[patch] sched: make cpu_clock() not use the rq clock

[patch] sched: make cpu_clock() not use the rq clock

[Ingo Molnar]

Subject: sched: make cpu_clock() not use the rq clock
From: Ingo Molnar <mingo@elte.hu>

it is enough to disable interrupts to get the precise rq-clock
of the local CPU.

this also solves an NMI watchdog regression: the NMI watchdog
calls touch_softlockup_watchdog(), which might deadlock on
rq->lock if the NMI hits an rq-locked critical section.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
---
 kernel/sched.c |    7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

Index: linux-rt-rebase.q/kernel/sched.c
===================================================================
--- linux-rt-rebase.q.orig/kernel/sched.c
+++ linux-rt-rebase.q/kernel/sched.c
@@ -446,13 +446,12 @@ static inline unsigned long long rq_cloc
  */
 unsigned long long cpu_clock(int cpu)
 {
-	struct rq *rq = cpu_rq(cpu);
 	unsigned long long now;
 	unsigned long flags;
 
-	spin_lock_irqsave(&rq->lock, flags);
-	now = rq_clock(rq);
-	spin_unlock_irqrestore(&rq->lock, flags);
+	local_irq_save(flags);
+	now = rq_clock(cpu_rq(cpu));
+	local_irq_restore(flags);
 
 	return now;
 }

Re: [patch] sched: make cpu_clock() not use the rq clock

[Eric St-Laurent]

On Thu, 2007-26-07 at 11:00 +0200, Ingo Molnar wrote:
> Subject: sched: make cpu_clock() not use the rq clock
> From: Ingo Molnar <mingo@elte.hu>
> 
> it is enough to disable interrupts to get the precise rq-clock
> of the local CPU.

Hi Ingo,

Those new fast nanoseconds resolution clock APIs are nice but it seems
to me that their naming and _where_ they are implemented in the tree is
a little odd, IMO.

We have:

1. sched_clock() is in kernel/sched.c (weak implementation)
2. sched_clock() is in arch/i386/kernel/tsc.c (architecture override)
3. rq_clock() is in kernel/sched.c
4. cpu_clock() is in kernel/sched.c

I would suggest:

1. rename sched_clock() (remove sched_ as it's not sched specific
anymore) and place it in kernel/time/...
2. rename the architecture specific version of it too 

This first function is the basic fast ns clock

3. base your rq_clock() on cpu_clock() (#4) or use the later directly.

This is local to sched.c

4. move cpu_clock() in kernel/time/...

This the per-cpu monotonic version.


See my point? Base the scheduler clock from a general kernel API, not
the other way around.

Just a suggestion.


Best regards,

- Eric

Re: [patch] sched: make cpu_clock() not use the rq clock

[Jeremy Fitzhardinge]

Ingo Molnar wrote:
> Subject: sched: make cpu_clock() not use the rq clock
>   

This subject doesn't make much sense to me.  All this patch does is get
the rq's current time under irq_disable rather than spinlock, right?

    J

> From: Ingo Molnar <mingo@elte.hu>
>
> it is enough to disable interrupts to get the precise rq-clock
> of the local CPU.
>
> this also solves an NMI watchdog regression: the NMI watchdog
> calls touch_softlockup_watchdog(), which might deadlock on
> rq->lock if the NMI hits an rq-locked critical section.
>
> Signed-off-by: Ingo Molnar <mingo@elte.hu>
> ---
>  kernel/sched.c |    7 +++----
>  1 file changed, 3 insertions(+), 4 deletions(-)
>
> Index: linux-rt-rebase.q/kernel/sched.c
> ===================================================================
> --- linux-rt-rebase.q.orig/kernel/sched.c
> +++ linux-rt-rebase.q/kernel/sched.c
> @@ -446,13 +446,12 @@ static inline unsigned long long rq_cloc
>   */
>  unsigned long long cpu_clock(int cpu)
>  {
> -	struct rq *rq = cpu_rq(cpu);
>  	unsigned long long now;
>  	unsigned long flags;
>  
> -	spin_lock_irqsave(&rq->lock, flags);
> -	now = rq_clock(rq);
> -	spin_unlock_irqrestore(&rq->lock, flags);
> +	local_irq_save(flags);
> +	now = rq_clock(cpu_rq(cpu));
> +	local_irq_restore(flags);
>  
>  	return now;
>  }
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>   

Re: [patch] sched: make cpu_clock() not use the rq clock

[Ingo Molnar]

* Jeremy Fitzhardinge <jeremy@goop.org> wrote:

> Ingo Molnar wrote:
> > Subject: sched: make cpu_clock() not use the rq clock
> >   
> 
> This subject doesn't make much sense to me.  All this patch does is get
> the rq's current time under irq_disable rather than spinlock, right?

i typoed it: s/clock/lock

	Ingo