>-----Original Message----- >From: Kevin Hilman [mailto:[EMAIL PROTECTED] >Sent: Thursday, January 31, 2008 4:09 PM >To: Pallipadi, Venkatesh >Cc: [email protected]; Adam Belay >Subject: Re: [PATCH/RFC] powertop non-ACPI: use CPUidle for >C-state details > >"Pallipadi, Venkatesh" <[EMAIL PROTECTED]> writes: > >> Nice timing for the patch. I am working on cpuidle patches that will >> add things like C1 idle residency time in /sysfs and was asking >> Arjan about switching powertop to use cpuidle interfaces :-). > >Great! My platform reports residency for all the states in sysfs >already, so powertop using sysfs is very useful to me. > >> One major comment is that with the patch C-state timing in powertop >> seems to be off, Due to different units of time in /proc/acpi and >> sysfs cpuidle. /proc/acpi C-state timings are in ACPI PM timer >> ticks and cpuidle exports it in micro-seconds. > >Yes, the percentage numbers are indeed off. I was just noticing the >same problem. I'm working right now on changing those calculations to >use usecs instead of ticks. > >I assume the hardcoded frequency in powertop (3579.545) is the >frequency of the ACPI timer? I'm pretty ignorant of ACPI, but am >guessing that's what those calculations are for.
Yes. That hardcoded freq number is coming from ACPI PM timer frequency. Thanks, Venki _______________________________________________ Power mailing list [email protected] http://www.bughost.org/mailman/listinfo/power
