[Bug 219851] New: [amd-pstate 6.15] Frequency Scaling Will Not Scale Above Lowest Non-Linear Performance on Ryzen 7 5700U

[Bug 219851] New: [amd-pstate 6.15] Frequency Scaling Will Not Scale Above Lowest Non-Linear Performance on Ryzen 7 5700U

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=219851

            Bug ID: 219851
           Summary: [amd-pstate 6.15] Frequency Scaling Will Not Scale
                    Above Lowest Non-Linear Performance on Ryzen 7 5700U
           Product: Power Management
           Version: 2.5
          Hardware: AMD
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P3
         Component: cpufreq
          Assignee: linux-pm@vger.kernel.org
          Reporter: dnaim@cachyos.org
        Regression: No

Created attachment 307781
  --> https://bugzilla.kernel.org/attachment.cgi?id=307781&action=edit
amd-pstate-triage log from bad kernel

On a kernel build with 6.15 amd-pstate material taken from
https://web.git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git/tag/?h=amd-pstate-v6.15-2025-03-03,
a user noticed that frequency scaling will not scale above Lowest Non-Linear
Perf. Git bisect shows the below as the first bad commit.

620136ced35a9329f4d1ea90e51bee2dfd7ee5b0 is the first bad commit
commit 620136ced35a9329f4d1ea90e51bee2dfd7ee5b0
Author: Dhananjay Ugwekar <dhananjay.ugwekar@amd.com>
Date:   Wed Feb 5 11:25:18 2025 +0000

    cpufreq/amd-pstate: Modularize perf<->freq conversion

    Delegate the perf<->frequency conversion to helper functions to reduce
    code duplication, and improve readability.

    Signed-off-by: Dhananjay Ugwekar <dhananjay.ugwekar@amd.com>
    Reviewed-by: Mario Limonciello <mario.limonciello@amd.com>
    Reviewed-by: Gautham R. Shenoy <gautham.shenoy@amd.com>
    Link:
https://lore.kernel.org/r/20250205112523.201101-8-dhananjay.ugwekar@amd.com
    Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>

 drivers/cpufreq/amd-pstate.c | 57
++++++++++++++++++++++++++++++---------------------------
 1 file changed, 30 insertions(+), 27 deletions(-)

Attached are amd-pstate-triage logs from good and bad kernels. They are not
necessarily equal because a revert of the bad commit isn't trivial with the
whole 6.15 content so the "good" kernel only has patches from Dhananjay's
series and the bad commit reverted.

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

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

[Bug 219851] [amd-pstate 6.15] Frequency Scaling Will Not Scale Above Lowest Non-Linear Performance on Ryzen 7 5700U

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=219851

--- Comment #1 from Eric Naim (dnaim@cachyos.org) ---
Created attachment 307782
  --> https://bugzilla.kernel.org/attachment.cgi?id=307782&action=edit
amd-pstate-triage log from good kernel

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

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

[Bug 219851] [amd-pstate 6.15] Frequency Scaling Will Not Scale Above Lowest Non-Linear Performance on Ryzen 7 5700U

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=219851

--- Comment #2 from Dhananjay Ugwekar (AMD) (Dhananjay.Ugwekar@amd.com) ---
Hello Eric,

1. Could you please give more details on how to reproduce the issue? which
governor, amd-pstate mode (when the issue was observed), workload. Are you able
to reproduce the issue on your machine as well?
2. Just to confirm, does the user have
https://web.git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git/commit/?h=amd-pstate-v6.15-2025-03-03&id=f6c0b760290951688697d9debbb2b8462c423a48,
in his tree, it fixed a corner case in the culprit commit that you pointed to.

Thanks, 
Dhananjay

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

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

[Bug 219851] [amd-pstate 6.15] Frequency Scaling Will Not Scale Above Lowest Non-Linear Performance on Ryzen 7 5700U

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=219851

--- Comment #3 from Eric Naim (dnaim@cachyos.org) ---
(In reply to Dhananjay Ugwekar (AMD) from comment #2)
> Hello Eric,
> 
Hi Dhananjay,

Unfortunately the user that first reported this issue hasn't reached out to me
since, so for now I'll give you information regarding what I already know.

> 1. Could you please give more details on how to reproduce the issue? which
> governor, amd-pstate mode (when the issue was observed), workload. Are you
> able to reproduce the issue on your machine as well?

No STR was given, from the lack of info I believe it's just a general case,
e.g. if you run `stress-ng -c $(nproc)`, you *should* see the issue (entirely
speculation). This regression only happens with the active driver. Both passive
and guided mode was reported to behave correctly.

I can't reproduce this either because my laptop is a Zen3 laptop. It is likely
that it is limited to Zen2.

> 2. Just to confirm, does the user have
> https://web.git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git/commit/
> ?h=amd-pstate-v6.15-2025-03-03&id=f6c0b760290951688697d9debbb2b8462c423a48,
> in his tree, it fixed a corner case in the culprit commit that you pointed
> to.

Yes, you can see the related tree
https://github.com/CachyOS/linux/tree/6.14/amd-pstate. It's not 1:1 with
Mario's tree but I diffed both trees and the only difference was commit
34562659110a42276d63f8b5a9d522d2400b2df2 which is from the "Dynamic ITMT core
ranking support" series.

> 
> Thanks, 
> Dhananjay

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

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

[Bug 219851] [amd-pstate 6.15] Frequency Scaling Will Not Scale Above Lowest Non-Linear Performance on Ryzen 7 5700U

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=219851

--- Comment #4 from Eric Naim (dnaim@cachyos.org) ---
(In reply to Dhananjay Ugwekar (AMD) from comment #2)
Hi Dhananjay,
> Hello Eric,
> 
> 1. Could you please give more details on how to reproduce the issue? which
> governor, amd-pstate mode (when the issue was observed), workload. Are you
> able to reproduce the issue on your machine as well?
> 2. Just to confirm, does the user have
> https://web.git.kernel.org/pub/scm/linux/kernel/git/superm1/linux.git/commit/
> ?h=amd-pstate-v6.15-2025-03-03&id=f6c0b760290951688697d9debbb2b8462c423a48,
> in his tree, it fixed a corner case in the culprit commit that you pointed
> to.

It looks like the user didn't have that commit after all. They were using an
older revision of the patchset that didn't have the updated amd-pstate commits.
I should've asked them to crosscheck their issue with the kernel in our
repositories first. Sorry to have wasted your time. I'm going to close this as
invalid.

> 
> Thanks, 
> Dhananjay

Regards,
Eric

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

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

[Bug 219851] [amd-pstate 6.15] Frequency Scaling Will Not Scale Above Lowest Non-Linear Performance on Ryzen 7 5700U

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=219851

Eric Naim (dnaim@cachyos.org) changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|---                         |INVALID

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

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

[Bug 219851] [amd-pstate 6.15] Frequency Scaling Will Not Scale Above Lowest Non-Linear Performance on Ryzen 7 5700U

[bugzilla-daemon]

https://bugzilla.kernel.org/show_bug.cgi?id=219851

--- Comment #5 from Dhananjay Ugwekar (AMD) (Dhananjay.Ugwekar@amd.com) ---
No worries, happy to help.

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

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