Re: [RFC 2/2] io_uring: mark REQ_NOWAIT for a non-mq queue as unspported

[Jeff Moyer <jmoyer@redhat.com>]

Jens Axboe <axboe@kernel.dk> writes:

> On 5/28/20 4:12 PM, Jeff Moyer wrote:
>> Jens Axboe <axboe@kernel.dk> writes:
>> 
>>>> poll won't work over dm, so that looks correct. What happens if you edit
>>>> it and disable poll? Would be curious to see both buffered = 0 and
>>>> buffered = 1 runs with that.
>>>>
>>>> I'll try this here too.
>>>
>>> I checked, and with the offending commit reverted, it behaves exactly
>>> like it should - io_uring doesn't hit endless retries, and we still
>>> return -EAGAIN to userspace for preadv2(..., RFW_NOWAIT) if not supported.
>>> I've queued up the revert.
>> 
>> With that revert, I now see an issue with an xfs file system on top of
>> an nvme device when running the liburing test suite:
>> 
>> Running test 500f9fbadef8-test
>> Test 500f9fbadef8-test failed with ret 130
>> 
>> That means the test harness timed out, so we never received a
>> completion.
>
> I can't reproduce this. Can you try again, and enable io_uring tracing?
>
> # echo 1 > /sys/kernel/debug/tracing/events/io_uring/enable
>
> run test
>
> send the 'trace' file, or take a look and see what is going on.

I took a look, and it appeared as though the issue was not in the
kernel.  My liburing was not uptodate, and after grabbing the latest,
the test runs to completion.

Thanks!
Jeff