IO-Uring Archive on lore.kernel.org
 help / color / Atom feed
From: Peter Zijlstra <peterz@infradead.org>
To: Jens Axboe <axboe@kernel.dk>
Cc: "Carter Li 李通洲" <carter.li@eoitek.com>,
	"Pavel Begunkov" <asml.silence@gmail.com>,
	io-uring <io-uring@vger.kernel.org>,
	"Oleg Nesterov" <oleg@redhat.com>
Subject: Re: [ISSUE] The time cost of IOSQE_IO_LINK
Date: Tue, 18 Feb 2020 14:13:10 +0100
Message-ID: <20200218131310.GZ14914@hirez.programming.kicks-ass.net> (raw)
In-Reply-To: <592cf069-41ee-0bc1-1f83-e058e5dd53ff@kernel.dk>

On Mon, Feb 17, 2020 at 10:16:20AM -0800, Jens Axboe wrote:

> Enjoy! Thanks again for taking a look. I've pushed out the update:
> 
> https://git.kernel.dk/cgit/linux-block/log/?h=io_uring-task-poll

Oleg, we're looking at adding sched-out/sched-in work, based on
task_work. Either see the above git tree, or the folded patch at the
very end of this email.

But this has me wondering about task_work_run(), as it is it will
unconditionally take pi_lock, would not something like this make sense?

(with the caveat that try_cmpxchg() doesn't seem available on !x86 -- I
should go fix that)

---
diff --git a/kernel/task_work.c b/kernel/task_work.c
index 0fef395662a6..bf5a1f37c3c0 100644
--- a/kernel/task_work.c
+++ b/kernel/task_work.c
@@ -93,16 +93,20 @@ void task_work_run(void)
 	struct callback_head *work, *head, *next;
 
 	for (;;) {
+		work = READ_ONCE(task->task_work);
+		if (!work)
+			break
+
 		/*
 		 * work->func() can do task_work_add(), do not set
 		 * work_exited unless the list is empty.
 		 */
 		raw_spin_lock_irq(&task->pi_lock);
 		do {
-			work = READ_ONCE(task->task_works);
-			head = !work && (task->flags & PF_EXITING) ?
-				&work_exited : NULL;
-		} while (cmpxchg(&task->task_works, work, head) != work);
+			head = NULL;
+			if (unlikely(!work && (task->flags & PF_EXITING)))
+				head = &work_exited;
+		} while (!try_cmpxchg(&task->task_works, &work, head));
 		raw_spin_unlock_irq(&task->pi_lock);
 
 		if (!work)



---

Jens, I think you want something like this on top of what you have,
mostly it is adding sched_work_run() to exit_task_work().

---

diff --git a/include/linux/task_work.h b/include/linux/task_work.h
index c688c56ce01d..e0c56f461df6 100644
--- a/include/linux/task_work.h
+++ b/include/linux/task_work.h
@@ -17,13 +17,14 @@ int task_work_add(struct task_struct *task, struct callback_head *twork, bool);
 struct callback_head *task_work_cancel(struct task_struct *, task_work_func_t);
 void task_work_run(void);
 
+int sched_work_add(struct task_struct *task, struct callback_head *work);
+struct callback_head *sched_work_cancel(struct task_struct *, task_work_func_t);
+void sched_work_run(void);
+
 static inline void exit_task_work(struct task_struct *task)
 {
 	task_work_run();
+	sched_work_run();
 }
 
-int sched_work_add(struct task_struct *task, struct callback_head *work);
-void sched_work_run(struct task_struct *task);
-struct callback_head *sched_work_cancel(struct task_struct *, task_work_func_t);
-
 #endif	/* _LINUX_TASK_WORK_H */
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 6fda35072c2d..26d6ecf0e0cc 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -4106,7 +4106,7 @@ void __noreturn do_task_dead(void)
 static bool sched_out_update(struct task_struct *tsk)
 {
 	if (unlikely(tsk->sched_work)) {
-		sched_work_run(tsk);
+		sched_work_run();
 		return true;
 	}
 
@@ -4138,7 +4138,7 @@ static void sched_in_update(struct task_struct *tsk)
 			io_wq_worker_running(tsk);
 	}
 	if (unlikely(tsk->sched_work))
-		sched_work_run(tsk);
+		sched_work_run();
 }
 
 static inline void sched_submit_work(struct task_struct *tsk)
diff --git a/kernel/task_work.c b/kernel/task_work.c
index d73d1cd0c4dd..679550c8c7a4 100644
--- a/kernel/task_work.c
+++ b/kernel/task_work.c
@@ -117,10 +117,10 @@ sched_work_cancel(struct task_struct *task, task_work_func_t func)
 	return __task_work_cancel(task, &task->sched_work, func);
 }
 
-static void __task_work_run(struct task_struct *task,
-			    struct callback_head **headptr)
+static void __task_work_run(struct callback_head **headptr)
 {
 	struct callback_head *work, *head, *next;
+	struct task_struct *task = current;
 
 	for (;;) {
 		/*
@@ -157,10 +157,10 @@ static void __task_work_run(struct task_struct *task,
  */
 void task_work_run(void)
 {
-	__task_work_run(current, &current->task_works);
+	__task_work_run(&current->task_works);
 }
 
-void sched_work_run(struct task_struct *task)
+void sched_work_run()
 {
-	__task_work_run(task, &task->sched_work);
+	__task_work_run(&task->sched_work);
 }


----

folded patches from Jens

---
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 0918904c939d..8d3fbf6e815b 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -649,6 +649,7 @@ struct task_struct {
 	/* Per task flags (PF_*), defined further below: */
 	unsigned int			flags;
 	unsigned int			ptrace;
+	int				on_rq;
 
 #ifdef CONFIG_SMP
 	struct llist_node		wake_entry;
@@ -671,13 +672,14 @@ struct task_struct {
 	int				recent_used_cpu;
 	int				wake_cpu;
 #endif
-	int				on_rq;
 
 	int				prio;
 	int				static_prio;
 	int				normal_prio;
 	unsigned int			rt_priority;
 
+	struct callback_head		*sched_work;
+
 	const struct sched_class	*sched_class;
 	struct sched_entity		se;
 	struct sched_rt_entity		rt;
diff --git a/include/linux/task_work.h b/include/linux/task_work.h
index bd9a6a91c097..c688c56ce01d 100644
--- a/include/linux/task_work.h
+++ b/include/linux/task_work.h
@@ -22,4 +22,8 @@ static inline void exit_task_work(struct task_struct *task)
 	task_work_run();
 }
 
+int sched_work_add(struct task_struct *task, struct callback_head *work);
+void sched_work_run(struct task_struct *task);
+struct callback_head *sched_work_cancel(struct task_struct *, task_work_func_t);
+
 #endif	/* _LINUX_TASK_WORK_H */
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index e94819d573be..6fda35072c2d 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2678,6 +2678,7 @@ int wake_up_state(struct task_struct *p, unsigned int state)
 static void __sched_fork(unsigned long clone_flags, struct task_struct *p)
 {
 	p->on_rq			= 0;
+	p->sched_work			= NULL;
 
 	p->se.on_rq			= 0;
 	p->se.exec_start		= 0;
@@ -4102,10 +4103,12 @@ void __noreturn do_task_dead(void)
 		cpu_relax();
 }
 
-static inline void sched_submit_work(struct task_struct *tsk)
+static bool sched_out_update(struct task_struct *tsk)
 {
-	if (!tsk->state)
-		return;
+	if (unlikely(tsk->sched_work)) {
+		sched_work_run(tsk);
+		return true;
+	}
 
 	/*
 	 * If a worker went to sleep, notify and ask workqueue whether
@@ -4123,18 +4126,10 @@ static inline void sched_submit_work(struct task_struct *tsk)
 		preempt_enable_no_resched();
 	}
 
-	if (tsk_is_pi_blocked(tsk))
-		return;
-
-	/*
-	 * If we are going to sleep and we have plugged IO queued,
-	 * make sure to submit it to avoid deadlocks.
-	 */
-	if (blk_needs_flush_plug(tsk))
-		blk_schedule_flush_plug(tsk);
+	return false;
 }
 
-static void sched_update_worker(struct task_struct *tsk)
+static void sched_in_update(struct task_struct *tsk)
 {
 	if (tsk->flags & (PF_WQ_WORKER | PF_IO_WORKER)) {
 		if (tsk->flags & PF_WQ_WORKER)
@@ -4142,6 +4137,28 @@ static void sched_update_worker(struct task_struct *tsk)
 		else
 			io_wq_worker_running(tsk);
 	}
+	if (unlikely(tsk->sched_work))
+		sched_work_run(tsk);
+}
+
+static inline void sched_submit_work(struct task_struct *tsk)
+{
+	if (!tsk->state)
+		return;
+
+	/* if we processed work, we could be runnable again. check. */
+	if (sched_out_update(tsk) && !tsk->state)
+		return;
+
+	if (tsk_is_pi_blocked(tsk))
+		return;
+
+	/*
+	 * If we are going to sleep and we have plugged IO queued,
+	 * make sure to submit it to avoid deadlocks.
+	 */
+	if (blk_needs_flush_plug(tsk))
+		blk_schedule_flush_plug(tsk);
 }
 
 asmlinkage __visible void __sched schedule(void)
@@ -4154,7 +4171,7 @@ asmlinkage __visible void __sched schedule(void)
 		__schedule(false);
 		sched_preempt_enable_no_resched();
 	} while (need_resched());
-	sched_update_worker(tsk);
+	sched_in_update(tsk);
 }
 EXPORT_SYMBOL(schedule);
 
diff --git a/kernel/task_work.c b/kernel/task_work.c
index 0fef395662a6..d73d1cd0c4dd 100644
--- a/kernel/task_work.c
+++ b/kernel/task_work.c
@@ -3,7 +3,30 @@
 #include <linux/task_work.h>
 #include <linux/tracehook.h>
 
-static struct callback_head work_exited; /* all we need is ->next == NULL */
+static void task_exit_func(struct callback_head *head)
+{
+}
+
+static struct callback_head work_exited = {
+	.next	= NULL,
+	.func	= task_exit_func,
+};
+
+static int __task_work_add(struct task_struct *task,
+			   struct callback_head **headptr,
+			   struct callback_head *work)
+{
+	struct callback_head *head;
+
+	do {
+		head = READ_ONCE(*headptr);
+		if (unlikely(head == &work_exited))
+			return -ESRCH;
+		work->next = head;
+	} while (cmpxchg(headptr, head, work) != head);
+
+	return 0;
+}
 
 /**
  * task_work_add - ask the @task to execute @work->func()
@@ -27,39 +50,31 @@ static struct callback_head work_exited; /* all we need is ->next == NULL */
 int
 task_work_add(struct task_struct *task, struct callback_head *work, bool notify)
 {
-	struct callback_head *head;
+	int ret;
 
-	do {
-		head = READ_ONCE(task->task_works);
-		if (unlikely(head == &work_exited))
-			return -ESRCH;
-		work->next = head;
-	} while (cmpxchg(&task->task_works, head, work) != head);
+	ret = __task_work_add(task, &task->task_works, work);
 
 	if (notify)
 		set_notify_resume(task);
-	return 0;
+
+	return ret;
 }
 
-/**
- * task_work_cancel - cancel a pending work added by task_work_add()
- * @task: the task which should execute the work
- * @func: identifies the work to remove
- *
- * Find the last queued pending work with ->func == @func and remove
- * it from queue.
- *
- * RETURNS:
- * The found work or NULL if not found.
- */
-struct callback_head *
-task_work_cancel(struct task_struct *task, task_work_func_t func)
+int
+sched_work_add(struct task_struct *task, struct callback_head *work)
+{
+	return __task_work_add(task, &task->sched_work, work);
+}
+
+static struct callback_head *__task_work_cancel(struct task_struct *task,
+						struct callback_head **headptr,
+						task_work_func_t func)
 {
-	struct callback_head **pprev = &task->task_works;
+	struct callback_head **pprev = headptr;
 	struct callback_head *work;
 	unsigned long flags;
 
-	if (likely(!task->task_works))
+	if (likely(!(*headptr)))
 		return NULL;
 	/*
 	 * If cmpxchg() fails we continue without updating pprev.
@@ -80,16 +95,31 @@ task_work_cancel(struct task_struct *task, task_work_func_t func)
 }
 
 /**
- * task_work_run - execute the works added by task_work_add()
+ * task_work_cancel - cancel a pending work added by task_work_add()
+ * @task: the task which should execute the work
+ * @func: identifies the work to remove
  *
- * Flush the pending works. Should be used by the core kernel code.
- * Called before the task returns to the user-mode or stops, or when
- * it exits. In the latter case task_work_add() can no longer add the
- * new work after task_work_run() returns.
+ * Find the last queued pending work with ->func == @func and remove
+ * it from queue.
+ *
+ * RETURNS:
+ * The found work or NULL if not found.
  */
-void task_work_run(void)
+struct callback_head *
+task_work_cancel(struct task_struct *task, task_work_func_t func)
+{
+	return __task_work_cancel(task, &task->task_works, func);
+}
+
+struct callback_head *
+sched_work_cancel(struct task_struct *task, task_work_func_t func)
+{
+	return __task_work_cancel(task, &task->sched_work, func);
+}
+
+static void __task_work_run(struct task_struct *task,
+			    struct callback_head **headptr)
 {
-	struct task_struct *task = current;
 	struct callback_head *work, *head, *next;
 
 	for (;;) {
@@ -99,10 +129,10 @@ void task_work_run(void)
 		 */
 		raw_spin_lock_irq(&task->pi_lock);
 		do {
-			work = READ_ONCE(task->task_works);
+			work = READ_ONCE(*headptr);
 			head = !work && (task->flags & PF_EXITING) ?
 				&work_exited : NULL;
-		} while (cmpxchg(&task->task_works, work, head) != work);
+		} while (cmpxchg(headptr, work, head) != work);
 		raw_spin_unlock_irq(&task->pi_lock);
 
 		if (!work)
@@ -116,3 +146,21 @@ void task_work_run(void)
 		} while (work);
 	}
 }
+
+/**
+ * task_work_run - execute the works added by task_work_add()
+ *
+ * Flush the pending works. Should be used by the core kernel code.
+ * Called before the task returns to the user-mode or stops, or when
+ * it exits. In the latter case task_work_add() can no longer add the
+ * new work after task_work_run() returns.
+ */
+void task_work_run(void)
+{
+	__task_work_run(current, &current->task_works);
+}
+
+void sched_work_run(struct task_struct *task)
+{
+	__task_work_run(task, &task->sched_work);
+}

  reply index

Thread overview: 59+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-02-12 16:31 Carter Li 李通洲
2020-02-12 17:11 ` Jens Axboe
2020-02-12 17:22   ` Jens Axboe
2020-02-12 17:29     ` Jens Axboe
2020-02-13  0:33   ` Carter Li 李通洲
2020-02-13 15:08     ` Pavel Begunkov
2020-02-13 15:14       ` Jens Axboe
2020-02-13 15:51         ` Carter Li 李通洲
2020-02-14  1:25           ` Carter Li 李通洲
2020-02-14  2:45             ` Jens Axboe
2020-02-14  5:03               ` Jens Axboe
2020-02-14 15:32                 ` Peter Zijlstra
2020-02-14 15:47                   ` Jens Axboe
2020-02-14 16:18                     ` Jens Axboe
2020-02-14 17:52                       ` Jens Axboe
2020-02-14 20:44                         ` Jens Axboe
2020-02-15  0:16                           ` Carter Li 李通洲
2020-02-15  1:10                             ` Jens Axboe
2020-02-15  1:25                               ` Carter Li 李通洲
2020-02-15  1:27                                 ` Jens Axboe
2020-02-15  6:01                                   ` Jens Axboe
2020-02-15  6:32                                     ` Carter Li 李通洲
2020-02-15 15:11                                       ` Jens Axboe
2020-02-16 19:06                                     ` Pavel Begunkov
2020-02-16 22:23                                       ` Jens Axboe
2020-02-17 10:30                                         ` Pavel Begunkov
2020-02-17 19:30                                           ` Jens Axboe
2020-02-16 23:06                                       ` Jens Axboe
2020-02-16 23:07                                         ` Jens Axboe
2020-02-17 12:09                           ` Peter Zijlstra
2020-02-17 16:12                             ` Jens Axboe
2020-02-17 17:16                               ` Jens Axboe
2020-02-17 17:46                                 ` Peter Zijlstra
2020-02-17 18:16                                   ` Jens Axboe
2020-02-18 13:13                                     ` Peter Zijlstra [this message]
2020-02-18 14:27                                       ` [PATCH] asm-generic/atomic: Add try_cmpxchg() fallbacks Peter Zijlstra
2020-02-18 14:40                                         ` Peter Zijlstra
2020-02-20 10:30                                         ` Will Deacon
2020-02-20 10:37                                           ` Peter Zijlstra
2020-02-20 10:39                                             ` Will Deacon
2020-02-18 14:56                                       ` [ISSUE] The time cost of IOSQE_IO_LINK Oleg Nesterov
2020-02-18 15:07                                         ` Oleg Nesterov
2020-02-18 15:38                                           ` Peter Zijlstra
2020-02-18 16:33                                             ` Jens Axboe
2020-02-18 15:07                                         ` Peter Zijlstra
2020-02-18 15:50                                           ` [PATCH] task_work_run: don't take ->pi_lock unconditionally Oleg Nesterov
2020-02-20 16:39                                             ` Peter Zijlstra
2020-02-20 17:22                                               ` Oleg Nesterov
2020-02-20 17:49                                                 ` Peter Zijlstra
2020-02-21 14:52                                                   ` Oleg Nesterov
2020-02-24 18:47                                                     ` Jens Axboe
2020-02-28 19:17                                                       ` Jens Axboe
2020-02-28 19:25                                                         ` Peter Zijlstra
2020-02-28 19:28                                                           ` Jens Axboe
2020-02-28 20:06                                                             ` Peter Zijlstra
2020-02-28 20:15                                                               ` Jens Axboe
2020-02-18 16:46                                       ` [ISSUE] The time cost of IOSQE_IO_LINK Jens Axboe
2020-02-18 16:52                                         ` Jens Axboe
2020-02-18 13:13                               ` Peter Zijlstra

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=20200218131310.GZ14914@hirez.programming.kicks-ass.net \
    --to=peterz@infradead.org \
    --cc=asml.silence@gmail.com \
    --cc=axboe@kernel.dk \
    --cc=carter.li@eoitek.com \
    --cc=io-uring@vger.kernel.org \
    --cc=oleg@redhat.com \
    /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

IO-Uring Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/io-uring/0 io-uring/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 io-uring io-uring/ https://lore.kernel.org/io-uring \
		io-uring@vger.kernel.org
	public-inbox-index io-uring

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.io-uring


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git