kern/140361: [cpufreq] speed-stepping broken on PhenomII (acpi?)
freebsd at soulrebel.in-berlin.de
Wed Sep 15 15:10:04 UTC 2010
The following reply was made to PR kern/140361; it has been noted by GNATS.
From: Hannes <freebsd at soulrebel.in-berlin.de>
To: "Bruce Cran" <bruce at cran.org.uk>
Cc: bug-followup at FreeBSD.org
Subject: Re: kern/140361: [cpufreq] speed-stepping broken on PhenomII
Date: Wed, 15 Sep 2010 17:00:38 +0200
thanks for your reply.
> It looks like powerd is doing its job: the line "dev.cpu.0.freq: 800"
> shows that your CPU is running at 800MHz which is far below its
> maximum. By checking the value of dev.cpu.0.freq every so often you
> should see it reduce when the system is idle and increase when doing
> CPU-intensive work like running a build etc.
No, the point is that dev.cpu.0.freq is NOT adjusting automatically. =20
That's what got me starting this whole report. In the above example I =20
set it manually to 800, which actually works (e.g. the power =20
consumption goes down), but it only works manually and powerd does not =20
regulate it up, even if I put 100% load on all cores.
> The cx_usage lines (e.g. dev.cpu.0.cx_usage: 100.00% last 500us) are
> just reporting the amount of time the CPU was in a certain power state.
> Since your CPU doesn't support C2 and C3 all the time is spent in C1 -
> hence the "100%". That's normal, and would only be less than 100% if
> you had C2 or C3 and you'd set hw.acpi.cpu.cx_lowest.
The CPU should have C2 or C3.
> I don't think this is a bug in FreeBSD if you can confirm that the
> dev.cpu.0.freq value is changing. If so, I'd like to close this PR.
- on FreeBSD i386 dev.cpu.0.cx_usage adjusts automatically, I have est =20
devices and C2/C3-states
- on FreeBSD amd64 nothing adjusts automatically, est doesnt attach =20
and I dont get states
So unless there is some amd64-related vodoo you have to do that I =20
don't know of (these are my only amd64-machines I ever had), I am =20
pretty sure it's a bug.
I don't have access to the machines the next two weeks but can provide =20
further info then. I would gladly test patches aswell, if they apply =20
to RELENG_8. I will also try to get hold of an i386 install again and =20
then provide diffs of dmesg and sysctls.
Thanks you very much for your help!
More information about the freebsd-bugs