Skip to content

cloud-kernel
cloud-kernel

openanolis / cloud-kernel
1 年多 前同步成功

通知 160
Star 36
Fork 7
cloud-kernel
cloud-kernel
切换分支/标签
  1. 09 7月, 2018 2 次提交
    • B
      block: Document how blk_update_request() handles RQF_SPECIAL_PAYLOAD requests · 1954e9a9
      由 Bart Van Assche 提交于 
      The payload of struct request is stored in the request.bio chain if
the RQF_SPECIAL_PAYLOAD flag is not set and in request.special_vec if
RQF_SPECIAL_PAYLOAD has been set. However, blk_update_request()
iterates over req->bio whether or not RQF_SPECIAL_PAYLOAD has been
set. Additionally, the RQF_SPECIAL_PAYLOAD flag is ignored by
blk_rq_bytes() which means that the value returned by that function
is incorrect if the RQF_SPECIAL_PAYLOAD flag has been set. It is not
clear to me whether this is an oversight or whether this happened on
purpose. Anyway, document that it is known that both functions ignore
RQF_SPECIAL_PAYLOAD. See also commit f9d03f96 ("block: improve
handling of the magic discard payload").
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NBart Van Assche <bart.vanassche@wdc.com>
Cc: Ming Lei <ming.lei@redhat.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      1954e9a9
    • M
      blk-mq: avoid to synchronize rcu inside blk_cleanup_queue() · 1311326c
      由 Ming Lei 提交于 
      SCSI probing may synchronously create and destroy a lot of request_queues
for non-existent devices. Any synchronize_rcu() in queue creation or
destroy path may introduce long latency during booting, see detailed
description in comment of blk_register_queue().

This patch removes one synchronize_rcu() inside blk_cleanup_queue()
for this case, commit c2856ae2(blk-mq: quiesce queue before freeing queue)
needs synchronize_rcu() for implementing blk_mq_quiesce_queue(), but
when queue isn't initialized, it isn't necessary to do that since
only pass-through requests are involved, no original issue in
scsi_execute() at all.

Without this patch and previous one, it may take more 20+ seconds for
virtio-scsi to complete disk probe. With the two patches, the time becomes
less than 100ms.

Fixes: c2856ae2 ("blk-mq: quiesce queue before freeing queue")
Reported-by: NAndrew Jones <drjones@redhat.com>
Cc: Omar Sandoval <osandov@fb.com>
Cc: Bart Van Assche <bart.vanassche@wdc.com>
Cc: linux-scsi@vger.kernel.org
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Cc: Christoph Hellwig <hch@lst.de>
Tested-by: NAndrew Jones <drjones@redhat.com>
Signed-off-by: NMing Lei <ming.lei@redhat.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      1311326c
  3. 28 6月, 2018 1 次提交
  5. 20 6月, 2018 1 次提交
    • B
      Revert "block: Add warning for bi_next not NULL in bio_endio()" · 9c24c10a
      由 Bart Van Assche 提交于 
      Commit 0ba99ca4 ("block: Add warning for bi_next not NULL in
bio_endio()") breaks the dm driver. end_clone_bio() detects whether
or not a bio is the last bio associated with a request by checking
the .bi_next field. Commit 0ba99ca4 clears that field before
end_clone_bio() has had a chance to inspect that field. Hence revert
commit 0ba99ca4.

This patch avoids that KASAN reports the following complaint when
running the srp-test software (srp-test/run_tests -c -d -r 10 -t 02-mq):

==================================================================
BUG: KASAN: use-after-free in bio_advance+0x11b/0x1d0
Read of size 4 at addr ffff8801300e06d0 by task ksoftirqd/0/9

CPU: 0 PID: 9 Comm: ksoftirqd/0 Not tainted 4.18.0-rc1-dbg+ #1
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.0.0-prebuilt.qemu-project.org 04/01/2014
Call Trace:
 dump_stack+0xa4/0xf5
 print_address_description+0x6f/0x270
 kasan_report+0x241/0x360
 __asan_load4+0x78/0x80
 bio_advance+0x11b/0x1d0
 blk_update_request+0xa7/0x5b0
 scsi_end_request+0x56/0x320 [scsi_mod]
 scsi_io_completion+0x7d6/0xb20 [scsi_mod]
 scsi_finish_command+0x1c0/0x280 [scsi_mod]
 scsi_softirq_done+0x19a/0x230 [scsi_mod]
 blk_mq_complete_request+0x160/0x240
 scsi_mq_done+0x50/0x1a0 [scsi_mod]
 srp_recv_done+0x515/0x1330 [ib_srp]
 __ib_process_cq+0xa0/0xf0 [ib_core]
 ib_poll_handler+0x38/0xa0 [ib_core]
 irq_poll_softirq+0xe8/0x1f0
 __do_softirq+0x128/0x60d
 run_ksoftirqd+0x3f/0x60
 smpboot_thread_fn+0x352/0x460
 kthread+0x1c1/0x1e0
 ret_from_fork+0x24/0x30

Allocated by task 1918:
 save_stack+0x43/0xd0
 kasan_kmalloc+0xad/0xe0
 kasan_slab_alloc+0x11/0x20
 kmem_cache_alloc+0xfe/0x350
 mempool_alloc_slab+0x15/0x20
 mempool_alloc+0xfb/0x270
 bio_alloc_bioset+0x244/0x350
 submit_bh_wbc+0x9c/0x2f0
 __block_write_full_page+0x299/0x5a0
 block_write_full_page+0x16b/0x180
 blkdev_writepage+0x18/0x20
 __writepage+0x42/0x80
 write_cache_pages+0x376/0x8a0
 generic_writepages+0xbe/0x110
 blkdev_writepages+0xe/0x10
 do_writepages+0x9b/0x180
 __filemap_fdatawrite_range+0x178/0x1c0
 file_write_and_wait_range+0x59/0xc0
 blkdev_fsync+0x46/0x80
 vfs_fsync_range+0x66/0x100
 do_fsync+0x3d/0x70
 __x64_sys_fsync+0x21/0x30
 do_syscall_64+0x77/0x230
 entry_SYSCALL_64_after_hwframe+0x49/0xbe

Freed by task 9:
 save_stack+0x43/0xd0
 __kasan_slab_free+0x137/0x190
 kasan_slab_free+0xe/0x10
 kmem_cache_free+0xd3/0x380
 mempool_free_slab+0x17/0x20
 mempool_free+0x63/0x160
 bio_free+0x81/0xa0
 bio_put+0x59/0x60
 end_bio_bh_io_sync+0x5d/0x70
 bio_endio+0x1a7/0x360
 blk_update_request+0xd0/0x5b0
 end_clone_bio+0xa3/0xd0 [dm_mod]
 bio_endio+0x1a7/0x360
 blk_update_request+0xd0/0x5b0
 scsi_end_request+0x56/0x320 [scsi_mod]
 scsi_io_completion+0x7d6/0xb20 [scsi_mod]
 scsi_finish_command+0x1c0/0x280 [scsi_mod]
 scsi_softirq_done+0x19a/0x230 [scsi_mod]
 blk_mq_complete_request+0x160/0x240
 scsi_mq_done+0x50/0x1a0 [scsi_mod]
 srp_recv_done+0x515/0x1330 [ib_srp]
 __ib_process_cq+0xa0/0xf0 [ib_core]
 ib_poll_handler+0x38/0xa0 [ib_core]
 irq_poll_softirq+0xe8/0x1f0
 __do_softirq+0x128/0x60d

The buggy address belongs to the object at ffff8801300e0640
 which belongs to the cache bio-0 of size 200
The buggy address is located 144 bytes inside of
 200-byte region [ffff8801300e0640, ffff8801300e0708)
The buggy address belongs to the page:
page:ffffea0004c03800 count:1 mapcount:0 mapping:ffff88015a563a00 index:0x0 compound_mapcount: 0
flags: 0x8000000000008100(slab|head)
raw: 8000000000008100 dead000000000100 dead000000000200 ffff88015a563a00
raw: 0000000000000000 0000000000330033 00000001ffffffff 0000000000000000
page dumped because: kasan: bad access detected

Memory state around the buggy address:
 ffff8801300e0580: fb fb fb fb fb fb fb fb fb fc fc fc fc fc fc fc
 ffff8801300e0600: fc fc fc fc fc fc fc fc fb fb fb fb fb fb fb fb
>ffff8801300e0680: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
                                                 ^
 ffff8801300e0700: fb fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
 ffff8801300e0780: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
==================================================================

Cc: Kent Overstreet <kent.overstreet@gmail.com>
Fixes: 0ba99ca4 ("block: Add warning for bi_next not NULL in bio_endio()")
Acked-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NBart Van Assche <bart.vanassche@wdc.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      9c24c10a
  7. 07 6月, 2018 1 次提交
  9. 03 6月, 2018 1 次提交
    • J
      block: don't use blocking queue entered for recursive bio submits · cd4a4ae4
      由 Jens Axboe 提交于 
      If we end up splitting a bio and the queue goes away between
the initial submission and the later split submission, then we
can block forever in blk_queue_enter() waiting for the reference
to drop to zero. This will never happen, since we already hold
a reference.

Mark a split bio as already having entered the queue, so we can
just use the live non-blocking queue enter variant.

Thanks to Tetsuo Handa for the analysis.

Reported-by: syzbot+c4f9cebf9d651f6e54de@syzkaller.appspotmail.com
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      cd4a4ae4
  11. 01 6月, 2018 3 次提交
  13. 31 5月, 2018 1 次提交
  15. 29 5月, 2018 1 次提交
    • K
      blk-mq: Remove generation seqeunce · 12f5b931
      由 Keith Busch 提交于 
      This patch simplifies the timeout handling by relying on the request
reference counting to ensure the iterator is operating on an inflight
and truly timed out request. Since the reference counting prevents the
tag from being reallocated, the block layer no longer needs to prevent
drivers from completing their requests while the timeout handler is
operating on it: a driver completing a request is allowed to proceed to
the next state without additional syncronization with the block layer.

This also removes any need for generation sequence numbers since the
request lifetime is prevented from being reallocated as a new sequence
while timeout handling is operating on it.

To enables this a refcount is added to struct request so that request
users can be sure they're operating on the same request without it
changing while they're processing it.  The request's tag won't be
released for reuse until both the timeout handler and the completion
are done with it.
Signed-off-by: NKeith Busch <keith.busch@intel.com>
[hch: slight cleanups, added back submission side hctx lock, use cmpxchg
 for completions]
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      12f5b931
  17. 15 5月, 2018 2 次提交
  19. 14 5月, 2018 4 次提交
  21. 09 5月, 2018 3 次提交
  23. 08 5月, 2018 2 次提交
    • T
      block: Shorten interrupt disabled regions · 50864670
      由 Thomas Gleixner 提交于 
      Commit 9c40cef2 ("sched: Move blk_schedule_flush_plug() out of
__schedule()") moved the blk_schedule_flush_plug() call out of the
interrupt/preempt disabled region in the scheduler. This allows to replace
local_irq_save/restore(flags) by local_irq_disable/enable() in
blk_flush_plug_list().

But it makes more sense to disable interrupts explicitly when the request
queue is locked end reenable them when the request to is unlocked. This
shortens the interrupt disabled section which is important when the plug
list contains requests for more than one queue. The comment which claims
that disabling interrupts around the loop is misleading as the called
functions can reenable interrupts unconditionally anyway and obfuscates the
scope badly:

 local_irq_save(flags);
   spin_lock(q->queue_lock);
   ...
   queue_unplugged(q...);
     scsi_request_fn();
       spin_unlock_irq(q->queue_lock);

-------------------^^^ ????

       spin_lock_irq(q->queue_lock);
     spin_unlock(q->queue_lock);
 local_irq_restore(flags);

Aside of that the detached interrupt disabling is a constant pain for
PREEMPT_RT as it requires patching and special casing when RT is enabled
while with the spin_*_irq() variants this happens automatically.
Signed-off-by: NThomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/20110622174919.025446432@linutronix.deSigned-off-by: NSebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      50864670
    • A
      block: Remove redundant WARN_ON() · 656cb6d0
      由 Anna-Maria Gleixner 提交于 
      Commit 2fff8a92 ("block: Check locking assumptions at runtime") added a
lockdep_assert_held(q->queue_lock) which makes the WARN_ON() redundant
because lockdep will detect and warn about context violations.

The unconditional WARN_ON() does not provide real additional value, so it
can be removed.
Signed-off-by: NAnna-Maria Gleixner <anna-maria@linutronix.de>
Signed-off-by: NSebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      656cb6d0
  25. 17 4月, 2018 1 次提交
    • J
      blk-mq: start request gstate with gen 1 · f4560231
      由 Jianchao Wang 提交于 
      rq->gstate and rq->aborted_gstate both are zero before rqs are
allocated. If we have a small timeout, when the timer fires,
there could be rqs that are never allocated, and also there could
be rq that has been allocated but not initialized and started. At
the moment, the rq->gstate and rq->aborted_gstate both are 0, thus
the blk_mq_terminate_expired will identify the rq is timed out and
invoke .timeout early.

For scsi, this will cause scsi_times_out to be invoked before the
scsi_cmnd is not initialized, scsi_cmnd->device is still NULL at
the moment, then we will get crash.

Cc: Bart Van Assche <bart.vanassche@wdc.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: Martin Steigerwald <Martin@Lichtvoll.de>
Cc: stable@vger.kernel.org
Signed-off-by: NJianchao Wang <jianchao.w.wang@oracle.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      f4560231
  27. 15 4月, 2018 1 次提交
    • A
      block: do not use interruptible wait anywhere · 1dc3039b
      由 Alan Jenkins 提交于 
      When blk_queue_enter() waits for a queue to unfreeze, or unset the
PREEMPT_ONLY flag, do not allow it to be interrupted by a signal.

The PREEMPT_ONLY flag was introduced later in commit 3a0a5299
("block, scsi: Make SCSI quiesce and resume work reliably").  Note the SCSI
device is resumed asynchronously, i.e. after un-freezing userspace tasks.

So that commit exposed the bug as a regression in v4.15.  A mysterious
SIGBUS (or -EIO) sometimes happened during the time the device was being
resumed.  Most frequently, there was no kernel log message, and we saw Xorg
or Xwayland killed by SIGBUS.[1]

[1] E.g. https://bugzilla.redhat.com/show_bug.cgi?id=1553979

Without this fix, I get an IO error in this test:

# dd if=/dev/sda of=/dev/null iflag=direct & \
  while killall -SIGUSR1 dd; do sleep 0.1; done & \
  echo mem > /sys/power/state ; \
  sleep 5; killall dd  # stop after 5 seconds

The interruptible wait was added to blk_queue_enter in
commit 3ef28e83 ("block: generic request_queue reference counting").
Before then, the interruptible wait was only in blk-mq, but I don't think
it could ever have been correct.
Reviewed-by: NBart Van Assche <bart.vanassche@wdc.com>
Cc: stable@vger.kernel.org
Signed-off-by: NAlan Jenkins <alan.christopher.jenkins@gmail.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      1dc3039b
  29. 11 4月, 2018 1 次提交
  31. 20 3月, 2018 1 次提交
    • B
      block: Change a rcu_read_{lock,unlock}_sched() pair into rcu_read_{lock,unlock}() · 818e0fa2
      由 Bart Van Assche 提交于 
      scsi_device_quiesce() uses synchronize_rcu() to guarantee that the
effect of blk_set_preempt_only() will be visible for percpu_ref_tryget()
calls that occur after the queue unfreeze by using the approach
explained in https://lwn.net/Articles/573497/. The rcu read lock and
unlock calls in blk_queue_enter() form a pair with the synchronize_rcu()
call in scsi_device_quiesce(). Both scsi_device_quiesce() and
blk_queue_enter() must either use regular RCU or RCU-sched.
Since neither the RCU-protected code in blk_queue_enter() nor
blk_queue_usage_counter_release() sleeps, regular RCU protection
is sufficient. Note: scsi_device_quiesce() does not have to be
modified since it already uses synchronize_rcu().
Reported-by: NTejun Heo <tj@kernel.org>
Fixes: 3a0a5299 ("block, scsi: Make SCSI quiesce and resume work reliably")
Signed-off-by: NBart Van Assche <bart.vanassche@wdc.com>
Acked-by: NTejun Heo <tj@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Cc: Hannes Reinecke <hare@suse.com>
Cc: Ming Lei <ming.lei@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Johannes Thumshirn <jthumshirn@suse.de>
Cc: Oleksandr Natalenko <oleksandr@natalenko.name>
Cc: Martin Steigerwald <martin@lichtvoll.de>
Cc: stable@vger.kernel.org # v4.15
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      818e0fa2
  33. 18 3月, 2018 1 次提交
  35. 09 3月, 2018 2 次提交
  37. 01 3月, 2018 4 次提交
  39. 07 2月, 2018 1 次提交
    • H
      block: Add should_fail_bio() for bpf error injection · 30abb3a6
      由 Howard McLauchlan 提交于 
      The classic error injection mechanism, should_fail_request() does not
support use cases where more information is required (from the entire
struct bio, for example).

To that end, this patch introduces should_fail_bio(), which calls
should_fail_request() under the hood but provides a convenient
place for kprobes to hook into if they require the entire struct bio.
This patch also replaces some existing calls to should_fail_request()
with should_fail_bio() with no degradation in performance.
Signed-off-by: NHoward McLauchlan <hmclauchlan@fb.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      30abb3a6
  41. 02 2月, 2018 1 次提交
    • J
      blk-mq: fix discard merge with scheduler attached · 445251d0
      由 Jens Axboe 提交于 
      I ran into an issue on my laptop that triggered a bug on the
discard path:

WARNING: CPU: 2 PID: 207 at drivers/nvme/host/core.c:527 nvme_setup_cmd+0x3d3/0x430
 Modules linked in: rfcomm fuse ctr ccm bnep arc4 binfmt_misc snd_hda_codec_hdmi nls_iso8859_1 nls_cp437 vfat snd_hda_codec_conexant fat snd_hda_codec_generic iwlmvm snd_hda_intel snd_hda_codec snd_hwdep mac80211 snd_hda_core snd_pcm snd_seq_midi snd_seq_midi_event snd_rawmidi snd_seq x86_pkg_temp_thermal intel_powerclamp kvm_intel uvcvideo iwlwifi btusb snd_seq_device videobuf2_vmalloc btintel videobuf2_memops kvm snd_timer videobuf2_v4l2 bluetooth irqbypass videobuf2_core aesni_intel aes_x86_64 crypto_simd cryptd snd glue_helper videodev cfg80211 ecdh_generic soundcore hid_generic usbhid hid i915 psmouse e1000e ptp pps_core xhci_pci xhci_hcd intel_gtt
 CPU: 2 PID: 207 Comm: jbd2/nvme0n1p7- Tainted: G     U           4.15.0+ #176
 Hardware name: LENOVO 20FBCTO1WW/20FBCTO1WW, BIOS N1FET59W (1.33 ) 12/19/2017
 RIP: 0010:nvme_setup_cmd+0x3d3/0x430
 RSP: 0018:ffff880423e9f838 EFLAGS: 00010217
 RAX: 0000000000000000 RBX: ffff880423e9f8c8 RCX: 0000000000010000
 RDX: ffff88022b200010 RSI: 0000000000000002 RDI: 00000000327f0000
 RBP: ffff880421251400 R08: ffff88022b200000 R09: 0000000000000009
 R10: 0000000000000000 R11: 0000000000000000 R12: 000000000000ffff
 R13: ffff88042341e280 R14: 000000000000ffff R15: ffff880421251440
 FS:  0000000000000000(0000) GS:ffff880441500000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 000055b684795030 CR3: 0000000002e09006 CR4: 00000000001606e0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
 Call Trace:
  nvme_queue_rq+0x40/0xa00
  ? __sbitmap_queue_get+0x24/0x90
  ? blk_mq_get_tag+0xa3/0x250
  ? wait_woken+0x80/0x80
  ? blk_mq_get_driver_tag+0x97/0xf0
  blk_mq_dispatch_rq_list+0x7b/0x4a0
  ? deadline_remove_request+0x49/0xb0
  blk_mq_do_dispatch_sched+0x4f/0xc0
  blk_mq_sched_dispatch_requests+0x106/0x170
  __blk_mq_run_hw_queue+0x53/0xa0
  __blk_mq_delay_run_hw_queue+0x83/0xa0
  blk_mq_run_hw_queue+0x6c/0xd0
  blk_mq_sched_insert_request+0x96/0x140
  __blk_mq_try_issue_directly+0x3d/0x190
  blk_mq_try_issue_directly+0x30/0x70
  blk_mq_make_request+0x1a4/0x6a0
  generic_make_request+0xfd/0x2f0
  ? submit_bio+0x5c/0x110
  submit_bio+0x5c/0x110
  ? __blkdev_issue_discard+0x152/0x200
  submit_bio_wait+0x43/0x60
  ext4_process_freed_data+0x1cd/0x440
  ? account_page_dirtied+0xe2/0x1a0
  ext4_journal_commit_callback+0x4a/0xc0
  jbd2_journal_commit_transaction+0x17e2/0x19e0
  ? kjournald2+0xb0/0x250
  kjournald2+0xb0/0x250
  ? wait_woken+0x80/0x80
  ? commit_timeout+0x10/0x10
  kthread+0x111/0x130
  ? kthread_create_worker_on_cpu+0x50/0x50
  ? do_group_exit+0x3a/0xa0
  ret_from_fork+0x1f/0x30
 Code: 73 89 c1 83 ce 10 c1 e1 10 09 ca 83 f8 04 0f 87 0f ff ff ff 8b 4d 20 48 8b 7d 00 c1 e9 09 48 01 8c c7 00 08 00 00 e9 f8 fe ff ff <0f> ff 4c 89 c7 41 bc 0a 00 00 00 e8 0d 78 d6 ff e9 a1 fc ff ff
 ---[ end trace 50d361cc444506c8 ]---
 print_req_error: I/O error, dev nvme0n1, sector 847167488

Decoding the assembly, the request claims to have 0xffff segments,
while nvme counts two. This turns out to be because we don't check
for a data carrying request on the mq scheduler path, and since
blk_phys_contig_segment() returns true for a non-data request,
we decrement the initial segment count of 0 and end up with
0xffff in the unsigned short.

There are a few issues here:

1) We should initialize the segment count for a discard to 1.
2) The discard merging is currently using the data limits for
   segments and sectors.

Fix this up by having attempt_merge() correctly identify the
request, and by initializing the segment count correctly
for discards.

This can only be triggered with mq-deadline on discard capable
devices right now, which isn't a common configuration.
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      445251d0
  43. 31 1月, 2018 1 次提交
    • M
      blk-mq: introduce BLK_STS_DEV_RESOURCE · 86ff7c2a
      由 Ming Lei 提交于 
      This status is returned from driver to block layer if device related
resource is unavailable, but driver can guarantee that IO dispatch
will be triggered in future when the resource is available.

Convert some drivers to return BLK_STS_DEV_RESOURCE.  Also, if driver
returns BLK_STS_RESOURCE and SCHED_RESTART is set, rerun queue after
a delay (BLK_MQ_DELAY_QUEUE) to avoid IO stalls.  BLK_MQ_DELAY_QUEUE is
3 ms because both scsi-mq and nvmefc are using that magic value.

If a driver can make sure there is in-flight IO, it is safe to return
BLK_STS_DEV_RESOURCE because:

1) If all in-flight IOs complete before examining SCHED_RESTART in
blk_mq_dispatch_rq_list(), SCHED_RESTART must be cleared, so queue
is run immediately in this case by blk_mq_dispatch_rq_list();

2) if there is any in-flight IO after/when examining SCHED_RESTART
in blk_mq_dispatch_rq_list():
- if SCHED_RESTART isn't set, queue is run immediately as handled in 1)
- otherwise, this request will be dispatched after any in-flight IO is
  completed via blk_mq_sched_restart()

3) if SCHED_RESTART is set concurently in context because of
BLK_STS_RESOURCE, blk_mq_delay_run_hw_queue() will cover the above two
cases and make sure IO hang can be avoided.

One invariant is that queue will be rerun if SCHED_RESTART is set.
Suggested-by: NJens Axboe <axboe@kernel.dk>
Tested-by: NLaurence Oberman <loberman@redhat.com>
Signed-off-by: NMing Lei <ming.lei@redhat.com>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      86ff7c2a
  45. 20 1月, 2018 2 次提交
  47. 19 1月, 2018 1 次提交
  49. 18 1月, 2018 1 次提交
    • M
      blk-mq: improve DM's blk-mq IO merging via blk_insert_cloned_request feedback · 396eaf21
      由 Ming Lei 提交于 
      blk_insert_cloned_request() is called in the fast path of a dm-rq driver
(e.g. blk-mq request-based DM mpath).  blk_insert_cloned_request() uses
blk_mq_request_bypass_insert() to directly append the request to the
blk-mq hctx->dispatch_list of the underlying queue.

1) This way isn't efficient enough because the hctx spinlock is always
used.

2) With blk_insert_cloned_request(), we completely bypass underlying
queue's elevator and depend on the upper-level dm-rq driver's elevator
to schedule IO.  But dm-rq currently can't get the underlying queue's
dispatch feedback at all.  Without knowing whether a request was issued
or not (e.g. due to underlying queue being busy) the dm-rq elevator will
not be able to provide effective IO merging (as a side-effect of dm-rq
currently blindly destaging a request from its elevator only to requeue
it after a delay, which kills any opportunity for merging).  This
obviously causes very bad sequential IO performance.

Fix this by updating blk_insert_cloned_request() to use
blk_mq_request_direct_issue().  blk_mq_request_direct_issue() allows a
request to be issued directly to the underlying queue and returns the
dispatch feedback (blk_status_t).  If blk_mq_request_direct_issue()
returns BLK_SYS_RESOURCE the dm-rq driver will now use DM_MAPIO_REQUEUE
to _not_ destage the request.  Whereby preserving the opportunity to
merge IO.

With this, request-based DM's blk-mq sequential IO performance is vastly
improved (as much as 3X in mpath/virtio-scsi testing).
Signed-off-by: NMing Lei <ming.lei@redhat.com>
[blk-mq.c changes heavily influenced by Ming Lei's initial solution, but
they were refactored to make them less fragile and easier to read/review]
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NJens Axboe <axboe@kernel.dk>
      396eaf21