[1/2] OMAP3+: PM: SR: use put_sync_suspend for disabling

Commit Message

Nishanth Menon July 22, 2011, 5:55 a.m. UTC

From: Colin Cross <[e-mail]>

omap_sr_disable_reset_volt is called with irqs off in omapx_enter_sleep,
as part of idle sequence, this eventually calls sr_disable and
pm_runtime_put_sync. pm_runtime_put_sync calls rpm_idle, which will
enable interrupts in order to call the callback. In this short interval
when interrupts are enabled, scenarios such as the following can occur:
while interrupts are enabled, the timer interrupt that is supposed to
wake the device out of idle occurs and is acked, so when the CPU finally
goes to off, the timer is already gone, missing a wakeup event.

Further, as the documentation for runtime states:"
 However, subsystems can use the pm_runtime_irq_safe() helper function
 to tell the PM core that a device's ->runtime_suspend() and ->runtime_resume()
 callbacks should be invoked in atomic context with interrupts disabled
 (->runtime_idle() is still invoked the default way)."

Hence, replace pm_runtime_put_sync with pm_runtime_put_sync_suspend
to invoke the suspend handler and shut off the fclk for SmartReflex
module instead of using the idle handler in interrupt disabled context.

Signed-off-by: Nishanth Menon <[e-mail]>
Signed-off-by: Colin Cross <[e-mail]>
---
 arch/arm/mach-omap2/smartreflex.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

Comments

Kevin Hilman July 22, 2011, 8:14 p.m. UTC | #1

Nishanth Menon <[e-mail]> writes:

> From: Colin Cross <[e-mail]>
>
> omap_sr_disable_reset_volt is called with irqs off in omapx_enter_sleep,
> as part of idle sequence, this eventually calls sr_disable and
> pm_runtime_put_sync. pm_runtime_put_sync calls rpm_idle, which will
> enable interrupts in order to call the callback. In this short interval
> when interrupts are enabled, scenarios such as the following can occur:
> while interrupts are enabled, the timer interrupt that is supposed to
> wake the device out of idle occurs and is acked, so when the CPU finally
> goes to off, the timer is already gone, missing a wakeup event.
>
> Further, as the documentation for runtime states:"
>  However, subsystems can use the pm_runtime_irq_safe() helper function
>  to tell the PM core that a device's ->runtime_suspend() and ->runtime_resume()
>  callbacks should be invoked in atomic context with interrupts disabled
>  (->runtime_idle() is still invoked the default way)."
>
> Hence, replace pm_runtime_put_sync with pm_runtime_put_sync_suspend
> to invoke the suspend handler and shut off the fclk for SmartReflex
> module instead of using the idle handler in interrupt disabled context.
>
> Signed-off-by: Nishanth Menon <[e-mail]>
> Signed-off-by: Colin Cross <[e-mail]>

Great catch!

Looking through Documentation/power/runtime_pm.txt, I see (now) that it
is well documented that _put_sync_suspend() is safe to use from
interrupts-disabled context, but _put_sync() is not on that list.

Queuing this as a fix for v3.1.

Kevin

Patch

diff --git a/arch/arm/mach-omap2/smartreflex.c b/arch/arm/mach-omap2/smartreflex.c
index fb7dc52..33a027f 100644
--- a/arch/arm/mach-omap2/smartreflex.c
+++ b/arch/arm/mach-omap2/smartreflex.c
@@ -622,7 +622,7 @@  void sr_disable(struct voltagedomain *voltdm)
 			sr_v2_disable(sr);
 	}
 
-	pm_runtime_put_sync(&sr->pdev->dev);
+	pm_runtime_put_sync_suspend(&sr->pdev->dev);
 }
 
 /**