4.8.1 regression with cpufreq governors
May or may not be related to similar reports, but here's what I've just observed on my system. Built a stock kernel from tags/v4.8.1, relevant cpufreq bits:
CONFIG_ACPI_CPU_FREQ_PSS=yCONFIG_CPU_FREQ=y CONFIG_CPU_FREQ_GOV_ATTR_SET=y CONFIG_CPU_FREQ_GOV_COMMON=y CONFIG_CPU_FREQ_STAT=y CONFIG_CPU_FREQ_STAT_DETAILS=y
# CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE is not set # CONFIG_CPU_FREQ_DEFAULT_GOV_POWERSAVE is not set # CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACE is not set # CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND is not set CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE=y
# CONFIG_CPU_FREQ_DEFAULT_GOV_SCHEDUTIL is not set CONFIG_CPU_FREQ_GOV_PERFORMANCE=y CONFIG_CPU_FREQ_GOV_POWERSAVE=m CONFIG_CPU_FREQ_GOV_USERSPACE=m CONFIG_CPU_FREQ_GOV_ONDEMAND=m CONFIG_CPU_FREQ_GOV_CONSERVATIVE=y CONFIG_CPU_FREQ_GOV_SCHEDUTIL=m
# CONFIG_X86_PCC_CPUFREQ is not set CONFIG_X86_ACPI_CPUFREQ=m CONFIG_X86_ACPI_CPUFREQ_CPB=y
Conservative is set as default governer, yet when boot completes, all CPUs are pegged at the highest frequency. Changing governor to powersave knocks them all down to the lowest available frequency. Putting them back on conservative (or ondemand) results in no change in frequency, despite generating load. Switching to performance of course kicks them back up to high frequency. Basically, the governors don't seem to be ... governing.
Signature · 1 line
[email protected], [email protected]
Quoted 26 lines · Tim Walberg
May or may not be related to similar reports, but here's what I've just observed on my system. Built a stock kernel from tags/v4.8.1, relevant cpufreq bits:
CONFIG_ACPI_CPU_FREQ_PSS=yCONFIG_CPU_FREQ=yCONFIG_CPU_FREQ_GOV_ATTR_SET=yCONFIG_CPU_FREQ_GOV_COMMON=yCONFIG_CPU_FREQ_STAT=yCONFIG_CPU_FREQ_STAT_DETAILS=y
#CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCEis not set #CONFIG_CPU_FREQ_DEFAULT_GOV_POWERSAVEis not set #CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACEis not set #CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMANDis not setCONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE=y
#CONFIG_CPU_FREQ_DEFAULT_GOV_SCHEDUTILis not setCONFIG_CPU_FREQ_GOV_PERFORMANCE=yCONFIG_CPU_FREQ_GOV_POWERSAVE=mCONFIG_CPU_FREQ_GOV_USERSPACE=mCONFIG_CPU_FREQ_GOV_ONDEMAND=mCONFIG_CPU_FREQ_GOV_CONSERVATIVE=yCONFIG_CPU_FREQ_GOV_SCHEDUTIL=m
#CONFIG_X86_PCC_CPUFREQis not setCONFIG_X86_ACPI_CPUFREQ=mCONFIG_X86_ACPI_CPUFREQ_CPB=y
Quoted 6 lines
Conservative is set as default governer, yet when boot completes, all CPUs are pegged at the highest frequency. Changing governor to powersave knocks them all down to the lowest available frequency. Putting them back on conservative (or ondemand) results in no change in frequency, despite generating load. Switching to performance of course kicks them back up to high frequency. Basically, the governors don't seem to be ... governing.
The "convervative" governor issue seems to be the one fixed recently
(http://git.kernel.org/cgit/linux/kernel/git/rafael/linux-pm.git/commit/?h=pm-cpufreq&id=abb6627910a1e783c8e034b35b7c80e5e7f98f41).
I'm not sure why "ondemand" behaves incorrectly for you though.
Thanks,
Rafael
Quoted 34 lines · Rafael J. Wysocki
Quoted 26 lines · Tim Walberg
May or may not be related to similar reports, but here's what I've just observed on my system. Built a stock kernel from tags/v4.8.1, relevant cpufreq bits:
CONFIG_ACPI_CPU_FREQ_PSS=yCONFIG_CPU_FREQ=yCONFIG_CPU_FREQ_GOV_ATTR_SET=yCONFIG_CPU_FREQ_GOV_COMMON=yCONFIG_CPU_FREQ_STAT=yCONFIG_CPU_FREQ_STAT_DETAILS=y
#CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCEis not set #CONFIG_CPU_FREQ_DEFAULT_GOV_POWERSAVEis not set #CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACEis not set #CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMANDis not setCONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE=y
#CONFIG_CPU_FREQ_DEFAULT_GOV_SCHEDUTILis not setCONFIG_CPU_FREQ_GOV_PERFORMANCE=yCONFIG_CPU_FREQ_GOV_POWERSAVE=mCONFIG_CPU_FREQ_GOV_USERSPACE=mCONFIG_CPU_FREQ_GOV_ONDEMAND=mCONFIG_CPU_FREQ_GOV_CONSERVATIVE=yCONFIG_CPU_FREQ_GOV_SCHEDUTIL=m
#CONFIG_X86_PCC_CPUFREQis not setCONFIG_X86_ACPI_CPUFREQ=mCONFIG_X86_ACPI_CPUFREQ_CPB=yQuoted 6 lines
Conservative is set as default governer, yet when boot completes, all CPUs are pegged at the highest frequency. Changing governor to powersave knocks them all down to the lowest available frequency. Putting them back on conservative (or ondemand) results in no change in frequency, despite generating load. Switching to performance of course kicks them back up to high frequency. Basically, the governors don't seem to be ... governing.
Quoted 4 lines
The "convervative" governor issue seems to be the one fixed recently
(http://git.kernel.org/cgit/linux/kernel/git/rafael/linux-pm.git/commit/?h=pm-cpufreq&id=abb6627910a1e783c8e034b35b7c80e5e7f98f41).I'm not sure why "ondemand" behaves incorrectly for you though.
@Tim: Can you please try this as well:
https://marc.info/?l=linux-kernel&m=147624218132250&w=2
Subject: [PATCH V2] cpufreq: skip invalid entries when searching the frequency
Signature · 1 line
viresh
This indeed turned out to be the fix.
Quoted 34 lines · Rafael J. Wysocki
Quoted 34 lines · Tim Walberg
> May or may not be related to similar reports, but here's what I've just observed > on my system. Built a stock kernel from tags/v4.8.1, relevant cpufreq bits:
>
>CONFIG_ACPI_CPU_FREQ_PSS=y
>CONFIG_CPU_FREQ=y
>CONFIG_CPU_FREQ_GOV_ATTR_SET=y
>CONFIG_CPU_FREQ_GOV_COMMON=y
>CONFIG_CPU_FREQ_STAT=y
>CONFIG_CPU_FREQ_STAT_DETAILS=y
> #CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCEis not set
> #CONFIG_CPU_FREQ_DEFAULT_GOV_POWERSAVEis not set
> #CONFIG_CPU_FREQ_DEFAULT_GOV_USERSPACEis not set
> #CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMANDis not set
>CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE=y
> #CONFIG_CPU_FREQ_DEFAULT_GOV_SCHEDUTILis not set
>CONFIG_CPU_FREQ_GOV_PERFORMANCE=y
>CONFIG_CPU_FREQ_GOV_POWERSAVE=m
>CONFIG_CPU_FREQ_GOV_USERSPACE=m
>CONFIG_CPU_FREQ_GOV_ONDEMAND=m
>CONFIG_CPU_FREQ_GOV_CONSERVATIVE=y
>CONFIG_CPU_FREQ_GOV_SCHEDUTIL=m
> #CONFIG_X86_PCC_CPUFREQis not set
>CONFIG_X86_ACPI_CPUFREQ=m
>CONFIG_X86_ACPI_CPUFREQ_CPB=y
>
>
> Conservative is set as default governer, yet when boot completes, all CPUs are > pegged at the highest frequency. Changing governor to powersave knocks them all > down to the lowest available frequency. Putting them back on conservative (or > ondemand) results in no change in frequency, despite generating load. Switching > to performance of course kicks them back up to high frequency. Basically, the > governors don't seem to be ... governing.
Quoted 7 lines
Quoted 7 lines
The "convervative" governor issue seems to be the one fixed recently
(http://git.kernel.org/cgit/linux/kernel/git/rafael/linux-pm.git/commit/?h=pm-cpufreq&id=abb6627910a1e783c8e034b35b7c80e5e7f98f41).I'm not sure why "ondemand" behaves incorrectly for you though.
Thanks,
Rafael
End of included message
Signature · 1 line
[email protected], [email protected]
Quoted 1 line · Tim Walberg
This indeed turned out to be the fix.
Really? How can this fix the ondemand governor thing?
Or is it that Ondemand never broke ?
Signature · 1 line
viresh