Clearing errors or badblocks during a BTT write requires sending an ACPI
DSM, which means potentially sleeping. Since a BTT IO happens in atomic
context (preemption disabled, spinlocks may be held), we cannot perform
error clearing in the course of an IO. Due to this error clearing for
BTT IOs has hitherto been disabled.
In this patch we move error clearing out of the atomic section, and thus
re-enable error clearing with BTTs. When we are about to add a block to
the free list, we check if it was previously marked as an error, and if
it was, we add it to the freelist, but also set a flag that says error
clearing will be required. We then drop the lane (ending the atomic
context), and send a zero buffer so that the error can be cleared. The
error flag in the free list is protected by the nd 'lane', and is set
only be a thread while it holds that lane. When the error is cleared,
the flag is cleared, but while holding a mutex for that freelist index.
When writing, we check for two things -
1/ If the freelist mutex is held or if the error flag is set. If so,
this is an error block that is being (or about to be) cleared.
2/ If the block is a known badblock based on nsio->bb
The second check is required because the BTT map error flag for a map
entry only gets set when an error LBA is read. If we write to a new
location that may not have the map error flag set, but still might be in
the region's badblock list, we can trigger an EIO on the write, which is
undesirable and completely avoidable.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The IO context conversion for rw_bytes missed a case in the BTT write
path (btt_map_write) which should've been marked as atomic.
In reality this should not cause a problem, because map writes are to
small for nsio_rw_bytes to attempt error clearing, but it should be
fixed for posterity.
Add a might_sleep() in the non-atomic section of nsio_rw_bytes so that
things like the nfit unit tests, which don't actually sleep, can catch
bugs like this.
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
A leftover from the 'bandaid' fix that disabled BTT error clearing in
rw_bytes resulted in an incorrect check. After we converted these checks
over to use the NVDIMM_IO_ATOMIC flag, the ndns->claim check was both
redundant, and incorrect. Remove it.
Fixes: 3ae3d67ba7 ("libnvdimm: add an atomic vs process context flag to rw_bytes")
Cc: <stable@vger.kernel.org>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The UEFI 2.7 specification defines an updated BTT metadata format,
bumping the revision to 2.0. Add support for the new format, while
retaining compatibility for the old 1.1 format.
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Linda Knippers <linda.knippers@hpe.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Now that all callers of the pmem api have been converted to dax helpers that
call back to the pmem driver, we can remove include/linux/pmem.h and
asm/pmem.h.
Cc: <x86@kernel.org>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: Oliver O'Halloran <oohall@gmail.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Kill this globally defined wrapper and move to libnvdimm so that we can
ultimately remove include/linux/pmem.h and asm/pmem.h.
Cc: <x86@kernel.org>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Starting with v1.2 labels, 'address abstractions' can be hinted via an
address abstraction id that implies an info-block format. The standard
address abstraction in the specification is the v2 format of the
Block-Translation-Table (BTT). Support for that is saved for a later
patch, for now we add support for the Linux supported address
abstractions BTT (v1), PFN, and DAX.
The new 'holder_class' attribute for namespace devices is added for
tooling to specify the 'abstraction_guid' to store in the namespace label.
For v1.1 labels this field is undefined and any setting of
'holder_class' away from the default 'none' value will only have effect
until the driver is unloaded. Setting 'holder_class' requires that
whatever device tries to claim the namespace must be of the specified
class.
Cc: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The pmem driver has a need to transfer data with a persistent memory
destination and be able to rely on the fact that the destination writes are not
cached. It is sufficient for the writes to be flushed to a cpu-store-buffer
(non-temporal / "movnt" in x86 terms), as we expect userspace to call fsync()
to ensure data-writes have reached a power-fail-safe zone in the platform. The
fsync() triggers a REQ_FUA or REQ_FLUSH to the pmem driver which will turn
around and fence previous writes with an "sfence".
Implement a __copy_from_user_inatomic_flushcache, memcpy_page_flushcache, and
memcpy_flushcache, that guarantee that the destination buffer is not dirty in
the cpu cache on completion. The new copy_from_iter_flushcache and sub-routines
will be used to replace the "pmem api" (include/linux/pmem.h +
arch/x86/include/asm/pmem.h). The availability of copy_from_iter_flushcache()
and memcpy_flushcache() are gated by the CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE
config symbol, and fallback to copy_from_iter_nocache() and plain memcpy()
otherwise.
This is meant to satisfy the concern from Linus that if a driver wants to do
something beyond the normal nocache semantics it should be something private to
that driver [1], and Al's concern that anything uaccess related belongs with
the rest of the uaccess code [2].
The first consumer of this interface is a new 'copy_from_iter' dax operation so
that pmem can inject cache maintenance operations without imposing this
overhead on other dax-capable drivers.
[1]: https://lists.01.org/pipermail/linux-nvdimm/2017-January/008364.html
[2]: https://lists.01.org/pipermail/linux-nvdimm/2017-April/009942.html
Cc: <x86@kernel.org>
Cc: Jan Kara <jack@suse.cz>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Toshi Kani <toshi.kani@hpe.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
nsio_rw_bytes can clear media errors, but this cannot be done while we
are in an atomic context due to locking within ACPI. From the BTT,
->rw_bytes may be called either from atomic or process context depending
on whether the calls happen during initialization or during IO.
During init, we want to ensure error clearing happens, and the flag
marking process context allows nsio_rw_bytes to do that. When called
during IO, we're in atomic context, and error clearing can be skipped.
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
This continues the 4.11 status quo of disabling of error clearing from
the BTT I/O path. Toshi found that even though we have eliminated all
the libnvdimm sources of sleeping-while-atomic triggers, we still have
sleeping operations that will occur in the path to send the ACPI DSM to
the DIMM to clear the error:
BUG: sleeping function called from invalid context at mm/slab.h:432
in_atomic(): 1, irqs_disabled(): 0, pid: 13353, name: dd
Call Trace:
dump_stack+0x86/0xc3
___might_sleep+0x17d/0x250
__might_sleep+0x4a/0x80
__kmalloc+0x1c0/0x2e0
acpi_os_allocate_zeroed+0x2d/0x2f
acpi_evaluate_object+0x59/0x3b1
acpi_evaluate_dsm+0xbd/0x10c
acpi_nfit_ctl+0x1ef/0x7c0 [nfit]
? nsio_rw_bytes+0x152/0x280
nvdimm_clear_poison+0x77/0x140
nsio_rw_bytes+0x18f/0x280
btt_write_pg+0x1d4/0x3d0 [nd_btt]
btt_make_request+0x119/0x2d0 [nd_btt]
A solution for tracking and handling media errors natively in the BTT is
needed.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Reported-by: Toshi Kani <toshi.kani@hpe.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
A debug patch to turn the standard device_lock() into something that
lockdep can analyze yielded the following:
======================================================
[ INFO: possible circular locking dependency detected ]
4.11.0-rc4+ #106 Tainted: G O
-------------------------------------------------------
lt-libndctl/1898 is trying to acquire lock:
(&dev->nvdimm_mutex/3){+.+.+.}, at: [<ffffffffc023c948>] nd_attach_ndns+0x178/0x1b0 [libnvdimm]
but task is already holding lock:
(&nvdimm_bus->reconfig_mutex){+.+.+.}, at: [<ffffffffc022e0b1>] nvdimm_bus_lock+0x21/0x30 [libnvdimm]
which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #1 (&nvdimm_bus->reconfig_mutex){+.+.+.}:
lock_acquire+0xf6/0x1f0
__mutex_lock+0x88/0x980
mutex_lock_nested+0x1b/0x20
nvdimm_bus_lock+0x21/0x30 [libnvdimm]
nvdimm_namespace_capacity+0x1b/0x40 [libnvdimm]
nvdimm_namespace_common_probe+0x230/0x510 [libnvdimm]
nd_pmem_probe+0x14/0x180 [nd_pmem]
nvdimm_bus_probe+0xa9/0x260 [libnvdimm]
-> #0 (&dev->nvdimm_mutex/3){+.+.+.}:
__lock_acquire+0x1107/0x1280
lock_acquire+0xf6/0x1f0
__mutex_lock+0x88/0x980
mutex_lock_nested+0x1b/0x20
nd_attach_ndns+0x178/0x1b0 [libnvdimm]
nd_namespace_store+0x308/0x3c0 [libnvdimm]
namespace_store+0x87/0x220 [libnvdimm]
In this case '&dev->nvdimm_mutex/3' mirrors '&dev->mutex'.
Fix this by replacing the use of device_lock() with nvdimm_bus_lock() to protect
nd_{attach,detach}_ndns() operations.
Cc: <stable@vger.kernel.org>
Fixes: 8c2f7e8658 ("libnvdimm: infrastructure for btt devices")
Reported-by: Yi Zhang <yizhan@redhat.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
nvdimm_clear_poison() expects a physical address, not an offset.
Fix nsio_rw_bytes() to call nvdimm_clear_poison() with a physical
address.
Signed-off-by: Toshi Kani <toshi.kani@hpe.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Vishal Verma <vishal.l.verma@intel.com>
Reviewed-by: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
memcpy_from_pmem() maps directly to memcpy_mcsafe(). The wrapper
serves no real benefit aside from affording a more generic function name
than the x86-specific 'mcsafe'. However this would not be the first time
that x86 terminology leaked into the global namespace. For lack of
better name, just use memcpy_mcsafe() directly.
This conversion also catches a place where we should have been using
plain memcpy, acpi_nfit_blk_single_io().
Cc: <x86@kernel.org>
Cc: Jan Kara <jack@suse.cz>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Acked-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
This reverts commit 4aa5615e08 "libnvdimm: band aid btt vs clear
poison locking".
Now that poison list locking has been converted to a spinlock and poison
list entry allocation during i/o has been converted to GFP_NOWAIT,
revert the band-aid that disabled error clearing from btt i/o.
Cc: Vishal Verma <vishal.l.verma@intel.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The following warning results from holding a lane spinlock,
preempt_disable(), or the btt map spinlock and then trying to take the
reconfig_mutex to walk the poison list and potentially add new entries.
BUG: sleeping function called from invalid context at kernel/locking/mutex.c:747
in_atomic(): 1, irqs_disabled(): 0, pid: 17159, name: dd
[..]
Call Trace:
dump_stack+0x85/0xc8
___might_sleep+0x184/0x250
__might_sleep+0x4a/0x90
__mutex_lock+0x58/0x9b0
? nvdimm_bus_lock+0x21/0x30 [libnvdimm]
? __nvdimm_bus_badblocks_clear+0x2f/0x60 [libnvdimm]
? acpi_nfit_forget_poison+0x79/0x80 [nfit]
? _raw_spin_unlock+0x27/0x40
mutex_lock_nested+0x1b/0x20
nvdimm_bus_lock+0x21/0x30 [libnvdimm]
nvdimm_forget_poison+0x25/0x50 [libnvdimm]
nvdimm_clear_poison+0x106/0x140 [libnvdimm]
nsio_rw_bytes+0x164/0x270 [libnvdimm]
btt_write_pg+0x1de/0x3e0 [nd_btt]
? blk_queue_enter+0x30/0x290
btt_make_request+0x11a/0x310 [nd_btt]
? blk_queue_enter+0xb7/0x290
? blk_queue_enter+0x30/0x290
generic_make_request+0x118/0x3b0
As a minimal fix, disable error clearing when the BTT is enabled for the
namespace. For the final fix a larger rework of the poison list locking
is needed.
Note that this is not a problem in the blk case since that path never
calls nvdimm_clear_poison().
Cc: <stable@vger.kernel.org>
Fixes: 82bf1037f2 ("libnvdimm: check and clear poison before writing to pmem")
Cc: Dave Jiang <dave.jiang@intel.com>
[jeff: dynamically disable error clearing in the btt case]
Suggested-by: Jeff Moyer <jmoyer@redhat.com>
Reviewed-by: Jeff Moyer <jmoyer@redhat.com>
Reported-by: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Colin, via static analysis, reports that the length could be negative
from nvdimm_clear_poison() in the error case. There was a similar
problem with commit 0a3f27b9a6 "libnvdimm, namespace: avoid multiple
sector calculations" that I noticed when merging the for-4.10/libnvdimm
topic branch into libnvdimm-for-next, but I missed this one. Fix both of
them to the following procedure:
* if we clear a block's worth of media, clear that many blocks in
badblocks
* if we clear less than the requested size of the transfer return an
error
* always invalidate cache after any non-error / non-zero
nvdimm_clear_poison result
Fixes: 82bf1037f2 ("libnvdimm: check and clear poison before writing to pmem")
Fixes: 0a3f27b9a6 ("libnvdimm, namespace: avoid multiple sector calculations")
Cc: Fabian Frederick <fabf@skynet.be>
Cc: Dave Jiang <dave.jiang@intel.com>
Reported-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
For warnings that should only ever trigger during development and
testing replace WARN statements with lockdep_assert_held. The lockdep
pattern is prevalent, and these paths are are well covered by libnvdimm
unit tests.
Reported-by: Johannes Thumshirn <jthumshirn@suse.de>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
else after return is not needed.
Signed-off-by: Fabian Frederick <fabf@skynet.be>
[djbw: removed some now unnecessary newlines]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Here is an example /proc/iomem listing for a system with 2 namespaces,
one in "sector" mode and one in "memory" mode:
1fc000000-2fbffffff : Persistent Memory (legacy)
1fc000000-2fbffffff : namespace1.0
340000000-34fffffff : Persistent Memory
340000000-34fffffff : btt0.1
Here is the corresponding ndctl listing:
# ndctl list
[
{
"dev":"namespace1.0",
"mode":"memory",
"size":4294967296,
"blockdev":"pmem1"
},
{
"dev":"namespace0.0",
"mode":"sector",
"size":267091968,
"uuid":"f7594f86-badb-4592-875f-ded577da2eaf",
"sector_size":4096,
"blockdev":"pmem0s"
}
]
Notice that the ndctl listing is purely in terms of namespace devices,
while the iomem listing leaks the internal "btt0.1" implementation
detail. Given that ndctl requires the namespace device name to change
the mode, for example:
# ndctl create-namespace --reconfig=namespace0.0 --mode=raw --force
...use the namespace name in the iomem listing to keep the claiming
device name consistent across different mode settings.
Cc: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
We need to clear any poison when we are writing to pmem. The granularity
will be sector size. If it's less then we can't do anything about it
barring corruption.
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Vishal Verma <vishal.l.verma@intel.com>
[djbw: fixup 0-length write request to succeed]
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
nsio_rw_bytes() is used to write info block metadata to the namespace,
so it should trigger a flush after every write. Replace wmb_pmem() with
nvdimm_flush() in this path.
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Prompted by commit 287980e49f "remove lots of IS_ERR_VALUE abuses", I
ran make coccicheck against drivers/nvdimm/ and found that:
if (IS_ERR(x))
return PTR_ERR(x);
return 0;
...can be replaced with PTR_ERR_OR_ZERO().
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The ndctl unit tests discovered that the dax enabling omitted updates to
nd_detach_and_reset(). This routine clears device the configuration
when the namespace is detached. Without this clearing userspace may
assume that the device is in the process of being configured by another
agent in the system.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Device DAX is the device-centric analogue of Filesystem DAX
(CONFIG_FS_DAX). It allows persistent memory ranges to be allocated and
mapped without need of an intervening file system. This initial
infrastructure arranges for a libnvdimm pfn-device to be represented as
a different device-type so that it can be attached to a driver other
than the pmem driver.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
In preparation for providing an alternative (to block device) access
mechanism to persistent memory, convert pmem_rw_bytes() to
nsio_rw_bytes(). This allows ->rw_bytes() functionality without
requiring a 'struct pmem_device' to be instantiated.
In other words, when ->rw_bytes() is in use i/o is driven through
'struct nd_namespace_io', otherwise it is driven through 'struct
pmem_device' and the block layer. This consolidates the disjoint calls
to devm_exit_badblocks() and devm_memunmap() into a common
devm_nsio_disable() and cleans up the init path to use a unified
pmem_attach_disk() implementation.
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Implement the base infrastructure for libnvdimm PFN devices. Similar to
BTT devices they take a namespace as a backing device and layer
functionality on top. In this case the functionality is reserving space
for an array of 'struct page' entries to be handed out through
pfn_to_page(). For now this is just the basic libnvdimm-device-model for
configuring the base PFN device.
As the namespace claiming mechanism for PFN devices is mostly identical
to BTT devices drivers/nvdimm/claim.c is created to house the common
bits.
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Vishal Verma