cpufreq_basic_tests() reads and writes every policy file, offlines and onlines every non-boot CPU five times, and then shuffles all frequencies and governors of every policy. The amount of work scales with the number of CPUs, but the test runs under the default 45 second kselftest timeout. On large systems the hotplug loop alone does not fit. On a 256 CPU x86 machine each offline/online pair takes about 0.5 seconds, so the reboot_cpus 5 step needs more than 10 minutes. Even with the timeout raised to 180 seconds it is killed in the second of the five loops, and the frequency and governor tests never run: not ok 1 selftests: cpufreq: main.sh # TIMEOUT 180 seconds A larger fixed value would only move the problem to the next bigger machine. The basic test has a fixed number of iterations and does not wait on external events, so disable the timeout with timeout=0, as ftrace, timers and net/forwarding already do. Fixes: 852c8cbf34d3 ("selftests/kselftest/runner.sh: Add 45 second timeout per test") Cc: stable@vger.kernel.org Signed-off-by: Gyokhan Kochmarla --- tools/testing/selftests/cpufreq/settings | 5 +++++ 1 file changed, 5 insertions(+) create mode 100644 tools/testing/selftests/cpufreq/settings diff --git a/tools/testing/selftests/cpufreq/settings b/tools/testing/selftests/cpufreq/settings new file mode 100644 index 000000000000..d3e07370a7e4 --- /dev/null +++ b/tools/testing/selftests/cpufreq/settings @@ -0,0 +1,5 @@ +# The basic test hotplugs every non-boot CPU five times and then walks +# all frequencies and governors of every policy, so its runtime grows +# with the number of CPUs. No fixed timeout fits every machine, and the +# test always terminates on its own, so disable the runner timeout. +timeout=0 base-commit: 6628e7d4c6b25c8d18f1345e3a20d21829aae7e4 -- 2.50.1