Re: [PATCH 2/4] ACPI: PM: Remove redundant cache flushing

[Rafael J. Wysocki]

On Mon, Dec 6, 2021 at 1:30 PM Kirill A. Shutemov
<kirill.shutemov@xxxxxxxxxxxxxxx> wrote:
>
> ACPICA code takes care about cache flushing on S1/S2/S3 in
> acpi_hw_extended_sleep() and acpi_hw_legacy_sleep().
>
> acpi_suspend_enter() calls into ACPICA code via acpi_enter_sleep_state()
> for S1 or x86_acpi_suspend_lowlevel() for S3. It only need to flush
> cache for S2 (not sure if this call path is ever used for S2).
>
> acpi_sleep_prepare() call tree:
>   __acpi_pm_prepare()
>     acpi_pm_prepare()
>       acpi_suspend_ops::prepare_late()
>       acpi_hibernation_ops::pre_snapshot()
>       acpi_hibernation_ops::prepare()
>     acpi_suspend_begin_old()
>       acpi_suspend_begin_old::begin()
>   acpi_hibernation_begin_old()
>     acpi_hibernation_ops_old::acpi_hibernation_begin_old()
>   acpi_power_off_prepare()
>     pm_power_off_prepare()
>
> Hibernation (S4) and Power Off (S5) don't require cache flushing. So,
> the only interesting callsites are acpi_suspend_ops::prepare_late() and
> acpi_suspend_begin_old::begin(). Both of them have cache flush on
> ->enter() operation in acpi_suspend_enter().
>
> Remove redundant ACPI_FLUSH_CPU_CACHE() in acpi_sleep_prepare() and
> acpi_suspend_enter().
>
> Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
> ---
>  drivers/acpi/sleep.c | 7 +++----
>  1 file changed, 3 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/acpi/sleep.c b/drivers/acpi/sleep.c
> index eaa47753b758..14e8df0ac762 100644
> --- a/drivers/acpi/sleep.c
> +++ b/drivers/acpi/sleep.c
> @@ -73,7 +73,6 @@ static int acpi_sleep_prepare(u32 acpi_state)
>                 acpi_set_waking_vector(acpi_wakeup_address);
>
>         }
> -       ACPI_FLUSH_CPU_CACHE();
>  #endif
>         pr_info("Preparing to enter system sleep state S%d\n", acpi_state);
>         acpi_enable_wakeup_devices(acpi_state);
> @@ -566,15 +565,15 @@ static int acpi_suspend_enter(suspend_state_t pm_state)
>         u32 acpi_state = acpi_target_sleep_state;
>         int error;
>
> -       ACPI_FLUSH_CPU_CACHE();
> -
>         trace_suspend_resume(TPS("acpi_suspend"), acpi_state, true);
>         switch (acpi_state) {
>         case ACPI_STATE_S1:
>                 barrier();
>                 status = acpi_enter_sleep_state(acpi_state);
>                 break;
> -
> +       case ACPI_STATE_S2:
> +               ACPI_FLUSH_CPU_CACHE();
> +               break;

I don't think this is needed for S2, because the function doesn't do
anything low-level in that case and simply returns (IOW, S2 isn't
really supported).

>         case ACPI_STATE_S3:
>                 if (!acpi_suspend_lowlevel)
>                         return -ENOSYS;
> --
> 2.32.0
>