Re: [PATCH 1/3] blk-mq: optimise rq sort function

From: Nikolay Borisov
Date: Fri Nov 29 2019 - 05:07:14 EST




On 29.11.19 Ð. 11:45 Ñ., Pavel Begunkov wrote:
> On 11/29/2019 11:28 AM, Nikolay Borisov wrote:
>> On 28.11.19 Ð. 23:11 Ñ., Pavel Begunkov wrote:
>>> Check "!=" in multi-layer comparisons. The same memory usage, fewer
>>> instructions, and 2 from 4 jumps are replaced with SETcc.
>>>
>>> Note, that list_sort() doesn't differ 0 and <0.
>>>
>>> Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx>
>>
>> My first reaction was this is wrong since you no longer return negative
>> values. But then I looked into list_sort/merge and this branch
>> 'if (cmp(priv, a, b) <= 0) {' clearly shows this is correct.
>
> Yes, that's why there is a note in the patch description. The same is
> told by list_sort() description.

And of course I have missed your remark in the changelog :)

>
>>
>> So :
>>
>> Reviewed-by: Nikolay Borisov <nborisov@xxxxxxxx>
>
> Thanks for taking a look
>
>>
>>> ---
>>> block/blk-mq.c | 12 ++++--------
>>> 1 file changed, 4 insertions(+), 8 deletions(-)
>>>
>>> diff --git a/block/blk-mq.c b/block/blk-mq.c
>>> index 323c9cb28066..f32a3cfdd34e 100644
>>> --- a/block/blk-mq.c
>>> +++ b/block/blk-mq.c
>>> @@ -1668,14 +1668,10 @@ static int plug_rq_cmp(void *priv, struct list_head *a, struct list_head *b)
>>> struct request *rqa = container_of(a, struct request, queuelist);
>>> struct request *rqb = container_of(b, struct request, queuelist);
>>>
>>> - if (rqa->mq_ctx < rqb->mq_ctx)
>>> - return -1;
>>> - else if (rqa->mq_ctx > rqb->mq_ctx)
>>> - return 1;
>>> - else if (rqa->mq_hctx < rqb->mq_hctx)
>>> - return -1;
>>> - else if (rqa->mq_hctx > rqb->mq_hctx)
>>> - return 1;
>>> + if (rqa->mq_ctx != rqb->mq_ctx)
>>> + return rqa->mq_ctx > rqb->mq_ctx;
>>> + if (rqa->mq_hctx != rqb->mq_hctx)
>>> + return rqa->mq_hctx > rqb->mq_hctx;
>>>
>>> return blk_rq_pos(rqa) > blk_rq_pos(rqb);
>>> }
>>>
>