[PATCH 0/2] Fix energy-pkg event for AMD CPUs

From: Dhananjay Ugwekar
Date: Thu May 02 2024 - 05:51:57 EST


After commit 63edbaa48a57 ("x86/cpu/topology: Add support for the AMD 0x80000026 leaf"),
on AMD processors that support extended CPUID leaf 0x80000026, the
topology_die_cpumask() and topology_logical_die_id(), no longer return
the package cpumask and package id, instead they return the CCD (Core
Complex Die) mask and CCD id respectively. This leads to the energy-pkg
event scope to be modified to CCD instead of package.

This is functionally incorrect as the AMD energy-pkg is a package scope
event.

On a 12 CCD 1 Package AMD Zen4 Genoa machine:

$ cat /sys/devices/power/cpumask
0,8,16,24,32,40,48,56,64,72,80,88.

Expected cpumask for 1 Package AMD CPUs energy-pkg event is supposed to
be just "0".

This in turn leads to incorrect values of package energy consumption.

$ perf stat -C 100 -e power/energy-pkg/ sleep 1

Performance counter stats for 'CPU(s) 100':

36.60 Joules power/energy-pkg/

1.005821050 seconds time elapsed

$ perf stat -a -e power/energy-pkg/ sleep 1

Performance counter stats for 'system wide':

414.03 Joules power/energy-pkg/

1.005252132 seconds time elapsed

In both cases the energy consumption values should have been the same,
but instead, due to the incorrect cpumask, the values are being
aggregated at a CCD level in the second case.

Replacing the die macros with the package macros(for AMD CPUs) fixes
this issue.

This patch cleanly applies on top of tip/master after
commit acf68d98cae8 ("perf/x86/rapl: Add support for Intel Lunar Lake"),
but not on Linus' tree.

Dhananjay Ugwekar (2):
perf/x86/rapl: Rename maxdie to nr_rapl_pmu and dieid to rapl_pmu_idx
perf/x86/rapl: Fix the energy-pkg event for AMD CPUs

arch/x86/events/rapl.c | 44 +++++++++++++++++++++++++++++++-----------
1 file changed, 33 insertions(+), 11 deletions(-)

--
2.34.1