Re: [PATCH] nvme: don't memset() the normal read/write command

[Jens Axboe <axboe@kernel.dk>]

On 10/12/21 12:31 PM, Keith Busch wrote:
> On Tue, Oct 12, 2021 at 12:13:52PM -0600, Jens Axboe wrote:
>> This memset in the fast path costs a lot of cycles on my setup. Here's a
>> top-of-profile of doing ~6.7M IOPS:
>>
>> +    5.90%  io_uring  [nvme]            [k] nvme_queue_rq
>> +    5.32%  io_uring  [nvme_core]       [k] nvme_setup_cmd
>> +    5.17%  io_uring  [kernel.vmlinux]  [k] io_submit_sqes
>> +    4.97%  io_uring  [kernel.vmlinux]  [k] blkdev_direct_IO
>>
>> and a perf diff with this patch:
>>
>>      0.92%     +4.40%  [nvme_core]       [k] nvme_setup_cmd
>>
>> reducing it from 5.3% to only 0.9%. This takes it from the 2nd most
>> cycle consumer to something that's mostly irrelevant.
>>
>> Retain the full clear for the other commands to avoid doing any audits
>> there, and just clear the fields in the rw command manually that we
>> don't already fill.
> 
> Oo, we knew about this optimization *years* ago, yet didn't do anything
> about it! Better late than never.
> 
>   http://lists.infradead.org/pipermail/linux-nvme/2014-May/000837.html

Dang, 2014! Better late than never...

> Acked-by: Keith Busch <kbusch@kernel.org>

Added, thanks.

-- 
Jens Axboe