On Mon, Apr 4, 2022 at 8:25 PM Akihiko Odaki <akihiko.odaki@xxxxxxxxx> wrote:
On 2022/04/05 3:13, Rafael J. Wysocki wrote:
On Sun, Apr 3, 2022 at 8:25 AM Akihiko Odaki <akihiko.odaki@xxxxxxxxx> wrote:
This reverts commit 87ebbb8c612b1214f227ebb8f25442c6d163e802.
ACPI processor power states can be transitioned in two distinct
situations: 1. when CPU goes idle and 2. before CPU goes offline
("playing dead") to suspend or hibernate. Case 1 is handled by
acpi_idle_enter or acpi_idle_enter_s2idle. Case 2 is handled by
acpi_idle_play_dead.
It is necessary to flush CPU caches in case 2 even if it is not
required to transit ACPI processor power states as CPU will go
offline soon. However, the reverted commit incorrectly removed CPU
cache flushing in such a condition.
I think what you mean is that the CPU cache must always be flushed in
acpi_idle_play_dead(), regardless of the target C-state that is going
to be requested, because this is likely to be part of a CPU offline
procedure or preparation for entering a system-wide sleep state and
the stale cache contents may lead to problems going forward, for
example when the CPU is taken back online.
If so, I will put the above information into the patch changelog.
I guess it is causing problems because the dirty caches will not get
written back and the RAM becomes stale if they are not flushed. From my
understanding, the CPU should have an empty cache and read back contents
from RAM when it is taken back online.
OK, please see if the changelog I've added to the patch is looking good:
https://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm.git/commit/?h=bleeding-edge&id=dfbba2518aac4204203b0697a894d3b2f80134d3