From: Hao Xu <[email protected]>
To: Pavel Begunkov <[email protected]>, Jens Axboe <[email protected]>
Cc: [email protected], Joseph Qi <[email protected]>
Subject: Re: [PATCH 6/6] io_uring: batch completion in prior_task_list
Date: Thu, 18 Nov 2021 18:39:49 +0800 [thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>
在 2021/11/18 上午6:55, Pavel Begunkov 写道:
> On 10/29/21 13:22, Hao Xu wrote:
>> In previous patches, we have already gathered some tw with
>> io_req_task_complete() as callback in prior_task_list, let's complete
>> them in batch regardless uring lock. For instance, we are doing simple
>> direct read, most task work will be io_req_task_complete(), with this
>> patch we don't need to hold uring lock there for long time.
>>
>> Signed-off-by: Hao Xu <[email protected]>
>> ---
>> fs/io_uring.c | 52 ++++++++++++++++++++++++++++++++++++++++++---------
>> 1 file changed, 43 insertions(+), 9 deletions(-)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index 694195c086f3..565cd0b34f18 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -2166,6 +2166,37 @@ static inline unsigned int
>> io_put_rw_kbuf(struct io_kiocb *req)
>> return io_put_kbuf(req, req->kbuf);
>> }
>> +static void handle_prior_tw_list(struct io_wq_work_node *node)
>> +{
>> + struct io_ring_ctx *ctx = NULL;
>> +
>> + do {
>> + struct io_wq_work_node *next = node->next;
>> + struct io_kiocb *req = container_of(node, struct io_kiocb,
>> + io_task_work.node);
>> + if (req->ctx != ctx) {
>> + if (ctx) {
>> + io_commit_cqring(ctx);
>> + spin_unlock(&ctx->completion_lock);
>> + io_cqring_ev_posted(ctx);
>> + percpu_ref_put(&ctx->refs);
>> + }
>> + ctx = req->ctx;
>> + percpu_ref_get(&ctx->refs);
>> + spin_lock(&ctx->completion_lock);
>> + }
>> + __io_req_complete_post(req, req->result, io_put_rw_kbuf(req));
>> + node = next;
>> + } while (node);
>> +
>> + if (ctx) {
>> + io_commit_cqring(ctx);
>> + spin_unlock(&ctx->completion_lock);
>> + io_cqring_ev_posted(ctx);
>> + percpu_ref_put(&ctx->refs);
>> + }
>> +}
>> +
>> static void handle_tw_list(struct io_wq_work_node *node, struct
>> io_ring_ctx **ctx, bool *locked)
>> {
>> do {
>> @@ -2193,25 +2224,28 @@ static void tctx_task_work(struct
>> callback_head *cb)
>> task_work);
>> while (1) {
>> - struct io_wq_work_node *node;
>> - struct io_wq_work_list *merged_list;
>> + struct io_wq_work_node *node1, *node2;
>> - if (!tctx->prior_task_list.first &&
>> - !tctx->task_list.first && locked)
>> + if (!tctx->task_list.first &&
>> + !tctx->prior_task_list.first && locked)
>> io_submit_flush_completions(ctx);
>> spin_lock_irq(&tctx->task_lock);
>> - merged_list = wq_list_merge(&tctx->prior_task_list,
>> &tctx->task_list);
>> - node = merged_list->first;
>> + node1 = tctx->prior_task_list.first;
>> + node2 = tctx->task_list.first;
>> INIT_WQ_LIST(&tctx->task_list);
>> INIT_WQ_LIST(&tctx->prior_task_list);
>> - if (!node)
>> + if (!node2 && !node1)
>> tctx->task_running = false;
>> spin_unlock_irq(&tctx->task_lock);
>> - if (!node)
>> + if (!node2 && !node1)
>> break;
>> - handle_tw_list(node, &ctx, &locked);
>> + if (node1)
>> + handle_prior_tw_list(node1);
>
> IIUC, it moves all IRQ rw completions to this new path even when we already
> have the lock. One concern is that io_submit_flush_completions() is better
> optimised. Should probably be visible for one threaded apps and a bunch of
> other cases.
>
> How about a combined scheme? if we can grab the lock, go through the old
> path, otherwise handle_prior_tw_list(). The rest looks good, will formally
> review once we deal with this one.
Thanks Pavel, I'll look into this patchset soon after
finishing some tests to my io-wq patchset.
>
>> +
>> + if (node2)
>> + handle_tw_list(node2, &ctx, &locked);
>> cond_resched();
>> }
>>
>
next prev parent reply other threads:[~2021-11-18 10:40 UTC|newest]
Thread overview: 15+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-10-29 12:22 [PATCH for-5.16 v4 0/6] task work optimization Hao Xu
2021-10-29 12:22 ` [PATCH 1/6] io-wq: add helper to merge two wq_lists Hao Xu
2021-11-17 22:41 ` Pavel Begunkov
2021-10-29 12:22 ` [PATCH 2/6] io_uring: add a priority tw list for irq completion work Hao Xu
2021-11-17 23:03 ` Pavel Begunkov
2021-11-24 7:53 ` Hao Xu
2021-10-29 12:22 ` [PATCH 3/6] io_uring: add helper for task work execution code Hao Xu
2021-10-29 12:22 ` [PATCH 4/6] io_uring: split io_req_complete_post() and add a helper Hao Xu
2021-10-29 12:22 ` [PATCH 5/6] io_uring: move up io_put_kbuf() and io_put_rw_kbuf() Hao Xu
2021-10-29 12:22 ` [PATCH 6/6] io_uring: batch completion in prior_task_list Hao Xu
2021-11-17 22:55 ` Pavel Begunkov
2021-11-18 10:39 ` Hao Xu [this message]
-- strict thread matches above, loose matches on Subject: below --
2021-11-24 12:21 [PATCH v5 0/6] task work optimization Hao Xu
2021-11-24 12:22 ` [PATCH 6/6] io_uring: batch completion in prior_task_list Hao Xu
2021-11-26 10:07 [PATCH v6 0/6] task work optimization Hao Xu
2021-11-26 10:07 ` [PATCH 6/6] io_uring: batch completion in prior_task_list Hao Xu
2021-11-26 12:56 ` Hao Xu
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=d6d298b5-fbd5-a58d-8fa1-610517a4ca1d@linux.alibaba.com \
[email protected] \
[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