Re: [PATCH] block/mq-deadline: Speed up the dispatch of low-priority requests

[Bart Van Assche <bvanassche@acm.org>]

On 8/27/21 6:59 PM, Leizhen (ThunderTown) wrote:
>>> @@ -711,6 +712,8 @@ static void dd_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
>>>  
>>>  	prio = ioprio_class_to_prio[ioprio_class];
>>>  	dd_count(dd, inserted, prio);
>>> +	per_prio = &dd->per_prio[prio];
>>> +	per_prio->nr_queued++;
>>>  
>>>  	if (blk_mq_sched_try_insert_merge(q, rq, &free)) {
>>>  		blk_mq_free_requests(&free);
>>
>> I think the above is wrong - nr_queued should not be incremented if the
>> request is merged into another request. Please move the code that increments
>> nr_queued past the above if-statement.
> 
> So dd_count(dd, inserted, prio) needs to be moved behind "if-statement" as well?

dd_insert_request() is called if a request is inserted and also if it is
requeued. dd_finish_request() is called once per request. Keeping
dd_count() before blk_mq_sched_try_insert_merge() is fine since
blk_mq_free_requests() will call dd_finish_request() indirectly if a
request is merged. However, dd_count() must only happen once per request
and must not be used if a request is requeued.

Additionally, since dd_insert_request() is called with dd->lock held and
since dd_finish_request() is called directly from inside
dd_insert_request() if a request is merged, acquiring dd->lock from
inside dd_finish_request() may trigger a deadlock. A convenient way to
trigger this code path is by running test block/015 from
https://github.com/osandov/blktests/.

Bart.