On 04/20/2017 04:57 PM, Bart Van Assche wrote:
> On Thu, 2017-04-20 at 16:45 -0600, Jens Axboe wrote:
>> If we're doing a flush/fua insert, insertion might block on getting
>> a driver tag, if can_block == true. Add a might_sleep check for that,
>> since we just had a bug like that. This will help us catch a similar
>> issue quicker in the future.
>>
>> diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
>> index 9e3c0f92851b..57aec8462e93 100644
>> --- a/block/blk-mq-sched.c
>> +++ b/block/blk-mq-sched.c
>> @@ -372,6 +372,7 @@ void blk_mq_sched_insert_request(struct request *rq, bool at_head,
>> struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
>>
>> if (rq->tag == -1 && op_is_flush(rq->cmd_flags)) {
>> + might_sleep_if(can_block);
>> blk_mq_sched_insert_flush(hctx, rq, can_block);
>> return;
>> }
>
> Hello Jens,
>
> The above patch looks fine to me. But seeing that patch made me wonder
> whether it would be useful to move that might_sleep_if() call into
> blk_mq_get_driver_tag() such that if ever an additional call to
> blk_mq_get_driver_tag() is added that might sleep would also be covered?
Yes, that's not a bad suggestion. Below:
From: Jens Axboe <axboe@xxxxxx>
Subject: [PATCH] blk-mq: add might_sleep check to blk_mq_get_driver_tag()
If the caller passes in wait=true, it has to be able to block
for a driver tag. We just had a bug where flush insertion
would block on tag allocation, while we had preempt disabled.
Ensure that we catch cases like that earlier next time.
Signed-off-by: Jens Axboe <axboe@xxxxxx>
diff --git a/block/blk-mq.c b/block/blk-mq.c
index 992f09772f8a..dd6e5dd62804 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -866,6 +866,8 @@ bool blk_mq_get_driver_tag(struct request *rq, struct blk_mq_hw_ctx **hctx,
.flags = wait ? 0 : BLK_MQ_REQ_NOWAIT,
};
+ might_sleep_if(wait);
+
if (rq->tag != -1)
goto done;
--
Jens Axboe
