any updates for apmd exposing c-state preferences?

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

any updates for apmd exposing c-state preferences?

Nicholas Sielicki
Hello,

I'm curious if there are any planned updates to apmd with regards to exposing a means for a user to set which c-state a processor should go into.

See this prior mailing list thread for context: http://comments.gmane.org/gmane.os.openbsd.misc/209854

The reason I ask is that when looking at this comment from December 2014 [1], and doing some reading about what intel_powerclamp is doing [2], I think that it could be very beneficial for battery life to force the CPU into a higher c-state when idle. As of 5.4-current, MWAIT is going to C1 when idle in all cases [3], and I wish to know if that's changed since.

[1]: http://permalink.gmane.org/gmane.os.openbsd.misc/218394
[2]: https://www.kernel.org/doc/Documentation/thermal/intel_powerclamp.txt
[3]: http://permalink.gmane.org/gmane.os.openbsd.misc/209861

Thank you!
Nicky Sielicki

Reply | Threaded
Open this post in threaded view
|

Re: any updates for apmd exposing c-state preferences?

Philip Guenther-2
On Tue, Mar 10, 2015 at 9:51 PM, Nicholas Sielicki <[hidden email]> wrote:
> I'm curious if there are any planned updates to apmd with regards to exposing a means for a user to set which c-state a processor should go into.
...
> As of 5.4-current, MWAIT is going to C1 when idle in all cases [3], and I wish to know if that's changed since.

Actually, that was disabled as it caused *increased* power consumption
on many boxes.  There's work in progress to have the kernel used the
C-state information provided by ACPI so that it uses MWAIT properly
for the involved system, including the higher C-states.  The main
hold-up is getting it to work robustly on various systems; sthen@ has
one box which hangs as soon as it goes multiuser with the in-progress
work.  Once I figure out which rule I'm not following (or which rule
the _system_ isn't following and how to detect that...) there will be
a wider round of testing...

Philip Guenther