[PATCH v2 0/5] amd_pmc: Delay s2idle suspend for some devices

[Daniel Gibson <daniel@gibson.sh>]

On some AMD Zen3 and Zen3+-based Lenovo IdeaPad laptops the keyboard and
the lid switch stop working after the first suspend, until rebooted.

More specifically, they stop sending events when pressing a key or
closing the lid - it's still possible to toggle the capslock- and
numlock-LEDs with an external keyboard or read the lid state at
/proc/acpi/button/lid/LID/state.

See also https://bugzilla.kernel.org/show_bug.cgi?id=221383

It appears that suspending and/or resuming gets the EC into a broken
state. This problem doesn't happen on Windows and Mario Limonciello
mentioned that the Windows kernel gives hardware and software some time
before actually suspending (before activating HW DRIPS), while Linux
(or the amd_pmc module) does that immediately, so it may be worth trying
if calling msleep() in amd_pmc_s2idle_check() helps.

It turned out that sleeping for 2.5 seconds at that point indeed makes
the problems mostly disappear. Sleeping for 1.5 seconds wasn't enough.

"Mostly" because it turned out that they still occur (at least on my
device) when using a wakeup timer (wakealarm).

I could build on an existing quirk[1] that also sleeps for 2.5 seconds
under other circumstances; my first commit refactors that a bit so I
can integrate my further changes in a cleaner way.

I found several reports of these or similar issues on the web, for
different devices, so in a second commit I added a parameter to the
kernel module that allows enabling or disabling this, which will make
it easy for people whose devices aren't matched yet to test this quirk.

Thanks to Mario Limonciello for his support and to Sindre Henriksen
for testing my patch and to Ilpo Järvinen for reviewing v1!

[1] https://lore.kernel.org/platform-driver-x86/20250414162446.3853194-1-superm1@kernel.org/

Changes in v2:
- Documented this in Documentation/arch/x86/amd-debugging.rst
- Added example for reset register kernel message in same file
- In amd_pmc_quirk_need_suspend_delay(), avoid dereferencing a NULL
  pointer of devices not detected for any quirk - oops!
- Mention that timed resumes may still cause those keyboard/lid issues
- Various code changes requested or suggested in reviews of v1:
  - Some formatting changes (commas behind non-terminating entries)
  - Moved check for existing quirk (that OVP thing) into its own function
    amd_pmc_intermediate_wakeup_need_delay() in pmc.c, so the checks of
    the different quirks are separated more clearly.
  - Added function amd_pmc_want_suspend_delay() in pmc.c handling
    amd_pmc_quirk_need_suspend_delay() together with disable_workarounds
    and delay_suspend and also logging about the reason for the delay,
    also for cleaner separation.
  - If delay_suspend=1 is used to force-enable the fix on hardware that
    is not automatically detected as needing this fix, log message
    encouraging the user to report their device, including the most
    relevant DMI values that could be used for matching

v1: https://lore.kernel.org/platform-driver-x86/20260501032655.283789-1-daniel@gibson.sh/t/#u

Daniel Gibson (5):
  platform/x86/amd/pmc: Check for intermediate wakeup in function
  platform/x86/amd/pmc: Delay suspend for some Lenovo Laptops
  platform/x86/amd/pmc: Add delay_suspend module parameter
  Documentation/arch/x86/amd-debugging: Add example for reset register
  Documentation/arch/x86/amd-debugging: Add section about delay_suspend

 Documentation/arch/x86/amd-debugging.rst  | 30 ++++++++++++
 drivers/platform/x86/amd/pmc/pmc-quirks.c | 31 ++++++++++++
 drivers/platform/x86/amd/pmc/pmc.c        | 60 +++++++++++++++++++++--
 drivers/platform/x86/amd/pmc/pmc.h        |  1 +
 4 files changed, 119 insertions(+), 3 deletions(-)

-- 
2.48.1