Re: Commit 554c8aa8ecad causing severe performance degression with pcc-cpufreq

From: Andreas Herrmann <aherrmann@suse.com>
To: "Rafael J. Wysocki" <rafael@kernel.org>
Cc: "Rafael J. Wysocki" <rafael.j.wysocki@intel.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Frederic Weisbecker <frederic@kernel.org>,
	Viresh Kumar <viresh.kumar@linaro.org>,
	Linux PM <linux-pm@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: Commit 554c8aa8ecad causing severe performance degression with pcc-cpufreq
Date: Tue, 17 Jul 2018 12:18:05 +0200	[thread overview]
Message-ID: <20180717101805.ctr3u3vy4gciggw3@suselix> (raw)
In-Reply-To: <20180717093620.ym6phfmr3rfvsxyo@suselix>

On Tue, Jul 17, 2018 at 11:36:20AM +0200, Andreas Herrmann wrote:
> On Tue, Jul 17, 2018 at 11:27:21AM +0200, Andreas Herrmann wrote:
> > On Tue, Jul 17, 2018 at 11:23:25AM +0200, Rafael J. Wysocki wrote:
> > > On Tue, Jul 17, 2018 at 11:11 AM, Andreas Herrmann <aherrmann@suse.com> wrote:
> > > > On Tue, Jul 17, 2018 at 11:06:29AM +0200, Rafael J. Wysocki wrote:
> > > >> On Tue, Jul 17, 2018 at 10:50 AM, Andreas Herrmann <aherrmann@suse.com> wrote:
> > > >>
> > > >> [cut]
> > > >>
> > > >> >
> > > >> > On balance before this commit users could use pcc-cpufreq but had
> > > >> > already suboptimal performance (compared to say intel_pstate driver
> > > >> > which can be used changing BIOS options).
> > > >>
> > > >> BTW, I wonder why you need to change the BIOS options for intel_pstate to load.
> > > >
> > > > I think this is because of (in intel_pstate_init()):
> > > >
> > > >         /*
> > > >          * The Intel pstate driver will be ignored if the platform
> > > >          * firmware has its own power management modes.
> > > >          */
> > > >         if (intel_pstate_platform_pwr_mgmt_exists())
> > > >                 return -ENODEV;
> > > >
> > > 
> > > OK, because of the "Proliant" entry, right?
> > > 
> > > So it looks like we have an issue there.  We find the entry and we
> > > look for _PSS.  It is not there, so we assume that the firmware is
> > > expected to control performance, which is not the case.

> FYI, there is another BIOS setting on those systems. It's called
> "Collaborative Power Control" (AFAIK enabled by default).
> 
> Only if this is disabled, firmware is (alone) in control of
> performance. (And of course in this case neither pcc-cpufreq nor
> intel_pstate will be loaded).

To clarify:

(i)  When "Dynamic Power Savings Mode" is enabled _PSS objects are
     missing (in fact they seem to be renamed to "XPSS").
     pcc-cpufreq driver evaluates PCCH header and loads.

     Same when "Collaborative Power Control" is disabled. No _PSS but
     XPSS objects. pcc-cpufreq driver fails to evaluate PCCH header,
     no cpufreq driver is loaded.

(ii) There are _PSS objects when "OS Control Mode" is selected.

     (But I think when "Collaborative Power Control" is disabled there
     are no _PSS objects either and the "OS Control Mode" selection
     does not matter.)

> > > It looks like we also should look for the presence of the PCC
> > > interface in there.
> > >
> > > I can provide a patch for that, will you be able to test it?
> > 
> > Yes, I can test it.
> > 
> > > >> It should be initialized before pcc-cpufreq (according to their
> > > >> respective initcall levels), so in theory intel_pstate should be used
> > > >> by default on the affected systems anyway.
> > > >
> > > >> What BIOS settings need to be changed for that?
> > > >
> > > > Already answered in other mail.
> > > 
> > > Indeed.
> > 
> > 
> > Andreas
> > 