public inbox for linux-omap@vger.kernel.org
 help / color / mirror / Atom feed
* [PATCH] ARM: OMAP: sched_clock() corrected
@ 2009-02-17 18:21 Aaro Koskinen
  2009-02-18 23:39 ` Tony Lindgren
  0 siblings, 1 reply; 2+ messages in thread
From: Aaro Koskinen @ 2009-02-17 18:21 UTC (permalink / raw)
  To: linux-omap

After my OMAP3 board has been running for a while, I'm seeing weird
latency traces like this:

      sh-1574    0d.h2  153us : do_timer (tick_do_update_jiffies64)
      sh-1574    0d.h2  153us : update_wall_time (do_timer)
      sh-1574    0d.h2  153us!: omap_32k_read (update_wall_time)
      sh-1574    0d.h2 1883us : update_xtime_cache (update_wall_time)
      sh-1574    0d.h2 1883us : clocksource_get_next (update_wall_time)
      sh-1574    0d.h2 1883us+: _spin_lock_irqsave (clocksource_get_next)

and after a while:

      sh-17818   0d.h3  153us : do_timer (tick_do_update_jiffies64)
      sh-17818   0d.h3  153us : update_wall_time (do_timer)
      sh-17818   0d.h3  153us!: omap_32k_read (update_wall_time)
      sh-17818   0d.h3 1915us : update_xtime_cache (update_wall_time)
      sh-17818   0d.h3 1915us+: clocksource_get_next (update_wall_time)
      sh-17818   0d.h3 1945us : _spin_lock_irqsave (clocksource_get_next)

Turns out that sched_clock() is using cyc2ns(), which returns NTP
adjusted time. The sched_clock() frequency should not be adjusted. The
patch deletes omap_32k_ticks_to_nsecs() and rewrites sched_clock()
to do the conversion using the constant multiplier.

Signed-off-by: Aaro Koskinen <Aaro.Koskinen@nokia.com>
---
 arch/arm/plat-omap/common.c |   14 +++++---------
 1 files changed, 5 insertions(+), 9 deletions(-)

diff --git a/arch/arm/plat-omap/common.c b/arch/arm/plat-omap/common.c
index 8c53125..2866612 100644
--- a/arch/arm/plat-omap/common.c
+++ b/arch/arm/plat-omap/common.c
@@ -220,20 +220,16 @@ static struct clocksource clocksource_32k = {
 };
 
 /*
- * Rounds down to nearest nsec.
- */
-unsigned long long omap_32k_ticks_to_nsecs(unsigned long ticks_32k)
-{
-	return cyc2ns(&clocksource_32k, ticks_32k);
-}
-
-/*
  * Returns current time from boot in nsecs. It's OK for this to wrap
  * around for now, as it's just a relative time stamp.
  */
 unsigned long long sched_clock(void)
 {
-	return omap_32k_ticks_to_nsecs(omap_32k_read());
+	unsigned long long ret;
+
+	ret = (unsigned long long)omap_32k_read();
+	ret = (ret * clocksource_32k.mult_orig) >> clocksource_32k.shift;
+	return ret;
 }
 
 static int __init omap_init_clocksource_32k(void)
-- 
1.5.4.3


^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH] ARM: OMAP: sched_clock() corrected
  2009-02-17 18:21 [PATCH] ARM: OMAP: sched_clock() corrected Aaro Koskinen
@ 2009-02-18 23:39 ` Tony Lindgren
  0 siblings, 0 replies; 2+ messages in thread
From: Tony Lindgren @ 2009-02-18 23:39 UTC (permalink / raw)
  To: Aaro Koskinen; +Cc: linux-omap

* Aaro Koskinen <Aaro.Koskinen@nokia.com> [090217 10:27]:
> After my OMAP3 board has been running for a while, I'm seeing weird
> latency traces like this:
> 
>       sh-1574    0d.h2  153us : do_timer (tick_do_update_jiffies64)
>       sh-1574    0d.h2  153us : update_wall_time (do_timer)
>       sh-1574    0d.h2  153us!: omap_32k_read (update_wall_time)
>       sh-1574    0d.h2 1883us : update_xtime_cache (update_wall_time)
>       sh-1574    0d.h2 1883us : clocksource_get_next (update_wall_time)
>       sh-1574    0d.h2 1883us+: _spin_lock_irqsave (clocksource_get_next)
> 
> and after a while:
> 
>       sh-17818   0d.h3  153us : do_timer (tick_do_update_jiffies64)
>       sh-17818   0d.h3  153us : update_wall_time (do_timer)
>       sh-17818   0d.h3  153us!: omap_32k_read (update_wall_time)
>       sh-17818   0d.h3 1915us : update_xtime_cache (update_wall_time)
>       sh-17818   0d.h3 1915us+: clocksource_get_next (update_wall_time)
>       sh-17818   0d.h3 1945us : _spin_lock_irqsave (clocksource_get_next)
> 
> Turns out that sched_clock() is using cyc2ns(), which returns NTP
> adjusted time. The sched_clock() frequency should not be adjusted. The
> patch deletes omap_32k_ticks_to_nsecs() and rewrites sched_clock()
> to do the conversion using the constant multiplier.

Nice catch! Applying to linux-omap and omap-fixes.

Tony


> Signed-off-by: Aaro Koskinen <Aaro.Koskinen@nokia.com>
> ---
>  arch/arm/plat-omap/common.c |   14 +++++---------
>  1 files changed, 5 insertions(+), 9 deletions(-)
> 
> diff --git a/arch/arm/plat-omap/common.c b/arch/arm/plat-omap/common.c
> index 8c53125..2866612 100644
> --- a/arch/arm/plat-omap/common.c
> +++ b/arch/arm/plat-omap/common.c
> @@ -220,20 +220,16 @@ static struct clocksource clocksource_32k = {
>  };
>  
>  /*
> - * Rounds down to nearest nsec.
> - */
> -unsigned long long omap_32k_ticks_to_nsecs(unsigned long ticks_32k)
> -{
> -	return cyc2ns(&clocksource_32k, ticks_32k);
> -}
> -
> -/*
>   * Returns current time from boot in nsecs. It's OK for this to wrap
>   * around for now, as it's just a relative time stamp.
>   */
>  unsigned long long sched_clock(void)
>  {
> -	return omap_32k_ticks_to_nsecs(omap_32k_read());
> +	unsigned long long ret;
> +
> +	ret = (unsigned long long)omap_32k_read();
> +	ret = (ret * clocksource_32k.mult_orig) >> clocksource_32k.shift;
> +	return ret;
>  }
>  
>  static int __init omap_init_clocksource_32k(void)
> -- 
> 1.5.4.3
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-omap" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2009-02-18 23:39 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2009-02-17 18:21 [PATCH] ARM: OMAP: sched_clock() corrected Aaro Koskinen
2009-02-18 23:39 ` Tony Lindgren

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox