On Sun, May 13, 2007 at 02:01:26PM +0200, Petter Reinholdtsen wrote:
>> How accurate is this tool? I saw that hald-addon-cpuf (which I
>> assume is short for ...-cpufreq), more precisely in
>> "queue_delayed_work_on (delayed_work_timer_fn)" stood for a lot of
>> wakeups, so I killed it. Yet it keeps running up and down the
>> lists...
> It is claimed to be fairly accurate. I recommend asking on the
> #powertop IRC channel if you need details.
I asked, and the explanation seems to be that the timers get accounted to the
process responsible for firing off the events in the first place. In this
case, hald-addon-cpufreq was the one setting the ondemand governor, and thus
responsible even after it was killed. (If I did the same from bash, bash
would show up.)
Anyhow, ondemand has stopped generating interrupts in 2.6.22-rc1. There's
a bug in the timer stats so it _looks_ like it's still generating interrupts,
though -- in reality it's just piggybacking on whatever other processes or
drivers's interrupts.
>> Apart from that, most of my time seems to be spent in i945
>> interrupts, but I guess those are needed for my display somehow. :-)
> The Tips and Tricks page mentioned a bug in i945 running interrupts
> every frame even if the interrupt isn't used for anything. Patch to
> fix it is available on the project web page, though the project web
> page seem to be down at the moment.
There's another bug still, but Keith Packard is on the case.
19:46 < keithp> arjan: need fixes in the X.org 2D driver to tell the kernel
to turn off interrupts. and a fix in the DRM module to leave
them off on VT switch
>> - 517.2 wakeups-from-idle per second, with tickless kernel :-/
> That is rather high if the machine is idle. It is supposed to be
> around 10 on a well configured system. :)
Well, ipw3945 has 50-60 alone, and there's unpatched issues for Firefox, the
GNOME mixer applet, dhcdbd and even gnome-power-manager. Most of these have
been pushed through upstream, though, so eventually they'll be trickling down
into Debian as well.
One might consider doing strategic bug filings and tag them with some
"power-saving" usertag, but I'm not sure if there's a real worth to it.
/* Steinar */
--
Homepage: http://www.sesse.net/
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
