Message-ID: <CAJZ5v0gcgMJ-qihgc3_OF4djxAy8K0i-cmnjRe4AQrc_YEu4DQ@mail.gmail.com>
Date: Tue, 13 May 2025 16:01:08 +0200
From: "Rafael J. Wysocki" <rafael@...nel.org>
To: Linux PM <linux-pm@...r.kernel.org>
Cc: LKML <linux-kernel@...r.kernel.org>, Lukasz Luba <lukasz.luba@....com>, 
	Peter Zijlstra <peterz@...radead.org>, 
	Srinivas Pandruvada <srinivas.pandruvada@...ux.intel.com>, 
	Dietmar Eggemann <dietmar.eggemann@....com>, Morten Rasmussen <morten.rasmussen@....com>, 
	Vincent Guittot <vincent.guittot@...aro.org>, 
	Ricardo Neri <ricardo.neri-calderon@...ux.intel.com>, 
	Pierre Gondois <pierre.gondois@....com>, Christian Loehle <christian.loehle@....com>, 
	"Rafael J. Wysocki" <rjw@...ysocki.net>
Subject: Re: [PATCH v2 0/7] cpufreq: intel_pstate: Enable EAS on hybrid
 platforms without SMT

On Tue, May 13, 2025 at 2:51 PM Rafael J. Wysocki <rafael@...nel.org> wrote:
>
> On Tue, May 6, 2025 at 10:49 PM Rafael J. Wysocki <rjw@...ysocki.net> wrote:
> >
> > Hi Everyone,
> >
> > This is a new (and most likely final) version of
> >
> > https://lore.kernel.org/linux-pm/3344336.aeNJFYEL58@rjwysocki.net/
> >
> > The most significant difference between it and the above is that schedutil is
> > now required for EAS to be enabled, like on the other platforms using EAS,
> > which means that intel_pstate needs to operate in the passive mode in order
> > to use it (the most straightforward way to switch it over to the passive mode
> > is to write "passive" to /sys/devices/system/cpu/intel_pstate/status).
> >
> > Accordingly, the changes that were needed for EAS to work without schedutil are
> > not needed any more, so there is one patch less in the series because of that
> > and patch [5/7] is simpler than its previous version because some changes made
> > by it are not necessary any more.
> >
> > Another patch that has been dropped is
> >
> > https://lore.kernel.org/linux-pm/1964444.taCxCBeP46@rjwysocki.net/
> >
> > because it didn't take CPU online into account properly and it is not essential
> > for the current hardware anyway.
> >
> > There is a new patch, [7/7], which adds CAS/EAS/hybrid support description to
> > the intel_pstate admin-guide documentation.
> >
> > The following paragraph from the original cover letter still applies:
> >
> > "The underlying observation is that on the platforms targeted by these changes,
> > Lunar Lake at the time of this writing, the "small" CPUs (E-cores), when run at
> > the same performance level, are always more energy-efficient than the "big" or
> > "performance" CPUs (P-cores).  This means that, regardless of the scale-
> > invariant utilization of a task, as long as there is enough spare capacity on
> > E-cores, the relative cost of running it there is always lower."
> >
> > The first 2 patches depend on the current cpufreq material queued up for 6.16
> > in linux-pm.git/linux-next (and in linux-next proper) and are not really
> > depended on by the rest of the series, but I've decided to include them into
> > this series because they have been slightly updated since the previous version,
> > mostly to take review feedback into account (I'm going to queue them up for
> > 6.16 shortly because they don't appear to be objectionable).
> >
> > The next 2 patches (Energy Model code changes) were reviewed previously, but
> > they are only needed because of patch [5/7].
> >
> > Patch [5/7] has not changed much except that some changes made by the previous
> > version have been dropped from it.  Also its changelog has been updated.  It
> > causes perf domains to be registered per CPU and in addition to the primary cost
> > component, which is related to the CPU type, there is a small component
> > proportional to performance whose role is to help balance the load between CPUs
> > of the same type.
> >
> > The expected effect is still that the CPUs of the "low-cost" type will be
> > preferred so long as there is enough spare capacity on any of them.
> >
> > Patch [6/7] has been updated to walk all of the cache leaves and look for
> > the ones with level equal to 3 because the check used in the previous version
> > does not always work.
> >
> > The documentation patch, [7/7], is new.
> >
> > Please refer to the individual patch changelogs for details.
>
> This series along with the fix at
>
> https://lore.kernel.org/linux-pm/2806514.mvXUDI8C0e@rjwysocki.net/
>
> is now present in the experimental/intel_pstate/eas-final brangh in
> linux-pm.git:
>
> https://web.git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm.git/log/?h=experimental/intel_pstate/eas-final
>
> and it has been added to the bleeding-edge branch for 0-day build testing.

In order to test it, the kernel needs to be compiled with
CONFIG_ENERGY_MODEL set.

If at least some cores in the system are SMT-capable, SMT needs to be
turned off (either in the BIOS setup or by passing nosmt to the kernel
in the command line).

Finally, schedutil needs to be the cpufreq governor which requires
intel_pstate to operate in the passive mode (schedutil is the default
governor in that case).  The most straightforward way to switch it
into the passive mode is to write "passive" to
/sys/devices/system/cpu/intel_pstate/status (it may also be started in
the passive mode as described in
https://www.kernel.org/doc/html/latest/admin-guide/pm/intel_pstate.html).

Note that you can compare the default non-EAS configuration of
intel_pstate to the one with EAS enabled by switching it between the
"active" and "passive" modes via sysfs (no reboots needed).

Thanks!

Powered by blists - more mailing lists