From: Pavel Begunkov <[email protected]>
To: Jens Axboe <[email protected]>, [email protected]
Subject: Re: [PATCH 3/3] io_uring: tweak io_req_task_work_add
Date: Wed, 30 Jun 2021 23:10:18 +0100 [thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>
On 6/30/21 10:56 PM, Jens Axboe wrote:
> On 6/30/21 3:45 PM, Jens Axboe wrote:
>> On 6/30/21 3:38 PM, Pavel Begunkov wrote:
>>> On 6/30/21 10:22 PM, Jens Axboe wrote:
>>>> On 6/30/21 3:19 PM, Pavel Begunkov wrote:
>>>>> On 6/30/21 10:17 PM, Jens Axboe wrote:
>>>>>> On 6/30/21 2:54 PM, Pavel Begunkov wrote:
>>>>>>> Whenever possible we don't want to fallback a request. task_work_add()
>>>>>>> will be fine if the task is exiting, so don't check for PF_EXITING,
>>>>>>> there is anyway only a relatively small gap between setting the flag
>>>>>>> and doing the final task_work_run().
>>>>>>>
>>>>>>> Also add likely for the hot path.
>>>>>>
>>>>>> I'm not a huge fan of likely/unlikely, and in particular constructs like:
>>>>>>
>>>>>>> - if (test_bit(0, &tctx->task_state) ||
>>>>>>> + if (likely(test_bit(0, &tctx->task_state)) ||
>>>>>>> test_and_set_bit(0, &tctx->task_state))
>>>>>>> return 0;
>>>>>>
>>>>>> where the state is combined. In any case, it should be a separate
>>>>>> change. If there's an "Also" paragraph in a patch, then that's also
>>>>>> usually a good clue that that particular change should've been
>>>>>> separate :-)
>>>>>
>>>>> Not sure what's wrong with likely above, but how about drop
>>>>> this one then?
>>>>
>>>> Yep I did - we can do the exiting change separately, the commit message
>>>
>>> I think 1-2 is good enough for 5.14, I'll just send it for-next
>>>
>>>> just needs to be clarified a bit on why it's ok to do now. And that
>>>
>>> It should have been ok to do before those 2 patches, but
>>> haven't tracked where it lost actuality.
>>
>> Right, I was thinking it was related to the swapping of the signal
>> exit and task work run ordering. But didn't look that far yet...
>
> BTW, in usual testing, even just the one hunk removing the exit check
> seems to result in quite a lot of memory leaks running
> test/poll-mshot-update. So something is funky with the patch.
I guess you're positive that patches 1-2 have nothing to do
with that. Right?
--
Pavel Begunkov
next prev parent reply other threads:[~2021-06-30 22:10 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-06-30 20:54 [PATCH 5.14 0/3] fallback fix and task_work cleanup Pavel Begunkov
2021-06-30 20:54 ` [PATCH 1/3] io_uring: fix stuck fallback reqs Pavel Begunkov
2021-06-30 20:54 ` [PATCH 2/3] io_uring: simplify task_work func Pavel Begunkov
2021-06-30 20:54 ` [PATCH 3/3] io_uring: tweak io_req_task_work_add Pavel Begunkov
2021-06-30 21:17 ` Jens Axboe
2021-06-30 21:19 ` Pavel Begunkov
2021-06-30 21:22 ` Jens Axboe
2021-06-30 21:38 ` Pavel Begunkov
2021-06-30 21:45 ` Jens Axboe
2021-06-30 21:56 ` Jens Axboe
2021-06-30 22:04 ` Pavel Begunkov
2021-06-30 22:10 ` Pavel Begunkov [this message]
2021-06-30 22:11 ` Jens Axboe
2021-06-30 22:53 ` Pavel Begunkov
2021-06-30 21:57 ` Pavel Begunkov
2021-06-30 21:57 ` Jens Axboe
2021-06-30 21:18 ` [PATCH 5.14 0/3] fallback fix and task_work cleanup Jens Axboe
2021-06-30 21:20 ` Pavel Begunkov
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox