Re: [REGRESSION] fan always on with 3.10-rc2

From: Frederic Weisbecker
Date: Thu Jun 13 2013 - 10:00:05 EST


On Sun, Jun 09, 2013 at 08:34:10PM +0200, Martin Steigerwald wrote:
> Am Samstag, 8. Juni 2013, 22:34:44 schrieb Martin Steigerwald:
> > Am Freitag, 24. Mai 2013, 13:03:18 schrieb Martin Steigerwald:
> > > Hi!
> > >
> > > With 3.10-rc2 I see fan always or almost always on, even during extended
> > > periods of basically idling around. I did not notice this with 3.9. This is
> > > on an ThinkPad T520 with Intel Sandybridge i5-2520M dual core with
> > > hyperthreading at regularily 2,5 GhZ and Intel graphics (no nvidia).
> > >
> > > I am using full hz:
> > >
> > > martin@merkaba:~/Linux/Kernel/Mainline/Bugs/fan always on with 3.10.2-rc2> xzgrep NO_HZ config-3.10.0-rc2-tp520.xz
> > > CONFIG_NO_HZ_COMMON=y
> > > # CONFIG_NO_HZ_IDLE is not set
> > > CONFIG_NO_HZ_FULL=y
> > > CONFIG_NO_HZ_FULL_ALL=y
> > > CONFIG_NO_HZ=y
> > > CONFIG_RCU_FAST_NO_HZ=y
> > >
> > > And P-State driver (which I used in 3.9 already as well).
> > >
> > > Kernel config attached as xz. Use xzless or xzcat to display.
> > >
> > >
> > > What puzzles is output of powertop, especially:
> >
> > Still present in 3.10-rc4.
> >
> > I disabled P-State driver but then rpm seems to be even worse.
> >
> > Around 2800 rpm all the time, was about 2650 with Intel P State driver.
> >
> > Next I will try without CONFIG_NO_HZ_FULL and CONFIG_NO_HZ_FULL_ALL.
>
> Okay, without these both the behavior is better. Fan still does not seem
> to get into non running state, which it was able to with earlier kernels
> (or userspaces?), but its down to a quiet 1950 rpm with Intel P-State driver
> enabled again. CC´d to people from thread "power management regression
> in linux 3.10? (related to CONFIG_NO_HZ_FULL)" which I think is related.
>
> Powertop values make a lot more sense as well:
>
> PowerTOP v2.0 Overview Idle stats Frequency stats Device stats Tunables
>
>
> Package | Core | CPU 0 CPU 1
> | | C0 active 1,4% 0,9%
> | | POLL 0,0% 0,0 ms 0,0% 0,0 ms
> | | C1E-SNB 7,7% 3,0 ms 2,9% 5,3 ms
> C2 (pc2) 1,1% | |
> C3 (pc3) 7,2% | C3 (cc3) 8,4% | C3-SNB 8,2% 3,1 ms 3,0% 15,2 ms
> C6 (pc6) 4,9% | C6 (cc6) 9,3% | C6-SNB 7,0% 3,5 ms 6,2% 13,0 ms
> C7 (pc7) 9,9% | C7 (cc7) 64,3% | C7-SNB 73,4% 6,8 ms 84,0% 21,6 ms
>
> | Core | CPU 2 CPU 3
> | | C0 active 1,1% 0,9%
> | | POLL 0,0% 0,0 ms 0,0% 0,0 ms
> | | C1E-SNB 11,0% 6,9 ms 1,4% 4,8 ms
> | |
> | C3 (cc3) 9,3% | C3-SNB 7,7% 5,8 ms 1,9% 6,0 ms
> | C6 (cc6) 10,0% | C6-SNB 8,5% 7,4 ms 4,1% 4,9 ms
> | C7 (cc7) 64,9% | C7-SNB 71,4% 8,2 ms 91,1% 7,1 ms
>
>
>
> PowerTOP v2.0 Overview Idle stats Frequency stats Device stats Tunables
>
>
> Package | Core | CPU 0 CPU 1
> | | Actual 2,0 GHz 2,4 GHz
> Idle 54,4% | Idle 97,4% | Idle 98,9% 98,2%
> 2,90 GHz 0,6% | 2,90 GHz 0,1% | 2,90 GHz 0,1% 0,1%
> 2,81 GHz 2,6% | 3,10 GHz 0,2% | 3,10 GHz 0,1% 0,1%
> 3,10 GHz 1,2% | 2,10 GHz 0,1% | 2,10 GHz 0,0% 0,0%
> 3,21 GHz 7,3% | 3,21 GHz 0,9% | 3,21 GHz 0,1% 0,9%
> 2,10 GHz 2,3% | 2,81 GHz 0,2% | 2,81 GHz 0,1% 0,1%
> 3,00 GHz 3,5% | 3,00 GHz 0,0% | 3,00 GHz 0,0% 0,0%
> 2,40 GHz 3,5% | 2,40 GHz 0,1% | 2,40 GHz 0,1% 0,1%
> 1,91 GHz 1,0% | 1,91 GHz 0,1% | 1,91 GHz 0,0% 0,0%
> 2,21 GHz 0,6% | 2,21 GHz 0,1% | 2,21 GHz 0,1% 0,1%
>
> | Core | CPU 2 CPU 3
> | | Actual 2,4 GHz 2,5 GHz
> | Idle 56,3% | Idle 98,4% 57,7%
> | 2,90 GHz 0,5% | 2,90 GHz 0,1% 0,9%
> | 2,81 GHz 2,5% | 2,81 GHz 0,1% 6,4%
> | 3,10 GHz 1,1% | 3,10 GHz 0,2% 2,2%
> | 3,21 GHz 6,7% | 3,00 GHz 0,0% 2,4%
> | 3,00 GHz 3,5% | 3,21 GHz 0,5% 3,4%
> | 2,71 GHz 2,4% | 2,50 GHz 0,1% 3,3%
> | 2,50 GHz 1,1% | 2,31 GHz 0,1% 0,4%
> | 2,31 GHz 0,2% | 2,71 GHz 0,1% 1,0%
> | 2,60 GHz 1,0% | 2,60 GHz 0,1% 2,7%
>
>
> If I haven´t stated it. This is on a ThinkPad T520 with dual core
> Sandybridge i5-2620M at 2.5 GHz (overclockable to 3.2 GhZ), now with
> Kernel 3.10-rc5. I didn´t test this one with the full hz stuff enabled, only
> rc2 and rc4.
>
> > According to powertop CPU 0 is never idle

Note that CONFIG_NO_HZ_FULL is very power unfriendly because it forces CPU 0
to keep its periodic tick to handle timekeeping, even when idle. Paul McKenney is
working on a solution to solve this.

Thanks.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/