[Bug 217931] amd-pstate lacks crucial features: CPU frequency and boost control

[bugzilla-daemon@kernel.org]

https://bugzilla.kernel.org/show_bug.cgi?id=217931

--- Comment #74 from Mario Limonciello (AMD) (mario.limonciello@amd.com) ---
Created attachment 307680
  --> https://bugzilla.kernel.org/attachment.cgi?id=307680&action=edit
patch to invalidate cached CPPC request at suspend

> Mario, just to be clear, I've run all tests now on unmodified system. 

OK.

> I'm attaching two new s2idle reports run as root. On the second one I've set
> manually the /sys/power/pm_debug_messages to 1. Seems the script is reverting
> it to 0 after running.

Both of these reports look good.  It successfully made it down to HW sleep
state and spent a good amount of time there.

This is how s2idle should generally be looking.

> 2025-02-17 20:08:39,757 DEBUG:        Disabling GPIO #8 interrupt for
> suspend.
> 2025-02-17 20:08:39,805 INFO: ○ GPIOs active: ['8']

Just want to point this out that this GPIO appears to be active over suspend
and the Linux kernel intentionally ignores it.  It's not in _AEI() or _EVT()
ACPI methods.  It could be a red herring, but do you know what it's connected
to?  If not; can you share your full acpidump?  I might be able to figure it
out.

> I'm adding the amd_pstate.py reports as
> 2025.02.17_pstate_reports_Miroslav.zip.

I don't see a problem with CPPC enable register here, which is what my
suspicion was.

BUT what I can see is that there is a problem with the values programmed in
MSR_AMD_CPPC_REQ.  You can see comparing the two that all of the values for min
and max and EPP are totally wrong.

This is most likely an amd-pstate bug you're finding.  I think I see the issue.
 Here's a patch against 6.14-rc for you to try.

-- 
You may reply to this email to add a comment.

You are receiving this mail because:
You are the assignee for the bug.