[PATCH] arm: omap4: allow disabling MPU local timer if 32K timer is enabled

[PATCH] arm: omap4: allow disabling MPU local timer if 32K timer is enabled

[Ming Lei]

With 32K gp timer, tick can be driven and system can run well, so
allow MPU local timer to be disabled if someone requires it, otherwise
MPU local timer is always chosen as the default clock_event_device.

Signed-off-by: Ming Lei <tom.leiming@gmail.com>
---
 arch/arm/mach-omap2/Kconfig |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/arch/arm/mach-omap2/Kconfig b/arch/arm/mach-omap2/Kconfig
index d965da4..12cd602 100644
--- a/arch/arm/mach-omap2/Kconfig
+++ b/arch/arm/mach-omap2/Kconfig
@@ -46,7 +46,7 @@ config ARCH_OMAP4
 	select CPU_V7
 	select ARM_GIC
 	select HAVE_SMP
-	select LOCAL_TIMERS if SMP
+	select LOCAL_TIMERS if (SMP && !OMAP_32K_TIMER)
 	select PL310_ERRATA_588369
 	select PL310_ERRATA_727915
 	select ARM_ERRATA_720789
-- 
1.7.9

[PATCH] arm: omap4: allow disabling MPU local timer if 32K timer is enabled

[Russell King - ARM Linux]

On Sat, Feb 11, 2012 at 04:31:25PM +0800, Ming Lei wrote:
> With 32K gp timer, tick can be driven and system can run well, so
> allow MPU local timer to be disabled if someone requires it, otherwise
> MPU local timer is always chosen as the default clock_event_device.

The point being?

What if you want to use NO_HZ?

[PATCH] arm: omap4: allow disabling MPU local timer if 32K timer is enabled

[Ming Lei]

On Sat, Feb 11, 2012 at 6:03 PM, Russell King - ARM Linux
<linux@arm.linux.org.uk> wrote:
> On Sat, Feb 11, 2012 at 04:31:25PM +0800, Ming Lei wrote:
>> With 32K gp timer, tick can be driven and system can run well, so
>> allow MPU local timer to be disabled if someone requires it, otherwise
>> MPU local timer is always chosen as the default clock_event_device.
>
> The point being?

IMO, 32K gp timer may be more energy-saving than MPU local timer
because gp timer has much less clock frequency and only half interrupts
generated in tick mode compared with mpu local timer.

>
> What if you want to use NO_HZ?

32K gp timer supports oneshot mode, so it should support NO_HZ.  In my test
.config, NO_HZ is enabled and system can run well.

thanks,
-- 
Ming Lei

[PATCH] arm: omap4: allow disabling MPU local timer if 32K timer is enabled

[Russell King - ARM Linux]

On Sat, Feb 11, 2012 at 06:19:41PM +0800, Ming Lei wrote:
> On Sat, Feb 11, 2012 at 6:03 PM, Russell King - ARM Linux
> <linux@arm.linux.org.uk> wrote:
> > On Sat, Feb 11, 2012 at 04:31:25PM +0800, Ming Lei wrote:
> >> With 32K gp timer, tick can be driven and system can run well, so
> >> allow MPU local timer to be disabled if someone requires it, otherwise
> >> MPU local timer is always chosen as the default clock_event_device.
> >
> > The point being?
> 
> IMO, 32K gp timer may be more energy-saving than MPU local timer
> because gp timer has much less clock frequency and only half interrupts
> generated in tick mode compared with mpu local timer.
> 
> >
> > What if you want to use NO_HZ?
> 
> 32K gp timer supports oneshot mode, so it should support NO_HZ.  In my test
> .config, NO_HZ is enabled and system can run well.

Have you actually checked the tick rate of your timer in /proc/interrupts?
Have you checked /proc/timer_list to see what mode the system is in?

[PATCH] arm: omap4: allow disabling MPU local timer if 32K timer is enabled

[Ming Lei]

On Sat, Feb 11, 2012 at 6:26 PM, Russell King - ARM Linux
<linux@arm.linux.org.uk> wrote:
> Have you actually checked the tick rate of your timer in /proc/interrupts?
> Have you checked /proc/timer_list to see what mode the system is in?

After checking, the system works at periodic mode, but I still don't know why
it can't be in one shot mode.

Seems a single clock_event_device can't work well at system with more than
one CPU, so the patch doesn't make sense, please ignore it.


thanks,
-- 
Ming Lei