mapphones: look into disable one cpu (or better scheduling) to improve power management

[MerlijnWajer]

18:03 < Wizzup> tmlind: hm, I think uvos and I see quite a bit more, but I suppose we might have more
                subsystems active
18:04 < Wizzup> I should replace the d4 on this lab psu since it has some hw problem that makes it use
                a lot more power
18:08 < tmlind> Wizzup: hmm i put cpu1 offline when i blank the screen, maybe give that a try?
18:09 < tmlind> echo 0 > /sys/devices/system/cpu/cpu1/online
18:09 < tmlind> that seems to get rid of the constant ipi traffic
18:10 < Wizzup> ok, will try that now
18:10 < tmlind> i also have this looks like:
18:11 < tmlind> echo 600  > /proc/sys/vm/stat_interval
18:11 < Wizzup> my device is in 3g btw, and I think m-l generates quite a bit of debug (and thus mmc)
                traffic on modem signal strength, although we (should be) disabling that upon screen
                off
18:12 < tmlind> ok, will check my power consumption with 3g again at some point, don't want to mess
                with it right now, expecting a package..
18:12 < Wizzup> heh, makes sense
19:07 -!- pere [~pere@84.212.193.114] has joined #maemo-leste
19:19 < Wizzup> tmlind: I think that saves quite a bit (turning one cpu off)
19:20 < Wizzup> will try to properly measure, but from sleep and droid4-pm it looks like it saves
                maybe 200mW?
19:20 < Wizzup> 20mW
19:20 -!- alex1216 [~alex1216@95-28-49-175.broadband.corbina.ru] has joined #maemo-leste
19:35 < tmlind> yeah so it seems

[MerlijnWajer]

20:55 < uvos> we can probubly tune the scheduler to try and not have both cpus active when there is
              little load
20:55 < Wizzup> hmmm ofono uses something else no?
20:55 < uvos> Wizzup: ofono has never worked for me
20:55 < uvos> except by extream massaging
20:55 < uvos> by restarting it often etc
20:55 < uvos> so no idea
20:56 < Wizzup> can try later as well (need to make a list lol)
21:08 < uvos> speaking of powermanagemnt we should maybe import androids interactive scaling govenor
21:08 < uvos> i know its sortof a hack
21:08 < Wizzup> it might just mostly be in mainline
21:10 < uvos> sortof
21:10 < uvos> the mainline version dosent have the pivitol hack :P
21:11 < uvos> the pivitol hack being that it ramps up on ts events
21:11 < uvos> no matter what
21:11 < uvos> this avoids ui latency
21:14 < Wizzup> heh
21:14 < Wizzup> maybe we can turn off a core upon screen blank
21:14 < uvos> sure
21:14 < uvos> i could have mce do that as a hack
21:14 < uvos> but really
21:14 < uvos> the scheduler should be smart enough
21:24 < Wizzup> yeah I am a bit surprised that it wouldn't do that by defaulty
21:24 < uvos> maybe we can use cpuset to prevent kernel threads from running on the second cpu
21:25 -!- _inky [~inky_@46.36.116.208] has quit [Ping timeout: 260 seconds]
21:25 < Wizzup> yeah but that's not ideal imho
21:25 < Wizzup> also it would need to be more than just kernel threads probably
21:26 < uvos> i dont think our userspace dose very mutch
21:26 < uvos> when ideling
21:26 < uvos> but yeah its not ideal
21:27 < uvos> but its somewhere between "sched is smart enougth" and "mce disables all cpus except
              cpu0 when display is off"

[MerlijnWajer]

This is done now with mce and mapphone quirks