lkml.org 
[lkml]   [2011]   [Jul]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 2/3] ARM: Call idle notifiers
On Mon, Jul 11, 2011 at 09:50:04PM +0200, Frederic Weisbecker wrote:
> On Mon, Jun 27, 2011 at 07:46:29PM -0700, Todd Poynor wrote:
> > Change-Id: Id833e61c13baa1783705ac9e9046d1f0cc90c95e
> > Signed-off-by: Todd Poynor <toddpoynor@google.com>
> > ---
> > arch/arm/kernel/process.c | 2 ++
> > 1 files changed, 2 insertions(+), 0 deletions(-)
> >
> > diff --git a/arch/arm/kernel/process.c b/arch/arm/kernel/process.c
> > index 5e1e541..1b9101e 100644
> > --- a/arch/arm/kernel/process.c
> > +++ b/arch/arm/kernel/process.c
> > @@ -184,6 +184,7 @@ void cpu_idle(void)
> > while (1) {
> > tick_nohz_stop_sched_tick(1);
> > leds_event(led_idle_start);
> > + idle_notifier_call_chain(IDLE_START);
> > while (!need_resched()) {
> > #ifdef CONFIG_HOTPLUG_CPU
> > if (cpu_is_offline(smp_processor_id()))
> > @@ -208,6 +209,7 @@ void cpu_idle(void)
> > }
> > }
> > leds_event(led_idle_end);
> > + idle_notifier_call_chain(IDLE_END);
> > tick_nohz_restart_sched_tick();
> > preempt_enable_no_resched();
> > schedule();
>
> You seem to use this notifier with different semantics than x86.
> x86 notifies idle state when it knows it goes to sleep and exit it any
> time it gets interrupted. And it does that every time in the need_resched()
> loop.
>
> But here in ARM you enter idle only once before the loop (and you don't even
> know if you will enter the loop). And you don't notify idle exit state on interrupts.
>
> So if in the end this idle notifier is something that is really wanted, it needs
> to have a consistant behaviour across archs.

Yes, I didn't want to change the behavior of the existing notifiers
being converted in these patches, but eventually one would want
cross-arch idle callbacks with consistent behavior, and the
existing arch-specific notifications have different semantics.
My goal is to notify when the OS scheduler enters and exits its idle
loop, so the existing ARM semantics are what I'm aiming for; the x86_64
behavior is probably further evidence that it is really a
cpuidle-style operation being performed. If the idle notifiers do
find any traction then it sounds like moving the notification to the
common schedule code would be the right thing to do.


Thanks -- Todd


\
 
 \ /
  Last update: 2011-07-14 00:57    [from the cache]
©2003-2011 Jasper Spaans