Kernel 3.6.2 Power Regressions (Sandy Bridge)

LordChaos73 · 2013-01-12 12:55:05

I came to the conclusion my laptop suffers from two particular issues:

1. After a suspend/resume cycle, the discrete card (Radeon HD6470M) is powered on again, although /sys/kernel/debug/vgaswitcheroo/switch still reports the contrary. This explains why the laptop is using ~5W more after a suspend/resume cycle. I can blacklist the radeon module, but then the discrete card is powered on when the laptop is started. Unfortunately, blacklisting the module and manually loading the module after boot, then switching the card off and removing the module results in a kernel panic.
The only way to prevent the card from being powered on after a suspend/resume cycle is to make sure the radeon module is not in user before suspending the laptop.

2. After a suspend/resume cycle, the GPU (not sure which one is used) is continuously active which results in a greater discharge rate than before the cycle.

hendry · 2013-01-15 12:44:13

I plotted my system temperatures here: http://stackoverflow.com/questions/14314895/
https://github.com/kaihendry/laptemp is the source where I'm basically learning GNUplot.

donniezazen · 2013-01-15 16:46:27

This is amazing this bug has persisted a whole kernel cycle without being fixed. It is really annoying.

UPDATE:- Corrected typo. without

Last edited by donniezazen (2013-01-17 02:17:56)

hendry · 2013-01-16 02:07:00

It doesn't help when power states aren't accessible sanely through the /proc or /sys or some file. One has to use something bonkers like: `sudo powertop --csv` and figure out if you're in the "good" power saving state... C6.or RC6 or something else cryptic.

NIGHT MARE.

http://s.natalian.org/2013-01-16/135830 … 66x768.png
http://s.natalian.org/2013-01-16/135830 … 66x768.png

the_joker · 2013-01-16 07:21:10

donniezazen wrote:

This is amazing this bug has persisted a whole kernel cycle with being fixed. It is really annoying.

I really second that comment. ....

this is amazing ... this things keeps popping up ... and popping up ..

I simply cant believe that there is nobody out there with a fix! ... I mean a real fix. 3.8 rc2 seems at least to work stable but im really shuttered in my confidence and it will take a long time before i get this one back.

And this is a opinion to the folks at the LKM ... not to all the hard working folks in this thread who provide ideas and comments and scripts and rep's etc. this is marvelous ..

but this bug is SOOO annoying ...

dontbugme · 2013-01-16 11:07:58

donniezazen wrote:

This is amazing this bug has persisted a whole kernel cycle with being fixed. It is really annoying.

Sorry for beeing slightly off topic but, wait ..what?
Does not compute. Are you just beeing sarcastic or is there some subtle semantics in the English language I don't understand(not my mother tongue). Do you mean there is a "supposed-to-be" fix that just doesn't work or what?
Also, what I'm wondering is: as of my understanding, the commit causing the regression has been successfully found/bisected/w/e. Why not just rolling it back - I mean upstream, not as a patch in the repos - until the conflicting commit can be fixed (idk atm what the commit initially was supposed to do).

donniezazen · 2013-01-16 15:23:18

@dontbugme I was just being frustrated. This bug was first noticed in 3.6.x and now we are in 3.8.x and still not sure if its fixed.

Mindstormscreator · 2013-01-16 20:41:54

dontbugme wrote:

Are you just beeing sarcastic or is there some subtle semantics in the English language I don't understand(not my mother tongue). Do you mean there is a "supposed-to-be" fix that just doesn't work or what?

I think there is a typo:

donniezazen wrote:

This is amazing this bug has persisted a whole kernel cycle without being fixed. It is really annoying.

donniezazen · 2013-01-16 21:45:24

@Mindstormscreator Thanks. Fixed.

cfr · 2013-01-17 01:17:48

Please indicate you've edited the original post. As it is the responses to it make no sense at all at the moment!

dontbugme · 2013-01-17 09:36:03

donniezazen wrote:

@Mindstormscreator Thanks. Fixed.

Oh, lol. That's why your post didn't made sense to me . The obvious explanation was the right one.

KaiSforza · 2013-01-17 10:05:47

Hello, all,
Sorry for my lack of updating the linux-mainline repository. I'll be updating when rc4 rolls around with a nice rebased config file. Hope this can fix even more bugs and keep our [gc]pu scaling working.

Hspasta · 2013-01-17 15:49:27

Hspasta wrote:

Just discovered on linux 3.7.1-2 that the overheating only happens when I suspend with the AC attached.

Well that was actually false. It seemed to work most of the time, but not always.

Currently with 3.7.2-1, I still have this power regression.
With 3.8rc3, the problem appears to be gone, but my GPU hangs when I try to watch/do anything that uses it (i.e mplayer, flash, fancy websites, etc).

curson · 2013-01-17 15:52:49

https://bugs.archlinux.org/task/32025#comment104514 <-- this seems to suggest that 3.8 fixes this. I'm still holding on to 3.6.2-1 for now, and encountering this randomly at boot from time to time, but it's manageable. Hopefully the fix will be confirmed and we'll get it as soon as the 3.8 kernel is released and it hits stable.

655321 · 2013-01-17 21:31:57

Just for testing, I tried the linux-ck kernel from AUR and I don't get this problem anymore (In fact I get GOOD battery life). https://aur.archlinux.org/packages/linux-ck/

uname -a
Linux thinkparch 3.7.2-2-ck #1 SMP PREEMPT Thu Jan 17 01:17:30 CST 2013 x86_64 GNU/Linux

With the default Arch kernel my battery wont even last an hour. horrible!

hendry · 2013-01-18 02:50:41

The relevant bugs I think are here:

https://bugzilla.kernel.org/show_bug.cgi?id=48721
https://bugzilla.kernel.org/show_bug.cgi?id=48791

Notice:
sudo cat /sys/kernel/debug/dri/0/i915_cur_delayinfo | grep CAGF

I wonder which forum Mattias is referring to? https://bugzilla.kernel.org/show_bug.cgi?id=48721#c30

KaiSforza · 2013-01-18 07:17:42

Just built the 3.8-rc4 kernel, headers and docs and uploaded them for those that want to try rc4 out: MAINLINE REPOSITORY (all files signed)

Last edited by KaiSforza (2013-01-18 07:18:16)

dr_undecided · 2013-01-18 19:29:22

My rc6 mode is not enabled automatically on every boot. Sometimes when I do sudo powertop it shows 100% active - and my laptop heats up to 95C. I have to reboot. Other times active is less than 100% and RC6 is some positive number. In these cases everything works well. The rc6=1 flag does not do anything.
Is this all part of the same problem discussed here?

curson · 2013-01-18 19:58:27

I believe it is, dr_undecided. My laptop shows the same exact symptoms you're describing.

Yurlungur · 2013-01-19 23:40:45

KaiSforza wrote:

Just built the 3.8-rc4 kernel, headers and docs and uploaded them for those that want to try rc4 out: MAINLINE REPOSITORY (all files signed)

Been struggling with this problem for a while now, and I just tried your mainline kernel. I haven't ran it long, but now problems for the first two reboots. CPU temperature now 53C instead of 80C. Thanks!

Nanthiel · 2013-01-20 09:51:10

I had this problem on kernel 3.6.11-1 aswell (ASUS K53SV). I tried installing linux-ck from AUR and it seems to have gone, but I'll have to test a bit longer.

I also tried installing 3.8 from miffe's repositories but my wireless didn't work after booting into it. I'm not experienced in kernel configuring so I don't know whether the module (ath9k) wasn't there or didn't load for some reason. I also need Nvidia driver support (+ bumblebee and bbswitch) and don't know if it works for 3.8. The Nvidia driver is separate for the -ck version in AUR.

donniezazen · 2013-01-21 03:06:27

@Nanthiel I have an Optimus laptop. I don't think Nvidia drivers will work with latest kernel release (3.8rcX). Better stick to stock or linux-ck packages.

https://aur.archlinux.org/packages/nvidia-bumblebee/

linux>=3.6
linux<3.8
nvidia-utils-bumblebee>310

Rukiri · 2013-01-21 03:21:39

LordChaos73 wrote:

I came to the conclusion my laptop suffers from two particular issues:
1. After a suspend/resume cycle, the discrete card (Radeon HD6470M) is powered on again, although /sys/kernel/debug/vgaswitcheroo/switch still reports the contrary. This explains why the laptop is using ~5W more after a suspend/resume cycle. I can blacklist the radeon module, but then the discrete card is powered on when the laptop is started. Unfortunately, blacklisting the module and manually loading the module after boot, then switching the card off and removing the module results in a kernel panic.
The only way to prevent the card from being powered on after a suspend/resume cycle is to make sure the radeon module is not in user before suspending the laptop.
2. After a suspend/resume cycle, the GPU (not sure which one is used) is continuously active which results in a greater discharge rate than before the cycle.

You may want to give 3.7 a try as it's the next stable kernel release, I also ahve a sandy bridge-E processor (going with a xeon 8-core in a few months with a new motherboard)

Are you using an i3, i5, or an i7? I wouldn't worry too much about it unless it's say 10-20 watts more but 5 isn't that much.
But, have you tried powertop? It'll certainly give you an estimate of what your using.

I can also comment on 3.8, would love to use it may just wait until a final 3.8 release which isn't that far away as I've seen it go from rc1-rc4 rather quickly. But until nvidia is fixed I'm stuck on 3.7.3.1.

Last edited by Rukiri (2013-01-21 03:23:16)

Nanthiel · 2013-01-21 08:15:05

donniezazen wrote:

@Nanthiel I have an Optimus laptop. I don't think Nvidia drivers will work with latest kernel release (3.8rcX). Better stick to stock or linux-ck packages.

Well, to continue on 3.7 observations. Battery usage is better (by about 20 minutes), but the same problem persists.

It's like this for me: AC (poweroff) -> battery (poweron) ~30 W -> battery (reset) ~13 W. So basically, when I turn the computer on without AC, I have to immediately reboot and it works fine after that. The same was true for 3.6. I wonder, are some hardware settings not getting set before a reboot? I would still love to get it to 10 W idle.

LordChaos73 · 2013-01-21 09:41:35

Rukiri wrote:

LordChaos73 wrote:
I came to the conclusion my laptop suffers from two particular issues:
1. After a suspend/resume cycle, the discrete card (Radeon HD6470M) is powered on again, although /sys/kernel/debug/vgaswitcheroo/switch still reports the contrary. This explains why the laptop is using ~5W more after a suspend/resume cycle. I can blacklist the radeon module, but then the discrete card is powered on when the laptop is started. Unfortunately, blacklisting the module and manually loading the module after boot, then switching the card off and removing the module results in a kernel panic.
The only way to prevent the card from being powered on after a suspend/resume cycle is to make sure the radeon module is not in user before suspending the laptop.
2. After a suspend/resume cycle, the GPU (not sure which one is used) is continuously active which results in a greater discharge rate than before the cycle.
You may want to give 3.7 a try as it's the next stable kernel release, I also ahve a sandy bridge-E processor (going with a xeon 8-core in a few months with a new motherboard)
Are you using an i3, i5, or an i7? I wouldn't worry too much about it unless it's say 10-20 watts more but 5 isn't that much.
But, have you tried powertop? It'll certainly give you an estimate of what your using.
I can also comment on 3.8, would love to use it may just wait until a final 3.8 release which isn't that far away as I've seen it go from rc1-rc4 rather quickly. But until nvidia is fixed I'm stuck on 3.7.3.1.

I found a workaround in order to disable the discrete card again after a resume, see here

The laptop consumes 9~11W when idling and this is now also the case after a resume, unless the Intel GPU is going berserk when the laptop is woken up, but that rarely happens.

Last edited by LordChaos73 (2013-01-21 09:44:03)

Arch Linux

#276 2013-01-12 12:55:05

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#277 2013-01-15 12:44:13

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#278 2013-01-15 16:46:27

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#279 2013-01-16 02:07:00

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#280 2013-01-16 07:21:10

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#281 2013-01-16 11:07:58

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#282 2013-01-16 15:23:18

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#283 2013-01-16 20:41:54

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#284 2013-01-16 21:45:24

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#285 2013-01-17 01:17:48

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#286 2013-01-17 09:36:03

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#287 2013-01-17 10:05:47

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#288 2013-01-17 15:49:27

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#289 2013-01-17 15:52:49

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#290 2013-01-17 21:31:57

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#291 2013-01-18 02:50:41

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#292 2013-01-18 07:17:42

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#293 2013-01-18 19:29:22

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#294 2013-01-18 19:58:27

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#295 2013-01-19 23:40:45

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#296 2013-01-20 09:51:10

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#297 2013-01-21 03:06:27

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#298 2013-01-21 03:21:39

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#299 2013-01-21 08:15:05

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#300 2013-01-21 09:41:35

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

Board footer