[RFC] why do sensors break CPU scaling

[RFC] why do sensors break CPU scaling

[Tom Psyborg]

Hi

Recently I've needed to set lowest CPU scaling profile, running ubuntu
16.04.06 I used standard approach - echoing powersave to
/sys/devices/system/cpu/cpu*/cpufreq/scaling_governor.
This did not work as the
/sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq kept returning
max scaling freq.

During testing of ubuntu 19.10 I've found that the above approach
actually does work, but as long as there are no xsensors (or just
sensors from cli) being run.
cpuinfo_cur_freq in this case was returning variable values +- 1% of
around 1.4GHz.
As soon as I issue sensors command cpuinfo_cur_freq jumps to 3.5GHz
for a fraction of second and returns back to 1.4GHz afterwards. If
running xsensors it keeps bouncing between 1.4 and 3.5GHz all the
time.

Rebooted back to 16.04.6 and was able to set lowest CPU scaling freq
as well, but issuing sensors command here once just breaks CPU scaling
that now remains at about 3.5GHz.
It can be set to lowest scaling freq again without rebooting but it
needs to change scaling_governor for all cores to something else and
then back to powersave.

Why is this happening, shouldn't sensors command just read temp/fan
values without writing to any of the CPU control registers?

Re: [RFC] why do sensors break CPU scaling

[David Niklas]

On Wed, 20 Nov 2019 21:42:12 +0100
Tom Psyborg <pozega.tomislav@gmail.com> wrote:
> Hi
> 
> Recently I've needed to set lowest CPU scaling profile, running ubuntu
> 16.04.06 I used standard approach - echoing powersave to
> /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor.
> This did not work as the
> /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq kept returning
> max scaling freq.
> 
> During testing of ubuntu 19.10 I've found that the above approach
> actually does work, but as long as there are no xsensors (or just
> sensors from cli) being run.
> cpuinfo_cur_freq in this case was returning variable values +- 1% of
> around 1.4GHz.
> As soon as I issue sensors command cpuinfo_cur_freq jumps to 3.5GHz
> for a fraction of second and returns back to 1.4GHz afterwards. If
> running xsensors it keeps bouncing between 1.4 and 3.5GHz all the
> time.
> 
> Rebooted back to 16.04.6 and was able to set lowest CPU scaling freq
> as well, but issuing sensors command here once just breaks CPU scaling
> that now remains at about 3.5GHz.
> It can be set to lowest scaling freq again without rebooting but it
> needs to change scaling_governor for all cores to something else and
> then back to powersave.
> 
> Why is this happening, shouldn't sensors command just read temp/fan
> values without writing to any of the CPU control registers?

I don't know if the maintainers will notice your email, but I did. I
don't guarantee that they'll notice or help you, but this should assist
you in writing a proper question.
You need to include the output of uname -a on both ubuntu boxes. The
output of:
dpkg -l xsensors
dpkg -l lm-sensors

The information on which processor you own and the motherboard and
it's BIOS version just in case.

This is just my understanding and it might be wrong, but the CPU is
probably accessed to do the request to the fan values and so it ramps up
expecting to have to deal with a more intense workload (which is exactly
what Ryzen and several newer Intel processors are supposed to do), and so
you're seeing expected behavior.
I've no idea how you'd change this.
Alternatively, and this is just a theory, you could have some program on
the system changing the behavior of the CPU just after you change it to
what you want it to be. As in, inotify is involved.

You're welcome,
David

-------------------------------------------------
This free account was provided by VFEmail.net - report spam to abuse@vfemail.net
 
ONLY AT VFEmail! - Use our Metadata Mitigator to keep your email out of the NSA's hands!
$24.95 ONETIME Lifetime accounts with Privacy Features!  
15GB disk! No bandwidth quotas!
Commercial and Bulk Mail Options!  

Re: [RFC] why do sensors break CPU scaling

[Tom Psyborg]

On 21/11/2019, David Niklas <Hgntkwis@vfemail.net> wrote:
> On Wed, 20 Nov 2019 21:42:12 +0100
> Tom Psyborg <pozega.tomislav@gmail.com> wrote:
>> Hi
>>
>> Recently I've needed to set lowest CPU scaling profile, running ubuntu
>> 16.04.06 I used standard approach - echoing powersave to
>> /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor.
>> This did not work as the
>> /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq kept returning
>> max scaling freq.
>>
>> During testing of ubuntu 19.10 I've found that the above approach
>> actually does work, but as long as there are no xsensors (or just
>> sensors from cli) being run.
>> cpuinfo_cur_freq in this case was returning variable values +- 1% of
>> around 1.4GHz.
>> As soon as I issue sensors command cpuinfo_cur_freq jumps to 3.5GHz
>> for a fraction of second and returns back to 1.4GHz afterwards. If
>> running xsensors it keeps bouncing between 1.4 and 3.5GHz all the
>> time.
>>
>> Rebooted back to 16.04.6 and was able to set lowest CPU scaling freq
>> as well, but issuing sensors command here once just breaks CPU scaling
>> that now remains at about 3.5GHz.
>> It can be set to lowest scaling freq again without rebooting but it
>> needs to change scaling_governor for all cores to something else and
>> then back to powersave.
>>
>> Why is this happening, shouldn't sensors command just read temp/fan
>> values without writing to any of the CPU control registers?
>
> I don't know if the maintainers will notice your email, but I did. I
> don't guarantee that they'll notice or help you, but this should assist
> you in writing a proper question.
> You need to include the output of uname -a on both ubuntu boxes. The
> output of:
> dpkg -l xsensors
> dpkg -l lm-sensors

this is on 16.04.6

whtw46ww4@I5576:~$ uname -a
Linux I5576 4.15.0-51-generic #55~16.04.1-Ubuntu SMP Thu May 16
09:24:37 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux
whtw46ww4@I5576:~$ dpkg -l xsensors
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend
|/ Err?=(none)/Reinst-required (Status,Err: uppercase=bad)
||/ Name                                          Version
       Architecture                Description
+++-=============================================-===========================-===========================-===============================================================================================
ii  xsensors                                      0.70-3
       amd64                       hardware health information viewer
whtw46ww4@I5576:~$ dpkg -l lm-sensors
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend
|/ Err?=(none)/Reinst-required (Status,Err: uppercase=bad)
||/ Name                                          Version
       Architecture                Description
+++-=============================================-===========================-===========================-===============================================================================================
ii  lm-sensors                                    1:3.4.0-2
       amd64                       utilities to read
temperature/voltage/fan sensors


>
> The information on which processor you own and the motherboard and
> it's BIOS version just in case.
>
> This is just my understanding and it might be wrong, but the CPU is
> probably accessed to do the request to the fan values and so it ramps up
> expecting to have to deal with a more intense workload (which is exactly
> what Ryzen and several newer Intel processors are supposed to do), and so
> you're seeing expected behavior.

I don't think that is the case here, since I can run any kind of more
demanding app than xsensors is, even compile software with -j 5
(multithread) and CPU clocks don't exceed 1.40GHz. Only on xsensors
launch CPU clocks are reset.

Regards, Tom

Re: [RFC] why do sensors break CPU scaling

[Pavel Machek]

[-- Attachment #1: Type: text/plain, Size: 695 bytes --]

On Wed 2019-11-20 21:42:12, Tom Psyborg wrote:
> Hi
> 
> Recently I've needed to set lowest CPU scaling profile, running ubuntu
> 16.04.06 I used standard approach - echoing powersave to
> /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor.
> This did not work as the
> /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq kept returning
> max scaling freq.

Noone noticed, right?

If you still believe that's problem, you may want to look at
MAINTAINERS file, and put sensors maintainers in the Cc list.

Best regards,
									Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 181 bytes --]

Re: [RFC] why do sensors break CPU scaling

[Tom Psyborg]

On 31/01/2020, Pavel Machek <pavel@ucw.cz> wrote:
> On Wed 2019-11-20 21:42:12, Tom Psyborg wrote:
>> Hi
>>
>> Recently I've needed to set lowest CPU scaling profile, running ubuntu
>> 16.04.06 I used standard approach - echoing powersave to
>> /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor.
>> This did not work as the
>> /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq kept returning
>> max scaling freq.
>
> Noone noticed, right?
>
> If you still believe that's problem, you may want to look at
> MAINTAINERS file, and put sensors maintainers in the Cc list.
>
> Best regards,
> 									Pavel
> --
> (english) http://www.livejournal.com/~pavelmachek
> (cesky, pictures)
> http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html
>

it is definitely a problem. i've installed radeon-profile and can
monitor both CPU and GPU temps from there without CPU scaling
breakage. CCing related list