On 03/14/2017 06:59 AM, Adrian Hunter wrote:
> On 13/03/17 16:19, Jens Axboe wrote:
>> On 03/13/2017 03:25 AM, Adrian Hunter wrote:
>>> On 11/03/17 00:05, Jens Axboe wrote:
>>>> On 03/10/2017 07:21 AM, Adrian Hunter wrote:
>>>>>> Essentially I take out that thread and replace it with this one worker
>>>>>> introduced in this very patch. I agree the driver can block in many ways
>>>>>> and that is why I need to have it running in process context, and this
>>>>>> is what the worker introduced here provides.
>>>>>
>>>>> The last time I looked at the blk-mq I/O scheduler code, it pulled up to
>>>>> qdepth requests from the I/O scheduler and left them on a local list while
>>>>> running ->queue_rq(). That means blocking in ->queue_rq() leaves some
>>>>> number of requests in limbo (not issued but also not in the I/O scheduler)
>>>>> for that time.
>>>>
>>>> Look again, if we're not handling the requeued dispatches, we pull one
>>>> at the time from the scheduler.
>>>>
>>>
>>> That's good :-)
>>>
>>> Now the next thing ;-)
>>>
>>> It looks like we either set BLK_MQ_F_BLOCKING and miss the possibility of
>>> issuing synchronous requests immediately, or we don't set BLK_MQ_F_BLOCKING
>>> in which case we are never allowed to sleep in ->queue_rq(). Is that true?
>>
>> Only one of those statements is true - if you don't set BLK_MQ_F_BLOCKING,
>> then you may never block in your ->queue_rq() function. But if you do set it,
>> it does not preclude immediate issue of sync requests.
>
> I meant it gets put to the workqueue rather than issued in the context of
> the submitter.
There's one case that doesn't look like it was converted properly, but
that's a mistake. The general insert-and-run cases run inline if we can,
but the direct-issue needs a fixup, see below.
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 159187a28d66..4196d6bee92d 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -1434,7 +1434,8 @@ static blk_qc_t request_to_qc_t(struct blk_mq_hw_ctx *hctx, struct request *rq)
return blk_tag_to_qc_t(rq->internal_tag, hctx->queue_num, true);
}
-static void blk_mq_try_issue_directly(struct request *rq, blk_qc_t *cookie)
+static void blk_mq_try_issue_directly(struct request *rq, blk_qc_t *cookie,
+ bool can_block)
{
struct request_queue *q = rq->q;
struct blk_mq_queue_data bd = {
@@ -1475,7 +1476,7 @@ static void blk_mq_try_issue_directly(struct request *rq, blk_qc_t *cookie)
}
insert:
- blk_mq_sched_insert_request(rq, false, true, true, false);
+ blk_mq_sched_insert_request(rq, false, true, false, can_block);
}
/*
@@ -1569,11 +1570,11 @@ static blk_qc_t blk_mq_make_request(struct request_queue *q, struct bio *bio)
if (!(data.hctx->flags & BLK_MQ_F_BLOCKING)) {
rcu_read_lock();
- blk_mq_try_issue_directly(old_rq, &cookie);
+ blk_mq_try_issue_directly(old_rq, &cookie, false);
rcu_read_unlock();
} else {
srcu_idx = srcu_read_lock(&data.hctx->queue_rq_srcu);
- blk_mq_try_issue_directly(old_rq, &cookie);
+ blk_mq_try_issue_directly(old_rq, &cookie, true);
srcu_read_unlock(&data.hctx->queue_rq_srcu, srcu_idx);
}
goto done;
--
Jens Axboe
