[patch 0/2] hrtimer: generic sleeper infrastructure

[patch 0/2] hrtimer: generic sleeper infrastructure

[Thomas Gleixner]

Andrew,

the removal of the data field in the hrtimer structure introduced a
private sleeper implementation for nanosleep. wakeup is the most common
callback case for timers and we should use a generic implementation
instead of forcing users to reimplement it all over the place.
    
        tglx

--

[patch 1/2] hrtimer

[Thomas Gleixner]

[-- Attachment #1: hrtimer-create-generic-sleeper.patch --]
[-- Type: text/plain, Size: 2334 bytes --]


The removal of the data field in the hrtimer structure enforces the embedding
of the timer into another data structure. nanosleep now uses a private
implementation of the most common used timer callback function (simple task wakeup).
In order to avoid the reimplentation of such functionality all over the place
a generic hrtimer_sleeper functionality is created.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

 include/linux/hrtimer.h |   16 ++++++++++++++++
 kernel/hrtimer.c        |   19 +++++++++++++++++++
 2 files changed, 35 insertions(+)

Index: linux-2.6.16/include/linux/hrtimer.h
===================================================================
--- linux-2.6.16.orig/include/linux/hrtimer.h
+++ linux-2.6.16/include/linux/hrtimer.h
@@ -58,6 +58,19 @@ struct hrtimer {
 };
 
 /**
+ * struct hrtimer_sleeper - simple sleeper structure
+ *
+ * @timer:	embedded timer structure
+ * @task:	task to wake up
+ *
+ * task is set to NULL, when the timer expires.
+ */
+struct hrtimer_sleeper {
+	struct hrtimer timer;
+	struct task_struct *task;
+};
+
+/**
  * struct hrtimer_base - the timer base for a specific clock
  *
  * @index:		clock type index for per_cpu support when moving a timer
@@ -127,6 +140,9 @@ extern long hrtimer_nanosleep(struct tim
 			      const enum hrtimer_mode mode,
 			      const clockid_t clockid);
 
+extern void hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
+				 struct task_struct *tsk);
+
 /* Soft interrupt function to run the hrtimer queues: */
 extern void hrtimer_run_queues(void);
 
Index: linux-2.6.16/kernel/hrtimer.c
===================================================================
--- linux-2.6.16.orig/kernel/hrtimer.c
+++ linux-2.6.16/kernel/hrtimer.c
@@ -656,6 +656,25 @@ void hrtimer_run_queues(void)
  * Sleep related functions:
  */
 
+static int hrtimer_wakeup(struct hrtimer *timer)
+{
+	struct hrtimer_sleeper *t =
+		container_of(timer, struct hrtimer_sleeper, timer);
+	struct task_struct *task = t->task;
+
+	t->task = NULL;
+	if (task)
+		wake_up_process(task);
+
+	return HRTIMER_NORESTART;
+}
+
+void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, task_t *task)
+{
+	sl->timer.function = hrtimer_wakeup;
+	sl->task = task;
+}
+
 struct sleep_hrtimer {
 	struct hrtimer timer;
 	struct task_struct *task;

--

[patch 2/2] hrtimer

[Thomas Gleixner]

[-- Attachment #1: hrtimer-nanosleep-use-generic-sleeper.patch --]
[-- Type: text/plain, Size: 2197 bytes --]


Replace the nanosleep private sleeper functionality by the generic
hrtimer sleeper.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>


 kernel/hrtimer.c |   34 +++++++---------------------------
 1 file changed, 7 insertions(+), 27 deletions(-)

Index: linux-2.6.16/kernel/hrtimer.c
===================================================================
--- linux-2.6.16.orig/kernel/hrtimer.c
+++ linux-2.6.16/kernel/hrtimer.c
@@ -655,7 +655,6 @@ void hrtimer_run_queues(void)
 /*
  * Sleep related functions:
  */
-
 static int hrtimer_wakeup(struct hrtimer *timer)
 {
 	struct hrtimer_sleeper *t =
@@ -675,28 +674,9 @@ void hrtimer_init_sleeper(struct hrtimer
 	sl->task = task;
 }
 
-struct sleep_hrtimer {
-	struct hrtimer timer;
-	struct task_struct *task;
-	int expired;
-};
-
-static int nanosleep_wakeup(struct hrtimer *timer)
-{
-	struct sleep_hrtimer *t =
-		container_of(timer, struct sleep_hrtimer, timer);
-
-	t->expired = 1;
-	wake_up_process(t->task);
-
-	return HRTIMER_NORESTART;
-}
-
-static int __sched do_nanosleep(struct sleep_hrtimer *t, enum hrtimer_mode mode)
+static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
 {
-	t->timer.function = nanosleep_wakeup;
-	t->task = current;
-	t->expired = 0;
+	hrtimer_init_sleeper(t, current);
 
 	do {
 		set_current_state(TASK_INTERRUPTIBLE);
@@ -704,18 +684,18 @@ static int __sched do_nanosleep(struct s
 
 		schedule();
 
-		if (unlikely(!t->expired)) {
+		if (unlikely(t->task)) {
 			hrtimer_cancel(&t->timer);
 			mode = HRTIMER_ABS;
 		}
-	} while (!t->expired && !signal_pending(current));
+	} while (t->task && !signal_pending(current));
 
-	return t->expired;
+	return t->task == NULL;
 }
 
 static long __sched nanosleep_restart(struct restart_block *restart)
 {
-	struct sleep_hrtimer t;
+	struct hrtimer_sleeper t;
 	struct timespec __user *rmtp;
 	struct timespec tu;
 	ktime_t time;
@@ -748,7 +728,7 @@ long hrtimer_nanosleep(struct timespec *
 		       const enum hrtimer_mode mode, const clockid_t clockid)
 {
 	struct restart_block *restart;
-	struct sleep_hrtimer t;
+	struct hrtimer_sleeper t;
 	struct timespec tu;
 	ktime_t rem;
 

--

Re: [patch 2/2] hrtimer

[Andrew Morton]

Thomas Gleixner <tglx@linutronix.de> wrote:
>
> 
> Replace the nanosleep private sleeper functionality by the generic
> hrtimer sleeper.
> 
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> Signed-off-by: Ingo Molnar <mingo@elte.hu>
> 
> 
>  kernel/hrtimer.c |   34 +++++++---------------------------
>  1 file changed, 7 insertions(+), 27 deletions(-)
> 
> Index: linux-2.6.16/kernel/hrtimer.c
> ===================================================================
> --- linux-2.6.16.orig/kernel/hrtimer.c
> +++ linux-2.6.16/kernel/hrtimer.c
> @@ -655,7 +655,6 @@ void hrtimer_run_queues(void)
>  /*
>   * Sleep related functions:
>   */
> -
>  static int hrtimer_wakeup(struct hrtimer *timer)
>  {
>  	struct hrtimer_sleeper *t =
> @@ -675,28 +674,9 @@ void hrtimer_init_sleeper(struct hrtimer
>  	sl->task = task;
>  }
>  
> -struct sleep_hrtimer {
> -	struct hrtimer timer;
> -	struct task_struct *task;
> -	int expired;
> -};
> -
> -static int nanosleep_wakeup(struct hrtimer *timer)
> -{
> -	struct sleep_hrtimer *t =
> -		container_of(timer, struct sleep_hrtimer, timer);
> -
> -	t->expired = 1;
> -	wake_up_process(t->task);
> -
> -	return HRTIMER_NORESTART;
> -}
> -
> -static int __sched do_nanosleep(struct sleep_hrtimer *t, enum hrtimer_mode mode)
> +static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
>  {
> -	t->timer.function = nanosleep_wakeup;
> -	t->task = current;
> -	t->expired = 0;
> +	hrtimer_init_sleeper(t, current);
>  
>  	do {
>  		set_current_state(TASK_INTERRUPTIBLE);
> @@ -704,18 +684,18 @@ static int __sched do_nanosleep(struct s
>  
>  		schedule();
>  
> -		if (unlikely(!t->expired)) {
> +		if (unlikely(t->task)) {
>  			hrtimer_cancel(&t->timer);
>  			mode = HRTIMER_ABS;
>  		}
> -	} while (!t->expired && !signal_pending(current));
> +	} while (t->task && !signal_pending(current));
>  
> -	return t->expired;
> +	return t->task == NULL;
>  }

This all looks vaguely racy.  hrtimer_wakeup() will set t->task to NULL
without barriers, locks or anything.  And the waiter here can break out of
schedule() due to signal delivery while a wakeup is in progress.

So the value of t->task here is fairly meaningless.  Ot just depends on how
far the waker has got through hrtimer_wakeup().

Maybe that doesn't matter, because hrtimer_cancel() will spin until
hrtimer_wakeup() has completed anyway, but could you please recheck and
confirm that this is all solid?

Re: [patch 2/2] hrtimer

[Thomas Gleixner]

On Sat, 2006-03-25 at 18:32 -0800, Andrew Morton wrote:

> This all looks vaguely racy.  hrtimer_wakeup() will set t->task to NULL
> without barriers, locks or anything.  And the waiter here can break out of
> schedule() due to signal delivery while a wakeup is in progress.

We set task = NULL before wake_up_process() which acts as a barrier.

> So the value of t->task here is fairly meaningless.  Ot just depends on how
> far the waker has got through hrtimer_wakeup().
> 
> Maybe that doesn't matter, because hrtimer_cancel() will spin until
> hrtimer_wakeup() has completed anyway, but could you please recheck and
> confirm that this is all solid?

Right, either it waits for the running timer or in case the wakeup
happens between

if (unlikely(t->task)) {

and

	hrtimer_cancel(&t->timer);

then hrtimer_cancel will see that the timer is inactive and we drop out
of the loop because the while(t->task) condition is not longer true. 

	tglx

Re: [patch 2/2] hrtimer

[Oleg Nesterov]

I also think this is racy.

CPU_0					CPU_1

hrtimer_wakeup:

	task = t->task;
	t->task = NULL;

	<--- INTERRUPT --->

					task is woken by signal,
					do_nanosleep() sees t->task == NULL,
					returns without hrtimer_cancel(),
					and __exits__.

	<--- RESUME --->

	wake_up_process(task);

Instead of exit(), 'task' can go to TASK_STOPPED or TASK_UNINTERRUPTIBLE
after return from do_nanosleep(), it will be awakened by hrtimer_wakeup()
unexpectedly.

Oleg.

Re: [patch 2/2] hrtimer

[Roman Zippel]

Hi,

On Tue, 28 Mar 2006, Oleg Nesterov wrote:

> I also think this is racy.
> 
> CPU_0					CPU_1
> 
> hrtimer_wakeup:
> 
> 	task = t->task;
> 	t->task = NULL;
> 
> 	<--- INTERRUPT --->
> 
> 					task is woken by signal,
> 					do_nanosleep() sees t->task == NULL,
> 					returns without hrtimer_cancel(),
> 					and __exits__.
> 
> 	<--- RESUME --->
> 
> 	wake_up_process(task);
> 
> Instead of exit(), 'task' can go to TASK_STOPPED or TASK_UNINTERRUPTIBLE
> after return from do_nanosleep(), it will be awakened by hrtimer_wakeup()
> unexpectedly.

Indeed and my original patch did call hrtimer_cancel() unconditionally to 
synchronize with a possibly running timer.
Thomas, could you please document it a bit better, when you modify my 
patches?

bye, Roman

Re: [patch 2/2] hrtimer

[Thomas Gleixner]

On Tue, 2006-03-28 at 03:55 +0400, Oleg Nesterov wrote:
> Instead of exit(), 'task' can go to TASK_STOPPED or TASK_UNINTERRUPTIBLE
> after return from do_nanosleep(), it will be awakened by hrtimer_wakeup()
> unexpectedly.

Yep, you are right.

Index: linux-2.6.16/kernel/hrtimer.c
===================================================================
--- linux-2.6.16.orig/kernel/hrtimer.c
+++ linux-2.6.16/kernel/hrtimer.c
@@ -684,10 +684,9 @@ static int __sched do_nanosleep(struct h
 
 		schedule();
 
-		if (unlikely(t->task)) {
-			hrtimer_cancel(&t->timer);
-			mode = HRTIMER_ABS;
-		}
+		hrtimer_cancel(&t->timer);
+		mode = HRTIMER_ABS;
+
 	} while (t->task && !signal_pending(current));
 
 	return t->task == NULL;