Re: [PATCH v2] cpufreq: intel_pstate: Add Icelake servers support in no-HWP mode

[Giovanni Gherdovich <ggherdovich@suse.cz>]

On Fri, 2021-05-14 at 08:31 -0700, Doug Smythies wrote:
> Hi All,
> 
> Can I on-board to this patch or do you want me to submit another?
> I want to add COMETLAKE (tested), as below:
> 
> ... Doug

Hello Doug!

Wait, why you don't want to use HWP? It's such a fantastic technology!

:) I'm just teasing you.

More seriously: 

when COMETLAKE is not in that list, can you confirm that if you go into the
BIOS config at boot, and disable HWP from there, then intel_pstate does *not* load?

Does it say "intel_pstate: CPU model not supported" in the dmesg log?

The control may be somewhere around "power mangement" in the BIOS config, and
may be called "Enable/disable Intel Speed Shift".

I'm asking because I've just checked on two Dell laptops, one Skylake and the
other Kabylake, and the menu is there in the BIOS config to disable HWP,
but if I disable it... nothing happens. "lscpu" shows all the hwp flags as usual:

    # lscpu | grep Flags | tr ' ' '\n' | grep hwp
    hwp
    hwp_notify
    hwp_act_window
    hwp_epp

and turbostat gives me:

    # turbostat -Summary -i 1 : 2>&1 | grep MSR_PM_ENABLE
    cpu0: MSR_PM_ENABLE: 0x00000001 (HWP)

Which is to say, on the Intel client machines I have, the firmware doesn't
seem to be able to hide HWP from the OS. Buggy BIOS? Maybe, the fact of the
matter is, I wouldn't need to add, say, KABYLAKE to that list, based on my
experience.

The other side of the issue is that, from my understanding, the
preferred/supported way to disable HWP is to boot with intel_pstate=no_hwp,
and that list is a sort of "known exceptions" that people really can't live
without (it's mostly server CPUs, and mostly because of unfortunate firmware
defaults). Otherwise you'd see the entire intel-family.h file in there.


Cheers,
Giovanni

> 
> On Thu, May 13, 2021 at 6:21 AM Giovanni Gherdovich <ggherdovich@suse.cz> wrote:
> > Users may disable HWP in firmware, in which case intel_pstate wouldn't load
> > unless the CPU model is explicitly supported.
> > 
> > Add ICELAKE_X to the list of CPUs that can register intel_pstate while not
> > advertising the HWP capability. Without this change, an ICELAKE_X in no-HWP
> > mode could only use the acpi_cpufreq frequency scaling driver.
> > 
> > See also commit d8de7a44e11f ("cpufreq: intel_pstate: Add Skylake servers
> > support").
> > 
> > Signed-off-by: Giovanni Gherdovich <ggherdovich@suse.cz>
> > ---
> > This replaces https://lore.kernel.org/lkml/20210513075930.22657-1-ggherdovich@suse.cz
> > 
> >  drivers/cpufreq/intel_pstate.c | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
> > index f0401064d7aa..28c9733e0dce 100644
> > --- a/drivers/cpufreq/intel_pstate.c
> > +++ b/drivers/cpufreq/intel_pstate.c
> > @@ -2087,6 +2087,7 @@ static const struct x86_cpu_id intel_pstate_cpu_ids[] = {
> >         X86_MATCH(ATOM_GOLDMONT,        core_funcs),
> >         X86_MATCH(ATOM_GOLDMONT_PLUS,   core_funcs),
> >         X86_MATCH(SKYLAKE_X,            core_funcs),
> > +       X86_MATCH(ICELAKE_X,            core_funcs),
>    +       X86_MATCH(COMETLAKE,          core_funcs),
> >         {}
> >  };
> >  MODULE_DEVICE_TABLE(x86cpu, intel_pstate_cpu_ids);
> > --
> > 2.26.2
> >