io-uring.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
To: Pavel Begunkov <asml.silence@gmail.com>, io-uring@vger.kernel.org
Cc: axboe@kernel.dk
Subject: Re: [RFC 3/3] io_uring: try to batch poll request completion
Date: Fri, 24 Sep 2021 12:28:07 +0800	[thread overview]
Message-ID: <61f88377-9950-7b7f-c350-693c2305449e@linux.alibaba.com> (raw)
In-Reply-To: <a6806f4e-de9f-81b5-2c5e-3e59a6a6d318@gmail.com>

hi,


> On 9/22/21 1:34 PM, Xiaoguang Wang wrote:
>> For an echo-server based on io_uring's IORING_POLL_ADD_MULTI feature,
>> only poll request are completed in task work, normal read/write
>> requests are issued when user app sees cqes on corresponding poll
>> requests, and they will mostly read/write data successfully, which
>> don't need task work. So at least for echo-server model, batching
>> poll request completion properly will give benefits.
>>
>> Currently don't find any appropriate place to store batched poll
>> requests, put them in struct io_submit_state temporarily, which I
>> think it'll need rework in future.
>>
>> Signed-off-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
>> ---
>>   fs/io_uring.c | 64 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-
>>   1 file changed, 63 insertions(+), 1 deletion(-)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index 6fdfb688cf91..14118388bfc6 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -321,6 +321,11 @@ struct io_submit_state {
>>   	 */
>>   	struct io_kiocb		*compl_reqs[IO_COMPL_BATCH];
>>   	unsigned int		compl_nr;
>> +
>> +	struct io_kiocb		*poll_compl_reqs[IO_COMPL_BATCH];
>> +	bool			poll_req_status[IO_COMPL_BATCH];
>> +	unsigned int		poll_compl_nr;
>> +
>>   	/* inline/task_work completion list, under ->uring_lock */
>>   	struct list_head	free_list;
>>   
>> @@ -2093,6 +2098,8 @@ static void ctx_flush_and_put(struct io_ring_ctx *ctx, bool *locked)
>>   	percpu_ref_put(&ctx->refs);
>>   }
>>   
>> +static void io_poll_flush_completions(struct io_ring_ctx *ctx, bool *locked);
>> +
>>   static void tctx_task_work(struct callback_head *cb)
>>   {
>>   	bool locked = false;
>> @@ -2103,8 +2110,11 @@ static void tctx_task_work(struct callback_head *cb)
>>   	while (1) {
>>   		struct io_wq_work_node *node;
>>   
>> -		if (!tctx->task_list.first && locked && ctx->submit_state.compl_nr)
>> +		if (!tctx->task_list.first && locked && (ctx->submit_state.compl_nr ||
>> +		    ctx->submit_state.poll_compl_nr)) {
> io_submit_flush_completions() shouldn't be called if there are no requests... And the
> check is already inside for-next, will be
>
> if (... && locked) {
> 	io_submit_flush_completions();
> 	if (poll_compl_nr)
> 		io_poll_flush_completions();

OK, thanks for pointing this, and I have dropped the poll request 
completion batching patch, since

it shows performance fluctuations, hard to say whether it's useful.


Regards,

Xiaoguang Wang


> }
>
>>   			io_submit_flush_completions(ctx);
>> +			io_poll_flush_completions(ctx, &locked);
>> +		}
>>   
>>   		spin_lock_irq(&tctx->task_lock);
>>   		node = tctx->task_list.first;
>> @@ -5134,6 +5144,49 @@ static inline bool io_poll_complete(struct io_kiocb *req, __poll_t mask)
>>   static bool __io_poll_remove_one(struct io_kiocb *req,
>>   				 struct io_poll_iocb *poll, bool do_cancel);
>>   
>> +static void io_poll_flush_completions(struct io_ring_ctx *ctx, bool *locked)
>> +	__must_hold(&ctx->uring_lock)
>> +{
>> +	struct io_submit_state *state = &ctx->submit_state;
>> +	struct io_kiocb *req, *nxt;
>> +	int i, nr = state->poll_compl_nr;
>> +	bool done, skip_done = true;
>> +
>> +	spin_lock(&ctx->completion_lock);
>> +	for (i = 0; i < nr; i++) {
>> +		req = state->poll_compl_reqs[i];
>> +		done = __io_poll_complete(req, req->result);
> I believe we first need to fix all the poll problems and lay out something more intuitive
> than the current implementation, or it'd be pure hell to do afterwards.
>
> Can be a nice addition, curious about numbers as well.
>
>

  reply	other threads:[~2021-09-24  4:28 UTC|newest]

Thread overview: 11+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-09-22 12:34 [RFC 0/3] improvements for poll requests Xiaoguang Wang
2021-09-22 12:34 ` [RFC 1/3] io_uring: reduce frequent add_wait_queue() overhead for multi-shot poll request Xiaoguang Wang
2021-09-22 17:52   ` Hao Xu
2021-09-22 12:34 ` [RFC 2/3] io_uring: don't get completion_lock in io_poll_rewait() Xiaoguang Wang
2021-09-22 12:34 ` [RFC 3/3] io_uring: try to batch poll request completion Xiaoguang Wang
2021-09-22 16:24   ` Pavel Begunkov
2021-09-24  4:28     ` Xiaoguang Wang [this message]
2021-09-22 17:00   ` Hao Xu
2021-09-22 17:01     ` Hao Xu
2021-09-22 17:09       ` Hao Xu
2021-09-22 13:00 ` [RFC 0/3] improvements for poll requests Jens Axboe

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=61f88377-9950-7b7f-c350-693c2305449e@linux.alibaba.com \
    --to=xiaoguang.wang@linux.alibaba.com \
    --cc=asml.silence@gmail.com \
    --cc=axboe@kernel.dk \
    --cc=io-uring@vger.kernel.org \
    --subject='Re: [RFC 3/3] io_uring: try to batch poll request completion' \
    /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

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).