Re: 2.6.18-rc4-mm1

[Dave Jones <davej@redhat.com>]

On Mon, Aug 14, 2006 at 11:13:38PM +0200, Ben B wrote:
 > Dave Jones <davej@redhat.com> uttered the following thing:
 > >  > > > [  734.156000]  [<e01f2665>] cpufreq_governor_dbs+0x2b5/0x310 [cpufreq_ondemand]
 > >  > > 
 > >  > > This makes no sense at all, because in -mm __create_workqueue doesn't
 > >  > > call lock_cpu_hotplug().
 > >  > > 
 > >  > > Are you sure this was from a tree with -mm1 applied ?
 > >  > 
 > >  > Definitely 2.6.18-rc4-mm1, and I've done a clean rebuild + removal of
 > >  > all modules under /lib/modules beforehand.
 > > 
 > > It's a real mystery.  Andrew ?
 > 
 > This seems to be specific to the ondemand governor - I just tried with
 > conservative, and alternating it with performance, with no problems, but
 > as soon as I loaded ondemand, the message appeared. It seems to fire off
 > the message as soon as I either set the governor to ondemand, or revert
 > it from ondemand to something else. But going from, eg performance to
 > conservative, wont give the message, even with ondemand loaded.

on-demand is unique in the sense that its the only governor that
creates a workqueue.

 > I wonder if this might also be related to my 1.83GHz cpu only being set
 > to a maximum of 1.33GHz via cpufreq? cpuinfo_max_freq is correct, but
 > scaling_max_freq is wrong. Though doing "cat cpuinfo_max_freq >
 > scaling_max_freq" has fixed it up, it should be correct already.

That's come up a lot lately. I'm still of the opinion that something
changed in acpi that's the explanation for this.

		Dave

-- 
http://www.codemonkey.org.uk