Arjan van de Ven wrote: > Dale Pontius wrote: >> While I've been typing this, the laptop has been running Gentoo's >> "emerge --sync". It just finished and I started powertop, preparing to >> start unloading USB modules to see if I could get into C3. It's already >> there. The listing is tacked onto the end. What seems odd to me is >> that all of the numbers are essentially the same as the listing >> yesterday, except that yesterday it was in C2, and today it's in C3. >> >> Can you point me to the source that handles these state transitions, or >> any documentation? >> > > drivers/acpi/processor_idle.c handles this code. > The "old" code (mainline) isn't very smart, and doesn't deal with > tickless well. > > One thing to try is to use the -hrt patches, they contain a rewrite of > this code, specifically the "menu governer" which is designed for use > in a tickless idle kernel. I have the -hrt patches applied to this machine, so I'm grubbing my way through the rewritten processor_idle.c as real work makes bits of time.
What seems odd is that C3 is always available, it just isn't used under certain conditions. I've noticed that framebuffer interrupts are sometimes there, sometimes not, so I tried a little experiment. After having run a job that saturated the machine and gone back to idle, it was sitting in C2 under conditions that would otherwise get to C3. Using the KVM I switched to the laptop, logged into the text console, did a few things, logged out, and switched back to the deskside. The powertop on the laptop running under ssh showed the residue of my "native" activity, including framebuffer interrupts. Then as it calmed down, it went into C3. So from what I can see, it appears that C3 is always available, but there are some circumstances where the idle logic can't figure out that it can get there. If I can change the circumstances, in this case by using the native console, those circumstances change and it idles to C3. I'll keep looking at the code, as time allows. Today is the first time I've even looked. Thanks, Dale Pontius _______________________________________________ Power mailing list [email protected] http://www.bughost.org/mailman/listinfo/power
