[PATCH 0/3] blk-mq: misc fixes

[PATCH 0/3] blk-mq: misc fixes

[Ming Lei]

Hello,

Here are 3 small fixes against latest linus tree.


Ming Lei (3):
  blk-mq: only try to run plug merge if request has same queue with
    incoming bio
  blk-mq: add RQF_ELV debug entry
  blk-mq: update hctx->nr_active in blk_mq_end_request_batch()

 block/blk-merge.c      |  6 ++++--
 block/blk-mq-debugfs.c |  1 +
 block/blk-mq.c         | 15 +++++++++++++--
 block/blk-mq.h         | 12 +++++++++---
 4 files changed, 27 insertions(+), 7 deletions(-)

-- 
2.31.1

[PATCH 1/3] blk-mq: only try to run plug merge if request has same queue with incoming bio

[Ming Lei]

It is obvious that io merge can't be done between two different queues, so
just try to run io merge in case of same queue.

Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 block/blk-merge.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/block/blk-merge.c b/block/blk-merge.c
index df69f4bb7717..893c1a60b701 100644
--- a/block/blk-merge.c
+++ b/block/blk-merge.c
@@ -1101,9 +1101,11 @@ bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
 		 * the same queue, there should be only one such rq in a queue
 		 */
 		*same_queue_rq = true;
+
+		if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) ==
+				BIO_MERGE_OK)
+			return true;
 	}
-	if (blk_attempt_bio_merge(q, rq, bio, nr_segs, false) == BIO_MERGE_OK)
-		return true;
 	return false;
 }
 
-- 
2.31.1

[PATCH 2/3] blk-mq: add RQF_ELV debug entry

[Ming Lei]

Looks it is missed so add it.

Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 block/blk-mq-debugfs.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/block/blk-mq-debugfs.c b/block/blk-mq-debugfs.c
index f5076c173477..4f2cf8399f3d 100644
--- a/block/blk-mq-debugfs.c
+++ b/block/blk-mq-debugfs.c
@@ -308,6 +308,7 @@ static const char *const rqf_name[] = {
 	RQF_NAME(SPECIAL_PAYLOAD),
 	RQF_NAME(ZONE_WRITE_LOCKED),
 	RQF_NAME(MQ_POLL_SLEPT),
+	RQF_NAME(ELV),
 };
 #undef RQF_NAME
 
-- 
2.31.1

[PATCH 3/3] blk-mq: update hctx->nr_active in blk_mq_end_request_batch()

[Ming Lei]

In case of shared tags and none io sched, batched completion still may
be run into, and hctx->nr_active is accounted when getting driver tag,
so it has to be updated in blk_mq_end_request_batch().

Otherwise, hctx->nr_active may become same with queue depth, then
hctx_may_queue() always return false, then io hang is caused.

Fixes the issue by updating the counter in batched way.

Reported-by: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Fixes: f794f3351f26 ("block: add support for blk_mq_end_request_batch()")
Signed-off-by: Ming Lei <ming.lei@redhat.com>
---
 block/blk-mq.c | 15 +++++++++++++--
 block/blk-mq.h | 12 +++++++++---
 2 files changed, 22 insertions(+), 5 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 07eb1412760b..0dbe75034f61 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -825,6 +825,7 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
 	struct blk_mq_hw_ctx *cur_hctx = NULL;
 	struct request *rq;
 	u64 now = 0;
+	int active = 0;
 
 	if (iob->need_ts)
 		now = ktime_get_ns();
@@ -846,16 +847,26 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
 		rq_qos_done(rq->q, rq);
 
 		if (nr_tags == TAG_COMP_BATCH || cur_hctx != rq->mq_hctx) {
-			if (cur_hctx)
+			if (cur_hctx) {
+				if (active)
+					__blk_mq_sub_active_requests(cur_hctx,
+							active);
 				blk_mq_flush_tag_batch(cur_hctx, tags, nr_tags);
+			}
 			nr_tags = 0;
+			active = 0;
 			cur_hctx = rq->mq_hctx;
 		}
 		tags[nr_tags++] = rq->tag;
+		if (rq->rq_flags & RQF_MQ_INFLIGHT)
+			active++;
 	}
 
-	if (nr_tags)
+	if (nr_tags) {
+		if (active)
+			__blk_mq_sub_active_requests(cur_hctx, active);
 		blk_mq_flush_tag_batch(cur_hctx, tags, nr_tags);
+	}
 }
 EXPORT_SYMBOL_GPL(blk_mq_end_request_batch);
 
diff --git a/block/blk-mq.h b/block/blk-mq.h
index 28859fc5faee..cb0b5482ca5e 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -225,12 +225,18 @@ static inline void __blk_mq_inc_active_requests(struct blk_mq_hw_ctx *hctx)
 		atomic_inc(&hctx->nr_active);
 }
 
-static inline void __blk_mq_dec_active_requests(struct blk_mq_hw_ctx *hctx)
+static inline void __blk_mq_sub_active_requests(struct blk_mq_hw_ctx *hctx,
+		int val)
 {
 	if (blk_mq_is_shared_tags(hctx->flags))
-		atomic_dec(&hctx->queue->nr_active_requests_shared_tags);
+		atomic_sub(val, &hctx->queue->nr_active_requests_shared_tags);
 	else
-		atomic_dec(&hctx->nr_active);
+		atomic_sub(val, &hctx->nr_active);
+}
+
+static inline void __blk_mq_dec_active_requests(struct blk_mq_hw_ctx *hctx)
+{
+	__blk_mq_sub_active_requests(hctx, 1);
 }
 
 static inline int __blk_mq_active_requests(struct blk_mq_hw_ctx *hctx)
-- 
2.31.1

Re: [PATCH 3/3] blk-mq: update hctx->nr_active in blk_mq_end_request_batch()

[Jens Axboe]

On 11/2/21 7:35 AM, Ming Lei wrote:
> In case of shared tags and none io sched, batched completion still may
> be run into, and hctx->nr_active is accounted when getting driver tag,
> so it has to be updated in blk_mq_end_request_batch().
> 
> Otherwise, hctx->nr_active may become same with queue depth, then
> hctx_may_queue() always return false, then io hang is caused.
> 
> Fixes the issue by updating the counter in batched way.
> 
> Reported-by: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> Fixes: f794f3351f26 ("block: add support for blk_mq_end_request_batch()")
> Signed-off-by: Ming Lei <ming.lei@redhat.com>
> ---
>  block/blk-mq.c | 15 +++++++++++++--
>  block/blk-mq.h | 12 +++++++++---
>  2 files changed, 22 insertions(+), 5 deletions(-)
> 
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index 07eb1412760b..0dbe75034f61 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -825,6 +825,7 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
>  	struct blk_mq_hw_ctx *cur_hctx = NULL;
>  	struct request *rq;
>  	u64 now = 0;
> +	int active = 0;
>  
>  	if (iob->need_ts)
>  		now = ktime_get_ns();
> @@ -846,16 +847,26 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
>  		rq_qos_done(rq->q, rq);
>  
>  		if (nr_tags == TAG_COMP_BATCH || cur_hctx != rq->mq_hctx) {
> -			if (cur_hctx)
> +			if (cur_hctx) {
> +				if (active)
> +					__blk_mq_sub_active_requests(cur_hctx,
> +							active);
>  				blk_mq_flush_tag_batch(cur_hctx, tags, nr_tags);
> +			}
>  			nr_tags = 0;
> +			active = 0;
>  			cur_hctx = rq->mq_hctx;
>  		}
>  		tags[nr_tags++] = rq->tag;
> +		if (rq->rq_flags & RQF_MQ_INFLIGHT)
> +			active++;

Are there any cases where either none or all of requests have the flag set, and
hence active == nr_tags?

Apart from that, the patch looks conceptually good.

-- 
Jens Axboe

Re: [PATCH 3/3] blk-mq: update hctx->nr_active in blk_mq_end_request_batch()

[Ming Lei]

On Tue, Nov 02, 2021 at 07:47:44AM -0600, Jens Axboe wrote:
> On 11/2/21 7:35 AM, Ming Lei wrote:
> > In case of shared tags and none io sched, batched completion still may
> > be run into, and hctx->nr_active is accounted when getting driver tag,
> > so it has to be updated in blk_mq_end_request_batch().
> > 
> > Otherwise, hctx->nr_active may become same with queue depth, then
> > hctx_may_queue() always return false, then io hang is caused.
> > 
> > Fixes the issue by updating the counter in batched way.
> > 
> > Reported-by: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> > Fixes: f794f3351f26 ("block: add support for blk_mq_end_request_batch()")
> > Signed-off-by: Ming Lei <ming.lei@redhat.com>
> > ---
> >  block/blk-mq.c | 15 +++++++++++++--
> >  block/blk-mq.h | 12 +++++++++---
> >  2 files changed, 22 insertions(+), 5 deletions(-)
> > 
> > diff --git a/block/blk-mq.c b/block/blk-mq.c
> > index 07eb1412760b..0dbe75034f61 100644
> > --- a/block/blk-mq.c
> > +++ b/block/blk-mq.c
> > @@ -825,6 +825,7 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
> >  	struct blk_mq_hw_ctx *cur_hctx = NULL;
> >  	struct request *rq;
> >  	u64 now = 0;
> > +	int active = 0;
> >  
> >  	if (iob->need_ts)
> >  		now = ktime_get_ns();
> > @@ -846,16 +847,26 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
> >  		rq_qos_done(rq->q, rq);
> >  
> >  		if (nr_tags == TAG_COMP_BATCH || cur_hctx != rq->mq_hctx) {
> > -			if (cur_hctx)
> > +			if (cur_hctx) {
> > +				if (active)
> > +					__blk_mq_sub_active_requests(cur_hctx,
> > +							active);
> >  				blk_mq_flush_tag_batch(cur_hctx, tags, nr_tags);
> > +			}
> >  			nr_tags = 0;
> > +			active = 0;
> >  			cur_hctx = rq->mq_hctx;
> >  		}
> >  		tags[nr_tags++] = rq->tag;
> > +		if (rq->rq_flags & RQF_MQ_INFLIGHT)
> > +			active++;
> 
> Are there any cases where either none or all of requests have the flag set, and
> hence active == nr_tags?

none and BLK_MQ_F_TAG_QUEUE_SHARED, and Shinichiro only observed the
issue on two NSs.


Thanks,
Ming

Re: [PATCH 3/3] blk-mq: update hctx->nr_active in blk_mq_end_request_batch()

[Jens Axboe]

On 11/2/21 7:57 AM, Ming Lei wrote:
> On Tue, Nov 02, 2021 at 07:47:44AM -0600, Jens Axboe wrote:
>> On 11/2/21 7:35 AM, Ming Lei wrote:
>>> In case of shared tags and none io sched, batched completion still may
>>> be run into, and hctx->nr_active is accounted when getting driver tag,
>>> so it has to be updated in blk_mq_end_request_batch().
>>>
>>> Otherwise, hctx->nr_active may become same with queue depth, then
>>> hctx_may_queue() always return false, then io hang is caused.
>>>
>>> Fixes the issue by updating the counter in batched way.
>>>
>>> Reported-by: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com>
>>> Fixes: f794f3351f26 ("block: add support for blk_mq_end_request_batch()")
>>> Signed-off-by: Ming Lei <ming.lei@redhat.com>
>>> ---
>>>  block/blk-mq.c | 15 +++++++++++++--
>>>  block/blk-mq.h | 12 +++++++++---
>>>  2 files changed, 22 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/block/blk-mq.c b/block/blk-mq.c
>>> index 07eb1412760b..0dbe75034f61 100644
>>> --- a/block/blk-mq.c
>>> +++ b/block/blk-mq.c
>>> @@ -825,6 +825,7 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
>>>  	struct blk_mq_hw_ctx *cur_hctx = NULL;
>>>  	struct request *rq;
>>>  	u64 now = 0;
>>> +	int active = 0;
>>>  
>>>  	if (iob->need_ts)
>>>  		now = ktime_get_ns();
>>> @@ -846,16 +847,26 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
>>>  		rq_qos_done(rq->q, rq);
>>>  
>>>  		if (nr_tags == TAG_COMP_BATCH || cur_hctx != rq->mq_hctx) {
>>> -			if (cur_hctx)
>>> +			if (cur_hctx) {
>>> +				if (active)
>>> +					__blk_mq_sub_active_requests(cur_hctx,
>>> +							active);
>>>  				blk_mq_flush_tag_batch(cur_hctx, tags, nr_tags);
>>> +			}
>>>  			nr_tags = 0;
>>> +			active = 0;
>>>  			cur_hctx = rq->mq_hctx;
>>>  		}
>>>  		tags[nr_tags++] = rq->tag;
>>> +		if (rq->rq_flags & RQF_MQ_INFLIGHT)
>>> +			active++;
>>
>> Are there any cases where either none or all of requests have the
>> flag set, and hence active == nr_tags?
> 
> none and BLK_MQ_F_TAG_QUEUE_SHARED, and Shinichiro only observed the
> issue on two NSs.

Maybe I wasn't clear enough. What I'm saying is that either all of the
requests will have RQF_MQ_INFLIGHT set, or none of them. Hence active
should be either 0, or == nr_tags.

That's the hypothesis that I wanted to check, because if that's true,
then we can do this in a better way.

-- 
Jens Axboe

Re: [PATCH 3/3] blk-mq: update hctx->nr_active in blk_mq_end_request_batch()

[Ming Lei]

On Tue, Nov 02, 2021 at 08:57:41AM -0600, Jens Axboe wrote:
> On 11/2/21 7:57 AM, Ming Lei wrote:
> > On Tue, Nov 02, 2021 at 07:47:44AM -0600, Jens Axboe wrote:
> >> On 11/2/21 7:35 AM, Ming Lei wrote:
> >>> In case of shared tags and none io sched, batched completion still may
> >>> be run into, and hctx->nr_active is accounted when getting driver tag,
> >>> so it has to be updated in blk_mq_end_request_batch().
> >>>
> >>> Otherwise, hctx->nr_active may become same with queue depth, then
> >>> hctx_may_queue() always return false, then io hang is caused.
> >>>
> >>> Fixes the issue by updating the counter in batched way.
> >>>
> >>> Reported-by: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com>
> >>> Fixes: f794f3351f26 ("block: add support for blk_mq_end_request_batch()")
> >>> Signed-off-by: Ming Lei <ming.lei@redhat.com>
> >>> ---
> >>>  block/blk-mq.c | 15 +++++++++++++--
> >>>  block/blk-mq.h | 12 +++++++++---
> >>>  2 files changed, 22 insertions(+), 5 deletions(-)
> >>>
> >>> diff --git a/block/blk-mq.c b/block/blk-mq.c
> >>> index 07eb1412760b..0dbe75034f61 100644
> >>> --- a/block/blk-mq.c
> >>> +++ b/block/blk-mq.c
> >>> @@ -825,6 +825,7 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
> >>>  	struct blk_mq_hw_ctx *cur_hctx = NULL;
> >>>  	struct request *rq;
> >>>  	u64 now = 0;
> >>> +	int active = 0;
> >>>  
> >>>  	if (iob->need_ts)
> >>>  		now = ktime_get_ns();
> >>> @@ -846,16 +847,26 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
> >>>  		rq_qos_done(rq->q, rq);
> >>>  
> >>>  		if (nr_tags == TAG_COMP_BATCH || cur_hctx != rq->mq_hctx) {
> >>> -			if (cur_hctx)
> >>> +			if (cur_hctx) {
> >>> +				if (active)
> >>> +					__blk_mq_sub_active_requests(cur_hctx,
> >>> +							active);
> >>>  				blk_mq_flush_tag_batch(cur_hctx, tags, nr_tags);
> >>> +			}
> >>>  			nr_tags = 0;
> >>> +			active = 0;
> >>>  			cur_hctx = rq->mq_hctx;
> >>>  		}
> >>>  		tags[nr_tags++] = rq->tag;
> >>> +		if (rq->rq_flags & RQF_MQ_INFLIGHT)
> >>> +			active++;
> >>
> >> Are there any cases where either none or all of requests have the
> >> flag set, and hence active == nr_tags?
> > 
> > none and BLK_MQ_F_TAG_QUEUE_SHARED, and Shinichiro only observed the
> > issue on two NSs.
> 
> Maybe I wasn't clear enough. What I'm saying is that either all of the
> requests will have RQF_MQ_INFLIGHT set, or none of them. Hence active
> should be either 0, or == nr_tags.

Yeah, that is right since BLK_MQ_F_TAG_QUEUE_SHARED is updated after
queue is frozen. Meantime blk_mq_end_request_batch() is only called
for ending successfully completed requests.

Will do that in V2.

Thanks,
Ming

Re: [PATCH 3/3] blk-mq: update hctx->nr_active in blk_mq_end_request_batch()

[Jens Axboe]

On 11/2/21 9:08 AM, Ming Lei wrote:
> On Tue, Nov 02, 2021 at 08:57:41AM -0600, Jens Axboe wrote:
>> On 11/2/21 7:57 AM, Ming Lei wrote:
>>> On Tue, Nov 02, 2021 at 07:47:44AM -0600, Jens Axboe wrote:
>>>> On 11/2/21 7:35 AM, Ming Lei wrote:
>>>>> In case of shared tags and none io sched, batched completion still may
>>>>> be run into, and hctx->nr_active is accounted when getting driver tag,
>>>>> so it has to be updated in blk_mq_end_request_batch().
>>>>>
>>>>> Otherwise, hctx->nr_active may become same with queue depth, then
>>>>> hctx_may_queue() always return false, then io hang is caused.
>>>>>
>>>>> Fixes the issue by updating the counter in batched way.
>>>>>
>>>>> Reported-by: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com>
>>>>> Fixes: f794f3351f26 ("block: add support for blk_mq_end_request_batch()")
>>>>> Signed-off-by: Ming Lei <ming.lei@redhat.com>
>>>>> ---
>>>>>  block/blk-mq.c | 15 +++++++++++++--
>>>>>  block/blk-mq.h | 12 +++++++++---
>>>>>  2 files changed, 22 insertions(+), 5 deletions(-)
>>>>>
>>>>> diff --git a/block/blk-mq.c b/block/blk-mq.c
>>>>> index 07eb1412760b..0dbe75034f61 100644
>>>>> --- a/block/blk-mq.c
>>>>> +++ b/block/blk-mq.c
>>>>> @@ -825,6 +825,7 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
>>>>>  	struct blk_mq_hw_ctx *cur_hctx = NULL;
>>>>>  	struct request *rq;
>>>>>  	u64 now = 0;
>>>>> +	int active = 0;
>>>>>  
>>>>>  	if (iob->need_ts)
>>>>>  		now = ktime_get_ns();
>>>>> @@ -846,16 +847,26 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
>>>>>  		rq_qos_done(rq->q, rq);
>>>>>  
>>>>>  		if (nr_tags == TAG_COMP_BATCH || cur_hctx != rq->mq_hctx) {
>>>>> -			if (cur_hctx)
>>>>> +			if (cur_hctx) {
>>>>> +				if (active)
>>>>> +					__blk_mq_sub_active_requests(cur_hctx,
>>>>> +							active);
>>>>>  				blk_mq_flush_tag_batch(cur_hctx, tags, nr_tags);
>>>>> +			}
>>>>>  			nr_tags = 0;
>>>>> +			active = 0;
>>>>>  			cur_hctx = rq->mq_hctx;
>>>>>  		}
>>>>>  		tags[nr_tags++] = rq->tag;
>>>>> +		if (rq->rq_flags & RQF_MQ_INFLIGHT)
>>>>> +			active++;
>>>>
>>>> Are there any cases where either none or all of requests have the
>>>> flag set, and hence active == nr_tags?
>>>
>>> none and BLK_MQ_F_TAG_QUEUE_SHARED, and Shinichiro only observed the
>>> issue on two NSs.
>>
>> Maybe I wasn't clear enough. What I'm saying is that either all of the
>> requests will have RQF_MQ_INFLIGHT set, or none of them. Hence active
>> should be either 0, or == nr_tags.
> 
> Yeah, that is right since BLK_MQ_F_TAG_QUEUE_SHARED is updated after
> queue is frozen. Meantime blk_mq_end_request_batch() is only called
> for ending successfully completed requests.
> 
> Will do that in V2.

Thanks, then it just becomes a single check in blk_mq_flush_tag_batch(),
which is a lot better than per-request.

-- 
Jens Axboe

Re: [PATCH 3/3] blk-mq: update hctx->nr_active in blk_mq_end_request_batch()

[Jens Axboe]

On 11/2/21 9:20 AM, Jens Axboe wrote:
> On 11/2/21 9:08 AM, Ming Lei wrote:
>> On Tue, Nov 02, 2021 at 08:57:41AM -0600, Jens Axboe wrote:
>>> On 11/2/21 7:57 AM, Ming Lei wrote:
>>>> On Tue, Nov 02, 2021 at 07:47:44AM -0600, Jens Axboe wrote:
>>>>> On 11/2/21 7:35 AM, Ming Lei wrote:
>>>>>> In case of shared tags and none io sched, batched completion still may
>>>>>> be run into, and hctx->nr_active is accounted when getting driver tag,
>>>>>> so it has to be updated in blk_mq_end_request_batch().
>>>>>>
>>>>>> Otherwise, hctx->nr_active may become same with queue depth, then
>>>>>> hctx_may_queue() always return false, then io hang is caused.
>>>>>>
>>>>>> Fixes the issue by updating the counter in batched way.
>>>>>>
>>>>>> Reported-by: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com>
>>>>>> Fixes: f794f3351f26 ("block: add support for blk_mq_end_request_batch()")
>>>>>> Signed-off-by: Ming Lei <ming.lei@redhat.com>
>>>>>> ---
>>>>>>  block/blk-mq.c | 15 +++++++++++++--
>>>>>>  block/blk-mq.h | 12 +++++++++---
>>>>>>  2 files changed, 22 insertions(+), 5 deletions(-)
>>>>>>
>>>>>> diff --git a/block/blk-mq.c b/block/blk-mq.c
>>>>>> index 07eb1412760b..0dbe75034f61 100644
>>>>>> --- a/block/blk-mq.c
>>>>>> +++ b/block/blk-mq.c
>>>>>> @@ -825,6 +825,7 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
>>>>>>  	struct blk_mq_hw_ctx *cur_hctx = NULL;
>>>>>>  	struct request *rq;
>>>>>>  	u64 now = 0;
>>>>>> +	int active = 0;
>>>>>>  
>>>>>>  	if (iob->need_ts)
>>>>>>  		now = ktime_get_ns();
>>>>>> @@ -846,16 +847,26 @@ void blk_mq_end_request_batch(struct io_comp_batch *iob)
>>>>>>  		rq_qos_done(rq->q, rq);
>>>>>>  
>>>>>>  		if (nr_tags == TAG_COMP_BATCH || cur_hctx != rq->mq_hctx) {
>>>>>> -			if (cur_hctx)
>>>>>> +			if (cur_hctx) {
>>>>>> +				if (active)
>>>>>> +					__blk_mq_sub_active_requests(cur_hctx,
>>>>>> +							active);
>>>>>>  				blk_mq_flush_tag_batch(cur_hctx, tags, nr_tags);
>>>>>> +			}
>>>>>>  			nr_tags = 0;
>>>>>> +			active = 0;
>>>>>>  			cur_hctx = rq->mq_hctx;
>>>>>>  		}
>>>>>>  		tags[nr_tags++] = rq->tag;
>>>>>> +		if (rq->rq_flags & RQF_MQ_INFLIGHT)
>>>>>> +			active++;
>>>>>
>>>>> Are there any cases where either none or all of requests have the
>>>>> flag set, and hence active == nr_tags?
>>>>
>>>> none and BLK_MQ_F_TAG_QUEUE_SHARED, and Shinichiro only observed the
>>>> issue on two NSs.
>>>
>>> Maybe I wasn't clear enough. What I'm saying is that either all of the
>>> requests will have RQF_MQ_INFLIGHT set, or none of them. Hence active
>>> should be either 0, or == nr_tags.
>>
>> Yeah, that is right since BLK_MQ_F_TAG_QUEUE_SHARED is updated after
>> queue is frozen. Meantime blk_mq_end_request_batch() is only called
>> for ending successfully completed requests.
>>
>> Will do that in V2.
> 
> Thanks, then it just becomes a single check in blk_mq_flush_tag_batch(),
> which is a lot better than per-request.

Something like this, untested. FWIW, I did apply 1-2 from this series,
so just do a v2 of 3/3 and that should do it.


diff --git a/block/blk-mq.c b/block/blk-mq.c
index 02f70dc06ced..18dee9af4487 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -817,6 +817,8 @@ static inline void blk_mq_flush_tag_batch(struct blk_mq_hw_ctx *hctx,
 	struct request_queue *q = hctx->queue;
 
 	blk_mq_put_tags(hctx->tags, tag_array, nr_tags);
+	if (hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)
+		__blk_mq_sub_active_requests(hctx, nr_tags);
 	percpu_ref_put_many(&q->q_usage_counter, nr_tags);
 }
 
diff --git a/block/blk-mq.h b/block/blk-mq.h
index 28859fc5faee..cb0b5482ca5e 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -225,12 +225,18 @@ static inline void __blk_mq_inc_active_requests(struct blk_mq_hw_ctx *hctx)
 		atomic_inc(&hctx->nr_active);
 }
 
-static inline void __blk_mq_dec_active_requests(struct blk_mq_hw_ctx *hctx)
+static inline void __blk_mq_sub_active_requests(struct blk_mq_hw_ctx *hctx,
+		int val)
 {
 	if (blk_mq_is_shared_tags(hctx->flags))
-		atomic_dec(&hctx->queue->nr_active_requests_shared_tags);
+		atomic_sub(val, &hctx->queue->nr_active_requests_shared_tags);
 	else
-		atomic_dec(&hctx->nr_active);
+		atomic_sub(val, &hctx->nr_active);
+}
+
+static inline void __blk_mq_dec_active_requests(struct blk_mq_hw_ctx *hctx)
+{
+	__blk_mq_sub_active_requests(hctx, 1);
 }
 
 static inline int __blk_mq_active_requests(struct blk_mq_hw_ctx *hctx)


-- 
Jens Axboe