[Bug 292615] cpufreq changes causes AMD 4750U to run 11x slower

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=292615

Olivier Certner <olce@FreeBSD.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
 Attachment #267532|0                           |1
        is obsolete|                            |
           Assignee|bugs@FreeBSD.org            |olce@FreeBSD.org
             Status|Open                        |In Progress

--- Comment #35 from Olivier Certner <olce@FreeBSD.org> ---
Created attachment 267549
  --> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=267549&action=edit
Patch with candidate fix and more knobs

This new patch (untested on real hardware) essentially does two things:
1. A potential fix for Martin's case, where the CPPC_CAPABILITY_1 was not
filled up correctly after initialization (to 0).  The original values of min
and max performance in CPPC_REQUEST seem inconsistent (as the min value is
lower than the max one; unless AMD's APM is wrong...), so the plan of leaving
them untouched does not work (confirmed by Martin's experiments).  When the min
and max performance values in CPPC_CAPABILITY_1 are not sorted correctly or are
equal (covers the above-mentioned case of staying at 0), use values as
suggested by the ACPI spec for generic CPPC (0 for min, 255 for max).
2. I have added knobs to control the min, max and desired performance levels in
the CPPC_REQUEST register.

To follow, instructions on what to report.

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