RE: [PATCH v3] intel_idle: Add Comet Lake support

[Pan, Harry]

Hi Rafael,

Thanks for the comments.
I have some questions, I am wondering if you can share upstream thought w.r.t the future development of intel_idle.

 - It looks to me since v5.6 intel_idle will prefer _CST of ACPI rather than the general table embedded in this driver.
 - Any pros and cons of using the tables of _CST in firmware and embedded one in the kernel?
 - Can the table in _CST archive more optimal idle states management? 

If there is already any reference, documents I missed, kindly enlighten me then I would like to read it first before refining the questions.

Thanks and take care,
Harry
________________________________________
å: Rafael J. Wysocki [rafael@xxxxxxxxxx]
åäææ: 2020å3æ5æ äå 05:14
è: Pan, Harry
åæ: Rafael J. Wysocki; LKML; Harry Pan; Jacob Pan; Len Brown; Linux PM
äæ: Re: [PATCH v3] intel_idle: Add Comet Lake support

On Wed, Mar 4, 2020 at 12:57 PM Pan, Harry <harry.pan@xxxxxxxxx> wrote:
>
> Hi Rafael,
>
> Yes, I skipped it considering to align CML-U V0 and A0 stepping w/ the same table; I sent v4 for your review.

Skipping that flag is risky, because it may cause some C-states to be
enabled on systems where they have not been validated (e.g. systems
shipping with other OSes which only use _CST C-states).  There were
problems related to that in the past which is one of the reasons for
adding _CST support to intel_idle.  use_acpi should be set for all new
platforms going forward as a rule.

> In the other hand, I am proposing using _CST as long term plan in CrOS dev teams.

That I obviously agree with. :-)


> > On Mar 4, 2020, at 17:53, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote:
> >
> > ïOn Tue, Mar 3, 2020 at 10:10 AM Harry Pan <harry.pan@xxxxxxxxx> wrote:
> >>
> >> Add a general C-state table in order to support Comet Lake.
> >>
> >> Signed-off-by: Harry Pan <harry.pan@xxxxxxxxx>
> >>
> >> ---
> >>
> >> drivers/idle/intel_idle.c | 7 +++++++
> >> 1 file changed, 7 insertions(+)
> >>
> >> diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c
> >> index d55606608ac8..05bce595fafe 100644
> >> --- a/drivers/idle/intel_idle.c
> >> +++ b/drivers/idle/intel_idle.c
> >> @@ -1067,6 +1067,11 @@ static const struct idle_cpu idle_cpu_dnv = {
> >>        .use_acpi = true,
> >> };
> >>
> >> +static const struct idle_cpu idle_cpu_cml = {
> >> +       .state_table = skl_cstates,
> >> +       .disable_promotion_to_c1e = true,
> >
> > .use_acpi = true,
> >
> > missing?  Otherwise you can just use idle_cpu_skl as is, can't you?
> >
> >> +};
> >> +
> >> static const struct x86_cpu_id intel_idle_ids[] __initconst = {
> >>        INTEL_CPU_FAM6(NEHALEM_EP,              idle_cpu_nhx),
> >>        INTEL_CPU_FAM6(NEHALEM,                 idle_cpu_nehalem),
> >> @@ -1105,6 +1110,8 @@ static const struct x86_cpu_id intel_idle_ids[] __initconst = {
> >>        INTEL_CPU_FAM6(ATOM_GOLDMONT_PLUS,      idle_cpu_bxt),
> >>        INTEL_CPU_FAM6(ATOM_GOLDMONT_D,         idle_cpu_dnv),
> >>        INTEL_CPU_FAM6(ATOM_TREMONT_D,          idle_cpu_dnv),
> >> +       INTEL_CPU_FAM6(COMETLAKE_L,             idle_cpu_cml),
> >> +       INTEL_CPU_FAM6(COMETLAKE,               idle_cpu_cml),
> >>        {}
> >> };
> >>
> >> --
> >> 2.24.1
> >>