Re: [PATCH 3/3] ARM: tegra114: cpuidle: add powered-down state

[Joseph Lo <josephl-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>]

On Fri, 2013-05-31 at 18:47 +0800, Joseph Lo wrote:
> On Fri, 2013-05-31 at 17:27 +0800, Daniel Lezcano wrote:
> > On 05/31/2013 11:12 AM, Joseph Lo wrote:
> > > Hi Daniel,
> > > 
> > > Thanks for your review.
> > > 
> > > On Thu, 2013-05-30 at 22:35 +0800, Daniel Lezcano wrote:
> > >> On 05/30/2013 01:19 PM, Joseph Lo wrote:
> > >>> This supports CPU core power down on each CPU when CPU idle. When CPU go
> > >>> into this state, it saves it's context and needs a proper configuration
> > >>> in flow controller to power gate the CPU when CPU runs into WFI
> > >>> instruction. And the CPU also needs to set the IRQ as CPU power down idle
> > >>> wake up event in flow controller.
> > >>>
> > >>> Signed-off-by: Joseph Lo <josephl-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> > >>> ---
> > >>
> > >> Hi Joseph,
> > >>
> > >> a new flag has been added in the cpuidle framework to let this one to
> > >> handle the timer broadcast : CPUIDLE_FLAG_TIMER_STOP
> > >>
> > >> It is the commit b60e6a0eb0273132cbb60a9806abf5f47a4aee1c
> > >>
> > >> You can get rid of the clockevent notify stuff by adding this flag to
> > >> the state.
> > >>
> > > 
> > > I just tested this flag on Tegra20, 30 and 114. It is only OK on
> > > Tegra20. It caused a warning message below on Tegra30/114 at boot time.
> > > 
> > > I gave it a quick check I found it can't clean
> > > tick_broadcast_pending_mask sometimes if apply the flag. But I keep the
> > > code right now it's always OK. Not sure why? 
> > 
> > Is it possible you didn't intialized the timer broadcast with
> > CLOCK_EVT_NOTIFY_BROADCAST_ON in the driver and then this one is
> > actually disabled for you driver ?
> > 
> I found if I don't apply the CPUIDLE_FLAG_TIMER_STOP flag, the function
> "cpuidle_setup_broadcast_timer" (drivers/cpuidle/driver.c) would not be
> called. Then it's OK.
> 
> If I apply this flag (with no CONFIG_CPU_IDLE_MULTIPLE_DRIVERS), the
> cpuidle_register_driver only register for CPU0. Then the
> "cpuidle_setup_broadcast_timer" only turn on for CPU0. I think this
> might be the root cause. Not sure this also can reproduce on the other
> SoCs?
> 
Looks it's not related to CLOCK_EVT_NOTIFY_BROADCAST_ON. I still saw the
warning message even enable CONFIG_CPU_IDLE_MULTIPLE_DRIVERS. But if I
move the clockevent_notify of CLOCK_EVT_NOTIFY_BROADCAST_EXIT from idle
framework back to driver, then everything is OK.