On 26-06-21, 09:41, Qian Cai wrote:
Another date point is that set ACPI_CPPC_CPUFREQ_FIE=n fixed the issue that any CPU could run below the lowest freq.
schedutil:
# cat /sys/devices/system/cpu/*/cpufreq/cpuinfo_cur_freq | sort | uniq -c
80 1000000
78 1400000
1 2010000
1 2800000
userspace:
# cat /sys/devices/system/cpu/*/cpufreq/cpuinfo_cur_freq | sort | uniq -c
158 1000000
2 2000000
ACPI_CPPC_CPUFREQ_FIE can play a role with schedutil, but not with
userspace governor. Userspace doesn't use the values being updated by
ACPI_CPPC_CPUFREQ_FIE. So I think the CPUs may not have been idle,
just for some reason.
Also, now that you are able run on latest kernel (linux-next), it
would be better if we can talk in terms of that only going forward.
5.4 adds more to the already unstable results :)
--
viresh