public inbox for [email protected]
 help / color / mirror / Atom feed
From: Pavel Begunkov <[email protected]>
To: [email protected]
Cc: Jens Axboe <[email protected]>, [email protected]
Subject: [RFC 08/11] io_uring: wire io_uring specific task work
Date: Thu, 21 Apr 2022 14:44:21 +0100	[thread overview]
Message-ID: <3355bb0947b74a37c32d078f64ef7e9f16cf622b.1650548192.git.asml.silence@gmail.com> (raw)
In-Reply-To: <[email protected]>

Instead of relying on task_work infrastructure for io_uring, add
io_uring specific path. This removes 2 cmpxchg and a
spin_[un]lock_irq pair per batch.

Signed-off-by: Pavel Begunkov <[email protected]>
---
 fs/io_uring.c            | 50 +++++++++++++++++++++-------------------
 include/linux/io_uring.h |  8 +++----
 2 files changed, 30 insertions(+), 28 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 22dcd2fb9687..ea7b0c71ca8b 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -509,7 +509,6 @@ struct io_uring_task {
 	spinlock_t		task_lock;
 	struct io_wq_work_list	task_list;
 	struct io_wq_work_list	prior_task_list;
-	struct callback_head	task_work;
 	struct file		**registered_rings;
 	bool			task_running;
 };
@@ -2541,12 +2540,14 @@ static void handle_tw_list(struct io_wq_work_node *node,
 	} while (node);
 }
 
-static void tctx_task_work(struct callback_head *cb)
+void io_uring_task_work_run(void)
 {
-	bool uring_locked = false;
+	struct io_uring_task *tctx = current->io_uring;
 	struct io_ring_ctx *ctx = NULL;
-	struct io_uring_task *tctx = container_of(cb, struct io_uring_task,
-						  task_work);
+	bool uring_locked = false;
+
+	if (!tctx)
+		return;
 
 	while (1) {
 		struct io_wq_work_node *node1, *node2;
@@ -2581,7 +2582,6 @@ static void io_req_task_work_add(struct io_kiocb *req, bool priority)
 	struct io_wq_work_list *list;
 	struct task_struct *tsk = req->task;
 	struct io_uring_task *tctx = tsk->io_uring;
-	enum task_work_notify_mode notify;
 	struct io_wq_work_node *node;
 	unsigned long flags;
 	bool running;
@@ -2602,21 +2602,19 @@ static void io_req_task_work_add(struct io_kiocb *req, bool priority)
 	spin_unlock_irqrestore(&tctx->task_lock, flags);
 
 	/* task_work already pending, we're done */
-	if (running)
-		return;
-
-	/*
-	 * SQPOLL kernel thread doesn't need notification, just a wakeup. For
-	 * all other cases, use TWA_SIGNAL unconditionally to ensure we're
-	 * processing task_work. There's no reliable way to tell if TWA_RESUME
-	 * will do the job.
-	 */
-	notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL;
-	if (likely(!task_work_add(tsk, &tctx->task_work, notify))) {
-		if (notify == TWA_NONE)
+	if (!running) {
+		/*
+		 * SQPOLL kernel thread doesn't need notification, just a wakeup. For
+		 * all other cases, use TWA_SIGNAL unconditionally to ensure we're
+		 * processing task_work. There's no reliable way to tell if TWA_RESUME
+		 * will do the job.
+		 */
+		if (req->ctx->flags & IORING_SETUP_SQPOLL)
 			wake_up_process(tsk);
-		return;
+		else
+			task_work_notify(tsk, TWA_SIGNAL);
 	}
+	return;
 
 	spin_lock_irqsave(&tctx->task_lock, flags);
 cancel_locked:
@@ -2795,7 +2793,9 @@ static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
 
 static inline bool io_run_task_work(void)
 {
-	if (test_thread_flag(TIF_NOTIFY_SIGNAL) || task_work_pending(current)) {
+	struct io_uring_task *tctx = current->io_uring;
+
+	if (test_thread_flag(TIF_NOTIFY_SIGNAL) || (tctx && tctx->task_running)) {
 		__set_current_state(TASK_RUNNING);
 		clear_notify_signal();
 		io_uring_task_work_run();
@@ -7988,6 +7988,7 @@ static bool io_sqd_handle_event(struct io_sq_data *sqd)
 
 static int io_sq_thread(void *data)
 {
+	struct io_uring_task *tctx = current->io_uring;
 	struct io_sq_data *sqd = data;
 	struct io_ring_ctx *ctx;
 	unsigned long timeout = 0;
@@ -8022,8 +8023,10 @@ static int io_sq_thread(void *data)
 			if (!sqt_spin && (ret > 0 || !wq_list_empty(&ctx->iopoll_list)))
 				sqt_spin = true;
 		}
-		if (io_run_task_work())
+		if (tctx->task_running) {
+			io_uring_task_work_run();
 			sqt_spin = true;
+		}
 
 		if (sqt_spin || !time_after(jiffies, timeout)) {
 			cond_resched();
@@ -8033,7 +8036,7 @@ static int io_sq_thread(void *data)
 		}
 
 		prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
-		if (!io_sqd_events_pending(sqd) && !task_work_pending(current)) {
+		if (!io_sqd_events_pending(sqd) && !tctx->task_running) {
 			bool needs_sched = true;
 
 			list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
@@ -8074,7 +8077,6 @@ static int io_sq_thread(void *data)
 	sqd->thread = NULL;
 	list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
 		io_ring_set_wakeup_flag(ctx);
-	io_run_task_work();
 	mutex_unlock(&sqd->lock);
 
 	audit_free(current);
@@ -9135,7 +9137,6 @@ static __cold int io_uring_alloc_task_context(struct task_struct *task,
 	spin_lock_init(&tctx->task_lock);
 	INIT_WQ_LIST(&tctx->task_list);
 	INIT_WQ_LIST(&tctx->prior_task_list);
-	init_task_work(&tctx->task_work, tctx_task_work);
 	return 0;
 }
 
@@ -10356,6 +10357,7 @@ static __cold void io_uring_cancel_generic(bool cancel_all,
 
 	do {
 		io_uring_drop_tctx_refs(current);
+		io_run_task_work();
 		/* read completions before cancelations */
 		inflight = tctx_inflight(tctx, !cancel_all);
 		if (!inflight)
diff --git a/include/linux/io_uring.h b/include/linux/io_uring.h
index e87ed946214f..331bea825ee6 100644
--- a/include/linux/io_uring.h
+++ b/include/linux/io_uring.h
@@ -5,15 +5,12 @@
 #include <linux/sched.h>
 #include <linux/xarray.h>
 
-static inline void io_uring_task_work_run(void)
-{
-}
-
 #if defined(CONFIG_IO_URING)
 struct sock *io_uring_get_socket(struct file *file);
 void __io_uring_cancel(bool cancel_all);
 void __io_uring_free(struct task_struct *tsk);
 void io_uring_unreg_ringfd(void);
+void io_uring_task_work_run(void);
 
 static inline void io_uring_files_cancel(void)
 {
@@ -46,6 +43,9 @@ static inline void io_uring_files_cancel(void)
 static inline void io_uring_free(struct task_struct *tsk)
 {
 }
+static inline void io_uring_task_work_run(void)
+{
+}
 #endif
 
 #endif
-- 
2.36.0


  parent reply	other threads:[~2022-04-21 13:45 UTC|newest]

Thread overview: 15+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-04-21 13:44 [RFC 00/11] io_uring specific task_work infra Pavel Begunkov
2022-04-21 13:44 ` [RFC 01/11] io_uring: optimise io_req_task_work_add Pavel Begunkov
2022-04-21 13:44 ` [RFC 02/11] io_uringg: add io_should_fail_tw() helper Pavel Begunkov
2022-04-21 13:44 ` [RFC 03/11] io_uring: ban tw queue for exiting processes Pavel Begunkov
2022-04-21 13:44 ` [RFC 04/11] io_uring: don't take ctx refs in tctx_task_work() Pavel Begunkov
2022-04-21 13:44 ` [RFC 05/11] io_uring: add dummy io_uring_task_work_run() Pavel Begunkov
2022-04-21 13:44 ` [RFC 06/11] task_work: add helper for signalling a task Pavel Begunkov
2022-04-21 13:44 ` [RFC 07/11] io_uring: run io_uring task_works on TIF_NOTIFY_SIGNAL Pavel Begunkov
2022-04-21 13:44 ` Pavel Begunkov [this message]
2022-04-21 13:44 ` [RFC 09/11] io_uring: refactor io_run_task_work() Pavel Begunkov
2022-04-21 13:44 ` [RFC 10/11] io_uring: remove priority tw list Pavel Begunkov
2022-04-21 13:44 ` [RFC 11/11] io_uring: lock-free task_work stack Pavel Begunkov
2022-04-21 14:50 ` [RFC 00/11] io_uring specific task_work infra Pavel Begunkov
2022-04-22  8:45   ` Hao Xu
2022-04-22 11:54     ` 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=3355bb0947b74a37c32d078f64ef7e9f16cf622b.1650548192.git.asml.silence@gmail.com \
    [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