[Devel] [PATCH RH7] ms/workqueue: fix ghost PENDING flag while doing MQ IO
Pavel Tikhomirov
ptikhomirov at virtuozzo.com
Thu Aug 24 16:00:03 MSK 2017
From: Roman Pen <roman.penyaev at profitbricks.com>
We have the hole node hang, many processes hang on similar stack as here:
crash> ps -m ffff8802b7f00000
[0 00:20:36.663] [UN] PID: 22713 TASK: ffff8802b7f00000 CPU: 1 COMMAND: "worker"
crash> bt ffff8802b7f00000
PID: 22713 TASK: ffff8802b7f00000 CPU: 1 COMMAND: "worker"
#0 [ffff88031b04f980] __schedule at ffffffff8256cdd1
#1 [ffff88031b04f9f8] schedule at ffffffff8256e239
#2 [ffff88031b04fa18] schedule_timeout at ffffffff82561cea
#3 [ffff88031b04fb88] io_schedule_timeout at ffffffff8256c0d9
#4 [ffff88031b04fbb8] wait_for_completion_io at ffffffff8256f3e0
#5 [ffff88031b04fc90] blkdev_issue_flush at ffffffff8193a207
#6 [ffff88031b04fe08] ext4_sync_file at ffffffffa0af6d34 [ext4]
#7 [ffff88031b04fe68] vfs_fsync_range at ffffffff8173212c
#8 [ffff88031b04fec8] do_fsync at ffffffff817330dc
#9 [ffff88031b04ff68] sys_fdatasync at ffffffff8173437e
RIP: 00007f474a581ddd RSP: 00007f46ba3fe8a0 RFLAGS: 00000282
RAX: 000000000000004b RBX: ffffffff8258f609 RCX: ffffffffffffffff
RDX: 00007f4754ffd458 RSI: 0000000000000000 RDI: 0000000000000011
RBP: 0000000000000000 R8: 0000000000000000 R9: 00000000000058b9
R10: 00007f46ba3fe8b0 R11: 0000000000000293 R12: 00007f475be25d80
R13: ffffffff8173437e R14: ffff88031b04ff78 R15: 00007f4755141452
ORIG_RAX: 000000000000004b CS: 0033 SS: 002b
crash> ps -m ffff8802b7f00000
[0 00:20:36.663] [UN] PID: 22713 TASK: ffff8802b7f00000 CPU: 1 COMMAND: "worker"
Sleeps for 20 minutes on bio completion:
blkdev_issue_flush:
submit_bio(WRITE_FLUSH, bio);
here> wait_for_completion_io(&wait);
As bio->bi_rw = (WRITE | REQ_SYNC | REQ_NOIDLE | REQ_FLUSH), we had:
submit_bio->generic_make_request->dm_make_request->queue_io->queue_work
So in wait_for_completion_io we wait for dm_wq_work to complete these
bio. But work is not in the workqueue already, as work->entry is empty
list, so the work seem completed. That could happen only if md->flags
had DMF_BLOCK_IO_FOR_SUSPEND bit set. But it is already unset, when we
clear the bit we queue another dm_wq_work on these wq in dm_queue_flush.
So what could've happened here is that operation reordering loads
DMF_BLOCK_IO_FOR_SUSPEND bit in dm_wq_work before it was cleared in
dm_queue_flush. Adding smp_mb in set_work_pool_and_clear_pending should
order operations properly.
https://jira.sw.ru/browse/PSBM-69788
original commit message:
The bug in a workqueue leads to a stalled IO request in MQ ctx->rq_list
with the following backtrace:
[ 601.347452] INFO: task kworker/u129:5:1636 blocked for more than 120 seconds.
[ 601.347574] Tainted: G O 4.4.5-1-storage+ #6
[ 601.347651] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 601.348142] kworker/u129:5 D ffff880803077988 0 1636 2 0x00000000
[ 601.348519] Workqueue: ibnbd_server_fileio_wq ibnbd_dev_file_submit_io_worker [ibnbd_server]
[ 601.348999] ffff880803077988 ffff88080466b900 ffff8808033f9c80 ffff880803078000
[ 601.349662] ffff880807c95000 7fffffffffffffff ffffffff815b0920 ffff880803077ad0
[ 601.350333] ffff8808030779a0 ffffffff815b01d5 0000000000000000 ffff880803077a38
[ 601.350965] Call Trace:
[ 601.351203] [<ffffffff815b0920>] ? bit_wait+0x60/0x60
[ 601.351444] [<ffffffff815b01d5>] schedule+0x35/0x80
[ 601.351709] [<ffffffff815b2dd2>] schedule_timeout+0x192/0x230
[ 601.351958] [<ffffffff812d43f7>] ? blk_flush_plug_list+0xc7/0x220
[ 601.352208] [<ffffffff810bd737>] ? ktime_get+0x37/0xa0
[ 601.352446] [<ffffffff815b0920>] ? bit_wait+0x60/0x60
[ 601.352688] [<ffffffff815af784>] io_schedule_timeout+0xa4/0x110
[ 601.352951] [<ffffffff815b3a4e>] ? _raw_spin_unlock_irqrestore+0xe/0x10
[ 601.353196] [<ffffffff815b093b>] bit_wait_io+0x1b/0x70
[ 601.353440] [<ffffffff815b056d>] __wait_on_bit+0x5d/0x90
[ 601.353689] [<ffffffff81127bd0>] wait_on_page_bit+0xc0/0xd0
[ 601.353958] [<ffffffff81096db0>] ? autoremove_wake_function+0x40/0x40
[ 601.354200] [<ffffffff81127cc4>] __filemap_fdatawait_range+0xe4/0x140
[ 601.354441] [<ffffffff81127d34>] filemap_fdatawait_range+0x14/0x30
[ 601.354688] [<ffffffff81129a9f>] filemap_write_and_wait_range+0x3f/0x70
[ 601.354932] [<ffffffff811ced3b>] blkdev_fsync+0x1b/0x50
[ 601.355193] [<ffffffff811c82d9>] vfs_fsync_range+0x49/0xa0
[ 601.355432] [<ffffffff811cf45a>] blkdev_write_iter+0xca/0x100
[ 601.355679] [<ffffffff81197b1a>] __vfs_write+0xaa/0xe0
[ 601.355925] [<ffffffff81198379>] vfs_write+0xa9/0x1a0
[ 601.356164] [<ffffffff811c59d8>] kernel_write+0x38/0x50
The underlying device is a null_blk, with default parameters:
queue_mode = MQ
submit_queues = 1
Verification that nullb0 has something inflight:
root at pserver8:~# cat /sys/block/nullb0/inflight
0 1
root at pserver8:~# find /sys/block/nullb0/mq/0/cpu* -name rq_list -print -exec cat {} \;
...
/sys/block/nullb0/mq/0/cpu2/rq_list
CTX pending:
ffff8838038e2400
...
During debug it became clear that stalled request is always inserted in
the rq_list from the following path:
save_stack_trace_tsk + 34
blk_mq_insert_requests + 231
blk_mq_flush_plug_list + 281
blk_flush_plug_list + 199
wait_on_page_bit + 192
__filemap_fdatawait_range + 228
filemap_fdatawait_range + 20
filemap_write_and_wait_range + 63
blkdev_fsync + 27
vfs_fsync_range + 73
blkdev_write_iter + 202
__vfs_write + 170
vfs_write + 169
kernel_write + 56
So blk_flush_plug_list() was called with from_schedule == true.
If from_schedule is true, that means that finally blk_mq_insert_requests()
offloads execution of __blk_mq_run_hw_queue() and uses kblockd workqueue,
i.e. it calls kblockd_schedule_delayed_work_on().
That means, that we race with another CPU, which is about to execute
__blk_mq_run_hw_queue() work.
Further debugging shows the following traces from different CPUs:
CPU#0 CPU#1
---------------------------------- -------------------------------
reqeust A inserted
STORE hctx->ctx_map[0] bit marked
kblockd_schedule...() returns 1
<schedule to kblockd workqueue>
request B inserted
STORE hctx->ctx_map[1] bit marked
kblockd_schedule...() returns 0
*** WORK PENDING bit is cleared ***
flush_busy_ctxs() is executed, but
bit 1, set by CPU#1, is not observed
As a result request B pended forever.
This behaviour can be explained by speculative LOAD of hctx->ctx_map on
CPU#0, which is reordered with clear of PENDING bit and executed _before_
actual STORE of bit 1 on CPU#1.
The proper fix is an explicit full barrier <mfence>, which guarantees
that clear of PENDING bit is to be executed before all possible
speculative LOADS or STORES inside actual work function.
Signed-off-by: Roman Pen <roman.penyaev at profitbricks.com>
Cc: Gioh Kim <gi-oh.kim at profitbricks.com>
Cc: Michael Wang <yun.wang at profitbricks.com>
Cc: Tejun Heo <tj at kernel.org>
Cc: Jens Axboe <axboe at kernel.dk>
Cc: linux-block at vger.kernel.org
Cc: linux-kernel at vger.kernel.org
Cc: stable at vger.kernel.org
Signed-off-by: Tejun Heo <tj at kernel.org>
Signed-off-by: Pavel Tikhomirov <ptikhomirov at virtuozzo.com>
---
kernel/workqueue.c | 29 +++++++++++++++++++++++++++++
1 file changed, 29 insertions(+)
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index ec41322ac41d..ffd9f4eb8be8 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -638,6 +638,35 @@ static void set_work_pool_and_clear_pending(struct work_struct *work,
*/
smp_wmb();
set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0);
+ /*
+ * The following mb guarantees that previous clear of a PENDING bit
+ * will not be reordered with any speculative LOADS or STORES from
+ * work->current_func, which is executed afterwards. This possible
+ * reordering can lead to a missed execution on attempt to qeueue
+ * the same @work. E.g. consider this case:
+ *
+ * CPU#0 CPU#1
+ * ---------------------------- --------------------------------
+ *
+ * 1 STORE event_indicated
+ * 2 queue_work_on() {
+ * 3 test_and_set_bit(PENDING)
+ * 4 } set_..._and_clear_pending() {
+ * 5 set_work_data() # clear bit
+ * 6 smp_mb()
+ * 7 work->current_func() {
+ * 8 LOAD event_indicated
+ * }
+ *
+ * Without an explicit full barrier speculative LOAD on line 8 can
+ * be executed before CPU#0 does STORE on line 1. If that happens,
+ * CPU#0 observes the PENDING bit is still set and new execution of
+ * a @work is not queued in a hope, that CPU#1 will eventually
+ * finish the queued @work. Meanwhile CPU#1 does not see
+ * event_indicated is set, because speculative LOAD was executed
+ * before actual STORE.
+ */
+ smp_mb();
}
static void clear_work_data(struct work_struct *work)
--
2.13.3
More information about the Devel
mailing list