On Mon, Dec 6, 2021 at 4:03 PM Peter Zijlstra [off-list ref] wrote:

On Mon, Dec 06, 2021 at 03:29:51PM +0300, Kirill A. Shutemov wrote:

quoted

According to the ACPI spec v6.4, section 8.2, cache flushing required
on entering C3 power state.

Avoid flushing cache on entering other power states.

Signed-off-by: Kirill A. Shutemov <redacted>
---
 drivers/acpi/processor_idle.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/acpi/processor_idle.c b/drivers/acpi/processor_idle.c
index 76ef1bcc8848..01495aca850e 100644
--- a/drivers/acpi/processor_idle.c
+++ b/drivers/acpi/processor_idle.c

@@ -567,7 +567,8 @@ static int acpi_idle_play_dead(struct cpuidle_device *dev, int index)
 {
      struct acpi_processor_cx *cx = per_cpu(acpi_cstate[index], dev->cpu);

-     ACPI_FLUSH_CPU_CACHE();
+     if (cx->type == ACPI_STATE_C3)
+             ACPI_FLUSH_CPU_CACHE();

acpi_idle_enter() already does this, acpi_idle_enter_s2idle() has it
confused again,

No, they do the same thing: acpi_idle_enter_bm() if flags.bm_check is set.

Also, I think acpi_idle_enter() does it too late; consider
acpi_idle_enter_mb(). Either that or the BM crud needs more comments.

I think the latter.

Evidently, acpi_idle_play_dead(() doesn't support FFH and the BM
thing, so it is only necessary to flush the cache when using
ACPI_CSTATE_SYSTEMIO and when cx->type is C3.