blk-mq: clear stale inactive flag on hctx reuse
Subsystems (2)
PATCH 1/1 blk-mq: clear stale inactive flag on hctx reuse
When the last CPU mapped to an hctx goes offline, blk_mq_hctx_notify_offline() sets BLK_MQ_S_INACTIVE. Updating the number of hardware queues exits every hctx and parks it on q->unused_hctx_list. A later update can reuse a parked hctx through blk_mq_alloc_and_init_hctx(), at any queue index, and nothing clears the flag.
If the reused hctx is mapped to an online CPU, blk_mq_get_tag() keeps returning BLK_MQ_NO_TAG for it. __blk_mq_alloc_requests() sleeps and retries, expecting CPU hotplug to move the task to another CPU, so a task that stays on that CPU never gets a tag. Onlining the original CPU doesn't recover it either, because blk_mq_hctx_notify_online() only clears the flag on the hctx that CPU now maps to.
Clear BLK_MQ_S_INACTIVE in blk_mq_init_hctx() so a reused hctx starts like a freshly allocated one. A reused hctx has no cpuhp instances at that point, so the offline callback can't set the flag again before blk_mq_add_hw_queues_cpuhp() registers it. Clearing the flag in blk_mq_exit_hctx() is not enough, because a parked hctx keeps its cpuhp instances until blk_mq_remove_hw_queues_cpuhp(). Preserve the other hctx state flags.
Reproduced with null_blk on v7.3 in a four-vCPU arm64 QEMU/KVM guest:
modprobe null_blk nr_devices=0
mkdir /sys/kernel/config/nullb/nullb0
cd /sys/kernel/config/nullb/nullb0
echo 4 > submit_queues; echo 1 > memory_backed; echo 16 > size echo 1 > power
echo 0 > /sys/devices/system/cpu/cpu3/online
echo 2 > submit_queues
echo 4 > submit_queues
Offlining CPU3 marks hctx3 inactive. After the two updates it comes back as hctx1 and is mapped to CPU0:
# cat /sys/block/nullb0/mq/1/cpu_list 0 # cat /sys/kernel/debug/block/nullb0/hctx1/state INACTIVE
A 4K O_DIRECT read pinned to CPU0 hangs in D state, while the same read on CPU1 and CPU2 completes. Onlining CPU3 again leaves hctx1 inactive and the CPU0 read still hangs. With this change hctx1 state is empty and the reads on CPU0, CPU1 and CPU2 complete, before and after onlining CPU3.
--- block/blk-mq.c | 1 + 1 file changed, 1 insertion(+)
block/blk-mq.c +1 -0
@@ -4002,6 +4002,7 @@ static int blk_mq_init_hctx(struct request_queue *q, goto fail; hctx->queue_num = hctx_idx; +clear_bit (BLK_MQ_S_INACTIVE , &hctx->state); hctx->tags = set->tags[hctx_idx];
Signature · 1 line
2.43.0
block/blk-mq.c static int blk_mq_init_hctx(struct request_queue *q, · line 4007
Quoted 11 lines · Nirmoy Das
block/blk-mq.c +1
-0@@ -4002,6 +4002,7 @@ static int blk_mq_init_hctx(struct request_queue *q, goto fail; hctx->queue_num = hctx_idx; +clear_bit (BLK_MQ_S_INACTIVE , &hctx->state); hctx->tags = set->tags[hctx_idx];
Why clear_bit() instead of __clear_bit()?
Thanks,
Bart.
Quoted 13 lines · Bart Van Assche
block/blk-mq.c static int blk_mq_init_hctx(struct · line 4006
Quoted 10 lines · Nirmoy Das
block/blk-mq.c +1
-0@@ -4002,6 +4002,7 @@ static int blk_mq_init_hctx(structrequest_queue *q, goto fail; hctx->queue_num = hctx_idx; + clear_bit(BLK_MQ_S_INACTIVE, &hctx->state); hctx->tags = set->tags[hctx_idx];Why
clear_bit()instead of__clear_bit()?
No strong reason, I followed the other hctx->state updates. __clear_bit() should work too I think. Should I resend with that ?
Quoted 4 lines
Thanks,
Bart.
Jens Axboe
maintainerQuoted 12 lines · Bart Van Assche
block/blk-mq.c static int blk_mq_init_hctx(struct request_queue *q, · line 4005
Quoted 9 lines · Nirmoy Das
block/blk-mq.c +1
-0@@ -4002,6 +4002,7 @@ static int blk_mq_init_hctx(struct request_queue *q, goto fail; hctx->queue_num = hctx_idx; + clear_bit(BLK_MQ_S_INACTIVE, &hctx->state); hctx->tags = set->tags[hctx_idx];Why
clear_bit()instead of__clear_bit()?
Since this isn't a hot path at all, going into details like that does more harm than good. clear_bit() is totally fine and honestly preferred.
Signature · 1 line
Jens Axboe
Quoted 14 lines · Jens Axboe
Quoted 12 lines · Bart Van Assche
block/blk-mq.c static int blk_mq_init_hctx(struct request_queue *q, · line 4005
Quoted 9 lines · Nirmoy Das
block/blk-mq.c +1
-0@@ -4002,6 +4002,7 @@ static int blk_mq_init_hctx(struct request_queue *q, goto fail; hctx->queue_num = hctx_idx; + clear_bit(BLK_MQ_S_INACTIVE, &hctx->state); hctx->tags = set->tags[hctx_idx];Why
clear_bit()instead of__clear_bit()?
Quoted 3 lines
Since this isn't a hot path at all, going into details like that does more harm than good.
clear_bit()is totally fine and honestly preferred.
Although I'm fine with the patch as-is, I want to clarify that the concern behind my reply is code readability rather than a micro-optimization. No other context should be accessing a hctx while blk_mq_init_hctx() is in progress. Using clear_bit() instead of __clear_bit() might make readers of this code wonder whether any other context might be accessing the hctx concurrently with the blk_mq_init_hctx() call.
Thanks,
Bart.
Jens Axboe
maintainerQuoted 19 lines · Bart Van Assche
Quoted 14 lines · Jens Axboe
Quoted 12 lines · Bart Van Assche
block/blk-mq.c static int blk_mq_init_hctx(struct request_queue *q, · line 4005
Quoted 9 lines · Nirmoy Das
block/blk-mq.c +1
-0@@ -4002,6 +4002,7 @@ static int blk_mq_init_hctx(struct request_queue *q, goto fail; hctx->queue_num = hctx_idx; +clear_bit (BLK_MQ_S_INACTIVE , &hctx->state); hctx->tags = set->tags[hctx_idx];Why
clear_bit()instead of__clear_bit()?Quoted 3 lines
Since this isn't a hot path at all, going into details like that does more harm than good.
clear_bit()is totally fine and honestly preferred.
Quoted 7 lines
Although I'm fine with the patch as-is, I want to clarify that the concern behind my reply is code readability rather than a micro-optimization. No other context should be accessing a hctx while
blk_mq_init_hctx()is in progress. Usingclear_bit()instead of__clear_bit()might make readers of this code wonder whether any other context might be accessing the hctx concurrently with theblk_mq_init_hctx()call.
I know that was your intent, but it just reads the opposite to me. Whenever someone uses the __ variants it should only be for optimization purposes, and preferably with a comment explaining why it's safe. Unless there's a reason to do so, using the normal variants is just preferable to me.
Signature · 1 line
Jens Axboe
Quoted 8 lines · Nirmoy Das
When the last CPU mapped to an hctx goes offline,
blk_mq_hctx_notify_offline()setsBLK_MQ_S_INACTIVE. Updating the number of hardware queues exits every hctx and parks it onq->unused_hctx_list. A later update can reuse a parked hctx throughblk_mq_alloc_and_init_hctx(), at any queue index, and nothing clears the flag.[...]
Applied, thanks!
[1/1] blk-mq: clear stale inactive flag on hctx reuse
commit: afc1de47a7ab607ae6403c4bfda36b0bf3cd342f
Best regards,
Signature · 1 line
Jens Axboe
Quoted 8 lines · Jens Axboe
Quoted 7 lines · Nirmoy Das
When the last CPU mapped to an hctx goes offline,
blk_mq_hctx_notify_offline()setsBLK_MQ_S_INACTIVE. Updating the number of hardware queues exits every hctx and parks it onq->unused_hctx_list. A later update can reuse a parked hctx throughblk_mq_alloc_and_init_hctx(), at any queue index, and nothing clears the flag.
Quoted 7 lines
Quoted 1 line
[...]
Applied, thanks!
[1/1] blk-mq: clear stale inactive flag on hctx reuse
commit: afc1de47a7ab607ae6403c4bfda36b0bf3cd342fBest regards,
Thanks, Jens.