[PATCH] powerpc/pseries: fix idle sleep early exit state

[PATCH] powerpc/pseries: fix idle sleep early exit state

[Sebastien Dugue]

  In pseries_dedicated_idle_sleep(), if we need to exit idle during the
snooze period (i.e. need_resched or cpu has been offlined), then we should
re-disable the interrupts and clear TIF_POLLING_NRFLAG before leaving.

Signed-off-by: Sebastien Dugue <sebastien.dugue@bull.net>
Cc: Paul Mackerras <paulus@samba.org>
---
 arch/powerpc/platforms/pseries/setup.c |   10 ++++++++--
 1 files changed, 8 insertions(+), 2 deletions(-)

diff --git a/arch/powerpc/platforms/pseries/setup.c b/arch/powerpc/platforms/pseries/setup.c
index ec34170..8270f04 100644
--- a/arch/powerpc/platforms/pseries/setup.c
+++ b/arch/powerpc/platforms/pseries/setup.c
@@ -504,8 +504,13 @@ static void pseries_dedicated_idle_sleep(void)
 		set_thread_flag(TIF_POLLING_NRFLAG);
 
 		while (get_tb() < start_snooze) {
-			if (need_resched() || cpu_is_offline(cpu))
-				goto out;
+			if (need_resched() || cpu_is_offline(cpu)) {
+				HMT_medium();
+				clear_thread_flag(TIF_POLLING_NRFLAG);
+				smp_mb();
+				local_irq_disable();
+				goto out2;
+			}
 			ppc64_runlatch_off();
 			HMT_low();
 			HMT_very_low();
@@ -523,6 +528,7 @@ static void pseries_dedicated_idle_sleep(void)
 
 out:
 	HMT_medium();
+out2:
 	out_purr = mfspr(SPRN_PURR);
 	get_lppaca()->wait_state_cycles += out_purr - in_purr;
 	get_lppaca()->donate_dedicated_cpu = 0;
-- 
1.6.0.1.308.gede4c

Re: [PATCH] powerpc/pseries: fix idle sleep early exit state

[Paul Mackerras]

Sebastien Dugue writes:

>   In pseries_dedicated_idle_sleep(), if we need to exit idle during the
> snooze period (i.e. need_resched or cpu has been offlined), then we should
> re-disable the interrupts and clear TIF_POLLING_NRFLAG before leaving.

Are you doing this because of a bug you actually observed, or did you
just find this by inspection?  I don't believe it's necessary to do
these things (disabling interrupts, clearing TIF_POLLING_NRFLAG)
because the code that calls it immediately reenables interrupts and
sets TIF_POLLING_NRFLAG again.  From arch/powerpc/kernel/idle.c (line
73):

				if (!need_resched() && !cpu_should_die())
					ppc_md.power_save();

				local_irq_enable();
				set_thread_flag(TIF_POLLING_NRFLAG);

Paul.

Re: [PATCH] powerpc/pseries: fix idle sleep early exit state

[Sebastien Dugue]

On Wed, 3 Dec 2008 20:38:54 +1100 Paul Mackerras <paulus@samba.org> wrote:

> Sebastien Dugue writes:
> 
> >   In pseries_dedicated_idle_sleep(), if we need to exit idle during the
> > snooze period (i.e. need_resched or cpu has been offlined), then we should
> > re-disable the interrupts and clear TIF_POLLING_NRFLAG before leaving.
> 
> Are you doing this because of a bug you actually observed, or did you
> just find this by inspection?

  By inspection, and just to be consistent with the other way out. But
you're right, it's not really necessary here.

>  I don't believe it's necessary to do
> these things (disabling interrupts, clearing TIF_POLLING_NRFLAG)
> because the code that calls it immediately reenables interrupts and
> sets TIF_POLLING_NRFLAG again.  From arch/powerpc/kernel/idle.c (line
> 73):
> 
> 				if (!need_resched() && !cpu_should_die())
> 					ppc_md.power_save();
> 
> 				local_irq_enable();
> 				set_thread_flag(TIF_POLLING_NRFLAG);
> 
> Paul.
> 

  Thanks,

  Sebastien.