[PATCH 0/2] cpufreq: intel_pstate: HWP support on hybrid processors

From: "Rafael J. Wysocki" <rjw@rjwysocki.net>
To: Linux PM <linux-pm@vger.kernel.org>
Cc: Linux ACPI <linux-acpi@vger.kernel.org>,
	LKML <linux-kernel@vger.kernel.org>,
	Len Brown <len.brown@intel.com>, Chen Yu <yu.c.chen@intel.com>,
	Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>,
	Zhang Rui <rui.zhang@intel.com>
Subject: [PATCH 0/2] cpufreq: intel_pstate: HWP support on hybrid processors
Date: Wed, 12 May 2021 16:14:18 +0200	[thread overview]
Message-ID: <2212930.ElGaqSPkdT@kreacher> (raw)

Hi All,

In Intel Hybrid processors there are cores with different micro-architecture,
so the relationship between CPU frequency and performance is generally not
uniform across all CPUs.  Because the units of CPU performance used in the
MSR_HWP_CAPABILITIES and MSR_HWP_REQUEST registers are the same for all CPUs,
the scaling between HWP performance levels and frequency generally depends on
the core type and it is not uniform across all CPUs either, so some assumptions
made by intel_pstate are not applicable to hybrid processors.

As the scaling factor between HWP performance and frequency will now generally
depend on the CPU, in particular it cannot be assumed to be equal to the
scaling factor between P-states and frequency for all CPUs which has been one
of the assumptions made so far by intel_pstate.  Unfortunately, in general it
is not very straightforward to determine that scaling factor which is the
purpose of patch [2/2] (please refer to the changelog of that patch for
details).

Moreover, two of the global sysfs attributes exposed by intel_pstate only work
as expected if all CPUs have the same number of HWP performance levels, so it
is better to avoid exposing those attributes when running on hybrid processors,
which is taken care of by patch [1/2].

Thanks!