On 6/24/2021 6:48 AM, Ionela Voinescu wrote: > Not if the counters are implemented properly. The kernel considers that > both reference and delivered performance counters should stop or reset > during idle. The kernel would not account for idle itself. > > If the reference performance counter does not stop during idle, while > the core performance counter (delivered) does stop, the behavior above > should be seen very often. > > Qian, do you see these small delivered performance values often or > seldom? Ionela, so I managed to upgrade the kernel on the system to today's linux-next which suppose to include this series. The delivered perf is now 280. However, scaling_min_freq (200 MHz) is not equal to lowest_perf (100). scaling_driver: acpi_cppc scaling_governor: schedutil Is that normal because lowest_nonlinear_perf is 200? Also, on this pretty idle system, 158 of 160 CPUs are always running in max freq (280 MHz). The other 2 are running in 243 and 213 MHz according to scaling_cur_freq. Apparently, "schedutil" does not work proper on this system. I am going to try other governors to narrow down the issue a bit. FYI, here is the acpi_cppc registers reading: /sys/devices/system/cpu/cpu0/acpi_cppc/feedback_ctrs ref:160705801 del:449594095 /sys/devices/system/cpu/cpu0/acpi_cppc/highest_perf 300 /sys/devices/system/cpu/cpu0/acpi_cppc/lowest_freq 1000 /sys/devices/system/cpu/cpu0/acpi_cppc/lowest_nonlinear_perf 200 /sys/devices/system/cpu/cpu0/acpi_cppc/lowest_perf 100 /sys/devices/system/cpu/cpu0/acpi_cppc/nominal_freq 2800 /sys/devices/system/cpu/cpu0/acpi_cppc/nominal_perf 280 /sys/devices/system/cpu/cpu0/acpi_cppc/reference_perf 100 /sys/devices/system/cpu/cpu0/acpi_cppc/wraparound_time 18446744073709551615
