[RFC PATCH 0/3] timer: patchset focus on del_timer_sync()

[RFC PATCH 0/3] timer: patchset focus on del_timer_sync()

[Yong Zhang]

From: Yong Zhang <yong.zhang@windriver.com>

This is inspired by http://lkml.org/lkml/2010/8/16/291
which catch a lockdep false positive on fake_timer_lock.
When I go into del_timer_sync(), but don't find anything
which prevent del_timer_sync() from using in softirq context,
and indeed it's been used in softirq for some timer(such as
__dst_free()).

Thus, tell others it can't be used in softirq context, and
teach lockdep about that. It's realized by patch-0002.

Your comments are very appreciated.

Thanks,
Yong

---
Yong Zhang (3):
      timer: fix comments of try_to_del_timer_sync()
      timer: del_timer_sync() can be used in softirq context
      timer: warn when del_timer_sync() used in hardirq context

 kernel/timer.c |   16 ++++++++--------
 1 files changed, 8 insertions(+), 8 deletions(-)

[RFC PATCH 1/3] timer: fix comments of try_to_del_timer_sync()

[Yong Zhang]

From: Yong Zhang <yong.zhang@windriver.com>

In commit fd450b7318b75343fd76b3d95416853e34e72c95, it was saying
try_to_del_timer_sync() can be used in interrupt context. So make
the comments consistent to what it should be.

Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Oleg Nesterov <oleg@redhat.com>
---
 kernel/timer.c |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)

diff --git a/kernel/timer.c b/kernel/timer.c
index 97bf05b..55b6a2d 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -943,8 +943,6 @@ EXPORT_SYMBOL(del_timer);
  *
  * This function tries to deactivate a timer. Upon successful (ret >= 0)
  * exit the timer is not queued and the handler is not running on any CPU.
- *
- * It must not be called from interrupt contexts.
  */
 int try_to_del_timer_sync(struct timer_list *timer)
 {
-- 
1.7.0.4

[RFC PATCH 2/3] timer: del_timer_sync() can be used in softirq context

[Yong Zhang]

From: Yong Zhang <yong.zhang@windriver.com>

Actually we have used del_timer_sync() in softirq context for a long
time, such as: __dst_free()::cancel_delayed_work() and maybe the
DEBUG_OBJECTS_TIMERS fundamental code.

So change the comments of it to warn on hardirq context only,
and make lockdep know about this change.

Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
---
 kernel/timer.c |    8 +++-----
 1 files changed, 3 insertions(+), 5 deletions(-)

diff --git a/kernel/timer.c b/kernel/timer.c
index 55b6a2d..08c9559 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -981,7 +981,7 @@ EXPORT_SYMBOL(try_to_del_timer_sync);
  *
  * Synchronization rules: Callers must prevent restarting of the timer,
  * otherwise this function is meaningless. It must not be called from
- * interrupt contexts. The caller must not hold locks which would prevent
+ * hardirq contexts. The caller must not hold locks which would prevent
  * completion of the timer's handler. The timer's handler must not call
  * add_timer_on(). Upon exit the timer is not queued and the handler is
  * not running on any CPU.
@@ -991,12 +991,10 @@ EXPORT_SYMBOL(try_to_del_timer_sync);
 int del_timer_sync(struct timer_list *timer)
 {
 #ifdef CONFIG_LOCKDEP
-	unsigned long flags;
-
-	local_irq_save(flags);
+	local_bh_disable();
 	lock_map_acquire(&timer->lockdep_map);
 	lock_map_release(&timer->lockdep_map);
-	local_irq_restore(flags);
+	local_bh_enable();
 #endif
 
 	for (;;) {
-- 
1.7.0.4

[RFC PATCH 3/3] timer: warn when del_timer_sync() used in hardirq context

[Yong Zhang]

From: Yong Zhang <yong.zhang@windriver.com>

Add explict warning to prevent del_timer_sync() from using
in hardirq context.

Signed-off-by: Yong Zhang <yong.zhang0@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Ingo Molnar <mingo@elte.hu>
---
 kernel/timer.c |    6 +++++-
 1 files changed, 5 insertions(+), 1 deletions(-)

diff --git a/kernel/timer.c b/kernel/timer.c
index 08c9559..1bc4b4a 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -996,7 +996,11 @@ int del_timer_sync(struct timer_list *timer)
 	lock_map_release(&timer->lockdep_map);
 	local_bh_enable();
 #endif
-
+	/*
+	 * don't use it in hardirq context, because it
+	 * could lead to deadlock.
+	 */
+	WARN_ON(in_irq());
 	for (;;) {
 		int ret = try_to_del_timer_sync(timer);
 		if (ret >= 0)
-- 
1.7.0.4

Re: [RFC PATCH 1/3] timer: fix comments of try_to_del_timer_sync()

[Oleg Nesterov]

On 08/24, Yong Zhang wrote:
>
> From: Yong Zhang <yong.zhang@windriver.com>
>
> In commit fd450b7318b75343fd76b3d95416853e34e72c95, it was saying
> try_to_del_timer_sync() can be used in interrupt context.

Yes, but not in UP case.

Please remove "#ifdef CONFIG_SMP" from set_running_timer(), then iirc
it can be used from irq.

Oleg.

Re: [RFC PATCH 1/3] timer: fix comments of try_to_del_timer_sync()

[Yong Zhang]

On Tue, Aug 24, 2010 at 02:11:09PM +0200, Oleg Nesterov wrote:
> On 08/24, Yong Zhang wrote:
> >
> > From: Yong Zhang <yong.zhang@windriver.com>
> >
> > In commit fd450b7318b75343fd76b3d95416853e34e72c95, it was saying
> > try_to_del_timer_sync() can be used in interrupt context.
> 
> Yes, but not in UP case.

Yeah, but in UP case there is no try_to_del_timer_sync(), it's redefined
to del_timer().

> 
> Please remove "#ifdef CONFIG_SMP" from set_running_timer(), then iirc
> it can be used from irq.

I have noticed your comments in the commit log, but I think it's about
introducing the same semantic of try_to_del_timer_sync() on UP as well
as SMP. But this patch is focusing on the current code(SMP special).
Not about realizing try_to_del_timer_sync() on UP case. Do we need
to do that?

Thanks,
Yong

Re: [RFC PATCH 1/3] timer: fix comments of try_to_del_timer_sync()

[Oleg Nesterov]

On 08/24, Yong Zhang wrote:
>
> On Tue, Aug 24, 2010 at 02:11:09PM +0200, Oleg Nesterov wrote:
> > On 08/24, Yong Zhang wrote:
> > >
> > > From: Yong Zhang <yong.zhang@windriver.com>
> > >
> > > In commit fd450b7318b75343fd76b3d95416853e34e72c95, it was saying
> > > try_to_del_timer_sync() can be used in interrupt context.
> >
> > Yes, but not in UP case.
>
> Yeah, but in UP case there is no try_to_del_timer_sync(), it's redefined
> to del_timer().

Ah, indeed, I forgot. This was another reason for the comment.

> > Please remove "#ifdef CONFIG_SMP" from set_running_timer(), then iirc
> > it can be used from irq.
>
> I have noticed your comments in the commit log, but I think it's about
> introducing the same semantic of try_to_del_timer_sync() on UP as well
> as SMP. But this patch is focusing on the current code(SMP special).
> Not about realizing try_to_del_timer_sync() on UP case. Do we need
> to do that?

I dunno.

But look, currently try_to_del_timer_sync() is not allowed from interrupt
even if it works with CONFIG_SMP.

If we "officially" allow it to use from irq, it should work on UP too but
it doesn't. del_timer() can't hang, but it can never return -1 to indicate
we hit the running timer.

Consider:

	// runs in interrup context

	if (try_do_del_timer_sync(&TIMER) > 0)
		kfree(something_which_can_be_used_by_TIMER_func);		

This is unsafe on UP.


del_timer_sync == del_timer is fine on UP. Since it must not be called
from irq, it can never hit the running timer.

Oleg.

Re: [RFC PATCH 1/3] timer: fix comments of try_to_del_timer_sync()

[Yong Zhang]

On Wed, Aug 25, 2010 at 12:31 AM, Oleg Nesterov <oleg@redhat.com> wrote:
> Ah, indeed, I forgot. This was another reason for the comment.
>
>> > Please remove "#ifdef CONFIG_SMP" from set_running_timer(), then iirc
>> > it can be used from irq.
>>
>> I have noticed your comments in the commit log, but I think it's about
>> introducing the same semantic of try_to_del_timer_sync() on UP as well
>> as SMP. But this patch is focusing on the current code(SMP special).
>> Not about realizing try_to_del_timer_sync() on UP case. Do we need
>> to do that?
>
> I dunno.
>
> But look, currently try_to_del_timer_sync() is not allowed from interrupt
> even if it works with CONFIG_SMP.
>
> If we "officially" allow it to use from irq, it should work on UP too

OK. I see.

> but
> it doesn't. del_timer() can't hang, but it can never return -1 to indicate
> we hit the running timer.
>
> Consider:
>
>        // runs in interrup context
>
>        if (try_do_del_timer_sync(&TIMER) > 0)
>                kfree(something_which_can_be_used_by_TIMER_func);
>
> This is unsafe on UP.

You are right. It's unsafe here.

I will try to update this patch which introduce try_do_del_timer_sync()
on UP case.

Thanks,
Yong