linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Pavel Begunkov <asml.silence@gmail.com>
To: Jens Axboe <axboe@kernel.dk>,
	io-uring@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 1/2] io_uring: trigger timeout after any sqe->off CQEs
Date: Thu, 23 Apr 2020 01:20:47 +0300	[thread overview]
Message-ID: <cf991f17-ad5f-c80a-d993-544d8746ac72@gmail.com> (raw)
In-Reply-To: <3fe32d07-10e6-4a5a-1390-f03ec4a09c6f@gmail.com>

On 20/04/2020 23:15, Pavel Begunkov wrote:
> On 20/04/2020 23:12, Pavel Begunkov wrote:
>> On 20/04/2020 22:40, Jens Axboe wrote:
>>> On 4/18/20 11:20 AM, Pavel Begunkov wrote:
>>>> +static void __io_flush_timeouts(struct io_ring_ctx *ctx)
>>>> +{
>>>> +	u32 end, start;
>>>> +
>>>> +	start = end = ctx->cached_cq_tail;
>>>> +	do {
>>>> +		struct io_kiocb *req = list_first_entry(&ctx->timeout_list,
>>>> +							struct io_kiocb, list);
>>>> +
>>>> +		if (req->flags & REQ_F_TIMEOUT_NOSEQ)
>>>> +			break;
>>>> +		/*
>>>> +		 * multiple timeouts may have the same target,
>>>> +		 * check that @req is in [first_tail, cur_tail]
>>>> +		 */
>>>> +		if (!io_check_in_range(req->timeout.target_cq, start, end))
>>>> +			break;
>>>> +
>>>> +		list_del_init(&req->list);
>>>> +		io_kill_timeout(req);
>>>> +		end = ctx->cached_cq_tail;
>>>> +	} while (!list_empty(&ctx->timeout_list));
>>>> +}
>>>> +
>>>>  static void io_commit_cqring(struct io_ring_ctx *ctx)
>>>>  {
>>>>  	struct io_kiocb *req;
>>>>  
>>>> -	while ((req = io_get_timeout_req(ctx)) != NULL)
>>>> -		io_kill_timeout(req);
>>>> +	if (!list_empty(&ctx->timeout_list))
>>>> +		__io_flush_timeouts(ctx);
>>>>  
>>>>  	__io_commit_cqring(ctx);
>>>>  
>>>
>>> Any chance we can do this without having to iterate timeouts on the
>>> completion path?
>>>
>>
>> If you mean the one in __io_flush_timeouts(), then no, unless we forbid timeouts
>> with identical target sequences + some extra constraints. The loop there is not
>> new, it iterates only over timeouts, that need to be completed, and removes
>> them. That's amortised O(1).
> 
> We can think about adding unlock/lock, if that's what you are thinking about.
> 
> 
>> On the other hand, there was a loop in io_timeout_fn() doing in total O(n^2),
>> and it was killed by this patch.
> 

Any thoughts on this?

I'll return fixing the last timeout bug I saw, but I'd prefer to know on top of
what to do that.

-- 
Pavel Begunkov

  reply	other threads:[~2020-04-22 22:21 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-04-18 17:20 [PATCH 0/2] CQ vs SQ timeout accounting Pavel Begunkov
2020-04-18 17:20 ` [PATCH 1/2] io_uring: trigger timeout after any sqe->off CQEs Pavel Begunkov
2020-04-20 19:40   ` Jens Axboe
2020-04-20 20:12     ` Pavel Begunkov
2020-04-20 20:15       ` Pavel Begunkov
2020-04-22 22:20         ` Pavel Begunkov [this message]
2020-04-22 22:23           ` Jens Axboe
2020-04-22 22:51             ` Pavel Begunkov
2020-04-18 17:20 ` [PATCH 2/2] io_uring: don't trigger timeout with another t-out 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 \
    --in-reply-to=cf991f17-ad5f-c80a-d993-544d8746ac72@gmail.com \
    --to=asml.silence@gmail.com \
    --cc=axboe@kernel.dk \
    --cc=io-uring@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    /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;
as well as URLs for NNTP newsgroup(s).