Kernel 3.6.2 Power Regressions (Sandy Bridge)

hendry · 2013-02-11 14:40:11

Thanks for confirming the issue degeberg. I git cloned the kernel sources and have a look between `git log v3.5.6..v3.5.7 --oneline`, I can't see anything obvious.

The problem for me, doing a git bisect, is that building the kernel makes my machine overheat, and it takes WAY too long. I wonder if there are "generic" packaged builds of every commit of the kernel for bi-secting? Or some way to speed up compilation?

Yours feverishly,

the sad clown · 2013-02-11 15:31:56

Strangely, I'm no longer having any issues with power since 3.7.6, nor does my GPU hang after a suspend. Just checked powertop, my computer was idling at 9W on battery, which is good for me (i7-2630QM).

sensors
acpitz-virtual-0
Adapter: Virtual device
temp1:        +35.0°C  (crit = +99.0°C)

coretemp-isa-0000
Adapter: ISA adapter
Physical id 0:  +39.0°C  (high = +86.0°C, crit = +100.0°C)
Core 0:         +34.0°C  (high = +86.0°C, crit = +100.0°C)
Core 1:         +38.0°C  (high = +86.0°C, crit = +100.0°C)
Core 2:         +32.0°C  (high = +86.0°C, crit = +100.0°C)
Core 3:         +39.0°C  (high = +86.0°C, crit = +100.0°C)

KaiSforza · 2013-02-12 07:27:01

hendry wrote:

Thanks for confirming the issue degeberg. I git cloned the kernel sources and have a look between `git log v3.5.6..v3.5.7 --oneline`, I can't see anything obvious.
The problem for me, doing a git bisect, is that building the kernel makes my machine overheat, and it takes WAY too long. I wonder if there are "generic" packaged builds of every commit of the kernel for bi-secting? Or some way to speed up compilation?
Yours feverishly,

Gonna assume you meant `git log v3.7.6..3.7.7 --oneline`.

You could try changing the MAKEFLAGS variable in makepkg.conf (make it -jN, where N is the number of threads your CPU has plus 1). Also, install ccache and edit makepkg.conf to use it.

You could also compile outside. Keeps your computer nice and cool in the winter.

The Compiler · 2013-02-12 09:33:56

hendry wrote:

The problem for me, doing a git bisect, is that building the kernel makes my machine overheat, and it takes WAY too long. I wonder if there are "generic" packaged builds of every commit of the kernel for bi-secting? Or some way to speed up compilation?

Take some time to get your custom kernel config and don't compile all the stuff you don't need. There are things like USB-drivers for Lego light towers there...
For me, that got the time down from about 45 minutes to about 7.

Florian

neo007 · 2013-02-12 17:37:03

I've got two affected laptops:

Lenovo x220, Intel(R) Core(TM) i7-2640M CPU @ 2.80GHz, linux-3.7.6-1-x86_64, high temperatures and power consumption
Lenovo t420s, Intel(R) Core(TM) i5-2540M CPU @ 2.60GHz, linux-3.7.6-1-x86_64, high temperatures and power consumtion

What's more: sometimes x220 was ok and sometimes when I run it it started heating terribly even though it reported doing nothing. If I can help anyhow with any additional info - just tell.

hendry · 2013-02-13 03:11:28

The Compiler wrote:

Take some time to get your custom kernel config and don't compile all the stuff you don't need. There are things like USB-drivers for Lego light towers there...
For me, that got the time down from about 45 minutes to about 7.

Then based on past experience, I will spend DAYS figuring out a cruft-free Linux config.

Managed to downgrade to virtualbox-host-modules-4.2.6-5-x86_64.pkg.tar.xz & linux-3.7.5-1-x86_64.pkg.tar.xz which don't overheat. Yay.

Anyone else enjoying wifi drop outs on the Linux stable series of kernels? https://bugs.archlinux.org/task/33503

donniezazen · 2013-02-13 03:57:03

@hendry How are you logging your temperature data?

hendry · 2013-02-13 04:01:54

donniezazen wrote:

@hendry How are you logging your temperature data?

https://github.com/kaihendry/sg and see https://github.com/kaihendry/sg/blob/master/g/temp.sh for the plot generator.

Happy to help you use it :-)

donniezazen · 2013-02-13 08:25:20

@hendry Thanks for replying. I will be installing it on my productive laptop. Does it really need all that /var/sg or www-data stuff.

@KaiSforza I just installed linux-mainline from your repo but it didn't get me any fallback mode. Am I missing something?

hendry · 2013-02-13 08:27:51

donniezazen wrote:

@hendry Thanks for replying. I will be installing it on my productive laptop. Does it really need all that /var/sg or www-data stuff.

It needs to store the values it collects somewhere, so yes I'm assuming /var/sg

You could collect the CSV yourself and just run temp.sh upon it. Anyway the source (I like to think) is super simple. https://github.com/kaihendry/sg/blob/master/sg-client

Hspasta · 2013-02-13 17:05:10

donniezazen wrote:

@KaiSforza I just installed linux-mainline from your repo but it didn't get me any fallback mode. Am I missing something?

Look under /etc/mkinitcpio.d/linux-mainline.preset

For those still having power issues, try using uxa instead of sna in your xorg file.

donniezazen · 2013-02-13 17:19:43

@Hspasta Is that a fix? I don't have an Xorg file.

Nanthiel · 2013-02-15 13:05:08

Is it possible to get Bumblebee working with nouveau or the proprietary driver on the 3.8 kernel? My quick attempts failed.

eruditorum · 2013-02-15 16:23:03

It seems to me everything is fine with Intel GPU on new kernel (I use 3.7.7).
Now RC6 state is enaled by default and doesn't need to be entered on the kernel command line.
Also, suspend hangs have gone.

My system's power consumption is ~5.5W in idle state.

Thanks.

replax · 2013-02-15 16:50:52

eruditorum wrote:

My system's power consumption is ~5.5W in idle state..

Woha, that's low! What system are you using if you mind sharing?
If on an x220 or somewhat similar hardware - mind sharing your configs somewhere ?
Thanks!!

eruditorum · 2013-02-15 17:07:10

https://bbs.archlinux.org/viewtopic.php?id=153959

+

/etc/modprobe.d/powersave.conf:

options snd_hda_intel power_save=1

/etc/sysctl.d/powersave.conf:

vm.laptop_mode = 5
vm.dirty_writeback_centisecs = 1500

nmi_watchdog=0 in kernel command line

Laptop runs about 6-7 hours on battery (5hrs on Windows 7).

Last edited by eruditorum (2013-02-17 13:30:19)

ANOKNUSA · 2013-02-15 20:28:22

neo007 wrote:

I've got two affected laptops:
Lenovo x220, Intel(R) Core(TM) i7-2640M CPU @ 2.80GHz, linux-3.7.6-1-x86_64, high temperatures and power consumption
Lenovo t420s, Intel(R) Core(TM) i5-2540M CPU @ 2.60GHz, linux-3.7.6-1-x86_64, high temperatures and power consumtion
What's more: sometimes x220 was ok and sometimes when I run it it started heating terribly even though it reported doing nothing. If I can help anyhow with any additional info - just tell.

I'm in the same situation: Despite managing to avoid this issue for a while, my Lenovo T520 now experiences higher temps and a greatly shortened battery life (~30% of the norm). This seems to come and go; sometimes after resuming from suspend, normal CPU behavior resumes as well, but more often it stays hot.

This also seems to depend on whether AC is connected or not, but I haven't done any real test.

chx · 2013-02-15 21:37:03

Note to those that stayed on 3.5.6-1 like me: keep your xf86-video-intel at 2.20.19-1 the 2.21 driver seems to crash with this kernel. More https://bbs.archlinux.org/viewtopic.php?id=158112 here.

Mindstormscreator · 2013-02-17 12:45:51

chx wrote:

Note to those that stayed on 3.5.6-1 like me: keep your xf86-video-intel at 2.20.19-1 the 2.21 driver seems to crash with this kernel. More https://bbs.archlinux.org/viewtopic.php?id=158112 here.

Ah, thank you! I was pretty annoyed that my Linux computer was suddenly emulating a pesky Windows behavior...
By the way, are you still sticking with 3.5.6, or do you think the general consensus is to move to 3.7.8?

Nanthiel · 2013-02-17 13:42:29

3.7.8 fixed the power issues for me too.

Mindstormscreator · 2013-02-18 03:07:22

Kernel 3.7.8 alone does NOT fix this for me. I know that the kernel command line arguments could help, but really, shouldn't I be getting the same temperatures that I got with 3.5.6 without any added configuration? From my perspective, this bug isn't fixed...

jarda-wien · 2013-02-18 09:56:13

Kernel 3.7.8 doesn't fix this for me either. The problem manifests itself in i7z with high CPU frequencies even during idle.
Sometimes suspend helps, sometimes not.

Now running 3.7.9 and still hot.

illtech · 2013-02-18 23:21:25

Just installed 3.7.9-1 last night and rebooted this morning; GPU is now constantly under load (showing 100% active under powertop). Rebooting seems to make no difference for me - even if I reboot to a console login (no X session up yet).

Running on a SandyBridge Thinkpad X220.

Let me know if there is any useful information I may be able to provide.

edit: Just rebooted in to windows and back; and now the GPU is gone back to idling (RC6).

Last edited by illtech (2013-02-18 23:33:25)

jkluebs · 2013-02-18 23:38:30

Running 3.7.8-1-ARCH on a Lenovo T420 i5 power consumption seems to get as good (~9W idle) 3.5.6 sometimes. Don't have good numbers but seems like 30-40% of resumes from suspend will come up in a high (23W at idle, nvidia graphics off) power state. It seems that every cold boot will have excessive power draw until a suspend/resume is performed.

This problem is definitely not fixed in 3.7.8-1.

the_joker · 2013-02-19 11:55:50

Hi Folks,

same for me .. the issue is apearing / dispearing etc .. with each update ... almost ZERO consistency.

I would like to make the following proposal. For all those that are reporting a solution / fix / NON fix / issue etc ... can we agree on some minimal information provided. ? There is to much information to my humble opinion floating around and Im getting seriously distracted by this BUG.

1) please show if you are suing any BOOT parameters, like GRUB_CMDLINE_LINUX="pcie_aspm=force i915.i915_enable_rc6=1 i915.lvds_downclock=1"
2) do you load the i915 module /etc/mkinitcpio.conf ...
3) any other parameter you have tunned that might be of interest.

Right now im running 3.7.9-1-ARCH with i915 loaded and the commandline above.

without loading + without command line in grub i get 75 - 80 degrees right from the beginning and the "famous" GPU hang on 100% busy no RC6

with the above changes, ( and i have turned them off for testing and a couple of reboots) im at least

acpitz-virtual-0
Adapter: Virtual device
temp1: +47.0°C (crit = +99.0°C)

thinkpad-isa-0000
Adapter: ISA adapter
fan1: 2973 RPM

coretemp-isa-0000
Adapter: ISA adapter
Physical id 0: +50.0°C (high = +86.0°C, crit = +100.0°C)
Core 0: +47.0°C (high = +86.0°C, crit = +100.0°C)
Core 1: +50.0°C (high = +86.0°C, crit = +100.0°C)

... im really getting frustrated at this bug. and for over 2 cycles no real solution ... partially its working ....

I hope 3.8 brings a solution ... i cant upgrade right now .. to an unofficial mainline repo ...

thanks

the Joker

Arch Linux

#351 2013-02-11 14:40:11

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#352 2013-02-11 15:31:56

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#353 2013-02-12 07:27:01

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#354 2013-02-12 09:33:56

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#355 2013-02-12 17:37:03

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#356 2013-02-13 03:11:28

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#357 2013-02-13 03:57:03

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#358 2013-02-13 04:01:54

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#359 2013-02-13 08:25:20

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#360 2013-02-13 08:27:51

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#361 2013-02-13 17:05:10

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#362 2013-02-13 17:19:43

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#363 2013-02-15 13:05:08

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#364 2013-02-15 16:23:03

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#365 2013-02-15 16:50:52

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#366 2013-02-15 17:07:10

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#367 2013-02-15 20:28:22

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#368 2013-02-15 21:37:03

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#369 2013-02-17 12:45:51

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#370 2013-02-17 13:42:29

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#371 2013-02-18 03:07:22

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#372 2013-02-18 09:56:13

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#373 2013-02-18 23:21:25

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#374 2013-02-18 23:38:30

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#375 2013-02-19 11:55:50

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

Board footer