Re: [PATCH v2] PM: sleep: Fix runtime PM based cpuidle support

[Rafael J. Wysocki]

On Fri, Oct 29, 2021 at 2:39 PM Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
>
> In the cpuidle-psci case, runtime PM in combination with the generic PM
> domain (genpd), may be used when entering/exiting a shared idlestate. More
> precisely, genpd relies on runtime PM to be enabled for the attached device
> (in this case it belongs to a CPU), to properly manage the reference
> counting of its PM domain.
>
> This works fine most of the time, but during system suspend in
> dpm_suspend_late(), the PM core disables runtime PM for all devices. Beyond
> this point, calls to pm_runtime_get_sync() to runtime resume a device may
> fail and therefore it could also mess up the reference counting in genpd.
>
> To fix this problem, let's call wake_up_all_idle_cpus() in
> dpm_suspend_late(), prior to disabling runtime PM. In this way a device
> that belongs to a CPU, becomes runtime resumed through cpuidle-psci and
> stays like that, because the runtime PM usage count has been bumped in
> device_prepare().
>
> Diagnosed-by: Maulik Shah <mkshah@xxxxxxxxxxxxxx>
> Suggested-by: Rafael J. Wysocki <rafael@xxxxxxxxxx>
> Signed-off-by: Ulf Hansson <ulf.hansson@xxxxxxxxxx>
> ---
>
> Changes in v2:
>         - Moved away from using cpuidle_pause|resume() to solve the problem, but
>         instead just waking up idle CPUs is suffient, due to other recent merged
>         changes.
>
> ---
>  drivers/base/power/main.c | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/drivers/base/power/main.c b/drivers/base/power/main.c
> index ac4dde8fdb8b..2fb08d4f1aca 100644
> --- a/drivers/base/power/main.c
> +++ b/drivers/base/power/main.c
> @@ -1463,6 +1463,7 @@ int dpm_suspend_late(pm_message_t state)
>         int error = 0;
>
>         trace_suspend_resume(TPS("dpm_suspend_late"), state.event, true);
> +       wake_up_all_idle_cpus();
>         mutex_lock(&dpm_list_mtx);
>         pm_transition = state;
>         async_error = 0;
> --

Applied as 5.16-rc material, thanks!