platform/kernel/linux-rpi.git
6 years agolightnvm: use internal allocation for chunk log page
Javier González [Tue, 9 Oct 2018 11:12:01 +0000 (13:12 +0200)]
lightnvm: use internal allocation for chunk log page

The lightnvm subsystem provides helpers to retrieve chunk metadata,
where the target needs to provide a buffer to store the metadata. An
implicit assumption is that this buffer is contiguous and can be used to
retrieve the data from the device. If the device exposes too many
chunks, then kmalloc might fail, thus failing instance creation.

This patch removes this assumption by implementing an internal buffer in
the lightnvm subsystem to retrieve chunk metadata. Targets can then
use virtual memory allocations. Since this is a target API change, adapt
pblk accordingly.

Signed-off-by: Javier González <javier@cnexlabs.com>
Reviewed-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: fix two sleep-in-atomic-context bugs
Jia-Ju Bai [Tue, 9 Oct 2018 11:12:00 +0000 (13:12 +0200)]
lightnvm: pblk: fix two sleep-in-atomic-context bugs

The driver may sleep with holding a spinlock.

The function call paths (from bottom to top) in Linux-4.16 are:

[FUNC] nvm_dev_dma_alloc(GFP_KERNEL)
drivers/lightnvm/pblk-core.c, 754:
nvm_dev_dma_alloc in pblk_line_submit_smeta_io
drivers/lightnvm/pblk-core.c, 1048:
pblk_line_submit_smeta_io in pblk_line_init_bb
drivers/lightnvm/pblk-core.c, 1434:
pblk_line_init_bb in pblk_line_replace_data
drivers/lightnvm/pblk-recovery.c, 980:
pblk_line_replace_data in pblk_recov_l2p
drivers/lightnvm/pblk-recovery.c, 976:
spin_lock in pblk_recov_l2p

[FUNC] bio_map_kern(GFP_KERNEL)
drivers/lightnvm/pblk-core.c, 762:
bio_map_kern in pblk_line_submit_smeta_io
drivers/lightnvm/pblk-core.c, 1048:
pblk_line_submit_smeta_io in pblk_line_init_bb
drivers/lightnvm/pblk-core.c, 1434:
pblk_line_init_bb in pblk_line_replace_data
drivers/lightnvm/pblk-recovery.c, 980:
pblk_line_replace_data in pblk_recov_l2p
drivers/lightnvm/pblk-recovery.c, 976:
spin_lock in pblk_recov_l2p

To fix these bugs, the call to pblk_line_replace_data()
is moved out of the spinlock protection.

These bugs are found by my static analysis tool DSAC.

Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: fix mapping issue on failed writes
Hans Holmberg [Tue, 9 Oct 2018 11:11:59 +0000 (13:11 +0200)]
lightnvm: pblk: fix mapping issue on failed writes

On 1.2-devices, the mapping-out of remaning sectors in the
failed-write's block can result in an infinite loop,
stalling the write pipeline, fix this.

Fixes: 6a3abf5beef6 ("lightnvm: pblk: rework write error recovery path")
Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: stop recreating global caches
Hans Holmberg [Tue, 9 Oct 2018 11:11:58 +0000 (13:11 +0200)]
lightnvm: pblk: stop recreating global caches

Pblk should not create a set of global caches every time
a pblk instance is created. The global caches should be
made available only when there is one or more pblk instances.

This patch bundles the global caches together with a kref
keeping track of whether the caches should be available or not.

Also, turn the global pblk lock into a mutex that explicitly
protects the caches (as this was the only purpose of the lock).

Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: calculate line pad distance in helper
Javier González [Tue, 9 Oct 2018 11:11:57 +0000 (13:11 +0200)]
lightnvm: pblk: calculate line pad distance in helper

If a line is padded, calculate the pad distance directly on the helper
being used for this purpose.

Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: move ppa transformations to core
Javier González [Tue, 9 Oct 2018 11:11:56 +0000 (13:11 +0200)]
lightnvm: move ppa transformations to core

Continuing the effort of moving 1.2 and 2.0 specific code to core, move
64_to_32 and 32_to_64 ppa helpers from pblk to core.

Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: add tracing for chunk resets
Hans Holmberg [Tue, 9 Oct 2018 11:11:55 +0000 (13:11 +0200)]
lightnvm: pblk: add tracing for chunk resets

Trace state of chunk resets.

Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: add trace events for pblk state changes
Hans Holmberg [Tue, 9 Oct 2018 11:11:54 +0000 (13:11 +0200)]
lightnvm: pblk: add trace events for pblk state changes

Add trace events for tracking pblk state changes.

Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: add trace events for line state changes
Hans Holmberg [Tue, 9 Oct 2018 11:11:53 +0000 (13:11 +0200)]
lightnvm: pblk: add trace events for line state changes

Add trace events for logging for line state changes.

Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: add trace events for chunk states
Hans Holmberg [Tue, 9 Oct 2018 11:11:52 +0000 (13:11 +0200)]
lightnvm: pblk: add trace events for chunk states

Introduce trace points for tracking chunk states in pblk - this is
useful for inspection of the entire state of the drive, and real handy
for both fw and pblk debugging.

Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: remove debug from pblk_[down/up]_page
Matias Bjørling [Tue, 9 Oct 2018 11:11:51 +0000 (13:11 +0200)]
lightnvm: pblk: remove debug from pblk_[down/up]_page

Remove the debug only iteration within __pblk_down_page, which
then allows us to reduce the number of arguments down to pblk and
the parallel unit from the functions that calls it. Simplifying the
callers logic considerably.

Also, rename the functions pblk_[down/up]_page to
pblk_[down/up]_chunk, to communicate that it manages the write
pointer of the chunk. Note that it also protects the parallel unit
such that at most one chunk is active per parallel unit.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: fix write amplificiation calculation
Hans Holmberg [Tue, 9 Oct 2018 11:11:50 +0000 (13:11 +0200)]
lightnvm: pblk: fix write amplificiation calculation

When the user data counter exceeds 32 bits, the write amplification
calculation does not provide the right value. Fix this by using
div64_u64 in stead of div64.

Fixes: 76758390f83e ("lightnvm: pblk: export write amplification counters to sysfs")
Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: fix up prints in pblk_read_check_rand
Hans Holmberg [Tue, 9 Oct 2018 11:11:49 +0000 (13:11 +0200)]
lightnvm: pblk: fix up prints in pblk_read_check_rand

The prefix when printing ppas in pblk_read_check_rand should be "rnd"
not "seq", so fix this so we can differentiate between lba missmatches
in random and sequential reads. Also change the print order so
we align with pblk_read_check_seq, printing read lba first.

Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: remove unused parameters in pblk_up_rq
Hans Holmberg [Tue, 9 Oct 2018 11:11:48 +0000 (13:11 +0200)]
lightnvm: pblk: remove unused parameters in pblk_up_rq

The parameters nr_ppas and ppa_list are not used, so remove them.

Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: allocate line map bitmaps using a mempool
Hans Holmberg [Tue, 9 Oct 2018 11:11:47 +0000 (13:11 +0200)]
lightnvm: pblk: allocate line map bitmaps using a mempool

Line map bitmap allocations are fairly large and can fail. Allocation
failures are fatal to pblk, stopping the write pipeline. To avoid this,
allocate the bitmaps using a mempool instead.

Mempool allocations never fail if called from a process context,
and pblk *should* only allocate map bitmaps in process context,
but keep the failure handling for robustness sake.

Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: introduce nvm_rq_to_ppa_list
Hans Holmberg [Tue, 9 Oct 2018 11:11:46 +0000 (13:11 +0200)]
lightnvm: introduce nvm_rq_to_ppa_list

There is a number of places in the lightnvm subsystem where the user
iterates over the ppa list. Before iterating, the user must know if it
is a single or multiple LBAs due to vector commands using either the
nvm_rq ->ppa_addr or ->ppa_list fields on command submission, which
leads to open-coding the if/else statement.

Instead of having multiple if/else's, move it into a function that can
be called by its users.

A nice side effect of this cleanup is that this patch fixes up a
bunch of cases where we don't consider the single-ppa case in pblk.

Signed-off-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: guarantee emeta on line close
Javier González [Tue, 9 Oct 2018 11:11:45 +0000 (13:11 +0200)]
lightnvm: pblk: guarantee emeta on line close

If a line is recovered from open chunks, the memory structures for
emeta have not necessarily been properly set on line initialization.
When closing a line, make sure that emeta is consistent so that the line
can be recovered on the fast path on next reboot.

Also, remove a couple of empty lines at the end of the function.

Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: remove unused variable.
Javier González [Tue, 9 Oct 2018 11:11:44 +0000 (13:11 +0200)]
lightnvm: pblk: remove unused variable.

Removed unused struct ppa_addr variable.

Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: fix comment typo
Javier González [Tue, 9 Oct 2018 11:11:43 +0000 (13:11 +0200)]
lightnvm: pblk: fix comment typo

Fix comment typo Decrese -> Decrease

Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: improve line helpers
Javier González [Tue, 9 Oct 2018 11:11:42 +0000 (13:11 +0200)]
lightnvm: pblk: improve line helpers

The current helper to obtain a line from a ppa returns the line id,
which requires its users to explicitly retrieve the pointer to the line
with the id.

Make 2 different helpers: one returning the line id and one returning
the line directly.

Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: add helpers for chunk addresses
Javier González [Tue, 9 Oct 2018 11:11:41 +0000 (13:11 +0200)]
lightnvm: pblk: add helpers for chunk addresses

Implement helpers to go from ppas to a chunk within a line and an
address within a chunk.

These helpers will be used on the patches adding trace support in pblk,
which will be sent in this window.

Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: refactor put line fn on read completion
Matias Bjørling [Tue, 9 Oct 2018 11:11:40 +0000 (13:11 +0200)]
lightnvm: pblk: refactor put line fn on read completion

The read completion path uses the put_line variable to decide whether
the reference on a line should be released. The function name used for
that is pblk_read_put_rqd_kref, which could lead one to believe that it
is the rqd that is releasing the reference, while it is the line
reference that is put.

Rename and also split the function in two to account for either rqd or
single ppa callers and move it to core, such that it later can be used
in the write path as well.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Reviewed-by: Heiner Litz <hlitz@ucsc.edu>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: remove size and out of bounds read check
Matias Bjørling [Tue, 9 Oct 2018 11:11:39 +0000 (13:11 +0200)]
lightnvm: pblk: remove size and out of bounds read check

The I/O size and capacity checks are already done by the block layer.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: fix incorrect min_write_pgs
Matias Bjørling [Tue, 9 Oct 2018 11:11:38 +0000 (13:11 +0200)]
lightnvm: pblk: fix incorrect min_write_pgs

The calculation of pblk->min_write_pgs should only use the optimal
write size attribute provided by the drive, it does not correlate to
the memory page size of the system, which can be smaller or larger
than the LBA size reported.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: unify vector max req constants
Matias Bjørling [Tue, 9 Oct 2018 11:11:37 +0000 (13:11 +0200)]
lightnvm: pblk: unify vector max req constants

Both NVM_MAX_VLBA and PBLK_MAX_REQ_ADDRS define how many LBAs that
are available in a vector command. pblk uses them interchangeably
in its implementation. Use NVM_MAX_VLBA as the main one and remove
usages of PBLK_MAX_REQ_ADDRS.

Also remove the power representation that only has one user, and
instead calculate it at runtime.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: move bad block and chunk state logic to core
Matias Bjørling [Tue, 9 Oct 2018 11:11:36 +0000 (13:11 +0200)]
lightnvm: move bad block and chunk state logic to core

pblk implements two data paths for recovery line state. One for 1.2
and another for 2.0, instead of having pblk implement these, combine
them in the core to reduce complexity and make available to other
targets.

The new interface will adhere to the 2.0 chunk definition,
including managing open chunks with an active write pointer. To provide
this interface, a 1.2 device recovers the state of the chunks by
manually detecting if a chunk is either free/open/close/offline, and if
open, scanning the flash pages sequentially to find the next writeable
page. This process takes on average ~10 seconds on a device with 64 dies,
1024 blocks and 60us read access time. The process can be parallelized
but is left out for maintenance simplicity, as the 1.2 specification is
deprecated. For 2.0 devices, the logic is maintained internally in the
drive and retrieved through the 2.0 interface.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: fix race condition on metadata I/O
Javier González [Tue, 9 Oct 2018 11:11:35 +0000 (13:11 +0200)]
lightnvm: pblk: fix race condition on metadata I/O

In pblk, when a new line is allocated, metadata for the previously
written line is scheduled. This is done through a fixed memory region
that is shared through time and contexts across different lines and
therefore protected by a lock. Unfortunately, this lock is not properly
covering all the metadata used for sharing this memory regions,
resulting in a race condition.

This patch fixes this race condition by protecting this metadata
properly.

Fixes: dd2a43437337 ("lightnvm: pblk: sched. metadata on write thread")
Signed-off-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: move device L2P detection to core
Matias Bjørling [Tue, 9 Oct 2018 11:11:34 +0000 (13:11 +0200)]
lightnvm: move device L2P detection to core

A 1.2 device is able to manage the logical to physical mapping
table internally or leave it to the host.

A target only supports one of those approaches, and therefore must
check on initialization. Move this check to core to avoid each target
implement the check.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: pblk: fix rqd.error return value in pblk_blk_erase_sync
Matias Bjørling [Tue, 9 Oct 2018 11:11:33 +0000 (13:11 +0200)]
lightnvm: pblk: fix rqd.error return value in pblk_blk_erase_sync

rqd.error is masked by the return value of pblk_submit_io_sync.
The rqd structure is then passed on to the end_io function, which
assumes that any error should lead to a chunk being marked
offline/bad. Since the pblk_submit_io_sync can fail before the
command is issued to the device, the error value maybe not correspond
to a media failure, leading to chunks being immaturely retired.

Also, the pblk_blk_erase_sync function prints an error message in case
the erase fails. Since the caller prints an error message by itself,
remove the error message in this function.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Reviewed-by: Hans Holmberg <hans.holmberg@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: combine 1.2 and 2.0 command flags
Matias Bjørling [Tue, 9 Oct 2018 11:11:32 +0000 (13:11 +0200)]
lightnvm: combine 1.2 and 2.0 command flags

Add nvm_set_flags helper to enable core to appropriately
set the command flags for read/write/erase depending on which version
a drive supports.

The flags arguments can be distilled into the access hint,
scrambling, and program/erase suspend. Replace the access hint with
a "is_seq" parameter. The rest of the flags are dependent on the
command opcode, which is trivial to detect and set.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agolightnvm: remove dependencies on BLK_DEV_NVME and PCI
Matias Bjørling [Tue, 9 Oct 2018 11:11:31 +0000 (13:11 +0200)]
lightnvm: remove dependencies on BLK_DEV_NVME and PCI

No need to force NVMe device driver to be compiled in if the
lightnvm subsystem is selected. Also no need for PCI to be selected
as well, as it would be selected by the device driver that hooks into
the subsystem.

Signed-off-by: Matias Bjørling <mb@lightnvm.io>
Reviewed-by: Javier González <javier@cnexlabs.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-mq: complete req in softirq context in case of single queue
Ming Lei [Fri, 28 Sep 2018 08:42:20 +0000 (16:42 +0800)]
blk-mq: complete req in softirq context in case of single queue

Lot of controllers may have only one irq vector for completing IO
request. And usually affinity of the only irq vector is all possible
CPUs, however, on most of ARCH, there may be only one specific CPU
for handling this interrupt.

So if all IOs are completed in hardirq context, it is inevitable to
degrade IO performance because of increased irq latency.

This patch tries to address this issue by allowing to complete request
in softirq context, like the legacy IO path.

IOPS is observed as ~13%+ in the following randread test on raid0 over
virtio-scsi.

mdadm --create --verbose /dev/md0 --level=0 --chunk=1024 --raid-devices=8 /dev/sdb /dev/sdc /dev/sdd /dev/sde /dev/sdf /dev/sdg /dev/sdh /dev/sdi

fio --time_based --name=benchmark --runtime=30 --filename=/dev/md0 --nrfiles=1 --ioengine=libaio --iodepth=32 --direct=1 --invalidate=1 --verify=0 --verify_fatal=0 --numjobs=32 --rw=randread --blocksize=4k

Cc: Dongli Zhang <dongli.zhang@oracle.com>
Cc: Zach Marano <zmarano@google.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Bart Van Assche <bvanassche@acm.org>
Cc: Jianchao Wang <jianchao.w.wang@oracle.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: panic fix for making cache device
Dongbo Cao [Mon, 8 Oct 2018 12:41:21 +0000 (20:41 +0800)]
bcache: panic fix for making cache device

when the nbuckets of cache device is smaller than 1024, making cache
device will trigger BUG_ON in kernel, add a condition to avoid this.

Reported-by: nitroxis <n@nxs.re>
Signed-off-by: Dongbo Cao <cdbdyx@163.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: split combined if-condition code into separate ones
Dongbo Cao [Mon, 8 Oct 2018 12:41:20 +0000 (20:41 +0800)]
bcache: split combined if-condition code into separate ones

Split the combined '||' statements in if() check, to make the code easier
for debug.

Signed-off-by: Dongbo Cao <cdbdyx@163.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: use MAX_CACHES_PER_SET instead of magic number 8 in __bch_bucket_alloc_set
Shenghui Wang [Mon, 8 Oct 2018 12:41:19 +0000 (20:41 +0800)]
bcache: use MAX_CACHES_PER_SET instead of magic number 8 in __bch_bucket_alloc_set

Current cache_set has MAX_CACHES_PER_SET caches most, and the macro
is used for
"
struct cache *cache_by_alloc[MAX_CACHES_PER_SET];
"
in the define of struct cache_set.

Use MAX_CACHES_PER_SET instead of magic number 8 in
__bch_bucket_alloc_set.

Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: replace hard coded number with BUCKET_GC_GEN_MAX
Coly Li [Mon, 8 Oct 2018 12:41:18 +0000 (20:41 +0800)]
bcache: replace hard coded number with BUCKET_GC_GEN_MAX

In extents.c:bch_extent_bad(), number 96 is used as parameter to call
btree_bug_on(). The purpose is to check whether stale gen value exceeds
BUCKET_GC_GEN_MAX, so it is better to use macro BUCKET_GC_GEN_MAX to
make the code more understandable.

Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: remove useless parameter of bch_debug_init()
Dongbo Cao [Mon, 8 Oct 2018 12:41:17 +0000 (20:41 +0800)]
bcache: remove useless parameter of bch_debug_init()

Parameter "struct kobject *kobj" in bch_debug_init() is useless,
remove it in this patch.

Signed-off-by: Dongbo Cao <cdbdyx@163.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: remove unused bch_passthrough_cache
Shenghui Wang [Mon, 8 Oct 2018 12:41:16 +0000 (20:41 +0800)]
bcache: remove unused bch_passthrough_cache

struct kmem_cache *bch_passthrough_cache is not used in
bcache code. Remove it.

Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: recal cached_dev_sectors on detach
Shenghui Wang [Mon, 8 Oct 2018 12:41:15 +0000 (20:41 +0800)]
bcache: recal cached_dev_sectors on detach

Recal cached_dev_sectors on cached_dev detached, as recal done on
cached_dev attached.

Update the cached_dev_sectors before bcache_device_detach called
as bcache_device_detach will set bcache_device->c to NULL.

Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: fix miss key refill->end in writeback
Tang Junhui [Mon, 8 Oct 2018 12:41:14 +0000 (20:41 +0800)]
bcache: fix miss key refill->end in writeback

refill->end record the last key of writeback, for example, at the first
time, keys (1,128K) to (1,1024K) are flush to the backend device, but
the end key (1,1024K) is not included, since the bellow code:
if (bkey_cmp(k, refill->end) >= 0) {
ret = MAP_DONE;
goto out;
}
And in the next time when we refill writeback keybuf again, we searched
key start from (1,1024K), and got a key bigger than it, so the key
(1,1024K) missed.
This patch modify the above code, and let the end key to be included to
the writeback key buffer.

Signed-off-by: Tang Junhui <tang.junhui.linux@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: Populate writeback_rate_minimum attribute
Ben Peddell [Mon, 8 Oct 2018 12:41:13 +0000 (20:41 +0800)]
bcache: Populate writeback_rate_minimum attribute

Forgot to include the maintainers with my first email.

Somewhere between Michael Lyle's original
"bcache: PI controller for writeback rate V2" patch dated 07 Sep 2017
and 1d316e6 bcache: implement PI controller for writeback rate,
the mapping of the writeback_rate_minimum attribute was dropped.

Re-add the missing sysfs writeback_rate_minimum attribute mapping to
"allow the user to specify a minimum rate at which dirty blocks are
retired."

Fixes: 1d316e6 ("bcache: implement PI controller for writeback rate")
Signed-off-by: Ben Peddell <klightspeed@killerwolves.net>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: correct dirty data statistics
Tang Junhui [Mon, 8 Oct 2018 12:41:12 +0000 (20:41 +0800)]
bcache: correct dirty data statistics

When bcache device is clean, dirty keys may still exist after
journal replay, so we need to count these dirty keys even
device in clean status, otherwise after writeback, the amount
of dirty data would be incorrect.

Signed-off-by: Tang Junhui <tang.junhui.linux@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: fix typo in code comments of closure_return_with_destructor()
Coly Li [Mon, 8 Oct 2018 12:41:11 +0000 (20:41 +0800)]
bcache: fix typo in code comments of closure_return_with_destructor()

The code comments of closure_return_with_destructor() in closure.h makrs
function name as closure_return(). This patch fixes this type with the
correct name - closure_return_with_destructor.

Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: fix ioctl in flash device
Tang Junhui [Mon, 8 Oct 2018 12:41:10 +0000 (20:41 +0800)]
bcache: fix ioctl in flash device

When doing ioctl in flash device, it will call ioctl_dev() in super.c,
then we should not to get cached device since flash only device has
no backend device. This patch just move the jugement dc->io_disable
to cached_dev_ioctl() to make ioctl in flash device correctly.

Fixes: 0f0709e6bfc3c ("bcache: stop bcache device when backing device is offline")
Signed-off-by: Tang Junhui <tang.junhui.linux@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: use REQ_PRIO to indicate bio for metadata
Coly Li [Mon, 8 Oct 2018 12:41:09 +0000 (20:41 +0800)]
bcache: use REQ_PRIO to indicate bio for metadata

In cached_dev_cache_miss() and check_should_bypass(), REQ_META is used
to check whether a bio is for metadata request. REQ_META is used for
blktrace, the correct REQ_ flag should be REQ_PRIO. This flag means the
bio should be prior to other bio, and frequently be used to indicate
metadata io in file system code.

This patch replaces REQ_META with correct flag REQ_PRIO.

CC Adam Manzanares because he explains to me what REQ_PRIO is for.

Signed-off-by: Coly Li <colyli@suse.de>
Cc: Adam Manzanares <adam.manzanares@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: trace missed reading by cache_missed
Tang Junhui [Mon, 8 Oct 2018 12:41:08 +0000 (20:41 +0800)]
bcache: trace missed reading by cache_missed

Missed reading IOs are identified by s->cache_missed, not the
s->cache_miss, so in trace_bcache_read() using trace_bcache_read
to identify whether the IO is missed or not.

Signed-off-by: Tang Junhui <tang.junhui.linux@gmail.com>
Cc: stable@vger.kernel.org
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agobcache: account size of buckets used in uuid write to ca->meta_sectors_written
Shenghui Wang [Mon, 8 Oct 2018 12:41:07 +0000 (20:41 +0800)]
bcache: account size of buckets used in uuid write to ca->meta_sectors_written

UUIDs are considered as metadata. __uuid_write should add the number
of buckets (in sectors) written to disk to ca->meta_sectors_written.
Currently only 1 bucket is used in uuid write.

Steps to test:
1) create a fresh backing device and a fresh cache device separately.
   The backing device didn't attach to any cache set.
2) cd /sys/block/<cache device>/bcache
   cat metadata_written      // record the output value
   cat bucket_size
3) attach the backing device to cache set
4) cat metadata_written
   The output value is almost the same as the value in step 2
   before the change.
   After the change, the value is bigger about 1 bucket size.

Signed-off-by: Shenghui Wang <shhuiw@foxmail.com>
Reviewed-by: Tang Junhui <tang.junhui.linux@gmail.com>
Signed-off-by: Coly Li <colyli@suse.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-mq-debugfs: Also show requests that have not yet been started
Bart Van Assche [Thu, 4 Oct 2018 17:35:24 +0000 (10:35 -0700)]
blk-mq-debugfs: Also show requests that have not yet been started

When debugging e.g. the SCSI timeout handler it is important that
requests that have not yet been started or that already have
completed are also reported through debugfs.

Cc: Christoph Hellwig <hch@lst.de>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoMerge branch 'nvme-4.20' of git://git.infradead.org/nvme into for-4.20/block
Jens Axboe [Fri, 5 Oct 2018 14:15:12 +0000 (08:15 -0600)]
Merge branch 'nvme-4.20' of git://git.infradead.org/nvme into for-4.20/block

Pull NVMe updates from Christoph:

"A relatively boring merge window:

 - better AEN tracing (Chaitanya)
 - NUMA aware PCIe multipathing (me)
 - RDMA workqueue fixes (Sagi)
 - better bio usage in the target (Sagi)
 - FC rework for target removal (James)
 - better multipath handling of ->queue_rq failures (James)
 - various cleanups (Milan)"

* 'nvme-4.20' of git://git.infradead.org/nvme:
  nvmet-rdma: use a private workqueue for delete
  nvme: take node locality into account when selecting a path
  nvmet: don't split large I/Os unconditionally
  nvme: call nvme_complete_rq when nvmf_check_ready fails for mpath I/O
  nvme-core: add async event trace helper
  nvme_fc: add 'nvme_discovery' sysfs attribute to fc transport device
  nvmet_fc: support target port removal with nvmet layer
  nvme-fc: fix for a minor typos
  nvmet: remove redundant module prefix
  nvme: fix typo in nvme_identify_ns_descs

6 years agonvmet-rdma: use a private workqueue for delete
Sagi Grimberg [Thu, 27 Sep 2018 18:00:31 +0000 (11:00 -0700)]
nvmet-rdma: use a private workqueue for delete

Queue deletion is done asynchronous when the last reference on the queue
is dropped.  Thus, in order to make sure we don't over allocate under a
connect/disconnect storm, we let queue deletion complete before making
forward progress.

However, given that we flush the system_wq from rdma_cm context which
runs from a workqueue context, we can have a circular locking complaint
[1]. Fix that by using a private workqueue for queue deletion.

[1]:
======================================================
WARNING: possible circular locking dependency detected
4.19.0-rc4-dbg+ #3 Not tainted
------------------------------------------------------
kworker/5:0/39 is trying to acquire lock:
00000000a10b6db9 (&id_priv->handler_mutex){+.+.}, at: rdma_destroy_id+0x6f/0x440 [rdma_cm]

but task is already holding lock:
00000000331b4e2c ((work_completion)(&queue->release_work)){+.+.}, at: process_one_work+0x3ed/0xa20

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #3 ((work_completion)(&queue->release_work)){+.+.}:
       process_one_work+0x474/0xa20
       worker_thread+0x63/0x5a0
       kthread+0x1cf/0x1f0
       ret_from_fork+0x24/0x30

-> #2 ((wq_completion)"events"){+.+.}:
       flush_workqueue+0xf3/0x970
       nvmet_rdma_cm_handler+0x133d/0x1734 [nvmet_rdma]
       cma_ib_req_handler+0x72f/0xf90 [rdma_cm]
       cm_process_work+0x2e/0x110 [ib_cm]
       cm_req_handler+0x135b/0x1c30 [ib_cm]
       cm_work_handler+0x2b7/0x38cd [ib_cm]
       process_one_work+0x4ae/0xa20
nvmet_rdma:nvmet_rdma_cm_handler: nvmet_rdma: disconnected (10): status 0 id 0000000040357082
       worker_thread+0x63/0x5a0
       kthread+0x1cf/0x1f0
       ret_from_fork+0x24/0x30
nvme nvme0: Reconnecting in 10 seconds...

-> #1 (&id_priv->handler_mutex/1){+.+.}:
       __mutex_lock+0xfe/0xbe0
       mutex_lock_nested+0x1b/0x20
       cma_ib_req_handler+0x6aa/0xf90 [rdma_cm]
       cm_process_work+0x2e/0x110 [ib_cm]
       cm_req_handler+0x135b/0x1c30 [ib_cm]
       cm_work_handler+0x2b7/0x38cd [ib_cm]
       process_one_work+0x4ae/0xa20
       worker_thread+0x63/0x5a0
       kthread+0x1cf/0x1f0
       ret_from_fork+0x24/0x30

-> #0 (&id_priv->handler_mutex){+.+.}:
       lock_acquire+0xc5/0x200
       __mutex_lock+0xfe/0xbe0
       mutex_lock_nested+0x1b/0x20
       rdma_destroy_id+0x6f/0x440 [rdma_cm]
       nvmet_rdma_release_queue_work+0x8e/0x1b0 [nvmet_rdma]
       process_one_work+0x4ae/0xa20
       worker_thread+0x63/0x5a0
       kthread+0x1cf/0x1f0
       ret_from_fork+0x24/0x30

Fixes: 777dc82395de ("nvmet-rdma: occasionally flush ongoing controller teardown")
Reported-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Tested-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agoblock: Finish renaming REQ_DISCARD into REQ_OP_DISCARD
Bart Van Assche [Wed, 3 Oct 2018 20:56:25 +0000 (13:56 -0700)]
block: Finish renaming REQ_DISCARD into REQ_OP_DISCARD

Some time ago REQ_DISCARD was renamed into REQ_OP_DISCARD. Some comments
and documentation files were not updated however. Update these comments
and documentation files. See also commit 4e1b2d52a80d ("block, fs,
drivers: remove REQ_OP compat defs and related code").

Signed-off-by: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Mike Christie <mchristi@redhat.com>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Cc: Philipp Reisner <philipp.reisner@linbit.com>
Cc: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agocdrom: fix improper type cast, which can leat to information leak.
Young_X [Wed, 3 Oct 2018 12:54:29 +0000 (12:54 +0000)]
cdrom: fix improper type cast, which can leat to information leak.

There is another cast from unsigned long to int which causes
a bounds check to fail with specially crafted input. The value is
then used as an index in the slot array in cdrom_slot_status().

This issue is similar to CVE-2018-16658 and CVE-2018-10940.

Signed-off-by: Young_X <YangX92@hotmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agopktcdvd: fix fall-through annotation
Gustavo A. R. Silva [Tue, 2 Oct 2018 10:24:40 +0000 (12:24 +0200)]
pktcdvd: fix fall-through annotation

Replace "fallthru" with a proper "fall through" annotation.

This fix is part of the ongoing efforts to enabling
-Wimplicit-fallthrough

Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agonvme: take node locality into account when selecting a path
Christoph Hellwig [Tue, 11 Sep 2018 07:51:29 +0000 (09:51 +0200)]
nvme: take node locality into account when selecting a path

Make current_path an array with an entry for every possible node, and
cache the best path on a per-node basis.  Take the node distance into
account when selecting it.  This is primarily useful for dual-ported PCIe
devices which are connected to PCIe root ports on different sockets.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
6 years agonvmet: don't split large I/Os unconditionally
Sagi Grimberg [Fri, 28 Sep 2018 22:40:43 +0000 (15:40 -0700)]
nvmet: don't split large I/Os unconditionally

If we know that the I/O size exceeds our inline bio vec, no
point using it and split the rest to begin with. We could
in theory reuse the inline bio and only allocate the bio_vec,
but its really not worth optimizing for.

Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme: call nvme_complete_rq when nvmf_check_ready fails for mpath I/O
James Smart [Thu, 27 Sep 2018 23:58:54 +0000 (16:58 -0700)]
nvme: call nvme_complete_rq when nvmf_check_ready fails for mpath I/O

When an io is rejected by nvmf_check_ready() due to validation of the
controller state, the nvmf_fail_nonready_command() will normally return
BLK_STS_RESOURCE to requeue and retry.  However, if the controller is
dying or the I/O is marked for NVMe multipath, the I/O is failed so that
the controller can terminate or so that the io can be issued on a
different path.  Unfortunately, as this reject point is before the
transport has accepted the command, blk-mq ends up completing the I/O
and never calls nvme_complete_rq(), which is where multipath may preserve
or re-route the I/O. The end result is, the device user ends up seeing an
EIO error.

Example: single path connectivity, controller is under load, and a reset
is induced.  An I/O is received:

  a) while the reset state has been set but the queues have yet to be
     stopped; or
  b) after queues are started (at end of reset) but before the reconnect
     has completed.

The I/O finishes with an EIO status.

This patch makes the following changes:

  - Adds the HOST_PATH_ERROR pathing status from TP4028
  - Modifies the reject point such that it appears to queue successfully,
    but actually completes the io with the new pathing status and calls
    nvme_complete_rq().
  - nvme_complete_rq() recognizes the new status, avoids resetting the
    controller (likely was already done in order to get this new status),
    and calls the multipather to clear the current path that errored.
    This allows the next command (retry or new command) to select a new
    path if there is one.

Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme-core: add async event trace helper
Chaitanya Kulkarni [Mon, 17 Sep 2018 17:47:06 +0000 (10:47 -0700)]
nvme-core: add async event trace helper

This patch adds a new event for nvme async event notification.
We print the async event in the decoded format when we recognize
the event otherwise we just dump the result.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme_fc: add 'nvme_discovery' sysfs attribute to fc transport device
James Smart [Thu, 13 Sep 2018 23:17:38 +0000 (16:17 -0700)]
nvme_fc: add 'nvme_discovery' sysfs attribute to fc transport device

The fc transport device should allow for a rediscovery, as userspace
might have lost the events. Example is udev events not handled during
system startup.

This patch add a sysfs entry 'nvme_discovery' on the fc class to
have it replay all udev discovery events for all local port/remote
port address pairs.

Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvmet_fc: support target port removal with nvmet layer
James Smart [Thu, 9 Aug 2018 23:48:14 +0000 (16:48 -0700)]
nvmet_fc: support target port removal with nvmet layer

Currently, if a targetport has been connected to via the nvmet config
(in other words, the add_port() transport routine called, and the nvmet
port pointer stored for using in upcalls on new io), and if the
targetport is then removed (say the lldd driver decides to unload or
fully reset its hardware) and then re-added (the lldd driver reloads or
reinits its hardware), the port pointer has been lost so there's no way
to continue to post commands up to nvmet via the transport port.

Correct by allocating a small "port context" structure that will be
linked to by the targetport. The context will save the targetport WWN's
and the nvmet port pointer to use for it.  Initial allocation will occur
when the targetport is bound to via add_port.  The context will be
deallocated when remove_port() is called.  If a targetport is removed
while nvmet has the active port context, the targetport will be unlinked
from the port context before removal.  If a new targetport is registered,
the port contexts without a binding are looked through and if the WWN's
match (so it's the same as nvmet's port context) the port context is
linked to the new target port.  Thus new io can be received on the new
targetport and operation resumes with nvmet.

Additionally, this also resolves nvmet configuration changing out from
underneath of the nvme-fc target port (for example: a nvmetcli clear).

Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme-fc: fix for a minor typos
Milan P. Gandhi [Fri, 10 Aug 2018 09:24:02 +0000 (14:54 +0530)]
nvme-fc: fix for a minor typos

Signed-off-by: Milan P. Gandhi <mgandhi@redhat.com>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvmet: remove redundant module prefix
Chaitanya Kulkarni [Tue, 11 Sep 2018 00:39:33 +0000 (17:39 -0700)]
nvmet: remove redundant module prefix

This patch removes the redundant module prefix used in the pr_err() when
nvmet_get_smart_log_nsid() failed to find the namespace provided as a part
of smart-log command.

Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agonvme: fix typo in nvme_identify_ns_descs
Milan P. Gandhi [Thu, 9 Aug 2018 16:19:24 +0000 (21:49 +0530)]
nvme: fix typo in nvme_identify_ns_descs

Signed-off-by: Milan P. Gandhi <mgandhi@redhat.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
6 years agoMerge tag 'v4.19-rc6' into for-4.20/block
Jens Axboe [Mon, 1 Oct 2018 14:58:57 +0000 (08:58 -0600)]
Merge tag 'v4.19-rc6' into for-4.20/block

Merge -rc6 in, for two reasons:

1) Resolve a trivial conflict in the blk-mq-tag.c documentation
2) A few important regression fixes went into upstream directly, so
   they aren't in the 4.20 branch.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
* tag 'v4.19-rc6': (780 commits)
  Linux 4.19-rc6
  MAINTAINERS: fix reference to moved drivers/{misc => auxdisplay}/panel.c
  cpufreq: qcom-kryo: Fix section annotations
  perf/core: Add sanity check to deal with pinned event failure
  xen/blkfront: correct purging of persistent grants
  Revert "xen/blkfront: When purging persistent grants, keep them in the buffer"
  selftests/powerpc: Fix Makefiles for headers_install change
  blk-mq: I/O and timer unplugs are inverted in blktrace
  dax: Fix deadlock in dax_lock_mapping_entry()
  x86/boot: Fix kexec booting failure in the SEV bit detection code
  bcache: add separate workqueue for journal_write to avoid deadlock
  drm/amd/display: Fix Edid emulation for linux
  drm/amd/display: Fix Vega10 lightup on S3 resume
  drm/amdgpu: Fix vce work queue was not cancelled when suspend
  Revert "drm/panel: Add device_link from panel device to DRM device"
  xen/blkfront: When purging persistent grants, keep them in the buffer
  clocksource/drivers/timer-atmel-pit: Properly handle error cases
  block: fix deadline elevator drain for zoned block devices
  ACPI / hotplug / PCI: Don't scan for non-hotplug bridges if slot is not bridge
  drm/syncobj: Don't leak fences when WAIT_FOR_SUBMIT is set
  ...

Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoLinux 4.19-rc6
Greg Kroah-Hartman [Sun, 30 Sep 2018 14:15:35 +0000 (07:15 -0700)]
Linux 4.19-rc6

6 years agoMerge tag 'auxdisplay-for-greg-v4.19-rc6' of https://github.com/ojeda/linux
Greg Kroah-Hartman [Sun, 30 Sep 2018 13:20:33 +0000 (06:20 -0700)]
Merge tag 'auxdisplay-for-greg-v4.19-rc6' of https://github.com/ojeda/linux

Miguel writes:
  "A trivial fix for auxdisplay

    - MAINTAINERS reference fix for moved file
      Reported by Joe Perches"

* tag 'auxdisplay-for-greg-v4.19-rc6' of https://github.com/ojeda/linux:
  MAINTAINERS: fix reference to moved drivers/{misc => auxdisplay}/panel.c

6 years agoMerge tag 'libnvdimm-fixes2-4.19-rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Greg Kroah-Hartman [Sun, 30 Sep 2018 13:19:38 +0000 (06:19 -0700)]
Merge tag 'libnvdimm-fixes2-4.19-rc6' of git://git./linux/kernel/git/nvdimm/nvdimm

Dan writes:
  "filesystem-dax for 4.19-rc6

   Fix a deadlock in the new for 4.19 dax_lock_mapping_entry() routine."

* tag 'libnvdimm-fixes2-4.19-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
  dax: Fix deadlock in dax_lock_mapping_entry()

6 years agoMAINTAINERS: fix reference to moved drivers/{misc => auxdisplay}/panel.c
Miguel Ojeda [Sun, 30 Sep 2018 11:50:05 +0000 (13:50 +0200)]
MAINTAINERS: fix reference to moved drivers/{misc => auxdisplay}/panel.c

Commit 51c1e9b554c9 ("auxdisplay: Move panel.c to drivers/auxdisplay folder")
moved the file, but the MAINTAINERS reference was not updated.

Link: https://lore.kernel.org/lkml/20180928220131.31075-1-joe@perches.com/
Reported-by: Joe Perches <joe@perches.com>
Signed-off-by: Miguel Ojeda <miguel.ojeda.sandonis@gmail.com>
6 years agoMerge tag 'for-linus-20180929' of git://git.kernel.dk/linux-block
Greg Kroah-Hartman [Sat, 29 Sep 2018 21:52:14 +0000 (14:52 -0700)]
Merge tag 'for-linus-20180929' of git://git.kernel.dk/linux-block

Jens writes:
  "Block fixes for 4.19-rc6

   A set of fixes that should go into this release. This pull request
   contains:

   - A fix (hopefully) for the persistent grants for xen-blkfront. A
     previous fix from this series wasn't complete, hence reverted, and
     this one should hopefully be it. (Boris Ostrovsky)

   - Fix for an elevator drain warning with SMR devices, which is
     triggered when you switch schedulers (Damien)

   - bcache deadlock fix (Guoju Fang)

   - Fix for the block unplug tracepoint, which has had the
     timer/explicit flag reverted since 4.11 (Ilya)

   - Fix a regression in this series where the blk-mq timeout hook is
     invoked with the RCU read lock held, hence preventing it from
     blocking (Keith)

   - NVMe pull from Christoph, with a single multipath fix (Susobhan Dey)"

* tag 'for-linus-20180929' of git://git.kernel.dk/linux-block:
  xen/blkfront: correct purging of persistent grants
  Revert "xen/blkfront: When purging persistent grants, keep them in the buffer"
  blk-mq: I/O and timer unplugs are inverted in blktrace
  bcache: add separate workqueue for journal_write to avoid deadlock
  xen/blkfront: When purging persistent grants, keep them in the buffer
  block: fix deadline elevator drain for zoned block devices
  blk-mq: Allow blocking queue tag iter callbacks
  nvme: properly propagate errors in nvme_mpath_init

6 years agoMerge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Greg Kroah-Hartman [Sat, 29 Sep 2018 21:34:06 +0000 (14:34 -0700)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Thomas writes:
  "A single fix for the AMD memory encryption boot code so it does not
   read random garbage instead of the cached encryption bit when a kexec
   kernel is allocated above the 32bit address limit."

* 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/boot: Fix kexec booting failure in the SEV bit detection code

6 years agoMerge branch 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Greg Kroah-Hartman [Sat, 29 Sep 2018 21:32:49 +0000 (14:32 -0700)]
Merge branch 'timers-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Thomas writes:
  "Three small fixes for clocksource drivers:
   - Proper error handling in the Atmel PIT driver
   - Add CLOCK_SOURCE_SUSPEND_NONSTOP for TI SoCs so suspend works again
   - Fix the next event function for Facebook Backpack-CMM BMC chips so
     usleep(100) doesnt sleep several milliseconds"

* 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  clocksource/drivers/timer-atmel-pit: Properly handle error cases
  clocksource/drivers/fttmr010: Fix set_next_event handler
  clocksource/drivers/ti-32k: Add CLOCK_SOURCE_SUSPEND_NONSTOP flag for non-am43 SoCs

6 years agoMerge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
Greg Kroah-Hartman [Sat, 29 Sep 2018 18:32:03 +0000 (11:32 -0700)]
Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Thomas writes:
  "A single fix for a missing sanity check when a pinned event is tried
  to be read on the wrong CPU due to a legit event scheduling failure."

* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  perf/core: Add sanity check to deal with pinned event failure

6 years agoMerge tag 'pm-4.19-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Greg Kroah-Hartman [Sat, 29 Sep 2018 13:50:36 +0000 (06:50 -0700)]
Merge tag 'pm-4.19-rc6' of git://git./linux/kernel/git/rafael/linux-pm

Rafael writes:
  "Power management fix for 4.19-rc6

   Fix incorrect __init and __exit annotations in the Qualcomm
   Kryo cpufreq driver (Nathan Chancellor)."

* tag 'pm-4.19-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  cpufreq: qcom-kryo: Fix section annotations

6 years agocpufreq: qcom-kryo: Fix section annotations
Nathan Chancellor [Thu, 20 Sep 2018 00:22:21 +0000 (17:22 -0700)]
cpufreq: qcom-kryo: Fix section annotations

There is currently a warning when building the Kryo cpufreq driver into
the kernel image:

WARNING: vmlinux.o(.text+0x8aa424): Section mismatch in reference from
the function qcom_cpufreq_kryo_probe() to the function
.init.text:qcom_cpufreq_kryo_get_msm_id()
The function qcom_cpufreq_kryo_probe() references
the function __init qcom_cpufreq_kryo_get_msm_id().
This is often because qcom_cpufreq_kryo_probe lacks a __init
annotation or the annotation of qcom_cpufreq_kryo_get_msm_id is wrong.

Remove the '__init' annotation from qcom_cpufreq_kryo_get_msm_id
so that there is no more mismatch warning.

Additionally, Nick noticed that the remove function was marked as
'__init' when it should really be marked as '__exit'.

Fixes: 46e2856b8e18 (cpufreq: Add Kryo CPU scaling driver)
Fixes: 5ad7346b4ae2 (cpufreq: kryo: Add module remove and exit)
Reported-by: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Cc: 4.18+ <stable@vger.kernel.org> # 4.18+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
6 years agoMerge tag 'dma-mapping-4.19-3' of git://git.infradead.org/users/hch/dma-mapping
Greg Kroah-Hartman [Sat, 29 Sep 2018 09:52:24 +0000 (02:52 -0700)]
Merge tag 'dma-mapping-4.19-3' of git://git.infradead.org/users/hch/dma-mapping

Christoph writes:
  "dma mapping fix for 4.19-rc6

   fix a missing Kconfig symbol for commits introduced in 4.19-rc"

* tag 'dma-mapping-4.19-3' of git://git.infradead.org/users/hch/dma-mapping:
  dma-mapping: add the missing ARCH_HAS_SYNC_DMA_FOR_CPU_ALL declaration

6 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Greg Kroah-Hartman [Sat, 29 Sep 2018 01:04:50 +0000 (18:04 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

Dmitry writes:
  "Input updates for v4.19-rc5

   Just a few driver fixes"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
  Input: uinput - allow for max == min during input_absinfo validation
  Input: elantech - enable middle button of touchpad on ThinkPad P72
  Input: atakbd - fix Atari CapsLock behaviour
  Input: atakbd - fix Atari keymap
  Input: egalax_ts - add system wakeup support
  Input: gpio-keys - fix a documentation index issue

6 years agoMerge tag 'spi-fix-v4.19-rc5' of https://git.kernel.org/pub/scm/linux/kernel/git...
Greg Kroah-Hartman [Sat, 29 Sep 2018 01:04:06 +0000 (18:04 -0700)]
Merge tag 'spi-fix-v4.19-rc5' of https://git./linux/kernel/git/broonie/spi

Mark writes:
  "spi: Fixes for v4.19

   Quite a few fixes for the Renesas drivers in here, plus a fix for the
   Tegra driver and some documentation fixes for the recently added
   spi-mem code.  The Tegra fix is relatively large but fairly
   straightforward and mechanical, it runs on probe so it's been
   reasonably well covered in -next testing."

* tag 'spi-fix-v4.19-rc5' of https://git.kernel.org/pub/scm/linux/kernel/git/broonie/spi:
  spi: spi-mem: Move the DMA-able constraint doc to the kerneldoc header
  spi: spi-mem: Add missing description for data.nbytes field
  spi: rspi: Fix interrupted DMA transfers
  spi: rspi: Fix invalid SPI use during system suspend
  spi: sh-msiof: Fix handling of write value for SISTR register
  spi: sh-msiof: Fix invalid SPI use during system suspend
  spi: gpio: Fix copy-and-paste error
  spi: tegra20-slink: explicitly enable/disable clock

6 years agoMerge tag 'regulator-v4.19-rc5' of https://git.kernel.org/pub/scm/linux/kernel/git...
Greg Kroah-Hartman [Sat, 29 Sep 2018 01:02:25 +0000 (18:02 -0700)]
Merge tag 'regulator-v4.19-rc5' of https://git./linux/kernel/git/broonie/regulator

Mark writes:
  "regulator: Fixes for 4.19

   A collection of fairly minor bug fixes here, a couple of driver
   specific ones plus two core fixes.  There's one fix for the new
   suspend state code which fixes some confusion with constant values
   that are supposed to indicate noop operation and another fixing a
   race condition with the creation of sysfs files on new regulators."

* tag 'regulator-v4.19-rc5' of https://git.kernel.org/pub/scm/linux/kernel/git/broonie/regulator:
  regulator: fix crash caused by null driver data
  regulator: Fix 'do-nothing' value for regulators without suspend state
  regulator: da9063: fix DT probing with constraints
  regulator: bd71837: Disable voltage monitoring for LDO3/4

6 years agoMerge tag 'powerpc-4.19-3' of https://git.kernel.org/pub/scm/linux/kernel/git/powerpc...
Greg Kroah-Hartman [Sat, 29 Sep 2018 00:43:32 +0000 (17:43 -0700)]
Merge tag 'powerpc-4.19-3' of https://git./linux/kernel/git/powerpc/linux

Michael writes:
  "powerpc fixes for 4.19 #3

   A reasonably big batch of fixes due to me being away for a few weeks.

   A fix for the TM emulation support on Power9, which could result in
   corrupting the guest r11 when running under KVM.

   Two fixes to the TM code which could lead to userspace GPR corruption
   if we take an SLB miss at exactly the wrong time.

   Our dynamic patching code had a bug that meant we could patch freed
   __init text, which could lead to corrupting userspace memory.

   csum_ipv6_magic() didn't work on little endian platforms since we
   optimised it recently.

   A fix for an endian bug when reading a device tree property telling
   us how many storage keys the machine has available.

   Fix a crash seen on some configurations of PowerVM when migrating the
   partition from one machine to another.

   A fix for a regression in the setup of our CPU to NUMA node mapping
   in KVM guests.

   A fix to our selftest Makefiles to make them work since a recent
   change to the shared Makefile logic."

* tag 'powerpc-4.19-3' of https://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
  selftests/powerpc: Fix Makefiles for headers_install change
  powerpc/numa: Use associativity if VPHN hcall is successful
  powerpc/tm: Avoid possible userspace r1 corruption on reclaim
  powerpc/tm: Fix userspace r13 corruption
  powerpc/pseries: Fix unitialized timer reset on migration
  powerpc/pkeys: Fix reading of ibm, processor-storage-keys property
  powerpc: fix csum_ipv6_magic() on little endian platforms
  powerpc/powernv/ioda2: Reduce upper limit for DMA window size (again)
  powerpc: Avoid code patching freed init sections
  KVM: PPC: Book3S HV: Fix guest r11 corruption with POWER9 TM workarounds

6 years agoMerge tag 'pinctrl-v4.19-4' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
Greg Kroah-Hartman [Sat, 29 Sep 2018 00:42:44 +0000 (17:42 -0700)]
Merge tag 'pinctrl-v4.19-4' of git://git./linux/kernel/git/linusw/linux-pinctrl

Linus writes:
  "Pin control fixes for v4.19:
   - Fixes to x86 hardware:
   - AMD interrupt debounce issues
   - Faulty Intel cannonlake register offset
   - Revert pin translation IRQ locking"

* tag 'pinctrl-v4.19-4' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw/linux-pinctrl:
  Revert "pinctrl: intel: Do pin translation when lock IRQ"
  pinctrl: cannonlake: Fix HOSTSW_OWN register offset of H variant
  pinctrl/amd: poll InterruptEnable bits in amd_gpio_irq_set_type

6 years agoperf/core: Add sanity check to deal with pinned event failure
Reinette Chatre [Wed, 19 Sep 2018 17:29:06 +0000 (10:29 -0700)]
perf/core: Add sanity check to deal with pinned event failure

It is possible that a failure can occur during the scheduling of a
pinned event. The initial portion of perf_event_read_local() contains
the various error checks an event should pass before it can be
considered valid. Ensure that the potential scheduling failure
of a pinned event is checked for and have a credible error.

Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: fenghua.yu@intel.com
Cc: tony.luck@intel.com
Cc: acme@kernel.org
Cc: gavin.hindman@intel.com
Cc: jithu.joseph@intel.com
Cc: dave.hansen@intel.com
Cc: hpa@zytor.com
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/6486385d1f30336e9973b24c8c65f5079543d3d3.1537377064.git.reinette.chatre@intel.com
6 years agoblk-iolatency: keep track of previous windows stats
Josef Bacik [Fri, 28 Sep 2018 17:45:43 +0000 (13:45 -0400)]
blk-iolatency: keep track of previous windows stats

We apply a smoothing to the scale changes in order to keep sawtoothy
behavior from occurring.  However our window for checking if we've
missed our target can sometimes be lower than the smoothing interval
(500ms), especially on faster drives like ssd's.  In order to deal with
this keep track of the running tally of the previous intervals that we
threw away because we had already done a scale event recently.

This is needed for the ssd case as these low latency drives will have
bursts of latency, and if it happens to be ok for the window that
directly follows the opening of the scale window we could unthrottle
when previous windows we were missing our target.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-iolatency: use a percentile approache for ssd's
Josef Bacik [Fri, 28 Sep 2018 17:45:42 +0000 (13:45 -0400)]
blk-iolatency: use a percentile approache for ssd's

We use an average latency approach for determining if we're missing our
latency target.  This works well for rotational storage where we have
generally consistent latencies, but for ssd's and other low latency
devices you have more of a spikey behavior, which means we often won't
throttle misbehaving groups because a lot of IO completes at drastically
faster times than our latency target.  Instead keep track of how many
IO's miss our target and how many IO's are done in our time window.  If
the p(90) latency is above our target then we know we need to throttle.
With this change in place we are seeing the same throttling behavior
with our testcase on ssd's as we see with rotational drives.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-iolatency: deal with small samples
Josef Bacik [Fri, 28 Sep 2018 17:45:41 +0000 (13:45 -0400)]
blk-iolatency: deal with small samples

There is logic to keep cgroups that haven't done a lot of IO in the most
recent scale window from being punished for over-active higher priority
groups.  However for things like ssd's where the windows are pretty
short we'll end up with small numbers of samples, so 5% of samples will
come out to 0 if there aren't enough.  Make the floor 1 sample to keep
us from improperly bailing out of scaling down.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-iolatency: deal with nr_requests == 1
Josef Bacik [Fri, 28 Sep 2018 17:45:40 +0000 (13:45 -0400)]
blk-iolatency: deal with nr_requests == 1

Hitting the case where blk_queue_depth() returned 1 uncovered the fact
that iolatency doesn't actually handle this case properly, it simply
doesn't scale down anybody.  For this case we should go straight into
applying the time delay, which we weren't doing.  Since we already limit
the floor at 1 request this if statement is not needed, and this allows
us to set our depth to 1 which allows us to apply the delay if needed.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblk-iolatency: use q->nr_requests directly
Josef Bacik [Fri, 28 Sep 2018 17:45:39 +0000 (13:45 -0400)]
blk-iolatency: use q->nr_requests directly

We were using blk_queue_depth() assuming that it would return
nr_requests, but we hit a case in production on drives that had to have
NCQ turned off in order for them to not shit the bed which resulted in a
qd of 1, even though the nr_requests was much larger.  iolatency really
only cares about requests we are allowed to queue up, as any io that
get's onto the request list is going to be serviced soonish, so we want
to be throttling before the bio gets onto the request list.  To make
iolatency work as expected, simply use q->nr_requests instead of
blk_queue_depth() as that is what we actually care about.

Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoMerge tag 'drm-fixes-2018-09-28' of git://anongit.freedesktop.org/drm/drm
Greg Kroah-Hartman [Fri, 28 Sep 2018 16:55:17 +0000 (18:55 +0200)]
Merge tag 'drm-fixes-2018-09-28' of git://anongit.freedesktop.org/drm/drm

Dave writes:
  "drm fixes for 4.19-rc6

   Looks like a pretty normal week for graphics,

   core: syncobj fix, panel link regression revert
   amd: suspend/resume fixes, EDID emulation fix
   mali-dp: NV12 writeback and vblank reset fixes
   etnaviv: DMA setup fix"

* tag 'drm-fixes-2018-09-28' of git://anongit.freedesktop.org/drm/drm:
  drm/amd/display: Fix Edid emulation for linux
  drm/amd/display: Fix Vega10 lightup on S3 resume
  drm/amdgpu: Fix vce work queue was not cancelled when suspend
  Revert "drm/panel: Add device_link from panel device to DRM device"
  drm/syncobj: Don't leak fences when WAIT_FOR_SUBMIT is set
  drm/malidp: Fix writeback in NV12
  drm: mali-dp: Call drm_crtc_vblank_reset on device init
  drm/etnaviv: add DMA configuration for etnaviv platform device

6 years agoMerge tag 'riscv-for-linus-4.19-rc6' of git://git.kernel.org/pub/scm/linux/kernel...
Greg Kroah-Hartman [Fri, 28 Sep 2018 16:53:22 +0000 (18:53 +0200)]
Merge tag 'riscv-for-linus-4.19-rc6' of git://git./linux/kernel/git/palmer/riscv-linux

Palmer writes:
  "A Single RISC-V Update for 4.19-rc6

   The Debian guys have been pushing on our port and found some
   unversioned symbols leaking into modules.  This PR contains a single
   fix for that issue."

* tag 'riscv-for-linus-4.19-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/palmer/riscv-linux:
  RISC-V: include linux/ftrace.h in asm-prototypes.h

6 years agokyber: fix integer overflow of latency targets on 32-bit
Omar Sandoval [Fri, 28 Sep 2018 16:22:50 +0000 (09:22 -0700)]
kyber: fix integer overflow of latency targets on 32-bit

NSEC_PER_SEC has type long, so 5 * NSEC_PER_SEC is calculated as a long.
However, 5 seconds is 5,000,000,000 nanoseconds, which overflows a
32-bit long. Make sure all of the targets are calculated as 64-bit
values.

Fixes: 6e25cb01ea20 ("kyber: implement improved heuristics")
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoMerge tag 'pci-v4.19-fixes-2' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git...
Greg Kroah-Hartman [Fri, 28 Sep 2018 16:20:41 +0000 (18:20 +0200)]
Merge tag 'pci-v4.19-fixes-2' of ssh://gitolite./linux/kernel/git/helgaas/pci

Bjorn writes:
  "PCI fixes:

  - Fix ACPI hotplug issue that causes black screen crash at boot (Mika
    Westerberg)

  - Fix DesignWare "scheduling while atomic" issues (Jisheng Zhang)

  - Add PPC contacts to MAINTAINERS for PCI core error handling (Bjorn
    Helgaas)

  - Sort Mobiveil MAINTAINERS entry (Lorenzo Pieralisi)"

* tag 'pci-v4.19-fixes-2' of ssh://gitolite.kernel.org/pub/scm/linux/kernel/git/helgaas/pci:
  ACPI / hotplug / PCI: Don't scan for non-hotplug bridges if slot is not bridge
  PCI: dwc: Fix scheduling while atomic issues
  MAINTAINERS: Move mobiveil PCI driver entry where it belongs
  MAINTAINERS: Update PPC contacts for PCI core error handling

6 years agoMerge branch 'nvme-4.19' of git://git.infradead.org/nvme into for-linus
Jens Axboe [Fri, 28 Sep 2018 15:41:40 +0000 (09:41 -0600)]
Merge branch 'nvme-4.19' of git://git.infradead.org/nvme into for-linus

Pull NVMe fix from Christoph.

* 'nvme-4.19' of git://git.infradead.org/nvme:
  nvme: properly propagate errors in nvme_mpath_init

6 years agoxen/blkfront: correct purging of persistent grants
Juergen Gross [Fri, 28 Sep 2018 07:28:27 +0000 (09:28 +0200)]
xen/blkfront: correct purging of persistent grants

Commit a46b53672b2c2e3770b38a4abf90d16364d2584b ("xen/blkfront: cleanup
stale persistent grants") introduced a regression as purged persistent
grants were not pu into the list of free grants again. Correct that.

Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoRevert "xen/blkfront: When purging persistent grants, keep them in the buffer"
Jens Axboe [Fri, 28 Sep 2018 15:40:17 +0000 (09:40 -0600)]
Revert "xen/blkfront: When purging persistent grants, keep them in the buffer"

Fix didn't work for all cases, reverting to add a (hopefully)
better fix.

This reverts commit f151ba989d149bbdfc90e5405724bbea094f9b17.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agovirtio-blk: modernize sysfs attribute creation
Hannes Reinecke [Fri, 28 Sep 2018 06:17:23 +0000 (08:17 +0200)]
virtio-blk: modernize sysfs attribute creation

Use new-style DEVICE_ATTR_RO/DEVICE_ATTR_RW to create the sysfs attributes
and register the disk with default sysfs attribute groups.

Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agozram: register default groups with device_add_disk()
Hannes Reinecke [Fri, 28 Sep 2018 06:17:22 +0000 (08:17 +0200)]
zram: register default groups with device_add_disk()

Register default sysfs groups during device_add_disk() to avoid a
race condition with udev during startup.

Signed-off-by: Hannes Reinecke <hare@suse.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Nitin Gupta <ngupta@vflare.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoaoe: register default groups with device_add_disk()
Hannes Reinecke [Fri, 28 Sep 2018 06:17:21 +0000 (08:17 +0200)]
aoe: register default groups with device_add_disk()

Register default sysfs groups during device_add_disk() to avoid a
race condition with udev during startup.

Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Acked-by: Ed L. Cachin <ed.cashin@acm.org>
Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agonvme: register ns_id attributes as default sysfs groups
Hannes Reinecke [Fri, 28 Sep 2018 06:17:20 +0000 (08:17 +0200)]
nvme: register ns_id attributes as default sysfs groups

We should be registering the ns_id attribute as default sysfs
attribute groups, otherwise we have a race condition between
the uevent and the attributes appearing in sysfs.

Suggested-by: Bart van Assche <bvanassche@acm.org>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Signed-off-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoblock: genhd: add 'groups' argument to device_add_disk
Hannes Reinecke [Fri, 28 Sep 2018 06:17:19 +0000 (08:17 +0200)]
block: genhd: add 'groups' argument to device_add_disk

Update device_add_disk() to take an 'groups' argument so that
individual drivers can register a device with additional sysfs
attributes.
This avoids race condition the driver would otherwise have if these
groups were to be created with sysfs_add_groups().

Signed-off-by: Martin Wilck <martin.wilck@suse.com>
Signed-off-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agoselftests/powerpc: Fix Makefiles for headers_install change
Michael Ellerman [Fri, 28 Sep 2018 04:53:18 +0000 (14:53 +1000)]
selftests/powerpc: Fix Makefiles for headers_install change

Commit b2d35fa5fc80 ("selftests: add headers_install to lib.mk")
introduced a requirement that Makefiles more than one level below the
selftests directory need to define top_srcdir, but it didn't update
any of the powerpc Makefiles.

This broke building all the powerpc selftests with eg:

  make[1]: Entering directory '/src/linux/tools/testing/selftests/powerpc'
  BUILD_TARGET=/src/linux/tools/testing/selftests/powerpc/alignment; mkdir -p $BUILD_TARGET; make OUTPUT=$BUILD_TARGET -k -C alignment all
  make[2]: Entering directory '/src/linux/tools/testing/selftests/powerpc/alignment'
  ../../lib.mk:20: ../../../../scripts/subarch.include: No such file or directory
  make[2]: *** No rule to make target '../../../../scripts/subarch.include'.
  make[2]: Failed to remake makefile '../../../../scripts/subarch.include'.
  Makefile:38: recipe for target 'alignment' failed

Fix it by setting top_srcdir in the affected Makefiles.

Fixes: b2d35fa5fc80 ("selftests: add headers_install to lib.mk")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
6 years agokyber: add tracepoints
Omar Sandoval [Thu, 27 Sep 2018 22:55:55 +0000 (15:55 -0700)]
kyber: add tracepoints

When debugging Kyber, it's really useful to know what latencies we've
been having, how the domain depths have been adjusted, and if we've
actually been throttling. Add three tracepoints, kyber_latency,
kyber_adjust, and kyber_throttled, to record that.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
6 years agokyber: implement improved heuristics
Omar Sandoval [Thu, 27 Sep 2018 22:55:54 +0000 (15:55 -0700)]
kyber: implement improved heuristics

Kyber's current heuristics have a few flaws:

- It's based on the mean latency, but p99 latency tends to be more
  meaningful to anyone who cares about latency. The mean can also be
  skewed by rare outliers that the scheduler can't do anything about.
- The statistics calculations are purely time-based with a short window.
  This works for steady, high load, but is more sensitive to outliers
  with bursty workloads.
- It only considers the latency once an I/O has been submitted to the
  device, but the user cares about the time spent in the kernel, as
  well.

These are shortcomings of the generic blk-stat code which doesn't quite
fit the ideal use case for Kyber. So, this replaces the statistics with
a histogram used to calculate percentiles of total latency and I/O
latency, which we then use to adjust depths in a slightly more
intelligent manner:

- Sync and async writes are now the same domain.
- Discards are a separate domain.
- Domain queue depths are scaled by the ratio of the p99 total latency
  to the target latency (e.g., if the p99 latency is double the target
  latency, we will double the queue depth; if the p99 latency is half of
  the target latency, we can halve the queue depth).
- We use the I/O latency to determine whether we should scale queue
  depths down: we will only scale down if any domain's I/O latency
  exceeds the target latency, which is an indicator of congestion in the
  device.

These new heuristics are just as scalable as the heuristics they
replace.

Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>