000000f66400f0 R28 = 0000000000000001 R13 = 00000000ada71900 R29 = 00000000ade5a300 R14 = 00000000ac2185a8 R30 = 00000000f663f238 R15 = 0000000000000004 R31 = 00000000f6624918 pc = c000000000050074 restore_gprs+0xc0/0x148 cfar= c00000000004fe28 dont_restore_vec+0x1c/0x1a4 lr = 00000000b52a8184 msr = 8000000100201030 cr = 24804888 ctr = 0000000000000000 xer = 0000000000000000 trap = 700 This moves tm_recheckpoint to a C function and moves the tm_restore_sprs into that function. It then adds IRQ disabling over the trechkpt critical section. It also sets the TEXASR FS in the signals code to ensure this is never set now that we explictly write the TM sprs in tm_recheckpoint. Signed-off-by: Michael Neuling <mikey@neuling.org> cc: stable@vger.kernel.org Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-04-04 17:19:48 +08:00
if (!MSR_TM_ACTIVE(new->thread.regs->msr)){
tm_restore_sprs(&new->thread);
return;
powerpc/tm: Disable IRQ in tm_recheckpoint We can't take an IRQ when we're about to do a trechkpt as our GPR state is set to user GPR values. We've hit this when running some IBM Java stress tests in the lab resulting in the following dump: cpu 0x3f: Vector: 700 (Program Check) at [c000000007eb3d40] pc: c000000000050074: restore_gprs+0xc0/0x148 lr: 00000000b52a8184 sp: ac57d360 msr: 8000000100201030 current = 0xc00000002c500000 paca = 0xc000000007dbfc00 softe: 0 irq_happened: 0x00 pid = 34535, comm = Pooled Thread # R00 = 00000000b52a8184 R16 = 00000000b3e48fda R01 = 00000000ac57d360 R17 = 00000000ade79bd8 R02 = 00000000ac586930 R18 = 000000000fac9bcc R03 = 00000000ade60000 R19 = 00000000ac57f930 R04 = 00000000f6624918 R20 = 00000000ade79be8 R05 = 00000000f663f238 R21 = 00000000ac218a54 R06 = 0000000000000002 R22 = 000000000f956280 R07 = 0000000000000008 R23 = 000000000000007e R08 = 000000000000000a R24 = 000000000000000c R09 = 00000000b6e69160 R25 = 00000000b424cf00 R10 = 0000000000000181 R26 = 00000000f66256d4 R11 = 000000000f365ec0 R27 = 00000000b6fdcdd0 R12 = 00000000f66400f0 R28 = 0000000000000001 R13 = 00000000ada71900 R29 = 00000000ade5a300 R14 = 00000000ac2185a8 R30 = 00000000f663f238 R15 = 0000000000000004 R31 = 00000000f6624918 pc = c000000000050074 restore_gprs+0xc0/0x148 cfar= c00000000004fe28 dont_restore_vec+0x1c/0x1a4 lr = 00000000b52a8184 msr = 8000000100201030 cr = 24804888 ctr = 0000000000000000 xer = 0000000000000000 trap = 700 This moves tm_recheckpoint to a C function and moves the tm_restore_sprs into that function. It then adds IRQ disabling over the trechkpt critical section. It also sets the TEXASR FS in the signals code to ensure this is never set now that we explictly write the TM sprs in tm_recheckpoint. Signed-off-by: Michael Neuling <mikey@neuling.org> cc: stable@vger.kernel.org Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-04-04 17:19:48 +08:00
}
/* Recheckpoint to restore original checkpointed register state. */
powerpc: Always save/restore checkpointed regs during treclaim/trecheckpoint Lazy save and restore of FP/Altivec means that a userspace process can be sent to userspace with FP or Altivec disabled and loaded only as required (by way of an FP/Altivec unavailable exception). Transactional Memory complicates this situation as a transaction could be started without FP/Altivec being loaded up. This causes the hardware to checkpoint incorrect registers. Handling FP/Altivec unavailable exceptions while a thread is transactional requires a reclaim and recheckpoint to ensure the CPU has correct state for both sets of registers. tm_reclaim() has optimisations to not always save the FP/Altivec registers to the checkpointed save area. This was originally done because the caller might have information that the checkpointed registers aren't valid due to lazy save and restore. We've also been a little vague as to how tm_reclaim() leaves the FP/Altivec state since it doesn't necessarily always save it to the thread struct. This has lead to an (incorrect) assumption that it leaves the checkpointed state on the CPU. tm_recheckpoint() has similar optimisations in reverse. It may not always reload the checkpointed FP/Altivec registers from the thread struct before the trecheckpoint. It is therefore quite unclear where it expects to get the state from. This didn't help with the assumption made about tm_reclaim(). These optimisations sit in what is by definition a slow path. If a process has to go through a reclaim/recheckpoint then its transaction will be doomed on returning to userspace. This mean that the process will be unable to complete its transaction and be forced to its failure handler. This is already an out if line case for userspace. Furthermore, the cost of copying 64 times 128 bits from registers isn't very long[0] (at all) on modern processors. As such it appears these optimisations have only served to increase code complexity and are unlikely to have had a measurable performance impact. Our transactional memory handling has been riddled with bugs. A cause of this has been difficulty in following the code flow, code complexity has not been our friend here. It makes sense to remove these optimisations in favour of a (hopefully) more stable implementation. This patch does mean that some times the assembly will needlessly save 'junk' registers which will subsequently get overwritten with the correct value by the C code which calls the assembly function. This small inefficiency is far outweighed by the reduction in complexity for general TM code, context switching paths, and transactional facility unavailable exception handler. 0: I tried to measure it once for other work and found that it was hiding in the noise of everything else I was working with. I find it exceedingly likely this will be the case here. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-11-02 11:09:05 +08:00
TM_DEBUG("*** tm_recheckpoint of pid %d (new->msr 0x%lx)\n",
new->pid, new->thread.regs->msr);
powerpc: Always save/restore checkpointed regs during treclaim/trecheckpoint Lazy save and restore of FP/Altivec means that a userspace process can be sent to userspace with FP or Altivec disabled and loaded only as required (by way of an FP/Altivec unavailable exception). Transactional Memory complicates this situation as a transaction could be started without FP/Altivec being loaded up. This causes the hardware to checkpoint incorrect registers. Handling FP/Altivec unavailable exceptions while a thread is transactional requires a reclaim and recheckpoint to ensure the CPU has correct state for both sets of registers. tm_reclaim() has optimisations to not always save the FP/Altivec registers to the checkpointed save area. This was originally done because the caller might have information that the checkpointed registers aren't valid due to lazy save and restore. We've also been a little vague as to how tm_reclaim() leaves the FP/Altivec state since it doesn't necessarily always save it to the thread struct. This has lead to an (incorrect) assumption that it leaves the checkpointed state on the CPU. tm_recheckpoint() has similar optimisations in reverse. It may not always reload the checkpointed FP/Altivec registers from the thread struct before the trecheckpoint. It is therefore quite unclear where it expects to get the state from. This didn't help with the assumption made about tm_reclaim(). These optimisations sit in what is by definition a slow path. If a process has to go through a reclaim/recheckpoint then its transaction will be doomed on returning to userspace. This mean that the process will be unable to complete its transaction and be forced to its failure handler. This is already an out if line case for userspace. Furthermore, the cost of copying 64 times 128 bits from registers isn't very long[0] (at all) on modern processors. As such it appears these optimisations have only served to increase code complexity and are unlikely to have had a measurable performance impact. Our transactional memory handling has been riddled with bugs. A cause of this has been difficulty in following the code flow, code complexity has not been our friend here. It makes sense to remove these optimisations in favour of a (hopefully) more stable implementation. This patch does mean that some times the assembly will needlessly save 'junk' registers which will subsequently get overwritten with the correct value by the C code which calls the assembly function. This small inefficiency is far outweighed by the reduction in complexity for general TM code, context switching paths, and transactional facility unavailable exception handler. 0: I tried to measure it once for other work and found that it was hiding in the noise of everything else I was working with. I find it exceedingly likely this will be the case here. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2017-11-02 11:09:05 +08:00
tm_recheckpoint(&new->thread);
powerpc: tm: Always use fp_state and vr_state to store live registers There is currently an inconsistency as to how the entire CPU register state is saved and restored when a thread uses transactional memory (TM). Using transactional memory results in the CPU having duplicated (almost) all of its register state. This duplication results in a set of registers which can be considered 'live', those being currently modified by the instructions being executed and another set that is frozen at a point in time. On context switch, both sets of state have to be saved and (later) restored. These two states are often called a variety of different things. Common terms for the state which only exists after the CPU has entered a transaction (performed a TBEGIN instruction) in hardware are 'transactional' or 'speculative'. Between a TBEGIN and a TEND or TABORT (or an event that causes the hardware to abort), regardless of the use of TSUSPEND the transactional state can be referred to as the live state. The second state is often to referred to as the 'checkpointed' state and is a duplication of the live state when the TBEGIN instruction is executed. This state is kept in the hardware and will be rolled back to on transaction failure. Currently all the registers stored in pt_regs are ALWAYS the live registers, that is, when a thread has transactional registers their values are stored in pt_regs and the checkpointed state is in ckpt_regs. A strange opposite is true for fp_state/vr_state. When a thread is non transactional fp_state/vr_state holds the live registers. When a thread has initiated a transaction fp_state/vr_state holds the checkpointed state and transact_fp/transact_vr become the structure which holds the live state (at this point it is a transactional state). This method creates confusion as to where the live state is, in some circumstances it requires extra work to determine where to put the live state and prevents the use of common functions designed (probably before TM) to save the live state. With this patch pt_regs, fp_state and vr_state all represent the same thing and the other structures [pending rename] are for checkpointed state. Acked-by: Simon Guo <wei.guo.simon@gmail.com> Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-09-23 14:18:24 +08:00
/*
* The checkpointed state has been restored but the live state has
* not, ensure all the math functionality is turned off to trigger
* restore_math() to reload.
*/
new->thread.regs->msr &= ~(MSR_FP | MSR_VEC | MSR_VSX);
TM_DEBUG("*** tm_recheckpoint of pid %d complete "
"(kernel msr 0x%lx)\n",
new->pid, mfmsr());
}
powerpc: tm: Always use fp_state and vr_state to store live registers There is currently an inconsistency as to how the entire CPU register state is saved and restored when a thread uses transactional memory (TM). Using transactional memory results in the CPU having duplicated (almost) all of its register state. This duplication results in a set of registers which can be considered 'live', those being currently modified by the instructions being executed and another set that is frozen at a point in time. On context switch, both sets of state have to be saved and (later) restored. These two states are often called a variety of different things. Common terms for the state which only exists after the CPU has entered a transaction (performed a TBEGIN instruction) in hardware are 'transactional' or 'speculative'. Between a TBEGIN and a TEND or TABORT (or an event that causes the hardware to abort), regardless of the use of TSUSPEND the transactional state can be referred to as the live state. The second state is often to referred to as the 'checkpointed' state and is a duplication of the live state when the TBEGIN instruction is executed. This state is kept in the hardware and will be rolled back to on transaction failure. Currently all the registers stored in pt_regs are ALWAYS the live registers, that is, when a thread has transactional registers their values are stored in pt_regs and the checkpointed state is in ckpt_regs. A strange opposite is true for fp_state/vr_state. When a thread is non transactional fp_state/vr_state holds the live registers. When a thread has initiated a transaction fp_state/vr_state holds the checkpointed state and transact_fp/transact_vr become the structure which holds the live state (at this point it is a transactional state). This method creates confusion as to where the live state is, in some circumstances it requires extra work to determine where to put the live state and prevents the use of common functions designed (probably before TM) to save the live state. With this patch pt_regs, fp_state and vr_state all represent the same thing and the other structures [pending rename] are for checkpointed state. Acked-by: Simon Guo <wei.guo.simon@gmail.com> Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-09-23 14:18:24 +08:00
static inline void __switch_to_tm(struct task_struct *prev,
struct task_struct *new)
{
if (cpu_has_feature(CPU_FTR_TM)) {
if (tm_enabled(prev) || tm_enabled(new))
tm_enable();
if (tm_enabled(prev)) {
prev->thread.load_tm++;
tm_reclaim_task(prev);
if (!MSR_TM_ACTIVE(prev->thread.regs->msr) && prev->thread.load_tm == 0)
prev->thread.regs->msr &= ~MSR_TM;
}
powerpc: tm: Always use fp_state and vr_state to store live registers There is currently an inconsistency as to how the entire CPU register state is saved and restored when a thread uses transactional memory (TM). Using transactional memory results in the CPU having duplicated (almost) all of its register state. This duplication results in a set of registers which can be considered 'live', those being currently modified by the instructions being executed and another set that is frozen at a point in time. On context switch, both sets of state have to be saved and (later) restored. These two states are often called a variety of different things. Common terms for the state which only exists after the CPU has entered a transaction (performed a TBEGIN instruction) in hardware are 'transactional' or 'speculative'. Between a TBEGIN and a TEND or TABORT (or an event that causes the hardware to abort), regardless of the use of TSUSPEND the transactional state can be referred to as the live state. The second state is often to referred to as the 'checkpointed' state and is a duplication of the live state when the TBEGIN instruction is executed. This state is kept in the hardware and will be rolled back to on transaction failure. Currently all the registers stored in pt_regs are ALWAYS the live registers, that is, when a thread has transactional registers their values are stored in pt_regs and the checkpointed state is in ckpt_regs. A strange opposite is true for fp_state/vr_state. When a thread is non transactional fp_state/vr_state holds the live registers. When a thread has initiated a transaction fp_state/vr_state holds the checkpointed state and transact_fp/transact_vr become the structure which holds the live state (at this point it is a transactional state). This method creates confusion as to where the live state is, in some circumstances it requires extra work to determine where to put the live state and prevents the use of common functions designed (probably before TM) to save the live state. With this patch pt_regs, fp_state and vr_state all represent the same thing and the other structures [pending rename] are for checkpointed state. Acked-by: Simon Guo <wei.guo.simon@gmail.com> Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-09-23 14:18:24 +08:00
tm_recheckpoint_new_task(new);
}
}
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
/*
* This is called if we are on the way out to userspace and the
* TIF_RESTORE_TM flag is set. It checks if we need to reload
* FP and/or vector state and does so if necessary.
* If userspace is inside a transaction (whether active or
* suspended) and FP/VMX/VSX instructions have ever been enabled
* inside that transaction, then we have to keep them enabled
* and keep the FP/VMX/VSX state loaded while ever the transaction
* continues. The reason is that if we didn't, and subsequently
* got a FP/VMX/VSX unavailable interrupt inside a transaction,
* we don't know whether it's the same transaction, and thus we
* don't know which of the checkpointed state and the transactional
* state to use.
*/
void restore_tm_state(struct pt_regs *regs)
{
unsigned long msr_diff;
powerpc: tm: Always use fp_state and vr_state to store live registers There is currently an inconsistency as to how the entire CPU register state is saved and restored when a thread uses transactional memory (TM). Using transactional memory results in the CPU having duplicated (almost) all of its register state. This duplication results in a set of registers which can be considered 'live', those being currently modified by the instructions being executed and another set that is frozen at a point in time. On context switch, both sets of state have to be saved and (later) restored. These two states are often called a variety of different things. Common terms for the state which only exists after the CPU has entered a transaction (performed a TBEGIN instruction) in hardware are 'transactional' or 'speculative'. Between a TBEGIN and a TEND or TABORT (or an event that causes the hardware to abort), regardless of the use of TSUSPEND the transactional state can be referred to as the live state. The second state is often to referred to as the 'checkpointed' state and is a duplication of the live state when the TBEGIN instruction is executed. This state is kept in the hardware and will be rolled back to on transaction failure. Currently all the registers stored in pt_regs are ALWAYS the live registers, that is, when a thread has transactional registers their values are stored in pt_regs and the checkpointed state is in ckpt_regs. A strange opposite is true for fp_state/vr_state. When a thread is non transactional fp_state/vr_state holds the live registers. When a thread has initiated a transaction fp_state/vr_state holds the checkpointed state and transact_fp/transact_vr become the structure which holds the live state (at this point it is a transactional state). This method creates confusion as to where the live state is, in some circumstances it requires extra work to determine where to put the live state and prevents the use of common functions designed (probably before TM) to save the live state. With this patch pt_regs, fp_state and vr_state all represent the same thing and the other structures [pending rename] are for checkpointed state. Acked-by: Simon Guo <wei.guo.simon@gmail.com> Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-09-23 14:18:24 +08:00
/*
* This is the only moment we should clear TIF_RESTORE_TM as
* it is here that ckpt_regs.msr and pt_regs.msr become the same
* again, anything else could lead to an incorrect ckpt_msr being
* saved and therefore incorrect signal contexts.
*/
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
clear_thread_flag(TIF_RESTORE_TM);
if (!MSR_TM_ACTIVE(regs->msr))
return;
msr_diff = current->thread.ckpt_regs.msr & ~regs->msr;
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
msr_diff &= MSR_FP | MSR_VEC | MSR_VSX;
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
/* Ensure that restore_math() will restore */
if (msr_diff & MSR_FP)
current->thread.load_fp = 1;
#ifdef CONFIG_ALTIVEC
if (cpu_has_feature(CPU_FTR_ALTIVEC) && msr_diff & MSR_VEC)
current->thread.load_vec = 1;
#endif
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
restore_math(regs);
powerpc: Don't corrupt transactional state when using FP/VMX in kernel Currently, when we have a process using the transactional memory facilities on POWER8 (that is, the processor is in transactional or suspended state), and the process enters the kernel and the kernel then uses the floating-point or vector (VMX/Altivec) facility, we end up corrupting the user-visible FP/VMX/VSX state. This happens, for example, if a page fault causes a copy-on-write operation, because the copy_page function will use VMX to do the copy on POWER8. The test program below demonstrates the bug. The bug happens because when FP/VMX state for a transactional process is stored in the thread_struct, we store the checkpointed state in .fp_state/.vr_state and the transactional (current) state in .transact_fp/.transact_vr. However, when the kernel wants to use FP/VMX, it calls enable_kernel_fp() or enable_kernel_altivec(), which saves the current state in .fp_state/.vr_state. Furthermore, when we return to the user process we return with FP/VMX/VSX disabled. The next time the process uses FP/VMX/VSX, we don't know which set of state (the current register values, .fp_state/.vr_state, or .transact_fp/.transact_vr) we should be using, since we have no way to tell if we are still in the same transaction, and if not, whether the previous transaction succeeded or failed. Thus it is necessary to strictly adhere to the rule that if FP has been enabled at any point in a transaction, we must keep FP enabled for the user process with the current transactional state in the FP registers, until we detect that it is no longer in a transaction. Similarly for VMX; once enabled it must stay enabled until the process is no longer transactional. In order to keep this rule, we add a new thread_info flag which we test when returning from the kernel to userspace, called TIF_RESTORE_TM. This flag indicates that there is FP/VMX/VSX state to be restored before entering userspace, and when it is set the .tm_orig_msr field in the thread_struct indicates what state needs to be restored. The restoration is done by restore_tm_state(). The TIF_RESTORE_TM bit is set by new giveup_fpu/altivec_maybe_transactional helpers, which are called from enable_kernel_fp/altivec, giveup_vsx, and flush_fp/altivec_to_thread instead of giveup_fpu/altivec. The other thing to be done is to get the transactional FP/VMX/VSX state from .fp_state/.vr_state when doing reclaim, if that state has been saved there by giveup_fpu/altivec_maybe_transactional. Having done this, we set the FP/VMX bit in the thread's MSR after reclaim to indicate that that part of the state is now valid (having been reclaimed from the processor's checkpointed state). Finally, in the signal handling code, we move the clearing of the transactional state bits in the thread's MSR a bit earlier, before calling flush_fp_to_thread(), so that we don't unnecessarily set the TIF_RESTORE_TM bit. This is the test program: /* Michael Neuling 4/12/2013 * * See if the altivec state is leaked out of an aborted transaction due to * kernel vmx copy loops. * * gcc -m64 htm_vmxcopy.c -o htm_vmxcopy * */ /* We don't use all of these, but for reference: */ int main(int argc, char *argv[]) { long double vecin = 1.3; long double vecout; unsigned long pgsize = getpagesize(); int i; int fd; int size = pgsize*16; char tmpfile[] = "/tmp/page_faultXXXXXX"; char buf[pgsize]; char *a; uint64_t aborted = 0; fd = mkstemp(tmpfile); assert(fd >= 0); memset(buf, 0, pgsize); for (i = 0; i < size; i += pgsize) assert(write(fd, buf, pgsize) == pgsize); unlink(tmpfile); a = mmap(NULL, size, PROT_READ|PROT_WRITE, MAP_PRIVATE, fd, 0); assert(a != MAP_FAILED); asm __volatile__( "lxvd2x 40,0,%[vecinptr] ; " // set 40 to initial value TBEGIN "beq 3f ;" TSUSPEND "xxlxor 40,40,40 ; " // set 40 to 0 "std 5, 0(%[map]) ;" // cause kernel vmx copy page TABORT TRESUME TEND "li %[res], 0 ;" "b 5f ;" "3: ;" // Abort handler "li %[res], 1 ;" "5: ;" "stxvd2x 40,0,%[vecoutptr] ; " : [res]"=r"(aborted) : [vecinptr]"r"(&vecin), [vecoutptr]"r"(&vecout), [map]"r"(a) : "memory", "r0", "r3", "r4", "r5", "r6", "r7"); if (aborted && (vecin != vecout)){ printf("FAILED: vector state leaked on abort %f != %f\n", (double)vecin, (double)vecout); exit(1); } munmap(a, size); close(fd); printf("PASSED!\n"); return 0; } Signed-off-by: Paul Mackerras <paulus@samba.org> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2014-01-13 12:56:29 +08:00
regs->msr |= msr_diff;
}
#else /* !CONFIG_PPC_TRANSACTIONAL_MEM */
#define tm_recheckpoint_new_task(new)
powerpc: tm: Always use fp_state and vr_state to store live registers There is currently an inconsistency as to how the entire CPU register state is saved and restored when a thread uses transactional memory (TM). Using transactional memory results in the CPU having duplicated (almost) all of its register state. This duplication results in a set of registers which can be considered 'live', those being currently modified by the instructions being executed and another set that is frozen at a point in time. On context switch, both sets of state have to be saved and (later) restored. These two states are often called a variety of different things. Common terms for the state which only exists after the CPU has entered a transaction (performed a TBEGIN instruction) in hardware are 'transactional' or 'speculative'. Between a TBEGIN and a TEND or TABORT (or an event that causes the hardware to abort), regardless of the use of TSUSPEND the transactional state can be referred to as the live state. The second state is often to referred to as the 'checkpointed' state and is a duplication of the live state when the TBEGIN instruction is executed. This state is kept in the hardware and will be rolled back to on transaction failure. Currently all the registers stored in pt_regs are ALWAYS the live registers, that is, when a thread has transactional registers their values are stored in pt_regs and the checkpointed state is in ckpt_regs. A strange opposite is true for fp_state/vr_state. When a thread is non transactional fp_state/vr_state holds the live registers. When a thread has initiated a transaction fp_state/vr_state holds the checkpointed state and transact_fp/transact_vr become the structure which holds the live state (at this point it is a transactional state). This method creates confusion as to where the live state is, in some circumstances it requires extra work to determine where to put the live state and prevents the use of common functions designed (probably before TM) to save the live state. With this patch pt_regs, fp_state and vr_state all represent the same thing and the other structures [pending rename] are for checkpointed state. Acked-by: Simon Guo <wei.guo.simon@gmail.com> Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-09-23 14:18:24 +08:00
#define __switch_to_tm(prev, new)
void tm_reclaim_current(uint8_t cause) {}
#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
static inline void save_sprs(struct thread_struct *t)
{
#ifdef CONFIG_ALTIVEC
if (cpu_has_feature(CPU_FTR_ALTIVEC))
t->vrsave = mfspr(SPRN_VRSAVE);
#endif
#ifdef CONFIG_SPE
if (cpu_has_feature(CPU_FTR_SPE))
t->spefscr = mfspr(SPRN_SPEFSCR);
#endif
#ifdef CONFIG_PPC_BOOK3S_64
if (cpu_has_feature(CPU_FTR_DSCR))
t->dscr = mfspr(SPRN_DSCR);
if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
t->bescr = mfspr(SPRN_BESCR);
t->ebbhr = mfspr(SPRN_EBBHR);
t->ebbrr = mfspr(SPRN_EBBRR);
t->fscr = mfspr(SPRN_FSCR);
/*
* Note that the TAR is not available for use in the kernel.
* (To provide this, the TAR should be backed up/restored on
* exception entry/exit instead, and be in pt_regs. FIXME,
* this should be in pt_regs anyway (for debug).)
*/
t->tar = mfspr(SPRN_TAR);
}
#endif
}
static inline void restore_sprs(struct thread_struct *old_thread,
struct thread_struct *new_thread)
{
#ifdef CONFIG_ALTIVEC
if (cpu_has_feature(CPU_FTR_ALTIVEC) &&
old_thread->vrsave != new_thread->vrsave)
mtspr(SPRN_VRSAVE, new_thread->vrsave);
#endif
#ifdef CONFIG_SPE
if (cpu_has_feature(CPU_FTR_SPE) &&
old_thread->spefscr != new_thread->spefscr)
mtspr(SPRN_SPEFSCR, new_thread->spefscr);
#endif
#ifdef CONFIG_PPC_BOOK3S_64
if (cpu_has_feature(CPU_FTR_DSCR)) {
u64 dscr = get_paca()->dscr_default;
if (new_thread->dscr_inherit)
dscr = new_thread->dscr;
if (old_thread->dscr != dscr)
mtspr(SPRN_DSCR, dscr);
}
if (cpu_has_feature(CPU_FTR_ARCH_207S)) {
if (old_thread->bescr != new_thread->bescr)
mtspr(SPRN_BESCR, new_thread->bescr);
if (old_thread->ebbhr != new_thread->ebbhr)
mtspr(SPRN_EBBHR, new_thread->ebbhr);
if (old_thread->ebbrr != new_thread->ebbrr)
mtspr(SPRN_EBBRR, new_thread->ebbrr);
if (old_thread->fscr != new_thread->fscr)
mtspr(SPRN_FSCR, new_thread->fscr);
if (old_thread->tar != new_thread->tar)
mtspr(SPRN_TAR, new_thread->tar);
}
if (cpu_has_feature(CPU_FTR_P9_TIDR) &&
old_thread->tidr != new_thread->tidr)
mtspr(SPRN_TIDR, new_thread->tidr);
#endif
}
struct task_struct *__switch_to(struct task_struct *prev,
struct task_struct *new)
{
struct thread_struct *new_thread, *old_thread;
struct task_struct *last;
#ifdef CONFIG_PPC_BOOK3S_64
struct ppc64_tlb_batch *batch;
#endif
new_thread = &new->thread;
old_thread = &current->thread;
WARN_ON(!irqs_disabled());
#ifdef CONFIG_PPC_BOOK3S_64
powerpc: Replace __get_cpu_var uses This still has not been merged and now powerpc is the only arch that does not have this change. Sorry about missing linuxppc-dev before. V2->V2 - Fix up to work against 3.18-rc1 __get_cpu_var() is used for multiple purposes in the kernel source. One of them is address calculation via the form &__get_cpu_var(x). This calculates the address for the instance of the percpu variable of the current processor based on an offset. Other use cases are for storing and retrieving data from the current processors percpu area. __get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment. __get_cpu_var() is defined as : __get_cpu_var() always only does an address determination. However, store and retrieve operations could use a segment prefix (or global register on other platforms) to avoid the address calculation. this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use optimized assembly code to read and write per cpu variables. This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr() or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided and less registers are used when code is generated. At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too. The patch set includes passes over all arches as well. Once these operations are used throughout then specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by f.e. using a global register that may be set to the per cpu base. Transformations done to __get_cpu_var() 1. Determine the address of the percpu instance of the current processor. DEFINE_PER_CPU(int, y); int *x = &__get_cpu_var(y); Converts to int *x = this_cpu_ptr(&y); 2. Same as #1 but this time an array structure is involved. DEFINE_PER_CPU(int, y[20]); int *x = __get_cpu_var(y); Converts to int *x = this_cpu_ptr(y); 3. Retrieve the content of the current processors instance of a per cpu variable. DEFINE_PER_CPU(int, y); int x = __get_cpu_var(y) Converts to int x = __this_cpu_read(y); 4. Retrieve the content of a percpu struct DEFINE_PER_CPU(struct mystruct, y); struct mystruct x = __get_cpu_var(y); Converts to memcpy(&x, this_cpu_ptr(&y), sizeof(x)); 5. Assignment to a per cpu variable DEFINE_PER_CPU(int, y) __get_cpu_var(y) = x; Converts to __this_cpu_write(y, x); 6. Increment/Decrement etc of a per cpu variable DEFINE_PER_CPU(int, y); __get_cpu_var(y)++ Converts to __this_cpu_inc(y) Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> CC: Paul Mackerras <paulus@samba.org> Signed-off-by: Christoph Lameter <cl@linux.com> [mpe: Fix build errors caused by set/or_softirq_pending(), and rework assignment in __set_breakpoint() to use memcpy().] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-10-22 04:23:25 +08:00
batch = this_cpu_ptr(&ppc64_tlb_batch);
if (batch->active) {
current_thread_info()->local_flags |= _TLF_LAZY_MMU;
if (batch->index)
__flush_tlb_pending(batch);
batch->active = 0;
}
#endif /* CONFIG_PPC_BOOK3S_64 */
#ifdef CONFIG_PPC_ADV_DEBUG_REGS
switch_booke_debug_regs(&new->thread.debug);
#else
/*
* For PPC_BOOK3S_64, we use the hw-breakpoint interfaces that would
* schedule DABR
*/
#ifndef CONFIG_HAVE_HW_BREAKPOINT
switch_hw_breakpoint(new);
#endif /* CONFIG_HAVE_HW_BREAKPOINT */
#endif
/*
* We need to save SPRs before treclaim/trecheckpoint as these will
* change a number of them.
*/
save_sprs(&prev->thread);
/* Save FPU, Altivec, VSX and SPE state */
giveup_all(prev);
powerpc: tm: Always use fp_state and vr_state to store live registers There is currently an inconsistency as to how the entire CPU register state is saved and restored when a thread uses transactional memory (TM). Using transactional memory results in the CPU having duplicated (almost) all of its register state. This duplication results in a set of registers which can be considered 'live', those being currently modified by the instructions being executed and another set that is frozen at a point in time. On context switch, both sets of state have to be saved and (later) restored. These two states are often called a variety of different things. Common terms for the state which only exists after the CPU has entered a transaction (performed a TBEGIN instruction) in hardware are 'transactional' or 'speculative'. Between a TBEGIN and a TEND or TABORT (or an event that causes the hardware to abort), regardless of the use of TSUSPEND the transactional state can be referred to as the live state. The second state is often to referred to as the 'checkpointed' state and is a duplication of the live state when the TBEGIN instruction is executed. This state is kept in the hardware and will be rolled back to on transaction failure. Currently all the registers stored in pt_regs are ALWAYS the live registers, that is, when a thread has transactional registers their values are stored in pt_regs and the checkpointed state is in ckpt_regs. A strange opposite is true for fp_state/vr_state. When a thread is non transactional fp_state/vr_state holds the live registers. When a thread has initiated a transaction fp_state/vr_state holds the checkpointed state and transact_fp/transact_vr become the structure which holds the live state (at this point it is a transactional state). This method creates confusion as to where the live state is, in some circumstances it requires extra work to determine where to put the live state and prevents the use of common functions designed (probably before TM) to save the live state. With this patch pt_regs, fp_state and vr_state all represent the same thing and the other structures [pending rename] are for checkpointed state. Acked-by: Simon Guo <wei.guo.simon@gmail.com> Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-09-23 14:18:24 +08:00
__switch_to_tm(prev, new);
if (!radix_enabled()) {
/*
* We can't take a PMU exception inside _switch() since there
* is a window where the kernel stack SLB and the kernel stack
* are out of sync. Hard disable here.
*/
hard_irq_disable();
}
/*
* Call restore_sprs() before calling _switch(). If we move it after
* _switch() then we miss out on calling it for new tasks. The reason
* for this is we manually create a stack frame for new tasks that
* directly returns through ret_from_fork() or
* ret_from_kernel_thread(). See copy_thread() for details.
*/
restore_sprs(old_thread, new_thread);
#ifdef CONFIG_PPC32
kuap_assert_locked();
#endif
last = _switch(old_thread, new_thread);
#ifdef CONFIG_PPC_BOOK3S_64
if (current_thread_info()->local_flags & _TLF_LAZY_MMU) {
current_thread_info()->local_flags &= ~_TLF_LAZY_MMU;
powerpc: Replace __get_cpu_var uses This still has not been merged and now powerpc is the only arch that does not have this change. Sorry about missing linuxppc-dev before. V2->V2 - Fix up to work against 3.18-rc1 __get_cpu_var() is used for multiple purposes in the kernel source. One of them is address calculation via the form &__get_cpu_var(x). This calculates the address for the instance of the percpu variable of the current processor based on an offset. Other use cases are for storing and retrieving data from the current processors percpu area. __get_cpu_var() can be used as an lvalue when writing data or on the right side of an assignment. __get_cpu_var() is defined as : __get_cpu_var() always only does an address determination. However, store and retrieve operations could use a segment prefix (or global register on other platforms) to avoid the address calculation. this_cpu_write() and this_cpu_read() can directly take an offset into a percpu area and use optimized assembly code to read and write per cpu variables. This patch converts __get_cpu_var into either an explicit address calculation using this_cpu_ptr() or into a use of this_cpu operations that use the offset. Thereby address calculations are avoided and less registers are used when code is generated. At the end of the patch set all uses of __get_cpu_var have been removed so the macro is removed too. The patch set includes passes over all arches as well. Once these operations are used throughout then specialized macros can be defined in non -x86 arches as well in order to optimize per cpu access by f.e. using a global register that may be set to the per cpu base. Transformations done to __get_cpu_var() 1. Determine the address of the percpu instance of the current processor. DEFINE_PER_CPU(int, y); int *x = &__get_cpu_var(y); Converts to int *x = this_cpu_ptr(&y); 2. Same as #1 but this time an array structure is involved. DEFINE_PER_CPU(int, y[20]); int *x = __get_cpu_var(y); Converts to int *x = this_cpu_ptr(y); 3. Retrieve the content of the current processors instance of a per cpu variable. DEFINE_PER_CPU(int, y); int x = __get_cpu_var(y) Converts to int x = __this_cpu_read(y); 4. Retrieve the content of a percpu struct DEFINE_PER_CPU(struct mystruct, y); struct mystruct x = __get_cpu_var(y); Converts to memcpy(&x, this_cpu_ptr(&y), sizeof(x)); 5. Assignment to a per cpu variable DEFINE_PER_CPU(int, y) __get_cpu_var(y) = x; Converts to __this_cpu_write(y, x); 6. Increment/Decrement etc of a per cpu variable DEFINE_PER_CPU(int, y); __get_cpu_var(y)++ Converts to __this_cpu_inc(y) Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> CC: Paul Mackerras <paulus@samba.org> Signed-off-by: Christoph Lameter <cl@linux.com> [mpe: Fix build errors caused by set/or_softirq_pending(), and rework assignment in __set_breakpoint() to use memcpy().] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2014-10-22 04:23:25 +08:00
batch = this_cpu_ptr(&ppc64_tlb_batch);
batch->active = 1;
}
powerpc: Restore FPU/VEC/VSX if previously used Currently the FPU, VEC and VSX facilities are lazily loaded. This is not a problem unless a process is using these facilities. Modern versions of GCC are very good at automatically vectorising code, new and modernised workloads make use of floating point and vector facilities, even the kernel makes use of vectorised memcpy. All this combined greatly increases the cost of a syscall since the kernel uses the facilities sometimes even in syscall fast-path making it increasingly common for a thread to take an *_unavailable exception soon after a syscall, not to mention potentially taking all three. The obvious overcompensation to this problem is to simply always load all the facilities on every exit to userspace. Loading up all FPU, VEC and VSX registers every time can be expensive and if a workload does avoid using them, it should not be forced to incur this penalty. An 8bit counter is used to detect if the registers have been used in the past and the registers are always loaded until the value wraps to back to zero. Several versions of the assembly in entry_64.S were tested: 1. Always calling C. 2. Performing a common case check and then calling C. 3. A complex check in asm. After some benchmarking it was determined that avoiding C in the common case is a performance benefit (option 2). The full check in asm (option 3) greatly complicated that codepath for a negligible performance gain and the trade-off was deemed not worth it. Signed-off-by: Cyril Bur <cyrilbur@gmail.com> [mpe: Move load_vec in the struct to fill an existing hole, reword change log] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> fixup
2016-02-29 14:53:47 +08:00
if (current->thread.regs) {
restore_math(current->thread.regs);
/*
* On POWER9 the copy-paste buffer can only paste into
* foreign real addresses, so unprivileged processes can not
* see the data or use it in any way unless they have
* foreign real mappings. If the new process has the foreign
* real address mappings, we must issue a cp_abort to clear
* any state and prevent snooping, corruption or a covert
* channel. ISA v3.1 supports paste into local memory.
*/
powerpc: Use mm_context vas_windows counter to issue CP_ABORT set_thread_uses_vas() sets used_vas flag for a process that opened VAS window and issue CP_ABORT during context switch for only that process. In multi-thread application, windows can be shared. For example Thread A can open a window and Thread B can run COPY/PASTE instructions to send NX request which may cause corruption or snooping or a covert channel Also once this flag is set, continue to run CP_ABORT even the VAS window is closed. So define vas-windows counter in process mm_context, increment this counter for each window open and decrement it for window close. If vas-windows is set, issue CP_ABORT during context switch. It means clear the foreign real address mapping only if the process / thread uses COPY/PASTE. Then disable it for that process if windows are not open. Moved set_thread_uses_vas() code to vas_tx_win_open() as this functionality is needed only for userspace open windows. We are adding VAS userspace support along with this fix. So no need to include this fix in stable releases. Fixes: 9d2a4d71332c ("powerpc: Define set_thread_uses_vas()") Signed-off-by: Haren Myneni <haren@linux.ibm.com> Reported-by: Nicholas Piggin <npiggin@gmail.com> Suggested-by: Milton Miller <miltonm@us.ibm.com> Suggested-by: Nicholas Piggin <npiggin@gmail.com> Reviewed-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/1587017291.2275.1077.camel@hbabu-laptop
2020-04-16 14:08:11 +08:00
if (current->mm &&
(cpu_has_feature(CPU_FTR_ARCH_31) ||
atomic_read(&current->mm->context.vas_windows)))
asm volatile(PPC_CP_ABORT);
}
#endif /* CONFIG_PPC_BOOK3S_64 */
return last;
}
#define NR_INSN_TO_PRINT 16
static void show_instructions(struct pt_regs *regs)
{
int i;
unsigned long nip = regs->nip;
unsigned long pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
printk("Instruction dump:");
/*
* If we were executing with the MMU off for instructions, adjust pc
* rather than printing XXXXXXXX.
*/
if (!IS_ENABLED(CONFIG_BOOKE) && !(regs->msr & MSR_IR)) {
pc = (unsigned long)phys_to_virt(pc);
nip = (unsigned long)phys_to_virt(regs->nip);
}
for (i = 0; i < NR_INSN_TO_PRINT; i++) {
int instr;
if (!(i % 8))
pr_cont("\n");
if (!__kernel_text_address(pc) ||
get_kernel_nofault(instr, (const void *)pc)) {
pr_cont("XXXXXXXX ");
} else {
if (nip == pc)
pr_cont("<%08x> ", instr);
else
pr_cont("%08x ", instr);
}
pc += sizeof(int);
}
pr_cont("\n");
}
void show_user_instructions(struct pt_regs *regs)
{
unsigned long pc;
int n = NR_INSN_TO_PRINT;
struct seq_buf s;
char buf[96]; /* enough for 8 times 9 + 2 chars */
pc = regs->nip - (NR_INSN_TO_PRINT * 3 / 4 * sizeof(int));
seq_buf_init(&s, buf, sizeof(buf));
while (n) {
int i;
seq_buf_clear(&s);
for (i = 0; i < 8 && n; i++, n--, pc += sizeof(int)) {
int instr;
if (copy_from_user_nofault(&instr, (void __user *)pc,
sizeof(instr))) {
seq_buf_printf(&s, "XXXXXXXX ");
continue;
}
seq_buf_printf(&s, regs->nip == pc ? "<%08x> " : "%08x ", instr);
}
if (!seq_buf_has_overflowed(&s))
pr_info("%s[%d]: code: %s\n", current->comm,
current->pid, s.buffer);
}
}
struct regbit {
unsigned long bit;
const char *name;
};
static struct regbit msr_bits[] = {
#if defined(CONFIG_PPC64) && !defined(CONFIG_BOOKE)
{MSR_SF, "SF"},
{MSR_HV, "HV"},
#endif
{MSR_VEC, "VEC"},
{MSR_VSX, "VSX"},
#ifdef CONFIG_BOOKE
{MSR_CE, "CE"},
#endif
{MSR_EE, "EE"},
{MSR_PR, "PR"},
{MSR_FP, "FP"},
{MSR_ME, "ME"},
#ifdef CONFIG_BOOKE
{MSR_DE, "DE"},
#else
{MSR_SE, "SE"},
{MSR_BE, "BE"},
#endif
{MSR_IR, "IR"},
{MSR_DR, "DR"},
{MSR_PMM, "PMM"},
#ifndef CONFIG_BOOKE
{MSR_RI, "RI"},
{MSR_LE, "LE"},
#endif
{0, NULL}
};
static void print_bits(unsigned long val, struct regbit *bits, const char *sep)
{
const char *s = "";
for (; bits->bit; ++bits)
if (val & bits->bit) {
pr_cont("%s%s", s, bits->name);
s = sep;
}
}
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
static struct regbit msr_tm_bits[] = {
{MSR_TS_T, "T"},
{MSR_TS_S, "S"},
{MSR_TM, "E"},
{0, NULL}
};
static void print_tm_bits(unsigned long val)
{
/*
* This only prints something if at least one of the TM bit is set.
* Inside the TM[], the output means:
* E: Enabled (bit 32)
* S: Suspended (bit 33)
* T: Transactional (bit 34)
*/
if (val & (MSR_TM | MSR_TS_S | MSR_TS_T)) {
pr_cont(",TM[");
print_bits(val, msr_tm_bits, "");
pr_cont("]");
}
}
#else
static void print_tm_bits(unsigned long val) {}
#endif
static void print_msr_bits(unsigned long val)
{
pr_cont("<");
print_bits(val, msr_bits, ",");
print_tm_bits(val);
pr_cont(">");
}
#ifdef CONFIG_PPC64
#define REG "%016lx"
#define REGS_PER_LINE 4
#else
#define REG "%08lx"
#define REGS_PER_LINE 8
#endif
powerpc: show registers when unwinding interrupt frames It's often useful to know the register state for interrupts in the stack frame. In the below example (with this patch applied), the important information is the state of the page fault. A blatant case like this probably rather should have the page fault regs passed down to the warning, but quite often there are less obvious cases where an interrupt shows up that might give some more clues. The downside is longer and more complex bug output. Bug: Write fault blocked by AMR! WARNING: CPU: 0 PID: 72 at arch/powerpc/include/asm/book3s/64/kup-radix.h:164 __do_page_fault+0x880/0xa90 Modules linked in: CPU: 0 PID: 72 Comm: systemd-gpt-aut Not tainted NIP: c00000000006e2f0 LR: c00000000006e2ec CTR: 0000000000000000 REGS: c00000000a4f3420 TRAP: 0700 MSR: 8000000000021033 <SF,ME,IR,DR,RI,LE> CR: 28002840 XER: 20040000 CFAR: c000000000128be0 IRQMASK: 3 GPR00: c00000000006e2ec c00000000a4f36c0 c0000000014f0700 0000000000000020 GPR04: 0000000000000001 c000000001290f50 0000000000000001 c000000001290f80 GPR08: c000000001612b08 0000000000000000 0000000000000000 00000000ffffe0f7 GPR12: 0000000048002840 c0000000016e0000 c00c000000021c80 c000000000fd6f60 GPR16: 0000000000000000 c00000000a104698 0000000000000003 c0000000087f0000 GPR20: 0000000000000100 c0000000070330b8 0000000000000000 0000000000000004 GPR24: 0000000002000000 0000000000000300 0000000002000000 c00000000a5b0c00 GPR28: 0000000000000000 000000000a000000 00007fffb2a90038 c00000000a4f3820 NIP [c00000000006e2f0] __do_page_fault+0x880/0xa90 LR [c00000000006e2ec] __do_page_fault+0x87c/0xa90 Call Trace: [c00000000a4f36c0] [c00000000006e2ec] __do_page_fault+0x87c/0xa90 (unreliable) [c00000000a4f3780] [c000000000e1c034] do_page_fault+0x34/0x90 [c00000000a4f37b0] [c000000000008908] data_access_common_virt+0x158/0x1b0 --- interrupt: 300 at __copy_tofrom_user_base+0x9c/0x5a4 NIP: c00000000009b028 LR: c000000000802978 CTR: 0000000000000800 REGS: c00000000a4f3820 TRAP: 0300 MSR: 800000000280b033 <SF,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 24004840 XER: 00000000 CFAR: c00000000009aff4 DAR: 00007fffb2a90038 DSISR: 0a000000 IRQMASK: 0 GPR00: 0000000000000000 c00000000a4f3ac0 c0000000014f0700 00007fffb2a90028 GPR04: c000000008720010 0000000000010000 0000000000000000 0000000000000000 GPR08: 0000000000000000 0000000000000000 0000000000000000 0000000000000001 GPR12: 0000000000004000 c0000000016e0000 c00c000000021c80 c000000000fd6f60 GPR16: 0000000000000000 c00000000a104698 0000000000000003 c0000000087f0000 GPR20: 0000000000000100 c0000000070330b8 0000000000000000 0000000000000004 GPR24: c00000000a4f3c80 c000000008720000 0000000000010000 0000000000000000 GPR28: 0000000000010000 0000000008720000 0000000000010000 c000000001515b98 NIP [c00000000009b028] __copy_tofrom_user_base+0x9c/0x5a4 LR [c000000000802978] copyout+0x68/0xc0 --- interrupt: 300 [c00000000a4f3af0] [c0000000008074b8] copy_page_to_iter+0x188/0x540 [c00000000a4f3b50] [c00000000035c678] generic_file_buffered_read+0x358/0xd80 [c00000000a4f3c40] [c0000000004c1e90] blkdev_read_iter+0x50/0x80 [c00000000a4f3c60] [c00000000045733c] new_sync_read+0x12c/0x1c0 [c00000000a4f3d00] [c00000000045a1f0] vfs_read+0x1d0/0x240 [c00000000a4f3d50] [c00000000045a7f4] ksys_read+0x84/0x140 [c00000000a4f3da0] [c000000000033a60] system_call_exception+0x100/0x280 [c00000000a4f3e10] [c00000000000c508] system_call_common+0xf8/0x2f8 Instruction dump: eae10078 3be0000b 4bfff890 60420000 792917e1 4182ff18 3c82ffab 3884a5e0 3c62ffab 3863a6e8 480ba891 60000000 <0fe00000> 3be0000b 4bfff860 e93c0938 Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20201107023305.2384874-1-npiggin@gmail.com
2020-11-07 10:33:05 +08:00
static void __show_regs(struct pt_regs *regs)
{
int i, trap;
printk("NIP: "REG" LR: "REG" CTR: "REG"\n",
regs->nip, regs->link, regs->ctr);
printk("REGS: %px TRAP: %04lx %s (%s)\n",
regs, regs->trap, print_tainted(), init_utsname()->release);
printk("MSR: "REG" ", regs->msr);
print_msr_bits(regs->msr);
pr_cont(" CR: %08lx XER: %08lx\n", regs->ccr, regs->xer);
trap = TRAP(regs);
if (!trap_is_syscall(regs) && cpu_has_feature(CPU_FTR_CFAR))
pr_cont("CFAR: "REG" ", regs->orig_gpr3);
if (trap == INTERRUPT_MACHINE_CHECK ||
trap == INTERRUPT_DATA_STORAGE ||
trap == INTERRUPT_ALIGNMENT) {
if (IS_ENABLED(CONFIG_4xx) || IS_ENABLED(CONFIG_BOOKE))
pr_cont("DEAR: "REG" ESR: "REG" ", regs->dar, regs->dsisr);
else
pr_cont("DAR: "REG" DSISR: %08lx ", regs->dar, regs->dsisr);
}
#ifdef CONFIG_PPC64
pr_cont("IRQMASK: %lx ", regs->softe);
#endif
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
if (MSR_TM_ACTIVE(regs->msr))
pr_cont("\nPACATMSCRATCH: %016llx ", get_paca()->tm_scratch);
#endif
for (i = 0; i < 32; i++) {
if ((i % REGS_PER_LINE) == 0)
pr_cont("\nGPR%02d: ", i);
pr_cont(REG " ", regs->gpr[i]);
}
pr_cont("\n");
/*
* Lookup NIP late so we have the best change of getting the
* above info out without failing
*/
if (IS_ENABLED(CONFIG_KALLSYMS)) {
printk("NIP ["REG"] %pS\n", regs->nip, (void *)regs->nip);
printk("LR ["REG"] %pS\n", regs->link, (void *)regs->link);
}
powerpc: show registers when unwinding interrupt frames It's often useful to know the register state for interrupts in the stack frame. In the below example (with this patch applied), the important information is the state of the page fault. A blatant case like this probably rather should have the page fault regs passed down to the warning, but quite often there are less obvious cases where an interrupt shows up that might give some more clues. The downside is longer and more complex bug output. Bug: Write fault blocked by AMR! WARNING: CPU: 0 PID: 72 at arch/powerpc/include/asm/book3s/64/kup-radix.h:164 __do_page_fault+0x880/0xa90 Modules linked in: CPU: 0 PID: 72 Comm: systemd-gpt-aut Not tainted NIP: c00000000006e2f0 LR: c00000000006e2ec CTR: 0000000000000000 REGS: c00000000a4f3420 TRAP: 0700 MSR: 8000000000021033 <SF,ME,IR,DR,RI,LE> CR: 28002840 XER: 20040000 CFAR: c000000000128be0 IRQMASK: 3 GPR00: c00000000006e2ec c00000000a4f36c0 c0000000014f0700 0000000000000020 GPR04: 0000000000000001 c000000001290f50 0000000000000001 c000000001290f80 GPR08: c000000001612b08 0000000000000000 0000000000000000 00000000ffffe0f7 GPR12: 0000000048002840 c0000000016e0000 c00c000000021c80 c000000000fd6f60 GPR16: 0000000000000000 c00000000a104698 0000000000000003 c0000000087f0000 GPR20: 0000000000000100 c0000000070330b8 0000000000000000 0000000000000004 GPR24: 0000000002000000 0000000000000300 0000000002000000 c00000000a5b0c00 GPR28: 0000000000000000 000000000a000000 00007fffb2a90038 c00000000a4f3820 NIP [c00000000006e2f0] __do_page_fault+0x880/0xa90 LR [c00000000006e2ec] __do_page_fault+0x87c/0xa90 Call Trace: [c00000000a4f36c0] [c00000000006e2ec] __do_page_fault+0x87c/0xa90 (unreliable) [c00000000a4f3780] [c000000000e1c034] do_page_fault+0x34/0x90 [c00000000a4f37b0] [c000000000008908] data_access_common_virt+0x158/0x1b0 --- interrupt: 300 at __copy_tofrom_user_base+0x9c/0x5a4 NIP: c00000000009b028 LR: c000000000802978 CTR: 0000000000000800 REGS: c00000000a4f3820 TRAP: 0300 MSR: 800000000280b033 <SF,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 24004840 XER: 00000000 CFAR: c00000000009aff4 DAR: 00007fffb2a90038 DSISR: 0a000000 IRQMASK: 0 GPR00: 0000000000000000 c00000000a4f3ac0 c0000000014f0700 00007fffb2a90028 GPR04: c000000008720010 0000000000010000 0000000000000000 0000000000000000 GPR08: 0000000000000000 0000000000000000 0000000000000000 0000000000000001 GPR12: 0000000000004000 c0000000016e0000 c00c000000021c80 c000000000fd6f60 GPR16: 0000000000000000 c00000000a104698 0000000000000003 c0000000087f0000 GPR20: 0000000000000100 c0000000070330b8 0000000000000000 0000000000000004 GPR24: c00000000a4f3c80 c000000008720000 0000000000010000 0000000000000000 GPR28: 0000000000010000 0000000008720000 0000000000010000 c000000001515b98 NIP [c00000000009b028] __copy_tofrom_user_base+0x9c/0x5a4 LR [c000000000802978] copyout+0x68/0xc0 --- interrupt: 300 [c00000000a4f3af0] [c0000000008074b8] copy_page_to_iter+0x188/0x540 [c00000000a4f3b50] [c00000000035c678] generic_file_buffered_read+0x358/0xd80 [c00000000a4f3c40] [c0000000004c1e90] blkdev_read_iter+0x50/0x80 [c00000000a4f3c60] [c00000000045733c] new_sync_read+0x12c/0x1c0 [c00000000a4f3d00] [c00000000045a1f0] vfs_read+0x1d0/0x240 [c00000000a4f3d50] [c00000000045a7f4] ksys_read+0x84/0x140 [c00000000a4f3da0] [c000000000033a60] system_call_exception+0x100/0x280 [c00000000a4f3e10] [c00000000000c508] system_call_common+0xf8/0x2f8 Instruction dump: eae10078 3be0000b 4bfff890 60420000 792917e1 4182ff18 3c82ffab 3884a5e0 3c62ffab 3863a6e8 480ba891 60000000 <0fe00000> 3be0000b 4bfff860 e93c0938 Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20201107023305.2384874-1-npiggin@gmail.com
2020-11-07 10:33:05 +08:00
}
void show_regs(struct pt_regs *regs)
{
show_regs_print_info(KERN_DEFAULT);
__show_regs(regs);
show_stack(current, (unsigned long *) regs->gpr[1], KERN_DEFAULT);
if (!user_mode(regs))
show_instructions(regs);
}
void flush_thread(void)
{
#ifdef CONFIG_HAVE_HW_BREAKPOINT
flush_ptrace_hw_breakpoint(current);
#else /* CONFIG_HAVE_HW_BREAKPOINT */
set_debug_reg_defaults(&current->thread);
#endif /* CONFIG_HAVE_HW_BREAKPOINT */
}
void arch_setup_new_exec(void)
{
#ifdef CONFIG_PPC_BOOK3S_64
if (!radix_enabled())
hash__setup_new_exec();
#endif
/*
* If we exec out of a kernel thread then thread.regs will not be
* set. Do it now.
*/
if (!current->thread.regs) {
struct pt_regs *regs = task_stack_page(current) + THREAD_SIZE;
current->thread.regs = regs - 1;
}
#ifdef CONFIG_PPC_MEM_KEYS
current->thread.regs->amr = default_amr;
current->thread.regs->iamr = default_iamr;
#endif
}
#ifdef CONFIG_PPC64
/**
* Assign a TIDR (thread ID) for task @t and set it in the thread
* structure. For now, we only support setting TIDR for 'current' task.
*
* Since the TID value is a truncated form of it PID, it is possible
* (but unlikely) for 2 threads to have the same TID. In the unlikely event
* that 2 threads share the same TID and are waiting, one of the following
* cases will happen:
*
* 1. The correct thread is running, the wrong thread is not
* In this situation, the correct thread is woken and proceeds to pass it's
* condition check.
*
* 2. Neither threads are running
* In this situation, neither thread will be woken. When scheduled, the waiting
* threads will execute either a wait, which will return immediately, followed
* by a condition check, which will pass for the correct thread and fail
* for the wrong thread, or they will execute the condition check immediately.
*
* 3. The wrong thread is running, the correct thread is not
* The wrong thread will be woken, but will fail it's condition check and
* re-execute wait. The correct thread, when scheduled, will execute either
* it's condition check (which will pass), or wait, which returns immediately
* when called the first time after the thread is scheduled, followed by it's
* condition check (which will pass).
*
* 4. Both threads are running
* Both threads will be woken. The wrong thread will fail it's condition check
* and execute another wait, while the correct thread will pass it's condition
* check.
*
* @t: the task to set the thread ID for
*/
int set_thread_tidr(struct task_struct *t)
{
if (!cpu_has_feature(CPU_FTR_P9_TIDR))
return -EINVAL;
if (t != current)
return -EINVAL;
if (t->thread.tidr)
return 0;
t->thread.tidr = (u16)task_pid_nr(t);
mtspr(SPRN_TIDR, t->thread.tidr);
return 0;
}
EXPORT_SYMBOL_GPL(set_thread_tidr);
#endif /* CONFIG_PPC64 */
void
release_thread(struct task_struct *t)
{
}
/*
fork: move the real prepare_to_copy() users to arch_dup_task_struct() Historical prepare_to_copy() is mostly a no-op, duplicated for majority of the architectures and the rest following the x86 model of flushing the extended register state like fpu there. Remove it and use the arch_dup_task_struct() instead. Suggested-by: Oleg Nesterov <oleg@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Link: http://lkml.kernel.org/r/1336692811-30576-1-git-send-email-suresh.b.siddha@intel.com Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Koichi Yasutake <yasutake.koichi@jp.panasonic.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Chris Zankel <chris@zankel.net> Cc: Richard Henderson <rth@twiddle.net> Cc: Russell King <linux@arm.linux.org.uk> Cc: Haavard Skinnemoen <hskinnemoen@gmail.com> Cc: Mike Frysinger <vapier@gentoo.org> Cc: Mark Salter <msalter@redhat.com> Cc: Aurelien Jacquiot <a-jacquiot@ti.com> Cc: Mikael Starvik <starvik@axis.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: Tony Luck <tony.luck@intel.com> Cc: Michal Simek <monstr@monstr.eu> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Jonas Bonn <jonas@southpole.se> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Chen Liqin <liqin.chen@sunplusct.com> Cc: Lennox Wu <lennox.wu@gmail.com> Cc: David S. Miller <davem@davemloft.net> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Guan Xuetao <gxt@mprc.pku.edu.cn> Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2012-05-17 06:03:51 +08:00
* this gets called so that we can store coprocessor state into memory and
* copy the current task into the new thread.
*/
fork: move the real prepare_to_copy() users to arch_dup_task_struct() Historical prepare_to_copy() is mostly a no-op, duplicated for majority of the architectures and the rest following the x86 model of flushing the extended register state like fpu there. Remove it and use the arch_dup_task_struct() instead. Suggested-by: Oleg Nesterov <oleg@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Link: http://lkml.kernel.org/r/1336692811-30576-1-git-send-email-suresh.b.siddha@intel.com Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Koichi Yasutake <yasutake.koichi@jp.panasonic.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Chris Zankel <chris@zankel.net> Cc: Richard Henderson <rth@twiddle.net> Cc: Russell King <linux@arm.linux.org.uk> Cc: Haavard Skinnemoen <hskinnemoen@gmail.com> Cc: Mike Frysinger <vapier@gentoo.org> Cc: Mark Salter <msalter@redhat.com> Cc: Aurelien Jacquiot <a-jacquiot@ti.com> Cc: Mikael Starvik <starvik@axis.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: Tony Luck <tony.luck@intel.com> Cc: Michal Simek <monstr@monstr.eu> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Jonas Bonn <jonas@southpole.se> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Chen Liqin <liqin.chen@sunplusct.com> Cc: Lennox Wu <lennox.wu@gmail.com> Cc: David S. Miller <davem@davemloft.net> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Guan Xuetao <gxt@mprc.pku.edu.cn> Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2012-05-17 06:03:51 +08:00
int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
{
flush_all_to_thread(src);
/*
* Flush TM state out so we can copy it. __switch_to_tm() does this
* flush but it removes the checkpointed state from the current CPU and
* transitions the CPU out of TM mode. Hence we need to call
* tm_recheckpoint_new_task() (on the same task) to restore the
* checkpointed state back and the TM mode.
*
* Can't pass dst because it isn't ready. Doesn't matter, passing
* dst is only important for __switch_to()
*/
powerpc: tm: Always use fp_state and vr_state to store live registers There is currently an inconsistency as to how the entire CPU register state is saved and restored when a thread uses transactional memory (TM). Using transactional memory results in the CPU having duplicated (almost) all of its register state. This duplication results in a set of registers which can be considered 'live', those being currently modified by the instructions being executed and another set that is frozen at a point in time. On context switch, both sets of state have to be saved and (later) restored. These two states are often called a variety of different things. Common terms for the state which only exists after the CPU has entered a transaction (performed a TBEGIN instruction) in hardware are 'transactional' or 'speculative'. Between a TBEGIN and a TEND or TABORT (or an event that causes the hardware to abort), regardless of the use of TSUSPEND the transactional state can be referred to as the live state. The second state is often to referred to as the 'checkpointed' state and is a duplication of the live state when the TBEGIN instruction is executed. This state is kept in the hardware and will be rolled back to on transaction failure. Currently all the registers stored in pt_regs are ALWAYS the live registers, that is, when a thread has transactional registers their values are stored in pt_regs and the checkpointed state is in ckpt_regs. A strange opposite is true for fp_state/vr_state. When a thread is non transactional fp_state/vr_state holds the live registers. When a thread has initiated a transaction fp_state/vr_state holds the checkpointed state and transact_fp/transact_vr become the structure which holds the live state (at this point it is a transactional state). This method creates confusion as to where the live state is, in some circumstances it requires extra work to determine where to put the live state and prevents the use of common functions designed (probably before TM) to save the live state. With this patch pt_regs, fp_state and vr_state all represent the same thing and the other structures [pending rename] are for checkpointed state. Acked-by: Simon Guo <wei.guo.simon@gmail.com> Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-09-23 14:18:24 +08:00
__switch_to_tm(src, src);
fork: move the real prepare_to_copy() users to arch_dup_task_struct() Historical prepare_to_copy() is mostly a no-op, duplicated for majority of the architectures and the rest following the x86 model of flushing the extended register state like fpu there. Remove it and use the arch_dup_task_struct() instead. Suggested-by: Oleg Nesterov <oleg@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Link: http://lkml.kernel.org/r/1336692811-30576-1-git-send-email-suresh.b.siddha@intel.com Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Koichi Yasutake <yasutake.koichi@jp.panasonic.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Chris Zankel <chris@zankel.net> Cc: Richard Henderson <rth@twiddle.net> Cc: Russell King <linux@arm.linux.org.uk> Cc: Haavard Skinnemoen <hskinnemoen@gmail.com> Cc: Mike Frysinger <vapier@gentoo.org> Cc: Mark Salter <msalter@redhat.com> Cc: Aurelien Jacquiot <a-jacquiot@ti.com> Cc: Mikael Starvik <starvik@axis.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: Tony Luck <tony.luck@intel.com> Cc: Michal Simek <monstr@monstr.eu> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Jonas Bonn <jonas@southpole.se> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Chen Liqin <liqin.chen@sunplusct.com> Cc: Lennox Wu <lennox.wu@gmail.com> Cc: David S. Miller <davem@davemloft.net> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Guan Xuetao <gxt@mprc.pku.edu.cn> Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2012-05-17 06:03:51 +08:00
*dst = *src;
clear_task_ebb(dst);
fork: move the real prepare_to_copy() users to arch_dup_task_struct() Historical prepare_to_copy() is mostly a no-op, duplicated for majority of the architectures and the rest following the x86 model of flushing the extended register state like fpu there. Remove it and use the arch_dup_task_struct() instead. Suggested-by: Oleg Nesterov <oleg@redhat.com> Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com> Link: http://lkml.kernel.org/r/1336692811-30576-1-git-send-email-suresh.b.siddha@intel.com Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: David Howells <dhowells@redhat.com> Cc: Koichi Yasutake <yasutake.koichi@jp.panasonic.com> Cc: Paul Mackerras <paulus@samba.org> Cc: Paul Mundt <lethal@linux-sh.org> Cc: Chris Zankel <chris@zankel.net> Cc: Richard Henderson <rth@twiddle.net> Cc: Russell King <linux@arm.linux.org.uk> Cc: Haavard Skinnemoen <hskinnemoen@gmail.com> Cc: Mike Frysinger <vapier@gentoo.org> Cc: Mark Salter <msalter@redhat.com> Cc: Aurelien Jacquiot <a-jacquiot@ti.com> Cc: Mikael Starvik <starvik@axis.com> Cc: Yoshinori Sato <ysato@users.sourceforge.jp> Cc: Richard Kuo <rkuo@codeaurora.org> Cc: Tony Luck <tony.luck@intel.com> Cc: Michal Simek <monstr@monstr.eu> Cc: Ralf Baechle <ralf@linux-mips.org> Cc: Jonas Bonn <jonas@southpole.se> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: Chen Liqin <liqin.chen@sunplusct.com> Cc: Lennox Wu <lennox.wu@gmail.com> Cc: David S. Miller <davem@davemloft.net> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Jeff Dike <jdike@addtoit.com> Cc: Richard Weinberger <richard@nod.at> Cc: Guan Xuetao <gxt@mprc.pku.edu.cn> Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
2012-05-17 06:03:51 +08:00
return 0;
}
static void setup_ksp_vsid(struct task_struct *p, unsigned long sp)
{
#ifdef CONFIG_PPC_BOOK3S_64
unsigned long sp_vsid;
unsigned long llp = mmu_psize_defs[mmu_linear_psize].sllp;
if (radix_enabled())
return;
if (mmu_has_feature(MMU_FTR_1T_SEGMENT))
sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_1T)
<< SLB_VSID_SHIFT_1T;
else
sp_vsid = get_kernel_vsid(sp, MMU_SEGSIZE_256M)
<< SLB_VSID_SHIFT;
sp_vsid |= SLB_VSID_KERNEL | llp;
p->thread.ksp_vsid = sp_vsid;
#endif
}
/*
* Copy a thread..
*/
/*
* Copy architecture-specific thread state
*/
int copy_thread(unsigned long clone_flags, unsigned long usp,
unsigned long kthread_arg, struct task_struct *p,
unsigned long tls)
{
struct pt_regs *childregs, *kregs;
extern void ret_from_fork(void);
extern void ret_from_fork_scv(void);
extern void ret_from_kernel_thread(void);
void (*f)(void);
unsigned long sp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
struct thread_info *ti = task_thread_info(p);
#ifdef CONFIG_HAVE_HW_BREAKPOINT
int i;
#endif
klp_init_thread_info(p);
/* Copy registers */
sp -= sizeof(struct pt_regs);
childregs = (struct pt_regs *) sp;
if (unlikely(p->flags & (PF_KTHREAD | PF_IO_WORKER))) {
/* kernel thread */
memset(childregs, 0, sizeof(struct pt_regs));
childregs->gpr[1] = sp + sizeof(struct pt_regs);
/* function */
if (usp)
childregs->gpr[14] = ppc_function_entry((void *)usp);
#ifdef CONFIG_PPC64
clear_tsk_thread_flag(p, TIF_32BIT);
childregs->softe = IRQS_ENABLED;
#endif
childregs->gpr[15] = kthread_arg;
p->thread.regs = NULL; /* no user register state */
ti->flags |= _TIF_RESTOREALL;
f = ret_from_kernel_thread;
} else {
/* user thread */
struct pt_regs *regs = current_pt_regs();
*childregs = *regs;
if (usp)
childregs->gpr[1] = usp;
p->thread.regs = childregs;
/* 64s sets this in ret_from_fork */
if (!IS_ENABLED(CONFIG_PPC_BOOK3S_64))
childregs->gpr[3] = 0; /* Result from fork() */
if (clone_flags & CLONE_SETTLS) {
if (!is_32bit_task())
childregs->gpr[13] = tls;
else
childregs->gpr[2] = tls;
}
if (trap_is_scv(regs))
f = ret_from_fork_scv;
else
f = ret_from_fork;
}
childregs->msr &= ~(MSR_FP|MSR_VEC|MSR_VSX);
sp -= STACK_FRAME_OVERHEAD;
/*
* The way this works is that at some point in the future
* some task will call _switch to switch to the new task.
* That will pop off the stack frame created below and start
* the new task running at ret_from_fork. The new task will
* do some house keeping and then return from the fork or clone
* system call, using the stack frame created above.
*/
((unsigned long *)sp)[0] = 0;
sp -= sizeof(struct pt_regs);
kregs = (struct pt_regs *) sp;
sp -= STACK_FRAME_OVERHEAD;
p->thread.ksp = sp;
#ifdef CONFIG_HAVE_HW_BREAKPOINT
for (i = 0; i < nr_wp_slots(); i++)
p->thread.ptrace_bps[i] = NULL;
#endif
#ifdef CONFIG_PPC_FPU_REGS
p->thread.fp_save_area = NULL;
#endif
#ifdef CONFIG_ALTIVEC
p->thread.vr_save_area = NULL;
#endif
powerpc/32s: Rework Kernel Userspace Access Protection On book3s/32, KUAP is provided by toggling Ks bit in segment registers. One segment register addresses 256M of virtual memory. At the time being, KUAP implements a complex logic to apply the unlock/lock on the exact number of segments covering the user range to access, with saving the boundaries of the range of segments in a member of thread struct. But most if not all user accesses are within a single segment. Rework KUAP with a different approach: - Open only one segment, the one corresponding to the starting address of the range to be accessed. - If a second segment is involved, it will generate a page fault. The segment will then be open by the page fault handler. The kuap member of thread struct will now contain: - The start address of the current on going user access, that will be used to know which segment to lock at the end of the user access. - ~0 when no user access is open - ~1 when additionnal segments are opened by a page fault. Then, at lock time - When only one segment is open, close it. - When several segments are open, close all user segments. Almost 100% of the time, only one segment will be involved. In interrupts, inline the function that unlock/lock all segments, because not inlining them implies a lot of register save/restore. With the patch, writing value 128 in userspace in perf_copy_attr() is done with 16 instructions: 3890: 93 82 04 dc stw r28,1244(r2) 3894: 7d 20 e5 26 mfsrin r9,r28 3898: 55 29 00 80 rlwinm r9,r9,0,2,0 389c: 7d 20 e1 e4 mtsrin r9,r28 38a0: 4c 00 01 2c isync 38a4: 39 20 00 80 li r9,128 38a8: 91 3c 00 00 stw r9,0(r28) 38ac: 81 42 04 dc lwz r10,1244(r2) 38b0: 39 00 ff ff li r8,-1 38b4: 91 02 04 dc stw r8,1244(r2) 38b8: 2c 0a ff fe cmpwi r10,-2 38bc: 41 82 00 88 beq 3944 <perf_copy_attr+0x36c> 38c0: 7d 20 55 26 mfsrin r9,r10 38c4: 65 29 40 00 oris r9,r9,16384 38c8: 7d 20 51 e4 mtsrin r9,r10 38cc: 4c 00 01 2c isync ... 3944: 48 00 00 01 bl 3944 <perf_copy_attr+0x36c> 3944: R_PPC_REL24 kuap_lock_all_ool Before the patch it was 118 instructions. In reality only 42 are executed in most cases, but GCC is not able to see that a properly aligned user access cannot involve more than one segment. 5060: 39 1d 00 04 addi r8,r29,4 5064: 3d 20 b0 00 lis r9,-20480 5068: 7c 08 48 40 cmplw r8,r9 506c: 40 81 00 08 ble 5074 <perf_copy_attr+0x2cc> 5070: 3d 00 b0 00 lis r8,-20480 5074: 39 28 ff ff addi r9,r8,-1 5078: 57 aa 00 06 rlwinm r10,r29,0,0,3 507c: 55 29 27 3e rlwinm r9,r9,4,28,31 5080: 39 29 00 01 addi r9,r9,1 5084: 7d 29 53 78 or r9,r9,r10 5088: 91 22 04 dc stw r9,1244(r2) 508c: 7d 20 ed 26 mfsrin r9,r29 5090: 55 29 00 80 rlwinm r9,r9,0,2,0 5094: 7c 08 50 40 cmplw r8,r10 5098: 40 81 00 c0 ble 5158 <perf_copy_attr+0x3b0> 509c: 7d 46 50 f8 not r6,r10 50a0: 7c c6 42 14 add r6,r6,r8 50a4: 54 c6 27 be rlwinm r6,r6,4,30,31 50a8: 7d 20 51 e4 mtsrin r9,r10 50ac: 3c ea 10 00 addis r7,r10,4096 50b0: 39 29 01 11 addi r9,r9,273 50b4: 7f 88 38 40 cmplw cr7,r8,r7 50b8: 55 29 02 06 rlwinm r9,r9,0,8,3 50bc: 40 9d 00 9c ble cr7,5158 <perf_copy_attr+0x3b0> 50c0: 2f 86 00 00 cmpwi cr7,r6,0 50c4: 41 9e 00 4c beq cr7,5110 <perf_copy_attr+0x368> 50c8: 2f 86 00 01 cmpwi cr7,r6,1 50cc: 41 9e 00 2c beq cr7,50f8 <perf_copy_attr+0x350> 50d0: 2f 86 00 02 cmpwi cr7,r6,2 50d4: 41 9e 00 14 beq cr7,50e8 <perf_copy_attr+0x340> 50d8: 7d 20 39 e4 mtsrin r9,r7 50dc: 39 29 01 11 addi r9,r9,273 50e0: 3c e7 10 00 addis r7,r7,4096 50e4: 55 29 02 06 rlwinm r9,r9,0,8,3 50e8: 7d 20 39 e4 mtsrin r9,r7 50ec: 39 29 01 11 addi r9,r9,273 50f0: 3c e7 10 00 addis r7,r7,4096 50f4: 55 29 02 06 rlwinm r9,r9,0,8,3 50f8: 7d 20 39 e4 mtsrin r9,r7 50fc: 3c e7 10 00 addis r7,r7,4096 5100: 39 29 01 11 addi r9,r9,273 5104: 7f 88 38 40 cmplw cr7,r8,r7 5108: 55 29 02 06 rlwinm r9,r9,0,8,3 510c: 40 9d 00 4c ble cr7,5158 <perf_copy_attr+0x3b0> 5110: 7d 20 39 e4 mtsrin r9,r7 5114: 39 29 01 11 addi r9,r9,273 5118: 3c c7 10 00 addis r6,r7,4096 511c: 55 29 02 06 rlwinm r9,r9,0,8,3 5120: 7d 20 31 e4 mtsrin r9,r6 5124: 39 29 01 11 addi r9,r9,273 5128: 3c c6 10 00 addis r6,r6,4096 512c: 55 29 02 06 rlwinm r9,r9,0,8,3 5130: 7d 20 31 e4 mtsrin r9,r6 5134: 39 29 01 11 addi r9,r9,273 5138: 3c c7 30 00 addis r6,r7,12288 513c: 55 29 02 06 rlwinm r9,r9,0,8,3 5140: 7d 20 31 e4 mtsrin r9,r6 5144: 3c e7 40 00 addis r7,r7,16384 5148: 39 29 01 11 addi r9,r9,273 514c: 7f 88 38 40 cmplw cr7,r8,r7 5150: 55 29 02 06 rlwinm r9,r9,0,8,3 5154: 41 9d ff bc bgt cr7,5110 <perf_copy_attr+0x368> 5158: 4c 00 01 2c isync 515c: 39 20 00 80 li r9,128 5160: 91 3d 00 00 stw r9,0(r29) 5164: 38 e0 00 00 li r7,0 5168: 90 e2 04 dc stw r7,1244(r2) 516c: 7d 20 ed 26 mfsrin r9,r29 5170: 65 29 40 00 oris r9,r9,16384 5174: 40 81 00 c0 ble 5234 <perf_copy_attr+0x48c> 5178: 7d 47 50 f8 not r7,r10 517c: 7c e7 42 14 add r7,r7,r8 5180: 54 e7 27 be rlwinm r7,r7,4,30,31 5184: 7d 20 51 e4 mtsrin r9,r10 5188: 3d 4a 10 00 addis r10,r10,4096 518c: 39 29 01 11 addi r9,r9,273 5190: 7c 08 50 40 cmplw r8,r10 5194: 55 29 02 06 rlwinm r9,r9,0,8,3 5198: 40 81 00 9c ble 5234 <perf_copy_attr+0x48c> 519c: 2c 07 00 00 cmpwi r7,0 51a0: 41 82 00 4c beq 51ec <perf_copy_attr+0x444> 51a4: 2c 07 00 01 cmpwi r7,1 51a8: 41 82 00 2c beq 51d4 <perf_copy_attr+0x42c> 51ac: 2c 07 00 02 cmpwi r7,2 51b0: 41 82 00 14 beq 51c4 <perf_copy_attr+0x41c> 51b4: 7d 20 51 e4 mtsrin r9,r10 51b8: 39 29 01 11 addi r9,r9,273 51bc: 3d 4a 10 00 addis r10,r10,4096 51c0: 55 29 02 06 rlwinm r9,r9,0,8,3 51c4: 7d 20 51 e4 mtsrin r9,r10 51c8: 39 29 01 11 addi r9,r9,273 51cc: 3d 4a 10 00 addis r10,r10,4096 51d0: 55 29 02 06 rlwinm r9,r9,0,8,3 51d4: 7d 20 51 e4 mtsrin r9,r10 51d8: 3d 4a 10 00 addis r10,r10,4096 51dc: 39 29 01 11 addi r9,r9,273 51e0: 7c 08 50 40 cmplw r8,r10 51e4: 55 29 02 06 rlwinm r9,r9,0,8,3 51e8: 40 81 00 4c ble 5234 <perf_copy_attr+0x48c> 51ec: 7d 20 51 e4 mtsrin r9,r10 51f0: 39 29 01 11 addi r9,r9,273 51f4: 3c ea 10 00 addis r7,r10,4096 51f8: 55 29 02 06 rlwinm r9,r9,0,8,3 51fc: 7d 20 39 e4 mtsrin r9,r7 5200: 39 29 01 11 addi r9,r9,273 5204: 3c e7 10 00 addis r7,r7,4096 5208: 55 29 02 06 rlwinm r9,r9,0,8,3 520c: 7d 20 39 e4 mtsrin r9,r7 5210: 39 29 01 11 addi r9,r9,273 5214: 3c ea 30 00 addis r7,r10,12288 5218: 55 29 02 06 rlwinm r9,r9,0,8,3 521c: 7d 20 39 e4 mtsrin r9,r7 5220: 3d 4a 40 00 addis r10,r10,16384 5224: 39 29 01 11 addi r9,r9,273 5228: 7c 08 50 40 cmplw r8,r10 522c: 55 29 02 06 rlwinm r9,r9,0,8,3 5230: 41 81 ff bc bgt 51ec <perf_copy_attr+0x444> 5234: 4c 00 01 2c isync Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu> [mpe: Export the ool handlers to fix build errors] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/d9121f96a7c4302946839a0771f5d1daeeb6968c.1622708530.git.christophe.leroy@csgroup.eu
2021-06-03 16:41:44 +08:00
#if defined(CONFIG_PPC_BOOK3S_32) && defined(CONFIG_PPC_KUAP)
p->thread.kuap = KUAP_NONE;
#endif
setup_ksp_vsid(p, sp);
#ifdef CONFIG_PPC64
if (cpu_has_feature(CPU_FTR_DSCR)) {
p->thread.dscr_inherit = current->thread.dscr_inherit;
p->thread.dscr = mfspr(SPRN_DSCR);
}
if (cpu_has_feature(CPU_FTR_HAS_PPR))
childregs->ppr = DEFAULT_PPR;
p->thread.tidr = 0;
#endif
/*
* Run with the current AMR value of the kernel
*/
#ifdef CONFIG_PPC_PKEY
if (mmu_has_feature(MMU_FTR_BOOK3S_KUAP))
kregs->amr = AMR_KUAP_BLOCKED;
if (mmu_has_feature(MMU_FTR_BOOK3S_KUEP))
kregs->iamr = AMR_KUEP_BLOCKED;
#endif
kregs->nip = ppc_function_entry(f);
return 0;
}
void preload_new_slb_context(unsigned long start, unsigned long sp);
/*
* Set up a thread for executing a new program
*/
void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
{
#ifdef CONFIG_PPC64
unsigned long load_addr = regs->gpr[2]; /* saved by ELF_PLAT_INIT */
if (IS_ENABLED(CONFIG_PPC_BOOK3S_64) && !radix_enabled())
preload_new_slb_context(start, sp);
#endif
powerpc/tm: Always reclaim in start_thread() for exec() class syscalls Userspace can quite legitimately perform an exec() syscall with a suspended transaction. exec() does not return to the old process, rather it load a new one and starts that, the expectation therefore is that the new process starts not in a transaction. Currently exec() is not treated any differently to any other syscall which creates problems. Firstly it could allow a new process to start with a suspended transaction for a binary that no longer exists. This means that the checkpointed state won't be valid and if the suspended transaction were ever to be resumed and subsequently aborted (a possibility which is exceedingly likely as exec()ing will likely doom the transaction) the new process will jump to invalid state. Secondly the incorrect attempt to keep the transactional state while still zeroing state for the new process creates at least two TM Bad Things. The first triggers on the rfid to return to userspace as start_thread() has given the new process a 'clean' MSR but the suspend will still be set in the hardware MSR. The second TM Bad Thing triggers in __switch_to() as the processor is still transactionally suspended but __switch_to() wants to zero the TM sprs for the new process. This is an example of the outcome of calling exec() with a suspended transaction. Note the first 700 is likely the first TM bad thing decsribed earlier only the kernel can't report it as we've loaded userspace registers. c000000000009980 is the rfid in fast_exception_return() Bad kernel stack pointer 3fffcfa1a370 at c000000000009980 Oops: Bad kernel stack pointer, sig: 6 [#1] CPU: 0 PID: 2006 Comm: tm-execed Not tainted NIP: c000000000009980 LR: 0000000000000000 CTR: 0000000000000000 REGS: c00000003ffefd40 TRAP: 0700 Not tainted MSR: 8000000300201031 <SF,ME,IR,DR,LE,TM[SE]> CR: 00000000 XER: 00000000 CFAR: c0000000000098b4 SOFTE: 0 PACATMSCRATCH: b00000010000d033 GPR00: 0000000000000000 00003fffcfa1a370 0000000000000000 0000000000000000 GPR04: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR08: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR12: 00003fff966611c0 0000000000000000 0000000000000000 0000000000000000 NIP [c000000000009980] fast_exception_return+0xb0/0xb8 LR [0000000000000000] (null) Call Trace: Instruction dump: f84d0278 e9a100d8 7c7b03a6 e84101a0 7c4ff120 e8410170 7c5a03a6 e8010070 e8410080 e8610088 e8810090 e8210078 <4c000024> 48000000 e8610178 88ed023b Kernel BUG at c000000000043e80 [verbose debug info unavailable] Unexpected TM Bad Thing exception at c000000000043e80 (msr 0x201033) Oops: Unrecoverable exception, sig: 6 [#2] CPU: 0 PID: 2006 Comm: tm-execed Tainted: G D task: c0000000fbea6d80 ti: c00000003ffec000 task.ti: c0000000fb7ec000 NIP: c000000000043e80 LR: c000000000015a24 CTR: 0000000000000000 REGS: c00000003ffef7e0 TRAP: 0700 Tainted: G D MSR: 8000000300201033 <SF,ME,IR,DR,RI,LE,TM[SE]> CR: 28002828 XER: 00000000 CFAR: c000000000015a20 SOFTE: 0 PACATMSCRATCH: b00000010000d033 GPR00: 0000000000000000 c00000003ffefa60 c000000000db5500 c0000000fbead000 GPR04: 8000000300001033 2222222222222222 2222222222222222 00000000ff160000 GPR08: 0000000000000000 800000010000d033 c0000000fb7e3ea0 c00000000fe00004 GPR12: 0000000000002200 c00000000fe00000 0000000000000000 0000000000000000 GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR20: 0000000000000000 0000000000000000 c0000000fbea7410 00000000ff160000 GPR24: c0000000ffe1f600 c0000000fbea8700 c0000000fbea8700 c0000000fbead000 GPR28: c000000000e20198 c0000000fbea6d80 c0000000fbeab680 c0000000fbea6d80 NIP [c000000000043e80] tm_restore_sprs+0xc/0x1c LR [c000000000015a24] __switch_to+0x1f4/0x420 Call Trace: Instruction dump: 7c800164 4e800020 7c0022a6 f80304a8 7c0222a6 f80304b0 7c0122a6 f80304b8 4e800020 e80304a8 7c0023a6 e80304b0 <7c0223a6> e80304b8 7c0123a6 4e800020 This fixes CVE-2016-5828. Fixes: bc2a9408fa65 ("powerpc: Hook in new transactional memory code") Cc: stable@vger.kernel.org # v3.9+ Signed-off-by: Cyril Bur <cyrilbur@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2016-06-17 12:58:34 +08:00
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
/*
* Clear any transactional state, we're exec()ing. The cause is
* not important as there will never be a recheckpoint so it's not
* user visible.
*/
if (MSR_TM_SUSPENDED(mfmsr()))
tm_reclaim_current(0);
#endif
memset(regs->gpr, 0, sizeof(regs->gpr));
regs->ctr = 0;
regs->link = 0;
regs->xer = 0;
regs->ccr = 0;
regs->gpr[1] = sp;
#ifdef CONFIG_PPC32
regs->mq = 0;
regs->nip = start;
regs->msr = MSR_USER;
#else
if (!is_32bit_task()) {
unsigned long entry;
if (is_elf2_task()) {
/* Look ma, no function descriptors! */
entry = start;
/*
* Ulrich says:
* The latest iteration of the ABI requires that when
* calling a function (at its global entry point),
* the caller must ensure r12 holds the entry point
* address (so that the function can quickly
* establish addressability).
*/
regs->gpr[12] = start;
/* Make sure that's restored on entry to userspace. */
set_thread_flag(TIF_RESTOREALL);
} else {
unsigned long toc;
/* start is a relocated pointer to the function
* descriptor for the elf _start routine. The first
* entry in the function descriptor is the entry
* address of _start and the second entry is the TOC
* value we need to use.
*/
__get_user(entry, (unsigned long __user *)start);
__get_user(toc, (unsigned long __user *)start+1);
/* Check whether the e_entry function descriptor entries
* need to be relocated before we can use them.
*/
if (load_addr != 0) {
entry += load_addr;
toc += load_addr;
}
regs->gpr[2] = toc;
}
regs->nip = entry;
regs->msr = MSR_USER64;
} else {
regs->nip = start;
regs->gpr[2] = 0;
regs->msr = MSR_USER32;
}
#endif
#ifdef CONFIG_VSX
current->thread.used_vsr = 0;
#endif
current->thread.load_slb = 0;
current->thread.load_fp = 0;
#ifdef CONFIG_PPC_FPU_REGS
memset(&current->thread.fp_state, 0, sizeof(current->thread.fp_state));
current->thread.fp_save_area = NULL;
#endif
#ifdef CONFIG_ALTIVEC
memset(&current->thread.vr_state, 0, sizeof(current->thread.vr_state));
current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
current->thread.vr_save_area = NULL;
current->thread.vrsave = 0;
current->thread.used_vr = 0;
current->thread.load_vec = 0;
#endif /* CONFIG_ALTIVEC */
#ifdef CONFIG_SPE
memset(current->thread.evr, 0, sizeof(current->thread.evr));
current->thread.acc = 0;
current->thread.spefscr = 0;
current->thread.used_spe = 0;
#endif /* CONFIG_SPE */
#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
current->thread.tm_tfhar = 0;
current->thread.tm_texasr = 0;
current->thread.tm_tfiar = 0;
current->thread.load_tm = 0;
#endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
}
EXPORT_SYMBOL(start_thread);
#define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
| PR_FP_EXC_RES | PR_FP_EXC_INV)
int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
{
struct pt_regs *regs = tsk->thread.regs;
/* This is a bit hairy. If we are an SPE enabled processor
* (have embedded fp) we store the IEEE exception enable flags in
* fpexc_mode. fpexc_mode is also used for setting FP exception
* mode (asyn, precise, disabled) for 'Classic' FP. */
if (val & PR_FP_EXC_SW_ENABLE) {
if (cpu_has_feature(CPU_FTR_SPE)) {
powerpc: fix exception clearing in e500 SPE float emulation The e500 SPE floating-point emulation code clears existing exceptions (__FPU_FPSCR &= ~FP_EX_MASK;) before ORing in the exceptions from the emulated operation. However, these exception bits are the "sticky", cumulative exception bits, and should only be cleared by the user program setting SPEFSCR, not implicitly by any floating-point instruction (whether executed purely by the hardware or emulated). The spurious clearing of these bits shows up as missing exceptions in glibc testing. Fixing this, however, is not as simple as just not clearing the bits, because while the bits may be from previous floating-point operations (in which case they should not be cleared), the processor can also set the sticky bits itself before the interrupt for an exception occurs, and this can happen in cases when IEEE 754 semantics are that the sticky bit should not be set. Specifically, the "invalid" sticky bit is set in various cases with non-finite operands, where IEEE 754 semantics do not involve raising such an exception, and the "underflow" sticky bit is set in cases of exact underflow, whereas IEEE 754 semantics are that this flag is set only for inexact underflow. Thus, for correct emulation the kernel needs to know the setting of these two sticky bits before the instruction being emulated. When a floating-point operation raises an exception, the kernel can note the state of the sticky bits immediately afterwards. Some <fenv.h> functions that affect the state of these bits, such as fesetenv and feholdexcept, need to use prctl with PR_GET_FPEXC and PR_SET_FPEXC anyway, and so it is natural to record the state of those bits during that call into the kernel and so avoid any need for a separate call into the kernel to inform it of a change to those bits. Thus, the interface I chose to use (in this patch and the glibc port) is that one of those prctl calls must be made after any userspace change to those sticky bits, other than through a floating-point operation that traps into the kernel anyway. feclearexcept and fesetexceptflag duly make those calls, which would not be required were it not for this issue. The previous EGLIBC port, and the uClibc code copied from it, is fundamentally broken as regards any use of prctl for floating-point exceptions because it didn't use the PR_FP_EXC_SW_ENABLE bit in its prctl calls (and did various worse things, such as passing a pointer when prctl expected an integer). If you avoid anything where prctl is used, the clearing of sticky bits still means it will never give anything approximating correct exception semantics with existing kernels. I don't believe the patch makes things any worse for existing code that doesn't try to inform the kernel of changes to sticky bits - such code may get incorrect exceptions in some cases, but it would have done so anyway in other cases. Signed-off-by: Joseph Myers <joseph@codesourcery.com> Signed-off-by: Scott Wood <scottwood@freescale.com>
2013-12-11 07:07:45 +08:00
/*
* When the sticky exception bits are set
* directly by userspace, it must call prctl
* with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
* in the existing prctl settings) or
* PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
* the bits being set). <fenv.h> functions
* saving and restoring the whole
* floating-point environment need to do so
* anyway to restore the prctl settings from
* the saved environment.
*/
#ifdef CONFIG_SPE
powerpc: fix exception clearing in e500 SPE float emulation The e500 SPE floating-point emulation code clears existing exceptions (__FPU_FPSCR &= ~FP_EX_MASK;) before ORing in the exceptions from the emulated operation. However, these exception bits are the "sticky", cumulative exception bits, and should only be cleared by the user program setting SPEFSCR, not implicitly by any floating-point instruction (whether executed purely by the hardware or emulated). The spurious clearing of these bits shows up as missing exceptions in glibc testing. Fixing this, however, is not as simple as just not clearing the bits, because while the bits may be from previous floating-point operations (in which case they should not be cleared), the processor can also set the sticky bits itself before the interrupt for an exception occurs, and this can happen in cases when IEEE 754 semantics are that the sticky bit should not be set. Specifically, the "invalid" sticky bit is set in various cases with non-finite operands, where IEEE 754 semantics do not involve raising such an exception, and the "underflow" sticky bit is set in cases of exact underflow, whereas IEEE 754 semantics are that this flag is set only for inexact underflow. Thus, for correct emulation the kernel needs to know the setting of these two sticky bits before the instruction being emulated. When a floating-point operation raises an exception, the kernel can note the state of the sticky bits immediately afterwards. Some <fenv.h> functions that affect the state of these bits, such as fesetenv and feholdexcept, need to use prctl with PR_GET_FPEXC and PR_SET_FPEXC anyway, and so it is natural to record the state of those bits during that call into the kernel and so avoid any need for a separate call into the kernel to inform it of a change to those bits. Thus, the interface I chose to use (in this patch and the glibc port) is that one of those prctl calls must be made after any userspace change to those sticky bits, other than through a floating-point operation that traps into the kernel anyway. feclearexcept and fesetexceptflag duly make those calls, which would not be required were it not for this issue. The previous EGLIBC port, and the uClibc code copied from it, is fundamentally broken as regards any use of prctl for floating-point exceptions because it didn't use the PR_FP_EXC_SW_ENABLE bit in its prctl calls (and did various worse things, such as passing a pointer when prctl expected an integer). If you avoid anything where prctl is used, the clearing of sticky bits still means it will never give anything approximating correct exception semantics with existing kernels. I don't believe the patch makes things any worse for existing code that doesn't try to inform the kernel of changes to sticky bits - such code may get incorrect exceptions in some cases, but it would have done so anyway in other cases. Signed-off-by: Joseph Myers <joseph@codesourcery.com> Signed-off-by: Scott Wood <scottwood@freescale.com>
2013-12-11 07:07:45 +08:00
tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
tsk->thread.fpexc_mode = val &
(PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
#endif
return 0;
} else {
return -EINVAL;
}
}
/* on a CONFIG_SPE this does not hurt us. The bits that
* __pack_fe01 use do not overlap with bits used for
* PR_FP_EXC_SW_ENABLE. Additionally, the MSR[FE0,FE1] bits
* on CONFIG_SPE implementations are reserved so writing to
* them does not change anything */
if (val > PR_FP_EXC_PRECISE)
return -EINVAL;
tsk->thread.fpexc_mode = __pack_fe01(val);
if (regs != NULL && (regs->msr & MSR_FP) != 0)
regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
| tsk->thread.fpexc_mode;
return 0;
}
int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
{
unsigned int val = 0;
if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE) {
powerpc: fix exception clearing in e500 SPE float emulation The e500 SPE floating-point emulation code clears existing exceptions (__FPU_FPSCR &= ~FP_EX_MASK;) before ORing in the exceptions from the emulated operation. However, these exception bits are the "sticky", cumulative exception bits, and should only be cleared by the user program setting SPEFSCR, not implicitly by any floating-point instruction (whether executed purely by the hardware or emulated). The spurious clearing of these bits shows up as missing exceptions in glibc testing. Fixing this, however, is not as simple as just not clearing the bits, because while the bits may be from previous floating-point operations (in which case they should not be cleared), the processor can also set the sticky bits itself before the interrupt for an exception occurs, and this can happen in cases when IEEE 754 semantics are that the sticky bit should not be set. Specifically, the "invalid" sticky bit is set in various cases with non-finite operands, where IEEE 754 semantics do not involve raising such an exception, and the "underflow" sticky bit is set in cases of exact underflow, whereas IEEE 754 semantics are that this flag is set only for inexact underflow. Thus, for correct emulation the kernel needs to know the setting of these two sticky bits before the instruction being emulated. When a floating-point operation raises an exception, the kernel can note the state of the sticky bits immediately afterwards. Some <fenv.h> functions that affect the state of these bits, such as fesetenv and feholdexcept, need to use prctl with PR_GET_FPEXC and PR_SET_FPEXC anyway, and so it is natural to record the state of those bits during that call into the kernel and so avoid any need for a separate call into the kernel to inform it of a change to those bits. Thus, the interface I chose to use (in this patch and the glibc port) is that one of those prctl calls must be made after any userspace change to those sticky bits, other than through a floating-point operation that traps into the kernel anyway. feclearexcept and fesetexceptflag duly make those calls, which would not be required were it not for this issue. The previous EGLIBC port, and the uClibc code copied from it, is fundamentally broken as regards any use of prctl for floating-point exceptions because it didn't use the PR_FP_EXC_SW_ENABLE bit in its prctl calls (and did various worse things, such as passing a pointer when prctl expected an integer). If you avoid anything where prctl is used, the clearing of sticky bits still means it will never give anything approximating correct exception semantics with existing kernels. I don't believe the patch makes things any worse for existing code that doesn't try to inform the kernel of changes to sticky bits - such code may get incorrect exceptions in some cases, but it would have done so anyway in other cases. Signed-off-by: Joseph Myers <joseph@codesourcery.com> Signed-off-by: Scott Wood <scottwood@freescale.com>
2013-12-11 07:07:45 +08:00
if (cpu_has_feature(CPU_FTR_SPE)) {
/*
* When the sticky exception bits are set
* directly by userspace, it must call prctl
* with PR_GET_FPEXC (with PR_FP_EXC_SW_ENABLE
* in the existing prctl settings) or
* PR_SET_FPEXC (with PR_FP_EXC_SW_ENABLE in
* the bits being set). <fenv.h> functions
* saving and restoring the whole
* floating-point environment need to do so
* anyway to restore the prctl settings from
* the saved environment.
*/
#ifdef CONFIG_SPE
powerpc: fix exception clearing in e500 SPE float emulation The e500 SPE floating-point emulation code clears existing exceptions (__FPU_FPSCR &= ~FP_EX_MASK;) before ORing in the exceptions from the emulated operation. However, these exception bits are the "sticky", cumulative exception bits, and should only be cleared by the user program setting SPEFSCR, not implicitly by any floating-point instruction (whether executed purely by the hardware or emulated). The spurious clearing of these bits shows up as missing exceptions in glibc testing. Fixing this, however, is not as simple as just not clearing the bits, because while the bits may be from previous floating-point operations (in which case they should not be cleared), the processor can also set the sticky bits itself before the interrupt for an exception occurs, and this can happen in cases when IEEE 754 semantics are that the sticky bit should not be set. Specifically, the "invalid" sticky bit is set in various cases with non-finite operands, where IEEE 754 semantics do not involve raising such an exception, and the "underflow" sticky bit is set in cases of exact underflow, whereas IEEE 754 semantics are that this flag is set only for inexact underflow. Thus, for correct emulation the kernel needs to know the setting of these two sticky bits before the instruction being emulated. When a floating-point operation raises an exception, the kernel can note the state of the sticky bits immediately afterwards. Some <fenv.h> functions that affect the state of these bits, such as fesetenv and feholdexcept, need to use prctl with PR_GET_FPEXC and PR_SET_FPEXC anyway, and so it is natural to record the state of those bits during that call into the kernel and so avoid any need for a separate call into the kernel to inform it of a change to those bits. Thus, the interface I chose to use (in this patch and the glibc port) is that one of those prctl calls must be made after any userspace change to those sticky bits, other than through a floating-point operation that traps into the kernel anyway. feclearexcept and fesetexceptflag duly make those calls, which would not be required were it not for this issue. The previous EGLIBC port, and the uClibc code copied from it, is fundamentally broken as regards any use of prctl for floating-point exceptions because it didn't use the PR_FP_EXC_SW_ENABLE bit in its prctl calls (and did various worse things, such as passing a pointer when prctl expected an integer). If you avoid anything where prctl is used, the clearing of sticky bits still means it will never give anything approximating correct exception semantics with existing kernels. I don't believe the patch makes things any worse for existing code that doesn't try to inform the kernel of changes to sticky bits - such code may get incorrect exceptions in some cases, but it would have done so anyway in other cases. Signed-off-by: Joseph Myers <joseph@codesourcery.com> Signed-off-by: Scott Wood <scottwood@freescale.com>
2013-12-11 07:07:45 +08:00
tsk->thread.spefscr_last = mfspr(SPRN_SPEFSCR);
val = tsk->thread.fpexc_mode;
#endif
powerpc: fix exception clearing in e500 SPE float emulation The e500 SPE floating-point emulation code clears existing exceptions (__FPU_FPSCR &= ~FP_EX_MASK;) before ORing in the exceptions from the emulated operation. However, these exception bits are the "sticky", cumulative exception bits, and should only be cleared by the user program setting SPEFSCR, not implicitly by any floating-point instruction (whether executed purely by the hardware or emulated). The spurious clearing of these bits shows up as missing exceptions in glibc testing. Fixing this, however, is not as simple as just not clearing the bits, because while the bits may be from previous floating-point operations (in which case they should not be cleared), the processor can also set the sticky bits itself before the interrupt for an exception occurs, and this can happen in cases when IEEE 754 semantics are that the sticky bit should not be set. Specifically, the "invalid" sticky bit is set in various cases with non-finite operands, where IEEE 754 semantics do not involve raising such an exception, and the "underflow" sticky bit is set in cases of exact underflow, whereas IEEE 754 semantics are that this flag is set only for inexact underflow. Thus, for correct emulation the kernel needs to know the setting of these two sticky bits before the instruction being emulated. When a floating-point operation raises an exception, the kernel can note the state of the sticky bits immediately afterwards. Some <fenv.h> functions that affect the state of these bits, such as fesetenv and feholdexcept, need to use prctl with PR_GET_FPEXC and PR_SET_FPEXC anyway, and so it is natural to record the state of those bits during that call into the kernel and so avoid any need for a separate call into the kernel to inform it of a change to those bits. Thus, the interface I chose to use (in this patch and the glibc port) is that one of those prctl calls must be made after any userspace change to those sticky bits, other than through a floating-point operation that traps into the kernel anyway. feclearexcept and fesetexceptflag duly make those calls, which would not be required were it not for this issue. The previous EGLIBC port, and the uClibc code copied from it, is fundamentally broken as regards any use of prctl for floating-point exceptions because it didn't use the PR_FP_EXC_SW_ENABLE bit in its prctl calls (and did various worse things, such as passing a pointer when prctl expected an integer). If you avoid anything where prctl is used, the clearing of sticky bits still means it will never give anything approximating correct exception semantics with existing kernels. I don't believe the patch makes things any worse for existing code that doesn't try to inform the kernel of changes to sticky bits - such code may get incorrect exceptions in some cases, but it would have done so anyway in other cases. Signed-off-by: Joseph Myers <joseph@codesourcery.com> Signed-off-by: Scott Wood <scottwood@freescale.com>
2013-12-11 07:07:45 +08:00
} else
return -EINVAL;
} else {
val = __unpack_fe01(tsk->thread.fpexc_mode);
}
return put_user(val, (unsigned int __user *) adr);
}
int set_endian(struct task_struct *tsk, unsigned int val)
{
struct pt_regs *regs = tsk->thread.regs;
if ((val == PR_ENDIAN_LITTLE && !cpu_has_feature(CPU_FTR_REAL_LE)) ||
(val == PR_ENDIAN_PPC_LITTLE && !cpu_has_feature(CPU_FTR_PPC_LE)))
return -EINVAL;
if (regs == NULL)
return -EINVAL;
if (val == PR_ENDIAN_BIG)
regs->msr &= ~MSR_LE;
else if (val == PR_ENDIAN_LITTLE || val == PR_ENDIAN_PPC_LITTLE)
regs->msr |= MSR_LE;
else
return -EINVAL;
return 0;
}
int get_endian(struct task_struct *tsk, unsigned long adr)
{
struct pt_regs *regs = tsk->thread.regs;
unsigned int val;
if (!cpu_has_feature(CPU_FTR_PPC_LE) &&
!cpu_has_feature(CPU_FTR_REAL_LE))
return -EINVAL;
if (regs == NULL)
return -EINVAL;
if (regs->msr & MSR_LE) {
if (cpu_has_feature(CPU_FTR_REAL_LE))
val = PR_ENDIAN_LITTLE;
else
val = PR_ENDIAN_PPC_LITTLE;
} else
val = PR_ENDIAN_BIG;
return put_user(val, (unsigned int __user *)adr);
}
int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
{
tsk->thread.align_ctl = val;
return 0;
}
int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
{
return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
}
static inline int valid_irq_stack(unsigned long sp, struct task_struct *p,
unsigned long nbytes)
{
unsigned long stack_page;
unsigned long cpu = task_cpu(p);
stack_page = (unsigned long)hardirq_ctx[cpu];
if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
return 1;
stack_page = (unsigned long)softirq_ctx[cpu];
if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
return 1;
return 0;
}
powerpc/64: mark emergency stacks valid to unwind Before: WARNING: CPU: 0 PID: 494 at arch/powerpc/kernel/irq.c:343 CPU: 0 PID: 494 Comm: a Tainted: G W NIP: c00000000001ed2c LR: c000000000d13190 CTR: c00000000003f910 REGS: c0000001fffd3870 TRAP: 0700 Tainted: G W MSR: 8000000000021003 <SF,ME,RI,LE> CR: 28000488 XER: 00000000 CFAR: c00000000001ec90 IRQMASK: 0 GPR00: c000000000aeb12c c0000001fffd3b00 c0000000012ba300 0000000000000000 GPR04: 0000000000000000 0000000000000000 000000010bd207c8 6b00696e74657272 GPR08: 0000000000000000 0000000000000000 0000000000000000 efbeadde00000000 GPR12: 0000000000000000 c0000000014a0000 0000000000000000 0000000000000000 GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR24: 0000000000000000 0000000000000000 0000000000000000 000000010bd207bc GPR28: 0000000000000000 c00000000148a898 0000000000000000 c0000001ffff3f50 NIP [c00000000001ed2c] arch_local_irq_restore.part.0+0xac/0x100 LR [c000000000d13190] _raw_spin_unlock_irqrestore+0x50/0xc0 Call Trace: Instruction dump: 60000000 7d2000a6 71298000 41820068 39200002 7d210164 4bffff9c 60000000 60000000 7d2000a6 71298000 4c820020 <0fe00000> 4e800020 60000000 60000000 After: WARNING: CPU: 0 PID: 499 at arch/powerpc/kernel/irq.c:343 CPU: 0 PID: 499 Comm: a Not tainted NIP: c00000000001ed2c LR: c000000000d13210 CTR: c00000000003f980 REGS: c0000001fffd3870 TRAP: 0700 Not tainted MSR: 8000000000021003 <SF,ME,RI,LE> CR: 28000488 XER: 00000000 CFAR: c00000000001ec90 IRQMASK: 0 GPR00: c000000000aeb1ac c0000001fffd3b00 c0000000012ba300 0000000000000000 GPR04: 0000000000000000 0000000000000000 00000001347607c8 6b00696e74657272 GPR08: 0000000000000000 0000000000000000 0000000000000000 efbeadde00000000 GPR12: 0000000000000000 c0000000014a0000 0000000000000000 0000000000000000 GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR24: 0000000000000000 0000000000000000 0000000000000000 00000001347607bc GPR28: 0000000000000000 c00000000148a898 0000000000000000 c0000001ffff3f50 NIP [c00000000001ed2c] arch_local_irq_restore.part.0+0xac/0x100 LR [c000000000d13210] _raw_spin_unlock_irqrestore+0x50/0xc0 Call Trace: [c0000001fffd3b20] [c000000000aeb1ac] of_find_property+0x6c/0x90 [c0000001fffd3b70] [c000000000aeb1f0] of_get_property+0x20/0x40 [c0000001fffd3b90] [c000000000042cdc] rtas_token+0x3c/0x70 [c0000001fffd3bb0] [c0000000000dc318] fwnmi_release_errinfo+0x28/0x70 [c0000001fffd3c10] [c0000000000dcd8c] pseries_machine_check_realmode+0x1dc/0x540 [c0000001fffd3cd0] [c00000000003fe04] machine_check_early+0x54/0x70 [c0000001fffd3d00] [c000000000008384] machine_check_early_common+0x134/0x1f0 --- interrupt: 200 at 0x1347607c8 LR = 0x7fffafbd8328 Instruction dump: 60000000 7d2000a6 71298000 41820068 39200002 7d210164 4bffff9c 60000000 60000000 7d2000a6 71298000 4c820020 <0fe00000> 4e800020 60000000 60000000 Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200325104144.158362-1-npiggin@gmail.com
2020-03-25 18:41:44 +08:00
static inline int valid_emergency_stack(unsigned long sp, struct task_struct *p,
unsigned long nbytes)
{
#ifdef CONFIG_PPC64
unsigned long stack_page;
unsigned long cpu = task_cpu(p);
if (!paca_ptrs)
return 0;
powerpc/64: mark emergency stacks valid to unwind Before: WARNING: CPU: 0 PID: 494 at arch/powerpc/kernel/irq.c:343 CPU: 0 PID: 494 Comm: a Tainted: G W NIP: c00000000001ed2c LR: c000000000d13190 CTR: c00000000003f910 REGS: c0000001fffd3870 TRAP: 0700 Tainted: G W MSR: 8000000000021003 <SF,ME,RI,LE> CR: 28000488 XER: 00000000 CFAR: c00000000001ec90 IRQMASK: 0 GPR00: c000000000aeb12c c0000001fffd3b00 c0000000012ba300 0000000000000000 GPR04: 0000000000000000 0000000000000000 000000010bd207c8 6b00696e74657272 GPR08: 0000000000000000 0000000000000000 0000000000000000 efbeadde00000000 GPR12: 0000000000000000 c0000000014a0000 0000000000000000 0000000000000000 GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR24: 0000000000000000 0000000000000000 0000000000000000 000000010bd207bc GPR28: 0000000000000000 c00000000148a898 0000000000000000 c0000001ffff3f50 NIP [c00000000001ed2c] arch_local_irq_restore.part.0+0xac/0x100 LR [c000000000d13190] _raw_spin_unlock_irqrestore+0x50/0xc0 Call Trace: Instruction dump: 60000000 7d2000a6 71298000 41820068 39200002 7d210164 4bffff9c 60000000 60000000 7d2000a6 71298000 4c820020 <0fe00000> 4e800020 60000000 60000000 After: WARNING: CPU: 0 PID: 499 at arch/powerpc/kernel/irq.c:343 CPU: 0 PID: 499 Comm: a Not tainted NIP: c00000000001ed2c LR: c000000000d13210 CTR: c00000000003f980 REGS: c0000001fffd3870 TRAP: 0700 Not tainted MSR: 8000000000021003 <SF,ME,RI,LE> CR: 28000488 XER: 00000000 CFAR: c00000000001ec90 IRQMASK: 0 GPR00: c000000000aeb1ac c0000001fffd3b00 c0000000012ba300 0000000000000000 GPR04: 0000000000000000 0000000000000000 00000001347607c8 6b00696e74657272 GPR08: 0000000000000000 0000000000000000 0000000000000000 efbeadde00000000 GPR12: 0000000000000000 c0000000014a0000 0000000000000000 0000000000000000 GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR24: 0000000000000000 0000000000000000 0000000000000000 00000001347607bc GPR28: 0000000000000000 c00000000148a898 0000000000000000 c0000001ffff3f50 NIP [c00000000001ed2c] arch_local_irq_restore.part.0+0xac/0x100 LR [c000000000d13210] _raw_spin_unlock_irqrestore+0x50/0xc0 Call Trace: [c0000001fffd3b20] [c000000000aeb1ac] of_find_property+0x6c/0x90 [c0000001fffd3b70] [c000000000aeb1f0] of_get_property+0x20/0x40 [c0000001fffd3b90] [c000000000042cdc] rtas_token+0x3c/0x70 [c0000001fffd3bb0] [c0000000000dc318] fwnmi_release_errinfo+0x28/0x70 [c0000001fffd3c10] [c0000000000dcd8c] pseries_machine_check_realmode+0x1dc/0x540 [c0000001fffd3cd0] [c00000000003fe04] machine_check_early+0x54/0x70 [c0000001fffd3d00] [c000000000008384] machine_check_early_common+0x134/0x1f0 --- interrupt: 200 at 0x1347607c8 LR = 0x7fffafbd8328 Instruction dump: 60000000 7d2000a6 71298000 41820068 39200002 7d210164 4bffff9c 60000000 60000000 7d2000a6 71298000 4c820020 <0fe00000> 4e800020 60000000 60000000 Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200325104144.158362-1-npiggin@gmail.com
2020-03-25 18:41:44 +08:00
stack_page = (unsigned long)paca_ptrs[cpu]->emergency_sp - THREAD_SIZE;
if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
return 1;
# ifdef CONFIG_PPC_BOOK3S_64
stack_page = (unsigned long)paca_ptrs[cpu]->nmi_emergency_sp - THREAD_SIZE;
if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
return 1;
stack_page = (unsigned long)paca_ptrs[cpu]->mc_emergency_sp - THREAD_SIZE;
if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
return 1;
# endif
#endif
return 0;
}
int validate_sp(unsigned long sp, struct task_struct *p,
unsigned long nbytes)
{
unsigned long stack_page = (unsigned long)task_stack_page(p);
if (sp < THREAD_SIZE)
return 0;
if (sp >= stack_page && sp <= stack_page + THREAD_SIZE - nbytes)
return 1;
powerpc/64: mark emergency stacks valid to unwind Before: WARNING: CPU: 0 PID: 494 at arch/powerpc/kernel/irq.c:343 CPU: 0 PID: 494 Comm: a Tainted: G W NIP: c00000000001ed2c LR: c000000000d13190 CTR: c00000000003f910 REGS: c0000001fffd3870 TRAP: 0700 Tainted: G W MSR: 8000000000021003 <SF,ME,RI,LE> CR: 28000488 XER: 00000000 CFAR: c00000000001ec90 IRQMASK: 0 GPR00: c000000000aeb12c c0000001fffd3b00 c0000000012ba300 0000000000000000 GPR04: 0000000000000000 0000000000000000 000000010bd207c8 6b00696e74657272 GPR08: 0000000000000000 0000000000000000 0000000000000000 efbeadde00000000 GPR12: 0000000000000000 c0000000014a0000 0000000000000000 0000000000000000 GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR24: 0000000000000000 0000000000000000 0000000000000000 000000010bd207bc GPR28: 0000000000000000 c00000000148a898 0000000000000000 c0000001ffff3f50 NIP [c00000000001ed2c] arch_local_irq_restore.part.0+0xac/0x100 LR [c000000000d13190] _raw_spin_unlock_irqrestore+0x50/0xc0 Call Trace: Instruction dump: 60000000 7d2000a6 71298000 41820068 39200002 7d210164 4bffff9c 60000000 60000000 7d2000a6 71298000 4c820020 <0fe00000> 4e800020 60000000 60000000 After: WARNING: CPU: 0 PID: 499 at arch/powerpc/kernel/irq.c:343 CPU: 0 PID: 499 Comm: a Not tainted NIP: c00000000001ed2c LR: c000000000d13210 CTR: c00000000003f980 REGS: c0000001fffd3870 TRAP: 0700 Not tainted MSR: 8000000000021003 <SF,ME,RI,LE> CR: 28000488 XER: 00000000 CFAR: c00000000001ec90 IRQMASK: 0 GPR00: c000000000aeb1ac c0000001fffd3b00 c0000000012ba300 0000000000000000 GPR04: 0000000000000000 0000000000000000 00000001347607c8 6b00696e74657272 GPR08: 0000000000000000 0000000000000000 0000000000000000 efbeadde00000000 GPR12: 0000000000000000 c0000000014a0000 0000000000000000 0000000000000000 GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000000 GPR24: 0000000000000000 0000000000000000 0000000000000000 00000001347607bc GPR28: 0000000000000000 c00000000148a898 0000000000000000 c0000001ffff3f50 NIP [c00000000001ed2c] arch_local_irq_restore.part.0+0xac/0x100 LR [c000000000d13210] _raw_spin_unlock_irqrestore+0x50/0xc0 Call Trace: [c0000001fffd3b20] [c000000000aeb1ac] of_find_property+0x6c/0x90 [c0000001fffd3b70] [c000000000aeb1f0] of_get_property+0x20/0x40 [c0000001fffd3b90] [c000000000042cdc] rtas_token+0x3c/0x70 [c0000001fffd3bb0] [c0000000000dc318] fwnmi_release_errinfo+0x28/0x70 [c0000001fffd3c10] [c0000000000dcd8c] pseries_machine_check_realmode+0x1dc/0x540 [c0000001fffd3cd0] [c00000000003fe04] machine_check_early+0x54/0x70 [c0000001fffd3d00] [c000000000008384] machine_check_early_common+0x134/0x1f0 --- interrupt: 200 at 0x1347607c8 LR = 0x7fffafbd8328 Instruction dump: 60000000 7d2000a6 71298000 41820068 39200002 7d210164 4bffff9c 60000000 60000000 7d2000a6 71298000 4c820020 <0fe00000> 4e800020 60000000 60000000 Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20200325104144.158362-1-npiggin@gmail.com
2020-03-25 18:41:44 +08:00
if (valid_irq_stack(sp, p, nbytes))
return 1;
return valid_emergency_stack(sp, p, nbytes);
}
EXPORT_SYMBOL(validate_sp);
static unsigned long __get_wchan(struct task_struct *p)
{
unsigned long ip, sp;
int count = 0;
if (!p || p == current || p->state == TASK_RUNNING)
return 0;
sp = p->thread.ksp;
if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD))
return 0;
do {
sp = *(unsigned long *)sp;
if (!validate_sp(sp, p, STACK_FRAME_OVERHEAD) ||
p->state == TASK_RUNNING)
return 0;
if (count > 0) {
ip = ((unsigned long *)sp)[STACK_FRAME_LR_SAVE];
if (!in_sched_functions(ip))
return ip;
}
} while (count++ < 16);
return 0;
}
unsigned long get_wchan(struct task_struct *p)
{
unsigned long ret;
if (!try_get_task_stack(p))
return 0;
ret = __get_wchan(p);
put_task_stack(p);
return ret;
}
static int kstack_depth_to_print = CONFIG_PRINT_STACK_DEPTH;
void __no_sanitize_address show_stack(struct task_struct *tsk,
unsigned long *stack,
const char *loglvl)
{
unsigned long sp, ip, lr, newsp;
int count = 0;
int firstframe = 1;
powerpc: Use ftrace_graph_ret_addr() when unwinding With support for HAVE_FUNCTION_GRAPH_RET_ADDR_PTR, ftrace_graph_ret_addr() provides more robust unwinding when function graph is in use. Update show_stack() to use the same. With dump_stack() added to sysrq_sysctl_handler(), before this patch: root@(none):/sys/kernel/debug/tracing# cat /proc/sys/kernel/sysrq CPU: 0 PID: 218 Comm: cat Not tainted 5.3.0-rc7-00868-g8453ad4a078c-dirty #20 Call Trace: [c0000000d1e13c30] [c00000000006ab98] return_to_handler+0x0/0x40 (dump_stack+0xe8/0x164) (unreliable) [c0000000d1e13c80] [c000000000145680] sysrq_sysctl_handler+0x48/0xb8 [c0000000d1e13cd0] [c00000000006ab98] return_to_handler+0x0/0x40 (proc_sys_call_handler+0x274/0x2a0) [c0000000d1e13d60] [c00000000006ab98] return_to_handler+0x0/0x40 (return_to_handler+0x0/0x40) [c0000000d1e13d80] [c00000000006ab98] return_to_handler+0x0/0x40 (__vfs_read+0x3c/0x70) [c0000000d1e13dd0] [c00000000006ab98] return_to_handler+0x0/0x40 (vfs_read+0xb8/0x1b0) [c0000000d1e13e20] [c00000000006ab98] return_to_handler+0x0/0x40 (ksys_read+0x7c/0x140) After this patch: Call Trace: [c0000000d1e33c30] [c00000000006ab58] return_to_handler+0x0/0x40 (dump_stack+0xe8/0x164) (unreliable) [c0000000d1e33c80] [c000000000145680] sysrq_sysctl_handler+0x48/0xb8 [c0000000d1e33cd0] [c00000000006ab58] return_to_handler+0x0/0x40 (proc_sys_call_handler+0x274/0x2a0) [c0000000d1e33d60] [c00000000006ab58] return_to_handler+0x0/0x40 (__vfs_read+0x3c/0x70) [c0000000d1e33d80] [c00000000006ab58] return_to_handler+0x0/0x40 (vfs_read+0xb8/0x1b0) [c0000000d1e33dd0] [c00000000006ab58] return_to_handler+0x0/0x40 (ksys_read+0x7c/0x140) [c0000000d1e33e20] [c00000000006ab58] return_to_handler+0x0/0x40 (system_call+0x5c/0x68) Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/dc89c9a887121342d9c7819482c3dabdece2a323.1567707399.git.naveen.n.rao@linux.vnet.ibm.com
2019-09-06 02:20:30 +08:00
unsigned long ret_addr;
int ftrace_idx = 0;
if (tsk == NULL)
tsk = current;
if (!try_get_task_stack(tsk))
return;
sp = (unsigned long) stack;
if (sp == 0) {
if (tsk == current)
sp = current_stack_frame();
else
sp = tsk->thread.ksp;
}
lr = 0;
printk("%sCall Trace:\n", loglvl);
do {
if (!validate_sp(sp, tsk, STACK_FRAME_OVERHEAD))
break;
stack = (unsigned long *) sp;
newsp = stack[0];
ip = stack[STACK_FRAME_LR_SAVE];
if (!firstframe || ip != lr) {
printk("%s["REG"] ["REG"] %pS",
loglvl, sp, ip, (void *)ip);
powerpc: Use ftrace_graph_ret_addr() when unwinding With support for HAVE_FUNCTION_GRAPH_RET_ADDR_PTR, ftrace_graph_ret_addr() provides more robust unwinding when function graph is in use. Update show_stack() to use the same. With dump_stack() added to sysrq_sysctl_handler(), before this patch: root@(none):/sys/kernel/debug/tracing# cat /proc/sys/kernel/sysrq CPU: 0 PID: 218 Comm: cat Not tainted 5.3.0-rc7-00868-g8453ad4a078c-dirty #20 Call Trace: [c0000000d1e13c30] [c00000000006ab98] return_to_handler+0x0/0x40 (dump_stack+0xe8/0x164) (unreliable) [c0000000d1e13c80] [c000000000145680] sysrq_sysctl_handler+0x48/0xb8 [c0000000d1e13cd0] [c00000000006ab98] return_to_handler+0x0/0x40 (proc_sys_call_handler+0x274/0x2a0) [c0000000d1e13d60] [c00000000006ab98] return_to_handler+0x0/0x40 (return_to_handler+0x0/0x40) [c0000000d1e13d80] [c00000000006ab98] return_to_handler+0x0/0x40 (__vfs_read+0x3c/0x70) [c0000000d1e13dd0] [c00000000006ab98] return_to_handler+0x0/0x40 (vfs_read+0xb8/0x1b0) [c0000000d1e13e20] [c00000000006ab98] return_to_handler+0x0/0x40 (ksys_read+0x7c/0x140) After this patch: Call Trace: [c0000000d1e33c30] [c00000000006ab58] return_to_handler+0x0/0x40 (dump_stack+0xe8/0x164) (unreliable) [c0000000d1e33c80] [c000000000145680] sysrq_sysctl_handler+0x48/0xb8 [c0000000d1e33cd0] [c00000000006ab58] return_to_handler+0x0/0x40 (proc_sys_call_handler+0x274/0x2a0) [c0000000d1e33d60] [c00000000006ab58] return_to_handler+0x0/0x40 (__vfs_read+0x3c/0x70) [c0000000d1e33d80] [c00000000006ab58] return_to_handler+0x0/0x40 (vfs_read+0xb8/0x1b0) [c0000000d1e33dd0] [c00000000006ab58] return_to_handler+0x0/0x40 (ksys_read+0x7c/0x140) [c0000000d1e33e20] [c00000000006ab58] return_to_handler+0x0/0x40 (system_call+0x5c/0x68) Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/dc89c9a887121342d9c7819482c3dabdece2a323.1567707399.git.naveen.n.rao@linux.vnet.ibm.com
2019-09-06 02:20:30 +08:00
ret_addr = ftrace_graph_ret_addr(current,
&ftrace_idx, ip, stack);
if (ret_addr != ip)
pr_cont(" (%pS)", (void *)ret_addr);
if (firstframe)
pr_cont(" (unreliable)");
pr_cont("\n");
}
firstframe = 0;
/*
* See if this is an exception frame.
* We look for the "regshere" marker in the current frame.
*/
powerpc/64: Fix stack trace not displaying final frame In commit bf13718bc57a ("powerpc: show registers when unwinding interrupt frames") we changed our stack dumping logic to show the full registers whenever we find an interrupt frame on the stack. However we didn't notice that on 64-bit this doesn't show the final frame, ie. the interrupt that brought us in from userspace, whereas on 32-bit it does. That is due to confusion about the size of that last frame. The code in show_stack() calls validate_sp(), passing it STACK_INT_FRAME_SIZE to check the sp is at least that far below the top of the stack. However on 64-bit that size is too large for the final frame, because it includes the red zone, but we don't allocate a red zone for the first frame. So add a new define that encodes the correct size for 32-bit and 64-bit, and use it in show_stack(). This results in the full trace being shown on 64-bit, eg: sysrq: Trigger a crash Kernel panic - not syncing: sysrq triggered crash CPU: 0 PID: 83 Comm: sh Not tainted 5.11.0-rc2-gcc-8.2.0-00188-g571abcb96b10-dirty #649 Call Trace: [c00000000a1c3ac0] [c000000000897b70] dump_stack+0xc4/0x114 (unreliable) [c00000000a1c3b00] [c00000000014334c] panic+0x178/0x41c [c00000000a1c3ba0] [c00000000094e600] sysrq_handle_crash+0x40/0x50 [c00000000a1c3c00] [c00000000094ef98] __handle_sysrq+0xd8/0x210 [c00000000a1c3ca0] [c00000000094f820] write_sysrq_trigger+0x100/0x188 [c00000000a1c3ce0] [c0000000005559dc] proc_reg_write+0x10c/0x1b0 [c00000000a1c3d10] [c000000000479950] vfs_write+0xf0/0x360 [c00000000a1c3d60] [c000000000479d9c] ksys_write+0x7c/0x140 [c00000000a1c3db0] [c00000000002bf5c] system_call_exception+0x19c/0x2c0 [c00000000a1c3e10] [c00000000000d35c] system_call_common+0xec/0x278 --- interrupt: c00 at 0x7fff9fbab428 NIP: 00007fff9fbab428 LR: 000000001000b724 CTR: 0000000000000000 REGS: c00000000a1c3e80 TRAP: 0c00 Not tainted (5.11.0-rc2-gcc-8.2.0-00188-g571abcb96b10-dirty) MSR: 900000000280f033 <SF,HV,VEC,VSX,EE,PR,FP,ME,IR,DR,RI,LE> CR: 22002884 XER: 00000000 IRQMASK: 0 GPR00: 0000000000000004 00007fffc3cb8960 00007fff9fc59900 0000000000000001 GPR04: 000000002a4b32d0 0000000000000002 0000000000000063 0000000000000063 GPR08: 000000002a4b32d0 0000000000000000 0000000000000000 0000000000000000 GPR12: 0000000000000000 00007fff9fcca9a0 0000000000000000 0000000000000000 GPR16: 0000000000000000 0000000000000000 0000000000000000 00000000100b8fd0 GPR20: 000000002a4b3485 00000000100b8f90 0000000000000000 0000000000000000 GPR24: 000000002a4b0440 00000000100e77b8 0000000000000020 000000002a4b32d0 GPR28: 0000000000000001 0000000000000002 000000002a4b32d0 0000000000000001 NIP [00007fff9fbab428] 0x7fff9fbab428 LR [000000001000b724] 0x1000b724 --- interrupt: c00 Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20210209141627.2898485-1-mpe@ellerman.id.au
2021-02-09 21:59:20 +08:00
if (validate_sp(sp, tsk, STACK_FRAME_WITH_PT_REGS)
&& stack[STACK_FRAME_MARKER] == STACK_FRAME_REGS_MARKER) {
struct pt_regs *regs = (struct pt_regs *)
(sp + STACK_FRAME_OVERHEAD);
powerpc: show registers when unwinding interrupt frames It's often useful to know the register state for interrupts in the stack frame. In the below example (with this patch applied), the important information is the state of the page fault. A blatant case like this probably rather should have the page fault regs passed down to the warning, but quite often there are less obvious cases where an interrupt shows up that might give some more clues. The downside is longer and more complex bug output. Bug: Write fault blocked by AMR! WARNING: CPU: 0 PID: 72 at arch/powerpc/include/asm/book3s/64/kup-radix.h:164 __do_page_fault+0x880/0xa90 Modules linked in: CPU: 0 PID: 72 Comm: systemd-gpt-aut Not tainted NIP: c00000000006e2f0 LR: c00000000006e2ec CTR: 0000000000000000 REGS: c00000000a4f3420 TRAP: 0700 MSR: 8000000000021033 <SF,ME,IR,DR,RI,LE> CR: 28002840 XER: 20040000 CFAR: c000000000128be0 IRQMASK: 3 GPR00: c00000000006e2ec c00000000a4f36c0 c0000000014f0700 0000000000000020 GPR04: 0000000000000001 c000000001290f50 0000000000000001 c000000001290f80 GPR08: c000000001612b08 0000000000000000 0000000000000000 00000000ffffe0f7 GPR12: 0000000048002840 c0000000016e0000 c00c000000021c80 c000000000fd6f60 GPR16: 0000000000000000 c00000000a104698 0000000000000003 c0000000087f0000 GPR20: 0000000000000100 c0000000070330b8 0000000000000000 0000000000000004 GPR24: 0000000002000000 0000000000000300 0000000002000000 c00000000a5b0c00 GPR28: 0000000000000000 000000000a000000 00007fffb2a90038 c00000000a4f3820 NIP [c00000000006e2f0] __do_page_fault+0x880/0xa90 LR [c00000000006e2ec] __do_page_fault+0x87c/0xa90 Call Trace: [c00000000a4f36c0] [c00000000006e2ec] __do_page_fault+0x87c/0xa90 (unreliable) [c00000000a4f3780] [c000000000e1c034] do_page_fault+0x34/0x90 [c00000000a4f37b0] [c000000000008908] data_access_common_virt+0x158/0x1b0 --- interrupt: 300 at __copy_tofrom_user_base+0x9c/0x5a4 NIP: c00000000009b028 LR: c000000000802978 CTR: 0000000000000800 REGS: c00000000a4f3820 TRAP: 0300 MSR: 800000000280b033 <SF,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 24004840 XER: 00000000 CFAR: c00000000009aff4 DAR: 00007fffb2a90038 DSISR: 0a000000 IRQMASK: 0 GPR00: 0000000000000000 c00000000a4f3ac0 c0000000014f0700 00007fffb2a90028 GPR04: c000000008720010 0000000000010000 0000000000000000 0000000000000000 GPR08: 0000000000000000 0000000000000000 0000000000000000 0000000000000001 GPR12: 0000000000004000 c0000000016e0000 c00c000000021c80 c000000000fd6f60 GPR16: 0000000000000000 c00000000a104698 0000000000000003 c0000000087f0000 GPR20: 0000000000000100 c0000000070330b8 0000000000000000 0000000000000004 GPR24: c00000000a4f3c80 c000000008720000 0000000000010000 0000000000000000 GPR28: 0000000000010000 0000000008720000 0000000000010000 c000000001515b98 NIP [c00000000009b028] __copy_tofrom_user_base+0x9c/0x5a4 LR [c000000000802978] copyout+0x68/0xc0 --- interrupt: 300 [c00000000a4f3af0] [c0000000008074b8] copy_page_to_iter+0x188/0x540 [c00000000a4f3b50] [c00000000035c678] generic_file_buffered_read+0x358/0xd80 [c00000000a4f3c40] [c0000000004c1e90] blkdev_read_iter+0x50/0x80 [c00000000a4f3c60] [c00000000045733c] new_sync_read+0x12c/0x1c0 [c00000000a4f3d00] [c00000000045a1f0] vfs_read+0x1d0/0x240 [c00000000a4f3d50] [c00000000045a7f4] ksys_read+0x84/0x140 [c00000000a4f3da0] [c000000000033a60] system_call_exception+0x100/0x280 [c00000000a4f3e10] [c00000000000c508] system_call_common+0xf8/0x2f8 Instruction dump: eae10078 3be0000b 4bfff890 60420000 792917e1 4182ff18 3c82ffab 3884a5e0 3c62ffab 3863a6e8 480ba891 60000000 <0fe00000> 3be0000b 4bfff860 e93c0938 Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20201107023305.2384874-1-npiggin@gmail.com
2020-11-07 10:33:05 +08:00
lr = regs->link;
powerpc: show registers when unwinding interrupt frames It's often useful to know the register state for interrupts in the stack frame. In the below example (with this patch applied), the important information is the state of the page fault. A blatant case like this probably rather should have the page fault regs passed down to the warning, but quite often there are less obvious cases where an interrupt shows up that might give some more clues. The downside is longer and more complex bug output. Bug: Write fault blocked by AMR! WARNING: CPU: 0 PID: 72 at arch/powerpc/include/asm/book3s/64/kup-radix.h:164 __do_page_fault+0x880/0xa90 Modules linked in: CPU: 0 PID: 72 Comm: systemd-gpt-aut Not tainted NIP: c00000000006e2f0 LR: c00000000006e2ec CTR: 0000000000000000 REGS: c00000000a4f3420 TRAP: 0700 MSR: 8000000000021033 <SF,ME,IR,DR,RI,LE> CR: 28002840 XER: 20040000 CFAR: c000000000128be0 IRQMASK: 3 GPR00: c00000000006e2ec c00000000a4f36c0 c0000000014f0700 0000000000000020 GPR04: 0000000000000001 c000000001290f50 0000000000000001 c000000001290f80 GPR08: c000000001612b08 0000000000000000 0000000000000000 00000000ffffe0f7 GPR12: 0000000048002840 c0000000016e0000 c00c000000021c80 c000000000fd6f60 GPR16: 0000000000000000 c00000000a104698 0000000000000003 c0000000087f0000 GPR20: 0000000000000100 c0000000070330b8 0000000000000000 0000000000000004 GPR24: 0000000002000000 0000000000000300 0000000002000000 c00000000a5b0c00 GPR28: 0000000000000000 000000000a000000 00007fffb2a90038 c00000000a4f3820 NIP [c00000000006e2f0] __do_page_fault+0x880/0xa90 LR [c00000000006e2ec] __do_page_fault+0x87c/0xa90 Call Trace: [c00000000a4f36c0] [c00000000006e2ec] __do_page_fault+0x87c/0xa90 (unreliable) [c00000000a4f3780] [c000000000e1c034] do_page_fault+0x34/0x90 [c00000000a4f37b0] [c000000000008908] data_access_common_virt+0x158/0x1b0 --- interrupt: 300 at __copy_tofrom_user_base+0x9c/0x5a4 NIP: c00000000009b028 LR: c000000000802978 CTR: 0000000000000800 REGS: c00000000a4f3820 TRAP: 0300 MSR: 800000000280b033 <SF,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 24004840 XER: 00000000 CFAR: c00000000009aff4 DAR: 00007fffb2a90038 DSISR: 0a000000 IRQMASK: 0 GPR00: 0000000000000000 c00000000a4f3ac0 c0000000014f0700 00007fffb2a90028 GPR04: c000000008720010 0000000000010000 0000000000000000 0000000000000000 GPR08: 0000000000000000 0000000000000000 0000000000000000 0000000000000001 GPR12: 0000000000004000 c0000000016e0000 c00c000000021c80 c000000000fd6f60 GPR16: 0000000000000000 c00000000a104698 0000000000000003 c0000000087f0000 GPR20: 0000000000000100 c0000000070330b8 0000000000000000 0000000000000004 GPR24: c00000000a4f3c80 c000000008720000 0000000000010000 0000000000000000 GPR28: 0000000000010000 0000000008720000 0000000000010000 c000000001515b98 NIP [c00000000009b028] __copy_tofrom_user_base+0x9c/0x5a4 LR [c000000000802978] copyout+0x68/0xc0 --- interrupt: 300 [c00000000a4f3af0] [c0000000008074b8] copy_page_to_iter+0x188/0x540 [c00000000a4f3b50] [c00000000035c678] generic_file_buffered_read+0x358/0xd80 [c00000000a4f3c40] [c0000000004c1e90] blkdev_read_iter+0x50/0x80 [c00000000a4f3c60] [c00000000045733c] new_sync_read+0x12c/0x1c0 [c00000000a4f3d00] [c00000000045a1f0] vfs_read+0x1d0/0x240 [c00000000a4f3d50] [c00000000045a7f4] ksys_read+0x84/0x140 [c00000000a4f3da0] [c000000000033a60] system_call_exception+0x100/0x280 [c00000000a4f3e10] [c00000000000c508] system_call_common+0xf8/0x2f8 Instruction dump: eae10078 3be0000b 4bfff890 60420000 792917e1 4182ff18 3c82ffab 3884a5e0 3c62ffab 3863a6e8 480ba891 60000000 <0fe00000> 3be0000b 4bfff860 e93c0938 Signed-off-by: Nicholas Piggin <npiggin@gmail.com> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au> Link: https://lore.kernel.org/r/20201107023305.2384874-1-npiggin@gmail.com
2020-11-07 10:33:05 +08:00
printk("%s--- interrupt: %lx at %pS\n",
loglvl, regs->trap, (void *)regs->nip);
__show_regs(regs);
printk("%s--- interrupt: %lx\n",
loglvl, regs->trap);
firstframe = 1;
}
sp = newsp;
} while (count++ < kstack_depth_to_print);
put_task_stack(tsk);
}
#ifdef CONFIG_PPC64
/* Called with hard IRQs off */
powerpc: Fix stack overflow crash in resume_kernel when ftracing It's possible for us to crash when running with ftrace enabled, eg: Bad kernel stack pointer bffffd12 at c00000000000a454 cpu 0x3: Vector: 300 (Data Access) at [c00000000ffe3d40] pc: c00000000000a454: resume_kernel+0x34/0x60 lr: c00000000000335c: performance_monitor_common+0x15c/0x180 sp: bffffd12 msr: 8000000000001032 dar: bffffd12 dsisr: 42000000 If we look at current's stack (paca->__current->stack) we see it is equal to c0000002ecab0000. Our stack is 16K, and comparing to paca->kstack (c0000002ecab3e30) we can see that we have overflowed our kernel stack. This leads to us writing over our struct thread_info, and in this case we have corrupted thread_info->flags and set _TIF_EMULATE_STACK_STORE. Dumping the stack we see: 3:mon> t c0000002ecab0000 [c0000002ecab0000] c00000000002131c .performance_monitor_exception+0x5c/0x70 [c0000002ecab0080] c00000000000335c performance_monitor_common+0x15c/0x180 --- Exception: f01 (Performance Monitor) at c0000000000fb2ec .trace_hardirqs_off+0x1c/0x30 [c0000002ecab0370] c00000000016fdb0 .trace_graph_entry+0xb0/0x280 (unreliable) [c0000002ecab0410] c00000000003d038 .prepare_ftrace_return+0x98/0x130 [c0000002ecab04b0] c00000000000a920 .ftrace_graph_caller+0x14/0x28 [c0000002ecab0520] c0000000000d6b58 .idle_cpu+0x18/0x90 [c0000002ecab05a0] c00000000000a934 .return_to_handler+0x0/0x34 [c0000002ecab0620] c00000000001e660 .timer_interrupt+0x160/0x300 [c0000002ecab06d0] c0000000000025dc decrementer_common+0x15c/0x180 --- Exception: 901 (Decrementer) at c0000000000104d4 .arch_local_irq_restore+0x74/0xa0 [c0000002ecab09c0] c0000000000fe044 .trace_hardirqs_on+0x14/0x30 (unreliable) [c0000002ecab0fb0] c00000000016fe3c .trace_graph_entry+0x13c/0x280 [c0000002ecab1050] c00000000003d038 .prepare_ftrace_return+0x98/0x130 [c0000002ecab10f0] c00000000000a920 .ftrace_graph_caller+0x14/0x28 [c0000002ecab1160] c0000000000161f0 .__ppc64_runlatch_on+0x10/0x40 [c0000002ecab11d0] c00000000000a934 .return_to_handler+0x0/0x34 --- Exception: 901 (Decrementer) at c0000000000104d4 .arch_local_irq_restore+0x74/0xa0 ... and so on __ppc64_runlatch_on() is called from RUNLATCH_ON in the exception entry path. At that point the irq state is not consistent, ie. interrupts are hard disabled (by the exception entry), but the paca soft-enabled flag may be out of sync. This leads to the local_irq_restore() in trace_graph_entry() actually enabling interrupts, which we do not want. Because we have not yet reprogrammed the decrementer we immediately take another decrementer exception, and recurse. The fix is twofold. Firstly make sure we call DISABLE_INTS before calling RUNLATCH_ON. The badly named DISABLE_INTS actually reconciles the irq state in the paca with the hardware, making it safe again to call local_irq_save/restore(). Although that should be sufficient to fix the bug, we also mark the runlatch routines as notrace. They are called very early in the exception entry and we are asking for trouble tracing them. They are also fairly uninteresting and tracing them just adds unnecessary overhead. [ This regression was introduced by fe1952fc0afb9a2e4c79f103c08aef5d13db1873 "powerpc: Rework runlatch code" by myself --BenH ] CC: <stable@vger.kernel.org> [v3.4+] Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-06-13 19:04:56 +08:00
void notrace __ppc64_runlatch_on(void)
{
struct thread_info *ti = current_thread_info();
if (cpu_has_feature(CPU_FTR_ARCH_206)) {
/*
* Least significant bit (RUN) is the only writable bit of
* the CTRL register, so we can avoid mfspr. 2.06 is not the
* earliest ISA where this is the case, but it's convenient.
*/
mtspr(SPRN_CTRLT, CTRL_RUNLATCH);
} else {
unsigned long ctrl;
/*
* Some architectures (e.g., Cell) have writable fields other
* than RUN, so do the read-modify-write.
*/
ctrl = mfspr(SPRN_CTRLF);
ctrl |= CTRL_RUNLATCH;
mtspr(SPRN_CTRLT, ctrl);
}
ti->local_flags |= _TLF_RUNLATCH;
}
/* Called with hard IRQs off */
powerpc: Fix stack overflow crash in resume_kernel when ftracing It's possible for us to crash when running with ftrace enabled, eg: Bad kernel stack pointer bffffd12 at c00000000000a454 cpu 0x3: Vector: 300 (Data Access) at [c00000000ffe3d40] pc: c00000000000a454: resume_kernel+0x34/0x60 lr: c00000000000335c: performance_monitor_common+0x15c/0x180 sp: bffffd12 msr: 8000000000001032 dar: bffffd12 dsisr: 42000000 If we look at current's stack (paca->__current->stack) we see it is equal to c0000002ecab0000. Our stack is 16K, and comparing to paca->kstack (c0000002ecab3e30) we can see that we have overflowed our kernel stack. This leads to us writing over our struct thread_info, and in this case we have corrupted thread_info->flags and set _TIF_EMULATE_STACK_STORE. Dumping the stack we see: 3:mon> t c0000002ecab0000 [c0000002ecab0000] c00000000002131c .performance_monitor_exception+0x5c/0x70 [c0000002ecab0080] c00000000000335c performance_monitor_common+0x15c/0x180 --- Exception: f01 (Performance Monitor) at c0000000000fb2ec .trace_hardirqs_off+0x1c/0x30 [c0000002ecab0370] c00000000016fdb0 .trace_graph_entry+0xb0/0x280 (unreliable) [c0000002ecab0410] c00000000003d038 .prepare_ftrace_return+0x98/0x130 [c0000002ecab04b0] c00000000000a920 .ftrace_graph_caller+0x14/0x28 [c0000002ecab0520] c0000000000d6b58 .idle_cpu+0x18/0x90 [c0000002ecab05a0] c00000000000a934 .return_to_handler+0x0/0x34 [c0000002ecab0620] c00000000001e660 .timer_interrupt+0x160/0x300 [c0000002ecab06d0] c0000000000025dc decrementer_common+0x15c/0x180 --- Exception: 901 (Decrementer) at c0000000000104d4 .arch_local_irq_restore+0x74/0xa0 [c0000002ecab09c0] c0000000000fe044 .trace_hardirqs_on+0x14/0x30 (unreliable) [c0000002ecab0fb0] c00000000016fe3c .trace_graph_entry+0x13c/0x280 [c0000002ecab1050] c00000000003d038 .prepare_ftrace_return+0x98/0x130 [c0000002ecab10f0] c00000000000a920 .ftrace_graph_caller+0x14/0x28 [c0000002ecab1160] c0000000000161f0 .__ppc64_runlatch_on+0x10/0x40 [c0000002ecab11d0] c00000000000a934 .return_to_handler+0x0/0x34 --- Exception: 901 (Decrementer) at c0000000000104d4 .arch_local_irq_restore+0x74/0xa0 ... and so on __ppc64_runlatch_on() is called from RUNLATCH_ON in the exception entry path. At that point the irq state is not consistent, ie. interrupts are hard disabled (by the exception entry), but the paca soft-enabled flag may be out of sync. This leads to the local_irq_restore() in trace_graph_entry() actually enabling interrupts, which we do not want. Because we have not yet reprogrammed the decrementer we immediately take another decrementer exception, and recurse. The fix is twofold. Firstly make sure we call DISABLE_INTS before calling RUNLATCH_ON. The badly named DISABLE_INTS actually reconciles the irq state in the paca with the hardware, making it safe again to call local_irq_save/restore(). Although that should be sufficient to fix the bug, we also mark the runlatch routines as notrace. They are called very early in the exception entry and we are asking for trouble tracing them. They are also fairly uninteresting and tracing them just adds unnecessary overhead. [ This regression was introduced by fe1952fc0afb9a2e4c79f103c08aef5d13db1873 "powerpc: Rework runlatch code" by myself --BenH ] CC: <stable@vger.kernel.org> [v3.4+] Signed-off-by: Michael Ellerman <michael@ellerman.id.au> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2013-06-13 19:04:56 +08:00
void notrace __ppc64_runlatch_off(void)
{
struct thread_info *ti = current_thread_info();
ti->local_flags &= ~_TLF_RUNLATCH;
if (cpu_has_feature(CPU_FTR_ARCH_206)) {
mtspr(SPRN_CTRLT, 0);
} else {
unsigned long ctrl;
ctrl = mfspr(SPRN_CTRLF);
ctrl &= ~CTRL_RUNLATCH;
mtspr(SPRN_CTRLT, ctrl);
}
}
#endif /* CONFIG_PPC64 */
unsigned long arch_align_stack(unsigned long sp)
{
if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
sp -= get_random_int() & ~PAGE_MASK;
return sp & ~0xf;
}
static inline unsigned long brk_rnd(void)
{
unsigned long rnd = 0;
/* 8MB for 32bit, 1GB for 64bit */
if (is_32bit_task())
rnd = (get_random_long() % (1UL<<(23-PAGE_SHIFT)));
else
rnd = (get_random_long() % (1UL<<(30-PAGE_SHIFT)));
return rnd << PAGE_SHIFT;
}
unsigned long arch_randomize_brk(struct mm_struct *mm)
{
unsigned long base = mm->brk;
unsigned long ret;
#ifdef CONFIG_PPC_BOOK3S_64
/*
* If we are using 1TB segments and we are allowed to randomise
* the heap, we can put it above 1TB so it is backed by a 1TB
* segment. Otherwise the heap will be in the bottom 1TB
* which always uses 256MB segments and this may result in a
* performance penalty. We don't need to worry about radix. For
* radix, mmu_highuser_ssize remains unchanged from 256MB.
*/
if (!is_32bit_task() && (mmu_highuser_ssize == MMU_SEGSIZE_1T))
base = max_t(unsigned long, mm->brk, 1UL << SID_SHIFT_1T);
#endif
ret = PAGE_ALIGN(base + brk_rnd());
if (ret < mm->brk)
return mm->brk;
return ret;
}