On 01/26/2017 04:14 PM, Bart Van Assche wrote:
> On Thu, 2017-01-26 at 14:51 -0700, Jens Axboe wrote:
>> That is exactly what it means, looks like that one path doesn't handle
>> that. You'd have to exhaust the pool with atomic allocs for this to
>> trigger, we don't do that at all in the normal IO path. So good catch,
>> must be the dm part that enables this since it does NOWAIT allocations.
>>
>>
>> diff --git a/block/blk-mq-sched.c b/block/blk-mq-sched.c
>> index 3136696f4991..c27613de80c5 100644
>> --- a/block/blk-mq-sched.c
>> +++ b/block/blk-mq-sched.c
>> @@ -134,7 +134,8 @@ struct request *blk_mq_sched_get_request(struct request_queue *q,
>> rq = __blk_mq_alloc_request(data, op);
>> } else {
>> rq = __blk_mq_alloc_request(data, op);
>> - data->hctx->tags->rqs[rq->tag] = rq;
>> + if (rq)
>> + data->hctx->tags->rqs[rq->tag] = rq;
>> }
>>
>> if (rq) {
>
> Hello Jens,
>
> With these two patches applied the scheduling-while-atomic complaint and
> the oops are gone. However, some tasks get stuck. Is the console output
> below enough to figure out what is going on or do you want me to bisect
> this? I don't think that any requests got stuck since no pending requests
> are shown in /sys/block/*/mq/*/{pending,*/rq_list}.
What device is stuck? Is it running with an mq scheduler attached, or
with "none"?
Would also be great to see the output of /sys/block/*/mq/*/tags and
sched_tags so we can see if they have anything pending.
>From a quick look at the below, it looks like a request leak. Bisection
would most likely be very helpful.
--
Jens Axboe
--
To unsubscribe from this list: send the line "unsubscribe linux-block" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
