From: Alexandre P. N. <al...@om...> - 2006-04-13 16:10:44
|
Dave Hylands escreveu: >Hi David, > >On 4/13/06, David Vescovi <dve...@ta...> wrote: > > >>Thanks for the info. >>I am aware of the clocking errata. The "TURBO" mode is really ment for >>throttling up in response to higher CPU usage as there is not the delay >>associated with the FCS (TURBO has its own issues too). >>This mode seems to have gone by the wayside when >>the 400mhz parts came out. >>In the CE world everything seems to be shifting to a "variable tic" >>frequency >>for this energy management type stuff. >> >> > >There is a patch called VST (tick elimination) available from over here: >http://sourceforge.net/project/showfiles.php?group_id=20460 >At work we use this on an ARM-9. It requires a little bit of work to integrate. > >We observed a few percent savings by incorporating this patch. >Probably not worth the effort, but worthwhile for squeezing out every >last drop. > >It basically looks at the timer queue and instead of scheduling a >timer tick 10 msec in the future, schedules it when the next timer >expires. > > Not sure if it's the same patch, but there are provisions on 2.6.15 to do that; You have to enable the kernel 2.6.15 option at Kernel Features -> Dynamic Tick Timer, and later you have to echo 1 >/sys/somedir/somefile . It may or may not help improve battery duration on some kind of batteries, but my amperimeter couldn't detect more tham 15ma reduction with this option enabled. You can see that the timer is not "ticking" regularly by comparing tick timer at /proc/interrupts with this option active versus the standard, inactive fashion. Alexandre |