On 1/21/2011 2:55 AM, Bernhart, Bryan wrote:
I’ve been noticing an excess amount of kernel (2.6.33.5-23-1006-ivi)
overhead contributed from volatile reads of the ACPI interface. Upon
investigation, the primary clocksource was using the rather less
efficient ACPI_PM after TSC fails the idle test, and switches over.
Interesting enough, a closer (on chip solution I believe) clocksource
called “tsc_constant” shows up when I run the query cpuinfo. But this
isn’t an available clocksource. The kernel was compiled with TSC &
HPET support, but I would like to know if anyone could shed some light
about enabling a faster clocksource or diagnosing the existing clock
overhead. This is a Tunnelcreek 1.5Ghz sku.
tsc will stop in idle (constant just means "fixed frequency", eg does
not change if the frequency of the cpu changes)
you should have HPET enabled in the bios to use it....
_______________________________________________
MeeGo-kernel mailing list
[email protected]
http://lists.meego.com/listinfo/meego-kernel
