svn commit: r326383 - head/sys/x86/cpufreq

[Sepherosa Ziehau]

On Fri, Dec 8, 2017 at 6:23 AM, Andriy Gapon <avg at freebsd.org> wrote:
> On 07/12/2017 12:03, Alexey Dokuchaev wrote:
>> On Thu, Nov 30, 2017 at 03:08:49PM -0500, Jung-uk Kim wrote:
>>> ...
>>> Probably.  However, I am just trying to fix my FX-8350 and A10-6800 and
>>> I don't have Zen processors to verify the MSRs are actually working on
>>> those CPUs.
>>
>> Ah, that's so lovely, thanks Jung-uk; I feel that our support for AMD
>> fam. 15h CPUs is lacking.  E.g. only four P-states are reported for my
>> A8-5550M, while it supports boosted P-states per BKDG, and reading MSRs
>> directly via `sysutils/amdmsrtweaker' reports eight of them (P0 .. P7),
>> with three turbo P-states P0 P1 P2.
>>
>> Since you have A10-6800 you might try to reproduce what I see here with
>> A8-5550M.
>
> I think that the boosted states are supposed to be hidden from the OS.
> It may be possible to query them though hardware specific registers, but they
> can not be set by software directly anyway.

For most of the modern Intel CPUs, boosted state is _not_ hidden from
OS, it normally shows as max_freq+1, e.g. if max_freq is 3400, then
the boosted state will have freq 3401.

> My impression is that the proper way to observe the boost states is via APERF /
> MPERF MSRs.  They are useful for other things, e.g. C0 residency, too.

Yeah, that's the only way to observe the current frequency of the cpu;
on most of the modern Intel CPUs, P-state just sets the frequency
upper limit.

Thanks,
sephe