On 18/12/2007, Tomas Carnecky <[EMAIL PROTECTED]> wrote: > > Alan Jenkins wrote: > > On 17/12/2007, Tomas Carnecky <[EMAIL PROTECTED]> wrote: > >> Oliver Neukum wrote:
>> - clocksource_register sometimes appears after a fresh boot, and > >> disappears later, accounting for two wakeups/s. It wasn't listed before > >> because I haven't rebooted the laptop, and after a few minutes that > >> function disappears (or after I start something, I'm not sure). > >> > >> I'll look into increasing the fbcon timer and getting rid of the > >> clocksource function to see if that improves my power consumption. > >> Thanks so far. > > > > The clocksource one sounds familiar. (No-one responded to my queries > > about it). My problem was caused by TSC consistency checks, which I > > believe could be safely avoided. IIRC my wakeups didn't go away after > > a while, so maybe your issue is different. > > > > Does yours default to the TSC? What happens if you change > > Default is hpet. > > > clocksources? Does switching back to TSC cause it to start again? > > It's unpredictable. Now I booted the laptop and clocksource_register > wasn't listed in powertop. After I echoed 'tsc' to current_clocksource, > the system shortly after locked up (hard). I also tried 'jiffies' and > the system locked up, too, albeit only softly, `date` stopped increasing > the time, and trying to revert it back to hpet didn't work (cat > current_cl.. still shows 'jiffies'). Seems like my system doesn't like > me changing the clocksource. I tried rebooting a few times now, no luck getting clocksource_register > back into powertop. Btw, that function is executed at a fixed HZ/2 rate > if 'watchdog_list' is non-empty. And it seems that every clocksource > that can be used as a clocksource watchdog is added to that list. That's > about how I understand the code. So why the function is listed only > sometimes in powertop is a mystery to me. > disclaimer: I am not a kernel hacker and this is mysterious to me too. Talking to me is unlikely to help get bug fixes into the kernel :-). I saw "clocksource watchdog" when I read the code for my problem, but my understanding was that it was a watchdog for the TSC clocksource, to check it against a more reliable (in some sense) clocksource. It was definitely at 2 wakeups/s. So I think you are only getting the clocksource_register wakeups when the TSC clocksource is used instead of HPET. The wakeups would go away later if the TSC became unreliable, because the kernel would switch to HPET. On some boots, the switch might happen immediately, so you don't see the watchdog timer at all. That should show up in the kernel log. E.g. Time: tsc clocksource has been installed. ... "several minutes" ... Marking TSC unstable due to: cpufreq changes. Time: acpi_pm clocksource has been installed. Probably you had syslog set up saving kernel logs from the previous boots, so if I'm right you'll be able to find something like that in there. Then you could disable the TSC with a kernel boot option if you wanted. I don't know why jiffies would break, though maybe it's to do with NO_IDLE_HZ. Alan
_______________________________________________ Discuss mailing list [email protected] http://mail.lesswatts.org/mailman/listinfo/discuss
