Re: [PATCHSET 0/2] io_uring: handle short reads internally

[Jens Axboe <axboe@kernel.dk>]

On 8/18/20 11:07 AM, Jeff Moyer wrote:
> Jens Axboe <axboe@kernel.dk> writes:
> 
>>> We must be hitting different problems, then.  I just tested your
>>> 5.7-stable branch (running the test suite from an xfs file system on an
>>> nvme partition with polling enabled), and the read-write test fails:
>>>
>>> Running test read-write:
>>> Non-vectored IO not supported, skipping
>>> cqe res -22, wanted 2048
>>> test_buf_select_short vec failed
>>> Test read-write failed with ret 1
>>>
>>> That's with this head: a451911d530075352fbc7ef9bb2df68145a747ad
>>
>> Not sure what this is, haven't seen that here and my regular liburing
>> runs include both xfs-on-nvme(with poll queues) as one of the test
>> points. Seems to me like there's two oddities in the above:
>>
>> 1) Saying that Non-vectored isn't supported, that is not true on 5.7.
>>    This is due to an -EINVAL return.
>> 2) The test_buf_select_short_vec failure
>>
>> I'll see if I can reproduce this. Anything special otherwise enabled?
>> Scheduler on the nvme device? nr_requests? XFS options?
> 
> No changes from defaults.
> 
> /dev/nvme0n1p1 on /mnt/test type xfs (rw,relatime,seclabel,attr2,inode64,logbufs=8,logbsize=32k,noquota)
> 
> # xfs_info /dev/nvme0n1p1
> meta-data=/dev/nvme0n1p1         isize=512    agcount=4, agsize=22893222 blks
>          =                       sectsz=4096  attr=2, projid32bit=1
>          =                       crc=1        finobt=1, sparse=1, rmapbt=0
>          =                       reflink=1
> data     =                       bsize=4096   blocks=91572885, imaxpct=25
>          =                       sunit=0      swidth=0 blks
> naming   =version 2              bsize=4096   ascii-ci=0, ftype=1
> log      =internal log           bsize=4096   blocks=44713, version=2
>          =                       sectsz=4096  sunit=1 blks, lazy-count=1
> realtime =none                   extsz=4096   blocks=0, rtextents=0
> 
> # cat /sys/block/nvme0n1/queue/scheduler 
> [none] mq-deadline kyber bfq 
> 
> # cat /sys/block/nvme0n1/queue/nr_requests 
> 1023
> 
> # cat /sys/module/nvme/parameters/poll_queues 
> 8
> 
> I'll see if I can figure out what's going on.

Thanks, I'll be a bit busy the next 24h, but let me know how it goes and I'll
dig into this too when I can.

-- 
Jens Axboe