[PATCH V5 16/20] ARM: exynos: cpuidle: Move the power sequence call in the cpu_pm notifier

Tomasz Figa t.figa at samsung.com
Thu Jun 26 02:48:24 PDT 2014

Hi Chander,

On 26.06.2014 11:07, Chander Kashyap wrote:
> On Fri, Apr 11, 2014 at 4:10 PM, Daniel Lezcano
> <daniel.lezcano at linaro.org> wrote:


>> @@ -359,6 +373,7 @@ static int exynos_cpu_pm_notifier(struct notifier_block *self,
>>         switch (cmd) {
>>         case CPU_PM_ENTER:
>>                 if (cpu == 0) {
>> +                       exynos_pm_central_suspend();
>>                         exynos_cpu_save_register();
>>                 }
>>                 break;
>> @@ -368,6 +383,7 @@ static int exynos_cpu_pm_notifier(struct notifier_block *self,
>>                         if (!soc_is_exynos5250())
>>                                 scu_enable(S5P_VA_SCU);
>>                         exynos_cpu_restore_register();
>> +                       exynos_pm_central_resume();
> This notifier is called for system wide suspend and cpuidle.
> In case of Exynos cpuidle only AFTR and LPA state need to program
> central_sequencer and store/restore the registers.
> But in 5420 (core-power-down), this is not required, and causing the regression.
> Hence need to remove this notifier, or need to find a way to
> differentiate the cpuidle state.

This patch is already present in v3.16. Moreover, Exynos5420 cpuidle has
not been merged yet. This means that this issue is not a regression and
I believe any further work on this should be carried out as further
patches on top of this change.

Anyway, this change has introduced a regression, though, but in another
area - it broke suspend, at least on Exynos4-based devices, because now
certain steps are performed twice. I've sent a patch for 3.16-rc3 to fix
this by dropping custom suspend-specific syscore ops, effectively moving
most of the handling to CPU PM notifier, which also matches requirements
of AFTR and lower power states. See [1].


However, in this case, moving back to suspend-specific syscore_ops and
simply duplicating some code for lower cpuidle states might be a better
option. Care to send a patch (fix for 3.16, replacing mine) or I should
do it?

Best regards,

More information about the linux-arm-kernel mailing list