[RFC PATCH v4 0/8] SRF: Fix offline CPU preventing pc6 entry

From: Patryk Wlazlyn
Date: Mon Nov 25 2024 - 08:20:50 EST


Sending an interim RFC patch. Rebased on top of Rafael's patches and
applied modified patch/suggestion from Gautham.

Please tell me what you think. Is this approach in general is
acceptable?

When playing dead we have now the preference ordered:
1. Idle driver.
- acpi_idle can now do FFH and calls mwait_play_dead_with_hint,
but we disable the handlers on AMD, because AMD doesn't want to
use mwait for play_dead.
- intel_idle have a handler for SRF. Intel platforms fallback to
cpuid leaf 0x5 on other platforms.
2. Old mwait_play_dead() based on cpuid leaf 0x5.
3. hlt_play_dead().


Changes since v3:
* Reworded commit messages authored by me as advised by Dave.
Expanded the context about the mwait_play_dead code, hint computation
and what is a problem we observe and the patch addresses.
* Different play_dead methods are called now one after the other,
because they only return on success.

Patryk Wlazlyn (5):
x86/smp: Allow calling mwait_play_dead with arbitrary hint
x86/smp native_play_dead: Prefer cpuidle_play_dead() over
mwait_play_dead()
intel_idle: Provide enter_dead() handler for SRF
acpi_idle: Add FFH cstate handling
acpi_idle: Disallow play_dead with FFH cstate on AMD platforms

Rafael J. Wysocki (3):
cpuidle: Do not return from cpuidle_play_dead() on callback failures
cpuidle: Change :enter_dead() driver callback return type to void
ACPI: processor_idle: Use acpi_idle_play_dead() for all C-states

arch/x86/include/asm/cpufeatures.h | 1 +
arch/x86/include/asm/smp.h | 3 +
arch/x86/kernel/acpi/cstate.c | 9 +++
arch/x86/kernel/smpboot.c | 91 ++++++++++++++++--------------
drivers/acpi/processor_idle.c | 23 ++++----
drivers/cpuidle/cpuidle.c | 10 +++-
drivers/idle/intel_idle.c | 15 +++++
include/acpi/processor.h | 5 ++
include/linux/cpuidle.h | 2 +-
9 files changed, 103 insertions(+), 56 deletions(-)

--
2.47.0