Re: [PATCH 6/6] blk-mq: remove REQ_ATOM_STARTED

[jianchao.wang]

ïï tejun

On 12/13/2017 01:26 AM, Tejun Heo wrote:
> Hello, again.
> 
> Sorry, I missed part of your comment in the previous reply.
> 
> On Tue, Dec 12, 2017 at 06:09:32PM +0800, jianchao.wang wrote:
>>>  static void __blk_mq_requeue_request(struct request *rq)
>>>  {
>>> @@ -679,7 +672,7 @@ static void __blk_mq_requeue_request(struct request *rq)
>>>  	wbt_requeue(q->rq_wb, &rq->issue_stat);
>>>  	blk_mq_sched_requeue_request(rq);
>>>  
>>> -	if (test_and_clear_bit(REQ_ATOM_STARTED, &rq->atomic_flags)) {
>>> +	if (blk_mq_rq_state(rq) != MQ_RQ_IDLE) {
>>>  		blk_mq_rq_update_state(rq, MQ_RQ_IDLE);
>>
>> The MQ_RQ_IDLE looks confused here. It is not freed , but idled.
>> And when the requeued request is started again, the generation
>> number will be increased.  But it is not a recycle instance of the
>> request. Maybe another state needed here ?
> 
> I don't quite follow it.  At this point, the request can't be
> in-flight on the device side and is scheduled for re-submission.  I'm
> not sure the distinction from IDLE is necessary.  Am I missing
> something?
Just don't quite understand the strict definition of "generation" you said.
A allocation->free cycle is a generation ? or a idle->in-flight cycle is a
generation? The idle->in-flight cycle could include the requeue case as above.
Can you clarify this?

Thanks
Jianchao
> 
> Thanks.
>