RE: [PATCH] ACPI: CPPC: Don't require _OSC if X86_FEATURE_CPPC is supported

[Limonciello, Mario]

[AMD Official Use Only - General]



> -----Original Message-----
> From: Rafael J. Wysocki <rafael@xxxxxxxxxx>
> Sent: Wednesday, June 29, 2022 14:09
> To: Limonciello, Mario <Mario.Limonciello@xxxxxxx>
> Cc: Rafael J. Wysocki <rafael@xxxxxxxxxx>; Len Brown <lenb@xxxxxxxxxx>;
> Pierre Gondois <pierre.gondois@xxxxxxx>; Sudeep Holla
> <sudeep.holla@xxxxxxx>; Yuan, Perry <Perry.Yuan@xxxxxxx>; ACPI Devel
> Maling List <linux-acpi@xxxxxxxxxxxxxxx>; Linux Kernel Mailing List <linux-
> kernel@xxxxxxxxxxxxxxx>
> Subject: Re: [PATCH] ACPI: CPPC: Don't require _OSC if X86_FEATURE_CPPC is
> supported
> 
> On Wed, Jun 29, 2022 at 8:49 PM Limonciello, Mario
> <Mario.Limonciello@xxxxxxx> wrote:
> >
> > [Public]
> >
> >
> >
> > > -----Original Message-----
> > > From: Rafael J. Wysocki <rafael@xxxxxxxxxx>
> > > Sent: Wednesday, June 29, 2022 13:42
> > > To: Limonciello, Mario <Mario.Limonciello@xxxxxxx>
> > > Cc: Rafael J. Wysocki <rafael@xxxxxxxxxx>; Len Brown <lenb@xxxxxxxxxx>;
> > > Pierre Gondois <pierre.gondois@xxxxxxx>; Sudeep Holla
> > > <sudeep.holla@xxxxxxx>; Yuan, Perry <Perry.Yuan@xxxxxxx>; ACPI Devel
> > > Maling List <linux-acpi@xxxxxxxxxxxxxxx>; Linux Kernel Mailing List <linux-
> > > kernel@xxxxxxxxxxxxxxx>
> > > Subject: Re: [PATCH] ACPI: CPPC: Don't require _OSC if X86_FEATURE_CPPC
> is
> > > supported
> > >
> > > On Mon, Jun 27, 2022 at 6:58 PM Mario Limonciello
> > > <mario.limonciello@xxxxxxx> wrote:
> > > >
> > > > commit 72f2ecb7ece7 ("ACPI: bus: Set CPPC _OSC bits for all and
> > > > when CPPC_LIB is supported") added support for claiming to
> > > > support CPPC in _OSC on non-Intel platforms.
> > > >
> > > > This unfortunately caused a regression on a vartiety of AMD
> > > > platforms in the field because a number of AMD platforms don't set
> > > > the `_OSC` bit 5 or 6 to indicate CPPC or CPPC v2 support.
> > > >
> > > > As these AMD platforms already claim CPPC support via
> `X86_FEATURE_CPPC`,
> > > > use this enable this feature rather than requiring the `_OSC`.
> > > >
> > > > Fixes: 72f2ecb7ece7 ("Set CPPC _OSC bits for all and when CPPC_LIB is
> > > supported")
> > > > Reported-by: Perry Yuan <perry.yuan@xxxxxxx>
> > > > Signed-off-by: Mario Limonciello <mario.limonciello@xxxxxxx>
> > > > ---
> > > >  drivers/acpi/cppc_acpi.c | 12 +++++++++---
> > > >  1 file changed, 9 insertions(+), 3 deletions(-)
> > > >
> > > > diff --git a/drivers/acpi/cppc_acpi.c b/drivers/acpi/cppc_acpi.c
> > > > index 903528f7e187..5463e6309b9a 100644
> > > > --- a/drivers/acpi/cppc_acpi.c
> > > > +++ b/drivers/acpi/cppc_acpi.c
> > > > @@ -629,6 +629,15 @@ static bool is_cppc_supported(int revision, int
> > > num_ent)
> > > >                 return false;
> > > >         }
> > > >
> > > > +       if (osc_sb_cppc_not_supported) {
> > > > +               pr_debug("Firmware missing _OSC support\n");
> > > > +#ifdef CONFIG_X86
> > > > +               return boot_cpu_has(X86_FEATURE_CPPC);
> > > > +#else
> > > > +               return false;
> > > > +#endif
> > >
> > > What about doing
> > >
> > > if (osc_sb_cppc_not_supported) {
> > >         pr_debug("Firmware missing _OSC support\n");
> > >         return IS_ENABLED(CONFIG_X86) &&
> boot_cpu_has(X86_FEATURE_CPPC);
> > > }
> > >
> > > instead for the sake of reducing #ifdeffery?
> >
> > I don't think that would compile on non-X86.  X86_FEATURE_CPPC comes as
> part of
> > arch/x86/include/asm/cpufeatures.h, which I wouldn't expect is included on
> !x86.
> 
> Good point.
> 
> Something like this would still look better though IMO:
> 
> if (!osc_sb_cppc_not_supported)
>         return true;
> 
> #ifdef CONFIG_X86
>         return boot_cpu_has(X86_FEATURE_CPPC);
> #else
>         return false;
> #endif
> }
> 

Thanks, I'll respin it with something similar to that.

> 
> >
> > >
> > > Also, this is somewhat risky, because even if the given processor has
> > > X86_FEATURE_CPPC set, the platform may still not want to expose CPPC
> > > through ACPI.  How's that going to work after this change?
> > >
> >
> > Well actually doing that through _OSC wouldn't have worked before
> 72f2ecb7ece7 either.
> > If desirable - a platform could avoid populating _CPC objects in ACPI tables in
> this case.
> >
> > I do know of OEM platforms that the underlying APU supports CPPC but the
> OEM doesn't
> > populate _CPC.  Presumably for this exact reason.
> 
> That is an option, but there is no requirement that _CPC must not be
> populated when CPPC is not supported.
> 
> _OSC is the proper mechanism for negotiating CPPC support.
> 

Right; I agree this should have been the proper mechanism.  I'll talk to
our internal BIOS team to double check reference BIOS is populated
with this correctly for programs going forward too.

> Still, if you know for a fact that on AMD systems X86_FEATURE_CPPC
> always means that CPPC is supported, I can live with an extra vendor
> check in the code above.

Thanks.  The definition of that CPUID 8000_0008 EBX bit 27 used
to populate X86_FEATURE_CPPC indicates whether the CPU/APU
supports the dedicated MSR.  There are also technically designs that can
work in shared memory mode that I think the only way to "safely" discover
will be via the _OSC.  If this same regression from 72f2ecb7ece7 crops up
on those we might need to look at changing the amd-pstate module parameter
override that enables it for shared memory into a general kernel command line
override for users to use.