Kernel 3.6.2 Power Regressions (Sandy Bridge)

Hspasta · 2012-10-15 20:37:26

After a recent kernel update, my laptop runs ~20C hotter and uses nearly twice as much power.

Boot parameters;

ro quiet i915.i915_enable_rc6=1 i915.i915_enable_fbc=1 i915.lvds_downclock=1 i915.semaphores=1

I see a bunch of these messages in my log:

Oct 15 16:30:06 HSP-ARCH kernel: CPU0: Package power limit notification (total events = 159)
Oct 15 16:30:06 HSP-ARCH kernel: CPU3: Package power limit notification (total events = 159)
Oct 15 16:30:06 HSP-ARCH kernel: CPU2: Package power limit notification (total events = 159)
Oct 15 16:30:06 HSP-ARCH kernel: CPU1: Package power limit notification (total events = 159)
Oct 15 16:30:06 HSP-ARCH kernel: CPU0: Package power limit normal
Oct 15 16:30:06 HSP-ARCH kernel: CPU3: Package power limit normal
Oct 15 16:30:06 HSP-ARCH kernel: CPU1: Package power limit normal
Oct 15 16:30:06 HSP-ARCH kernel: CPU2: Package power limit normal

I believe this happened after a resume from a suspend.

KaiSforza · 2012-10-16 03:12:14

All,

I am getting the same issues. Originally, I had essentially the same thing for boot parameters:

 ro init=/usr/lib/systemd/systemd quiet libahci.ignore_sss=1 i915.i915_enable_fbc=1 i915.i915_enable_rc6=1 i915.lvds_downclock=1

I removed all of the i915.*'s and that seemed to fix it, for a time. I am now getting much higher power usage (3.5.x: 7-10W, 3.6-2: 20-24W) and almost half the battery life, and upon a search of my journal, I got the same log mesages:

Oct 15 21:39:04 WST420 kernel: CPU3: Package power limit notification (total events = 1)
Oct 15 21:39:04 WST420 kernel: CPU1: Package power limit notification (total events = 1)
Oct 15 21:39:04 WST420 kernel: CPU0: Package power limit notification (total events = 1)
Oct 15 21:39:04 WST420 kernel: CPU2: Package power limit notification (total events = 1)
Oct 15 21:39:04 WST420 kernel: CPU0: Package power limit normal
Oct 15 21:39:04 WST420 kernel: CPU1: Package power limit normal
Oct 15 21:39:04 WST420 kernel: CPU2: Package power limit normal
Oct 15 21:39:04 WST420 kernel: CPU3: Package power limit normal

I think we should do something similar to the 3.3.5+ power regression thread. (see below)

Please report your status in the following form:

Affected?, Laptop brand, CPU-vendor, CPU-name, GPU-name (+kernel driver), chipset, current kernel version

Power usage statistics (wattage before -> after) would also be nice.

INFORMATION
BUG REPORTS
Arch FS#32025
Kernel Bug 48721
FreeDesktop.org Bug 54089 (Includes possible patch)
POSSIBLE FIXES
Has 3.8-rc4 fixed it? Test people, test!
Some precompiled mainline kernels
FreeDesktop.org Bug 54089
Kernel arguments to rc6=7
AFFECTED:
All kernels >= 3.6.2 and < 3.8 on Sandy (and rarely, Ivy) Bridge processors.
Lenovo X220, Intel i7 Sandy Bridge, Intel 3000 HD, Linux-3.6.2
Lenovo X220, Intel i5, Intel 3000 HD, Linux 3.6.2
Lenovo T520 with Intel i7-2640M (2.8Ghz), Intel 3000 GFX, Linux 3.6.2
Acer Timeline X, Intel i5-2450M, Linux 3.6.2
Asus UX32VD, Intel core i7-3517U Ivy Bridge(HM76 chipset?), intel HD4000(i915) + nvidia gt620m(optimus) nvidia+bbswitch for bumblebee, kernel 3.6.2-1-ARCH
HP Probook 4320s, Intel core i3 2.4ghz, ATI radeon 5470, linux 3.6.2-1
Dell Inspiron N5110, Intel, Intel Core i7-2670QM CPU, Intel/NVIDIA GeForce GT 540M (switchable), 3.6.2-1-ARCH
NOT Affected:
None consistently. See above.
Note: If I've assumed something wrong about your laptop, please notify me by email or on the board. (I tried to fill in as many blanks as I could, notably the graphics chips)

Last edited by KaiSforza (2013-01-18 06:53:35)

WonderWoofy · 2012-10-16 04:06:38

You know all those i915 parameters you two are using? Check out

# systool -m i915 -av

I set all those things at first, because I wasn't sure what the setting of -1 meant in those cases. Apparently those things are defaulting to being on. For instance, the i915_enable_rc6 can be set to 0, 1, 3, or 7. 1 will enable rc6, 3 will enable rc6p also, and 7 will enable rc6pp in addition to the other two. But without setting it to 1 on my own, it actually sits in rc6p most of the time (which is a lower power state than rc6, and rc6pp is even lower).

So I think that all the rc6 setting will actually set it to use ore power than the default. Frame buffer compression appears to be on by default, as does semaphores. The only one I think you are having a real effect with is the lvds_downclock setting.

I am not 100% on this. But I noticed my Ivy Bridge machine actually uses less power when I don't set all that crap, though I have not tried the lvds_downclock parameter.

ixnine · 2012-10-16 09:05:05

I have same boot parameters as OP

$ cat /sys/class/thermal/thermal_zone{0,1,2,3,4}/temp
43000
32900
50000
50000
43000

Is this normal temps? Laptop is idle

Last edited by ixnine (2012-10-16 09:05:29)

trsk · 2012-10-16 10:34:20

Lenovo X220, i7 Sandy Bridge:
before linux-3.6.2: 7 W idle
after linux-3.6.2: 10 W idle

I'm using the same kernel options. I'm also noticing higher CPU temperatures since 3.6.2.

EDIT: This is independent of whether I've just resumed or not.

EDIT2: Additionally, 70% of the time, after resuming from suspend, the CPU doesn't clock down, which leades to a power consumption of 25 W in idle. I've downgraded to linux-3.5...

Last edited by trsk (2012-10-17 09:21:15)

Gusar · 2012-10-16 10:37:43

As WonderWoofy mentioned, rc6=1 only activates part of the whole thing, rc6=7 activates everything. Though you shouldn't need to play with this stuff anymore, the kernel now handles it automatically.

jrk · 2012-10-16 10:38:57

Hi,

also affected: Thinkpad X220, Core-i5.

I've also written something here.
This sounds related too.

DIDI2002 · 2012-10-16 10:57:54

Hi,

Acer Timeline X (i5-2450M).
Before: 9W idle
Now: 29W idle
powertop shows 100% active GPU, dmesg says:

[    3.730181] [drm] Enabling RC6 states: RC6 on, RC6p on, RC6pp on

3.7 rc1 is not better, 3.5 is fine

dontbugme · 2012-10-16 13:10:07

Wow. 29W vs. 9W is HUUUGE. Thought my Ivy Bridge (Asus UX32VD) Laptop is affected by a massive power regression. Just compared Kernel 3.5.6-1 with 3.6.2-1. 3.5.6-1: Average: 7.8W; 3.6.2-1: somewhere between 12 and 13 (sorry I need to check again for more precise data, but the fact it's beyond 12 should be enogh .. <60% of the consumption of 3.6.2-1 on kernel 3.5.6-1).
Details:
Asus UX32VD, Intel core i7-3517U Ivy Bridge(HM76 chipset?), intel HD4000(i915) + nvidia gt620m(optimus) nvidia+bbswitch for bumblebee, kernel 3.6.2-1-ARCH.

donniezazen · 2012-10-16 14:42:25

Fresh installation here. No boot parameters or modifications. Power consumption is 20-30W and Temperature hangs between 65-90C.

Lenovo Thinkpad T420i, i7-2620M Dual-Core Sandy-Bridge, Intel 3000 HD / Nvidia NVS4200 Graphic Cards, Linux-3.6.2

Last edited by donniezazen (2012-10-16 15:50:02)

ixnine · 2012-10-16 15:03:35

Sorry to ask but how do you check the power consumption?

WonderWoofy · 2012-10-16 15:04:35

powertop should tell you that kind of stuff.

eduedix · 2012-10-16 15:16:51

lol what can be the problem here? im on x220 i7 and it is hot!

Hspasta · 2012-10-16 15:19:05

I don't know if I was clear enough in the OP. I only get this power regression after I resume from a suspend.

Are people having this issue just by default? I found this on the web. Same issue I believe.

l3nkz · 2012-10-16 15:50:28

Well it's the same issue for me.

After resuming from suspend my power consumption goes up to 25 W (around 12W before suspend).

I have a Lenovo Thinkpad T420s with a core i5 (i5-2410).

Using the lts kernel fixes the problem, but produces some other ones. So this isn't really an alternative for me.
I recognized previously when switching from 3.3 to 3.4 that the power consumption raised slightly (from 9W to ~12W).
Which was already a little bit annoying but this seems to be nothing compared to the latest situation.

Im using the following kernel parameters:
i915.i915_enable_rc6=7 i915.i915_enable_fbc=1 i915.lvds_downclock=1 drm.vblankoffdelay=1

Kagarrache · 2012-10-16 15:52:55

Hspasta wrote:

Are people having this issue just by default? I found this on the web. Same issue I believe.

Yeap, by default, at least here: Asus K53SV, i7-2670QM, Intel HD 3000 & Nvidia GT540M, Linux-(stock, ck or Bede)-3.6.2.

raimund · 2012-10-16 20:19:39

Hi,

I have the same problem, also on Thinkpad T420.
It's about the CPU Frequency scaling, which is not working correctly anymore (though it *seems* like it is).
Just compare the output from

cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq

with

cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq

.
The first one is "...the frequency the kernel thinks the CPU runs
at.", the second one " ...is the frequency
the CPU actually runs at." From http://www.kernel.org/doc/Documentation … -guide.txt

See https://bugzilla.kernel.org/show_bug.cgi?id=48721

pYrO1v1aniac · 2012-10-16 20:37:05

Same here. Fan spinning constantly, graphics driver also appears to have broken xf86-video-ati, gnome-shell now using 109% CPU.

Intel core i3 2.4ghz
ATI radeon 5470
HP Probook 4320s

donniezazen · 2012-10-16 21:09:43

I have power problems on default system; no relation to suspend. Both CPU and Memory are both minimal.

[donnie@arch ~]$ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq
800000
[donnie@arch ~]$ sudo cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq
2700000

KaiSforza · 2012-10-16 22:29:01

donniezazen wrote:

I have power problems on default system; no relation to suspend. Both CPU and Memory are both minimal.
[donnie@arch ~]$ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq
800000
[donnie@arch ~]$ sudo cat /sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_cur_freq
2700000
http://i.imgur.com/NwQ9zs.png http://i.imgur.com/KxQ5bs.png

I am getting almost the same reports (I've only got an i5, so 2.5, not 2.7ghz), and it seems that running things with sudo actually prints the correct frequency values.

Here is a screenshot of my system running powertop, htop, and showing the output of running cpupower as root and as a normal user. I'm going to take a wild guess that this is something to do with the kernel not respecting our frequency settings. All other power 'things' seem to be normal for me.

DIDI2002 · 2012-10-17 00:00:34

Seems to be 2 bugs (at least for me): CPU not clocking down and GPU not entering rc6 although rc6 is activated

KaiSforza · 2012-10-17 00:37:36

This is seeming to be somewhat random. I booted my laptop up with i915_enable_rc6=7, and that was working fine, but cpu scaling was still not happening.

Then I suspended my laptop to check if the results would be the same. Oddly enough, cpu scaling began working again.

There seems to be active work being done on kernel bug 48721, so hopefully something will be found.

Edit:
Just went through another sleep cycle, same results. All I changed was the i915_enable_rc6=7 and it seems to be using reasonable amounts of power so far.

Last edited by KaiSforza (2012-10-17 01:12:53)

donniezazen · 2012-10-17 02:15:11

@Kaisforza I messed around installed and removed linux-lts. Surprising, CPU Frequency problem is gone. Both reading by aforementioned commands are now same. I tried to suspend and resume. Power consumption is back to normal.

System temperature is now between 55-60C which should have been around 50C.

heavynetol · 2012-10-17 04:28:53

Affected=Yes
Dell Inspiron N5110
Intel, Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz
Intel Corporation 2nd Generation Core Processor Family Integrated Graphics Controller/NVIDIA Corporation GF108 [GeForce GT 540M]
3.6.2-1-ARCH

Current Idle Temp after update: 62.5ºC
Past Idle Temp before update: 45ºC-50ºC

ayr0 · 2012-10-17 06:56:16

I'm also affected

Thinkpad T520 with Intel i7-2640M (2.8Ghz)
Intel 3000 GFX

linux-3.5.6: ~10-12W
linux-3.6.2: ~16-18W

I have downgraded until a solution is found.

Arch Linux

#1 2012-10-15 20:37:26

Kernel 3.6.2 Power Regressions (Sandy Bridge)

#2 2012-10-16 03:12:14

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

INFORMATION

#3 2012-10-16 04:06:38

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#4 2012-10-16 09:05:05

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#5 2012-10-16 10:34:20

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#6 2012-10-16 10:37:43

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#7 2012-10-16 10:38:57

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#8 2012-10-16 10:57:54

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#9 2012-10-16 13:10:07

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#10 2012-10-16 14:42:25

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#11 2012-10-16 15:03:35

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#12 2012-10-16 15:04:35

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#13 2012-10-16 15:16:51

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#14 2012-10-16 15:19:05

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#15 2012-10-16 15:50:28

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#16 2012-10-16 15:52:55

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#17 2012-10-16 20:19:39

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#18 2012-10-16 20:37:05

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#19 2012-10-16 21:09:43

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#20 2012-10-16 22:29:01

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#21 2012-10-17 00:00:34

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#22 2012-10-17 00:37:36

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#23 2012-10-17 02:15:11

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#24 2012-10-17 04:28:53

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#25 2012-10-17 06:56:16

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

Board footer