Mark Hellman wrote:
Thomas Renninger wrote:
Try to boot with init=/bin/bash boot param
mount /proc
mount /sys
modprobe processor
watch -n1 cat /proc/acpi/processor/*/power
The processor should mostly stay in C3 and bus master activity should be
low. Battery lifetime should be really high (as high as under windows?).
Thanks for the tip.
I think I have found the main responsible for the processor not entering C3:
the 'radeon' kernel module. When I disable 3D Acceleration in SaX, the
processor then enters C3 state much more often.
Hmmm, I wonder how much difference there is with DynamicClock Mode on/off and
3D acceleration. Especially with C3 triggering.
While running watch -n1 cat /proc/acpi/processor/*/power I noticed that the
processor jumps between C2 and C3 quite often, even when the system is not
under any apparent load. Here goes the output of
cat /proc/acpi/processor/CPU0/power
25 minutes after the system has booted:
active state: C2
max_cstate: C8
bus master activity: 00000004
states:
C1: type[C1] promotion[C2] demotion[--] latency[001] usage[00000010]
*C2: type[C2] promotion[C3] demotion[C1] latency[001] usage[00159142]
C3: type[C3] promotion[--] demotion[C2] latency[085] usage[00301416]
Is it normal that C2 has so much usage even when the system is idle?
The code needs some tweaking, I hope to be able to work on this soon.
This is probably most of the difference you see to Windows XP.
Thomas
P.S.: If you want to test very recent C-state and dynamic tick code (disable HZ
and
timer interrupt if idle) you can try the most recent patches on this site:
http://www.muru.com/linux/dyntick/
and on top you should try this patch (unfortunately it seem to not patch
cleanly any
more. Tell me and I fix it for recent kernels):
ftp://ftp.suse.com/pub/people/trenn/dyn_tick_c_states/dynamic_tick_cstate_patch.diff
Tell me if you are not familiar with patching/compiling kernel sources and I
can help
you (but expect that you have to spend some time/hours to get it going then).
Maybe an ICQ chat or if you can give me remote ssh access would help a lot.
Thomas
_______________________________________________
powersave-devel mailing list
[email protected]
http://forge.novell.com/mailman/listinfo/powersave-devel
