[PATCH -next v2 0/2] block, bfq: make bfq_has_work() more accurate

[PATCH -next v2 0/2] block, bfq: make bfq_has_work() more accurate

[Yu Kuai]

Changes in v2:
 - add reviewed-by tag for patch 1
 - use WRITE_ONCE() for updating of 'bfqd->queued' in patch 2

This patchset try to make bfq_has_work() more accurate, patch 1 is a
small problem found by code review.

BTW, I not sure why blk_mq_run_hw_queues() is called with 'bfqd->lock'
held, I think this is not necessary. And bfq_has_work() can be more
accurate by reading 'bfqd->queued' with 'bfqd->lock' held after patch 2.

Previous versions:
v1: https://lore.kernel.org/all/20220510131629.1964415-1-yukuai3@huawei.com/

Yu Kuai (2):
  block, bfq: protect 'bfqd->queued' by 'bfqd->lock'
  block, bfq: make bfq_has_work() more accurate

 block/bfq-iosched.c | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

-- 
2.31.1

[PATCH -next v2 1/2] block, bfq: protect 'bfqd->queued' by 'bfqd->lock'

[Yu Kuai]

If bfq_schedule_dispatch() is called from bfq_idle_slice_timer_body(),
then 'bfqd->queued' is read without holding 'bfqd->lock'. This is
wrong since it can be wrote concurrently.

Fix the problem by holding 'bfqd->lock' in such case.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
---
 block/bfq-iosched.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
index 272d48d8f326..61750696e87f 100644
--- a/block/bfq-iosched.c
+++ b/block/bfq-iosched.c
@@ -456,6 +456,8 @@ static struct bfq_io_cq *bfq_bic_lookup(struct request_queue *q)
  */
 void bfq_schedule_dispatch(struct bfq_data *bfqd)
 {
+	lockdep_assert_held(&bfqd->lock);
+
 	if (bfqd->queued != 0) {
 		bfq_log(bfqd, "schedule dispatch");
 		blk_mq_run_hw_queues(bfqd->queue, true);
@@ -6898,8 +6900,8 @@ bfq_idle_slice_timer_body(struct bfq_data *bfqd, struct bfq_queue *bfqq)
 	bfq_bfqq_expire(bfqd, bfqq, true, reason);
 
 schedule_dispatch:
-	spin_unlock_irqrestore(&bfqd->lock, flags);
 	bfq_schedule_dispatch(bfqd);
+	spin_unlock_irqrestore(&bfqd->lock, flags);
 }
 
 /*
-- 
2.31.1

[PATCH -next v2 2/2] block, bfq: make bfq_has_work() more accurate

[Yu Kuai]

bfq_has_work() is using busy_queues currently, which is not accurate
because bfq_queue is busy doesn't represent that it has requests. Since
bfqd aready has a counter 'queued' to record how many requests are in
bfq, use it instead of busy_queues.

Noted that bfq_has_work() can be called with 'bfqd->lock' held, thus the
lock can't be held in bfq_has_work() to protect 'bfqd->queued'.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
---
 block/bfq-iosched.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
index 61750696e87f..740dd83853a6 100644
--- a/block/bfq-iosched.c
+++ b/block/bfq-iosched.c
@@ -2210,7 +2210,11 @@ static void bfq_add_request(struct request *rq)
 
 	bfq_log_bfqq(bfqd, bfqq, "add_request %d", rq_is_sync(rq));
 	bfqq->queued[rq_is_sync(rq)]++;
-	bfqd->queued++;
+	/*
+	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
+	 * may be read without holding the lock in bfq_has_work().
+	 */
+	WRITE_ONCE(bfqd->queued, bfqd->queued + 1);
 
 	if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_bfqq_sync(bfqq)) {
 		bfq_check_waker(bfqd, bfqq, now_ns);
@@ -2402,7 +2406,11 @@ static void bfq_remove_request(struct request_queue *q,
 	if (rq->queuelist.prev != &rq->queuelist)
 		list_del_init(&rq->queuelist);
 	bfqq->queued[sync]--;
-	bfqd->queued--;
+	/*
+	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
+	 * may be read without holding the lock in bfq_has_work().
+	 */
+	WRITE_ONCE(bfqd->queued, bfqd->queued - 1);
 	elv_rb_del(&bfqq->sort_list, rq);
 
 	elv_rqhash_del(q, rq);
@@ -5063,11 +5071,11 @@ static bool bfq_has_work(struct blk_mq_hw_ctx *hctx)
 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
 
 	/*
-	 * Avoiding lock: a race on bfqd->busy_queues should cause at
+	 * Avoiding lock: a race on bfqd->queued should cause at
 	 * most a call to dispatch for nothing
 	 */
 	return !list_empty_careful(&bfqd->dispatch) ||
-		bfq_tot_busy_queues(bfqd) > 0;
+		READ_ONCE(bfqd->queued);
 }
 
 static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
-- 
2.31.1

Re: [PATCH -next v2 1/2] block, bfq: protect 'bfqd->queued' by 'bfqd->lock'

[Chaitanya Kulkarni]

On 5/12/2022 7:35 PM, Yu Kuai wrote:
> If bfq_schedule_dispatch() is called from bfq_idle_slice_timer_body(),
> then 'bfqd->queued' is read without holding 'bfqd->lock'. This is
> wrong since it can be wrote concurrently.
> 
> Fix the problem by holding 'bfqd->lock' in such case.
> 
> Signed-off-by: Yu Kuai <yukuai3@huawei.com>
> Reviewed-by: Jan Kara <jack@suse.cz>
> ---

Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>

-ck

Re: [PATCH -next v2 2/2] block, bfq: make bfq_has_work() more accurate

[Jan Kara]

On Fri 13-05-22 10:35:07, Yu Kuai wrote:
> bfq_has_work() is using busy_queues currently, which is not accurate
> because bfq_queue is busy doesn't represent that it has requests. Since
> bfqd aready has a counter 'queued' to record how many requests are in
> bfq, use it instead of busy_queues.
> 
> Noted that bfq_has_work() can be called with 'bfqd->lock' held, thus the
> lock can't be held in bfq_has_work() to protect 'bfqd->queued'.
> 
> Signed-off-by: Yu Kuai <yukuai3@huawei.com>

Looks good. Feel free to add:

Reviewed-by: Jan Kara <jack@suse.cz>

								Honza

> ---
>  block/bfq-iosched.c | 16 ++++++++++++----
>  1 file changed, 12 insertions(+), 4 deletions(-)
> 
> diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
> index 61750696e87f..740dd83853a6 100644
> --- a/block/bfq-iosched.c
> +++ b/block/bfq-iosched.c
> @@ -2210,7 +2210,11 @@ static void bfq_add_request(struct request *rq)
>  
>  	bfq_log_bfqq(bfqd, bfqq, "add_request %d", rq_is_sync(rq));
>  	bfqq->queued[rq_is_sync(rq)]++;
> -	bfqd->queued++;
> +	/*
> +	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
> +	 * may be read without holding the lock in bfq_has_work().
> +	 */
> +	WRITE_ONCE(bfqd->queued, bfqd->queued + 1);
>  
>  	if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_bfqq_sync(bfqq)) {
>  		bfq_check_waker(bfqd, bfqq, now_ns);
> @@ -2402,7 +2406,11 @@ static void bfq_remove_request(struct request_queue *q,
>  	if (rq->queuelist.prev != &rq->queuelist)
>  		list_del_init(&rq->queuelist);
>  	bfqq->queued[sync]--;
> -	bfqd->queued--;
> +	/*
> +	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
> +	 * may be read without holding the lock in bfq_has_work().
> +	 */
> +	WRITE_ONCE(bfqd->queued, bfqd->queued - 1);
>  	elv_rb_del(&bfqq->sort_list, rq);
>  
>  	elv_rqhash_del(q, rq);
> @@ -5063,11 +5071,11 @@ static bool bfq_has_work(struct blk_mq_hw_ctx *hctx)
>  	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
>  
>  	/*
> -	 * Avoiding lock: a race on bfqd->busy_queues should cause at
> +	 * Avoiding lock: a race on bfqd->queued should cause at
>  	 * most a call to dispatch for nothing
>  	 */
>  	return !list_empty_careful(&bfqd->dispatch) ||
> -		bfq_tot_busy_queues(bfqd) > 0;
> +		READ_ONCE(bfqd->queued);
>  }
>  
>  static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
> -- 
> 2.31.1
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: [PATCH -next v2 0/2] block, bfq: make bfq_has_work() more accurate

[Jens Axboe]

On Fri, 13 May 2022 10:35:05 +0800, Yu Kuai wrote:
> Changes in v2:
>  - add reviewed-by tag for patch 1
>  - use WRITE_ONCE() for updating of 'bfqd->queued' in patch 2
> 
> This patchset try to make bfq_has_work() more accurate, patch 1 is a
> small problem found by code review.
> 
> [...]

Applied, thanks!

[1/2] block, bfq: protect 'bfqd->queued' by 'bfqd->lock'
      commit: 181490d5321806e537dc5386db5ea640b826bf78
[2/2] block, bfq: make bfq_has_work() more accurate
      commit: ddc25c86b466d2359b57bc7798f167baa1735a44

Best regards,
-- 
Jens Axboe

Re: [PATCH -next v2 2/2] block, bfq: make bfq_has_work() more accurate

[Paolo Valente]

> Il giorno 16 mag 2022, alle ore 11:56, Jan Kara <jack@suse.cz> ha scritto:
> 
> On Fri 13-05-22 10:35:07, Yu Kuai wrote:
>> bfq_has_work() is using busy_queues currently, which is not accurate
>> because bfq_queue is busy doesn't represent that it has requests. Since
>> bfqd aready has a counter 'queued' to record how many requests are in
>> bfq, use it instead of busy_queues.
>> 

The number of requests queued is not equal to the number of busy
queues (it is >=).  If this patch is based on this assumption then
unfortunately it is wrong :(

Paolo

>> Noted that bfq_has_work() can be called with 'bfqd->lock' held, thus the
>> lock can't be held in bfq_has_work() to protect 'bfqd->queued'.
>> 
>> Signed-off-by: Yu Kuai <yukuai3@huawei.com>
> 
> Looks good. Feel free to add:
> 
> Reviewed-by: Jan Kara <jack@suse.cz>
> 
> 								Honza
> 
>> ---
>> block/bfq-iosched.c | 16 ++++++++++++----
>> 1 file changed, 12 insertions(+), 4 deletions(-)
>> 
>> diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
>> index 61750696e87f..740dd83853a6 100644
>> --- a/block/bfq-iosched.c
>> +++ b/block/bfq-iosched.c
>> @@ -2210,7 +2210,11 @@ static void bfq_add_request(struct request *rq)
>> 
>> 	bfq_log_bfqq(bfqd, bfqq, "add_request %d", rq_is_sync(rq));
>> 	bfqq->queued[rq_is_sync(rq)]++;
>> -	bfqd->queued++;
>> +	/*
>> +	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
>> +	 * may be read without holding the lock in bfq_has_work().
>> +	 */
>> +	WRITE_ONCE(bfqd->queued, bfqd->queued + 1);
>> 
>> 	if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_bfqq_sync(bfqq)) {
>> 		bfq_check_waker(bfqd, bfqq, now_ns);
>> @@ -2402,7 +2406,11 @@ static void bfq_remove_request(struct request_queue *q,
>> 	if (rq->queuelist.prev != &rq->queuelist)
>> 		list_del_init(&rq->queuelist);
>> 	bfqq->queued[sync]--;
>> -	bfqd->queued--;
>> +	/*
>> +	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
>> +	 * may be read without holding the lock in bfq_has_work().
>> +	 */
>> +	WRITE_ONCE(bfqd->queued, bfqd->queued - 1);
>> 	elv_rb_del(&bfqq->sort_list, rq);
>> 
>> 	elv_rqhash_del(q, rq);
>> @@ -5063,11 +5071,11 @@ static bool bfq_has_work(struct blk_mq_hw_ctx *hctx)
>> 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
>> 
>> 	/*
>> -	 * Avoiding lock: a race on bfqd->busy_queues should cause at
>> +	 * Avoiding lock: a race on bfqd->queued should cause at
>> 	 * most a call to dispatch for nothing
>> 	 */
>> 	return !list_empty_careful(&bfqd->dispatch) ||
>> -		bfq_tot_busy_queues(bfqd) > 0;
>> +		READ_ONCE(bfqd->queued);
>> }
>> 
>> static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
>> -- 
>> 2.31.1
>> 
> -- 
> Jan Kara <jack@suse.com>
> SUSE Labs, CR

Re: [PATCH -next v2 2/2] block, bfq: make bfq_has_work() more accurate

[Paolo Valente]

> Il giorno 17 mag 2022, alle ore 16:21, Paolo Valente <paolo.valente@linaro.org> ha scritto:
> 
> 
> 
>> Il giorno 16 mag 2022, alle ore 11:56, Jan Kara <jack@suse.cz> ha scritto:
>> 
>> On Fri 13-05-22 10:35:07, Yu Kuai wrote:
>>> bfq_has_work() is using busy_queues currently, which is not accurate
>>> because bfq_queue is busy doesn't represent that it has requests. Since
>>> bfqd aready has a counter 'queued' to record how many requests are in
>>> bfq, use it instead of busy_queues.
>>> 
> 
> The number of requests queued is not equal to the number of busy
> queues (it is >=).

No, sorry. It is actually != in general.

In particular, if queued == 0 but there are busy queues (although
still waiting for I/O to arrive), then responding that there is no
work caused blk-mq to stop asking, and hence an I/O freeze.  IOW I/O
eventually arrives for a busy queue, but blk-mq does not ask for a new
request any longer.  But maybe things have changed around bfq since
then.

Paolo

>  If this patch is based on this assumption then
> unfortunately it is wrong :(
> 
> Paolo
> 
>>> Noted that bfq_has_work() can be called with 'bfqd->lock' held, thus the
>>> lock can't be held in bfq_has_work() to protect 'bfqd->queued'.
>>> 
>>> Signed-off-by: Yu Kuai <yukuai3@huawei.com>
>> 
>> Looks good. Feel free to add:
>> 
>> Reviewed-by: Jan Kara <jack@suse.cz>
>> 
>> 								Honza
>> 
>>> ---
>>> block/bfq-iosched.c | 16 ++++++++++++----
>>> 1 file changed, 12 insertions(+), 4 deletions(-)
>>> 
>>> diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
>>> index 61750696e87f..740dd83853a6 100644
>>> --- a/block/bfq-iosched.c
>>> +++ b/block/bfq-iosched.c
>>> @@ -2210,7 +2210,11 @@ static void bfq_add_request(struct request *rq)
>>> 
>>> 	bfq_log_bfqq(bfqd, bfqq, "add_request %d", rq_is_sync(rq));
>>> 	bfqq->queued[rq_is_sync(rq)]++;
>>> -	bfqd->queued++;
>>> +	/*
>>> +	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
>>> +	 * may be read without holding the lock in bfq_has_work().
>>> +	 */
>>> +	WRITE_ONCE(bfqd->queued, bfqd->queued + 1);
>>> 
>>> 	if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_bfqq_sync(bfqq)) {
>>> 		bfq_check_waker(bfqd, bfqq, now_ns);
>>> @@ -2402,7 +2406,11 @@ static void bfq_remove_request(struct request_queue *q,
>>> 	if (rq->queuelist.prev != &rq->queuelist)
>>> 		list_del_init(&rq->queuelist);
>>> 	bfqq->queued[sync]--;
>>> -	bfqd->queued--;
>>> +	/*
>>> +	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
>>> +	 * may be read without holding the lock in bfq_has_work().
>>> +	 */
>>> +	WRITE_ONCE(bfqd->queued, bfqd->queued - 1);
>>> 	elv_rb_del(&bfqq->sort_list, rq);
>>> 
>>> 	elv_rqhash_del(q, rq);
>>> @@ -5063,11 +5071,11 @@ static bool bfq_has_work(struct blk_mq_hw_ctx *hctx)
>>> 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
>>> 
>>> 	/*
>>> -	 * Avoiding lock: a race on bfqd->busy_queues should cause at
>>> +	 * Avoiding lock: a race on bfqd->queued should cause at
>>> 	 * most a call to dispatch for nothing
>>> 	 */
>>> 	return !list_empty_careful(&bfqd->dispatch) ||
>>> -		bfq_tot_busy_queues(bfqd) > 0;
>>> +		READ_ONCE(bfqd->queued);
>>> }
>>> 
>>> static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
>>> -- 
>>> 2.31.1
>>> 
>> -- 
>> Jan Kara <jack@suse.com>
>> SUSE Labs, CR
> 

Re: [PATCH -next v2 2/2] block, bfq: make bfq_has_work() more accurate

[yukuai (C)]

在 2022/05/17 23:06, Paolo Valente 写道:
> 
> 
>> Il giorno 17 mag 2022, alle ore 16:21, Paolo Valente <paolo.valente@linaro.org> ha scritto:
>>
>>
>>
>>> Il giorno 16 mag 2022, alle ore 11:56, Jan Kara <jack@suse.cz> ha scritto:
>>>
>>> On Fri 13-05-22 10:35:07, Yu Kuai wrote:
>>>> bfq_has_work() is using busy_queues currently, which is not accurate
>>>> because bfq_queue is busy doesn't represent that it has requests. Since
>>>> bfqd aready has a counter 'queued' to record how many requests are in
>>>> bfq, use it instead of busy_queues.
>>>>
>>
>> The number of requests queued is not equal to the number of busy
>> queues (it is >=).
> 
> No, sorry. It is actually != in general.
Hi, Paolo

I'm aware that number of requests queued is not equal to the number of
busy queues, and that is the motivation of this patch.

> 
> In particular, if queued == 0 but there are busy queues (although
> still waiting for I/O to arrive), then responding that there is no
> work caused blk-mq to stop asking, and hence an I/O freeze.  IOW I/O
> eventually arrives for a busy queue, but blk-mq does not ask for a new
> request any longer.  But maybe things have changed around bfq since
> then.

The problem is that if queued == 0 while there are busy queues, is there
any point to return true in bfq_has_work() ? IMO, it will only cause
unecessary run queue. And if new request arrives,
blk_mq_sched_insert_request() will trigger a run queue.

Thanks,
Kuai
> 
> Paolo
> 
>>   If this patch is based on this assumption then
>> unfortunately it is wrong :(
>>
>> Paolo
>>
>>>> Noted that bfq_has_work() can be called with 'bfqd->lock' held, thus the
>>>> lock can't be held in bfq_has_work() to protect 'bfqd->queued'.
>>>>
>>>> Signed-off-by: Yu Kuai <yukuai3@huawei.com>
>>>
>>> Looks good. Feel free to add:
>>>
>>> Reviewed-by: Jan Kara <jack@suse.cz>
>>>
>>> 								Honza
>>>
>>>> ---
>>>> block/bfq-iosched.c | 16 ++++++++++++----
>>>> 1 file changed, 12 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
>>>> index 61750696e87f..740dd83853a6 100644
>>>> --- a/block/bfq-iosched.c
>>>> +++ b/block/bfq-iosched.c
>>>> @@ -2210,7 +2210,11 @@ static void bfq_add_request(struct request *rq)
>>>>
>>>> 	bfq_log_bfqq(bfqd, bfqq, "add_request %d", rq_is_sync(rq));
>>>> 	bfqq->queued[rq_is_sync(rq)]++;
>>>> -	bfqd->queued++;
>>>> +	/*
>>>> +	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
>>>> +	 * may be read without holding the lock in bfq_has_work().
>>>> +	 */
>>>> +	WRITE_ONCE(bfqd->queued, bfqd->queued + 1);
>>>>
>>>> 	if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_bfqq_sync(bfqq)) {
>>>> 		bfq_check_waker(bfqd, bfqq, now_ns);
>>>> @@ -2402,7 +2406,11 @@ static void bfq_remove_request(struct request_queue *q,
>>>> 	if (rq->queuelist.prev != &rq->queuelist)
>>>> 		list_del_init(&rq->queuelist);
>>>> 	bfqq->queued[sync]--;
>>>> -	bfqd->queued--;
>>>> +	/*
>>>> +	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
>>>> +	 * may be read without holding the lock in bfq_has_work().
>>>> +	 */
>>>> +	WRITE_ONCE(bfqd->queued, bfqd->queued - 1);
>>>> 	elv_rb_del(&bfqq->sort_list, rq);
>>>>
>>>> 	elv_rqhash_del(q, rq);
>>>> @@ -5063,11 +5071,11 @@ static bool bfq_has_work(struct blk_mq_hw_ctx *hctx)
>>>> 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
>>>>
>>>> 	/*
>>>> -	 * Avoiding lock: a race on bfqd->busy_queues should cause at
>>>> +	 * Avoiding lock: a race on bfqd->queued should cause at
>>>> 	 * most a call to dispatch for nothing
>>>> 	 */
>>>> 	return !list_empty_careful(&bfqd->dispatch) ||
>>>> -		bfq_tot_busy_queues(bfqd) > 0;
>>>> +		READ_ONCE(bfqd->queued);
>>>> }
>>>>
>>>> static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
>>>> -- 
>>>> 2.31.1
>>>>
>>> -- 
>>> Jan Kara <jack@suse.com>
>>> SUSE Labs, CR
>>
> 
> .
> 

Re: [PATCH -next v2 2/2] block, bfq: make bfq_has_work() more accurate

[Paolo VALENTE]

> Il giorno 18 mag 2022, alle ore 03:17, yukuai (C) <yukuai3@huawei.com> ha scritto:
> 
> 在 2022/05/17 23:06, Paolo Valente 写道:
>>> Il giorno 17 mag 2022, alle ore 16:21, Paolo Valente <paolo.valente@linaro.org> ha scritto:
>>> 
>>> 
>>> 
>>>> Il giorno 16 mag 2022, alle ore 11:56, Jan Kara <jack@suse.cz> ha scritto:
>>>> 
>>>> On Fri 13-05-22 10:35:07, Yu Kuai wrote:
>>>>> bfq_has_work() is using busy_queues currently, which is not accurate
>>>>> because bfq_queue is busy doesn't represent that it has requests. Since
>>>>> bfqd aready has a counter 'queued' to record how many requests are in
>>>>> bfq, use it instead of busy_queues.
>>>>> 
>>> 
>>> The number of requests queued is not equal to the number of busy
>>> queues (it is >=).
>> No, sorry. It is actually != in general.
> Hi, Paolo
> 
> I'm aware that number of requests queued is not equal to the number of
> busy queues, and that is the motivation of this patch.
> 
>> In particular, if queued == 0 but there are busy queues (although
>> still waiting for I/O to arrive), then responding that there is no
>> work caused blk-mq to stop asking, and hence an I/O freeze.  IOW I/O
>> eventually arrives for a busy queue, but blk-mq does not ask for a new
>> request any longer.  But maybe things have changed around bfq since
>> then.
> 
> The problem is that if queued == 0 while there are busy queues, is there
> any point to return true in bfq_has_work() ? IMO, it will only cause
> unecessary run queue. And if new request arrives,
> blk_mq_sched_insert_request() will trigger a run queue.

Great, if this is the scheme now, then the patch is correct and optimizing.

Thanks,
Paolo

> 
> Thanks,
> Kuai
>> Paolo
>>>  If this patch is based on this assumption then
>>> unfortunately it is wrong :(
>>> 
>>> Paolo
>>> 
>>>>> Noted that bfq_has_work() can be called with 'bfqd->lock' held, thus the
>>>>> lock can't be held in bfq_has_work() to protect 'bfqd->queued'.
>>>>> 
>>>>> Signed-off-by: Yu Kuai <yukuai3@huawei.com>
>>>> 
>>>> Looks good. Feel free to add:
>>>> 
>>>> Reviewed-by: Jan Kara <jack@suse.cz>
>>>> 
>>>> 								Honza
>>>> 
>>>>> ---
>>>>> block/bfq-iosched.c | 16 ++++++++++++----
>>>>> 1 file changed, 12 insertions(+), 4 deletions(-)
>>>>> 
>>>>> diff --git a/block/bfq-iosched.c b/block/bfq-iosched.c
>>>>> index 61750696e87f..740dd83853a6 100644
>>>>> --- a/block/bfq-iosched.c
>>>>> +++ b/block/bfq-iosched.c
>>>>> @@ -2210,7 +2210,11 @@ static void bfq_add_request(struct request *rq)
>>>>> 
>>>>> 	bfq_log_bfqq(bfqd, bfqq, "add_request %d", rq_is_sync(rq));
>>>>> 	bfqq->queued[rq_is_sync(rq)]++;
>>>>> -	bfqd->queued++;
>>>>> +	/*
>>>>> +	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
>>>>> +	 * may be read without holding the lock in bfq_has_work().
>>>>> +	 */
>>>>> +	WRITE_ONCE(bfqd->queued, bfqd->queued + 1);
>>>>> 
>>>>> 	if (RB_EMPTY_ROOT(&bfqq->sort_list) && bfq_bfqq_sync(bfqq)) {
>>>>> 		bfq_check_waker(bfqd, bfqq, now_ns);
>>>>> @@ -2402,7 +2406,11 @@ static void bfq_remove_request(struct request_queue *q,
>>>>> 	if (rq->queuelist.prev != &rq->queuelist)
>>>>> 		list_del_init(&rq->queuelist);
>>>>> 	bfqq->queued[sync]--;
>>>>> -	bfqd->queued--;
>>>>> +	/*
>>>>> +	 * Updating of 'bfqd->queued' is protected by 'bfqd->lock', however, it
>>>>> +	 * may be read without holding the lock in bfq_has_work().
>>>>> +	 */
>>>>> +	WRITE_ONCE(bfqd->queued, bfqd->queued - 1);
>>>>> 	elv_rb_del(&bfqq->sort_list, rq);
>>>>> 
>>>>> 	elv_rqhash_del(q, rq);
>>>>> @@ -5063,11 +5071,11 @@ static bool bfq_has_work(struct blk_mq_hw_ctx *hctx)
>>>>> 	struct bfq_data *bfqd = hctx->queue->elevator->elevator_data;
>>>>> 
>>>>> 	/*
>>>>> -	 * Avoiding lock: a race on bfqd->busy_queues should cause at
>>>>> +	 * Avoiding lock: a race on bfqd->queued should cause at
>>>>> 	 * most a call to dispatch for nothing
>>>>> 	 */
>>>>> 	return !list_empty_careful(&bfqd->dispatch) ||
>>>>> -		bfq_tot_busy_queues(bfqd) > 0;
>>>>> +		READ_ONCE(bfqd->queued);
>>>>> }
>>>>> 
>>>>> static struct request *__bfq_dispatch_request(struct blk_mq_hw_ctx *hctx)
>>>>> -- 
>>>>> 2.31.1
>>>>> 
>>>> -- 
>>>> Jan Kara <jack@suse.com>
>>>> SUSE Labs, CR
>>> 
>> .