Kernel 3.6.2 Power Regressions (Sandy Bridge)

KaiSforza · 2013-03-10 09:00:31

@ywarlock: If you're going to update to a package in [testing], don't use pacman -U. Enable the testing repos (all of them) and use `pacman -Syu` so you get all the right dependencies. There's a tp_smapi package for the 3.8 series in [community-testing].

koz · 2013-03-11 03:57:04

KaiSforza: what happened to your repo for mainline??

KaiSforza · 2013-03-11 04:09:45

It's been taken down. As of 3.8, the bug has been fixed for me, and I don't have the time now with school really kicking up. You'd just get a lot of stale kernels and bad management in it. If you want the mainline kernels, miffe (the maintainer of the AUR package) has a repository for them.

PurpleAlert · 2013-03-14 13:48:56

Just hit this bug again on 3.8.2-1-ck on my X220

magillos · 2013-03-14 18:16:20

The last bug free version for me is 3.7.3-1 (and the only 3.7.x version) and I keep using it. No overheating at all. Powertop shows reasonable 12/13W.
I read here yesterday that 3.8.2 is fixed eventually. I tried it and it's definitely not. At least for me. So still waiting...

magillos · 2013-03-14 18:29:08

.

Last edited by magillos (2013-03-14 18:29:55)

kazimir · 2013-03-16 10:04:25

thinkpad x220 here.

The heating issue used to be better with KaiSforza's mainline kernel and I switched to the sandybridge ck kernel when 3.8.2 got out.

3.8.2-ck seemed to fix the issue, until this morning, when getting out of "suspend to ram" the fan instantly sped up and the laptop started feeling hot after a few seconds.

With cpu usage below 5%, there was no apparent reason for this sudden heating.

Here's sensors output:

thinkpad-isa-0000
Adapter: ISA adapter
fan1: 4460 RPM
acpitz-virtual-0
Adapter: Virtual device
temp1: +95.0 C (crit = +99.0 C)
coretemp-isa-0000
Adapter: ISA adapter
Physical id 0: +97.0 C (high = +86.0 C, crit = +100.0 C)
Core 0: +95.0 C (high = +86.0 C, crit = +100.0 C)
Core 1: +97.0 C (high = +86.0 C, crit = +100.0 C)

donniezazen · 2013-03-18 05:29:43

Wow will this bug be ever fixed. How can it persist whole kernel cycle?

ayr0 · 2013-03-18 16:11:29

I believe it has exhibited this behavior for two kernel cycles (3.6, 3.7, and now apparent we are doing it again 3.8).

Are the people experiencing this issue have Optimus enabled on their laptops?

donniezazen · 2013-03-18 16:36:34

@ayr0 I do have Optimus enabled which I just disabled to try 3.9 mainline kernel as it doesn't support bumblebee.

ayr0 · 2013-03-18 17:57:08

According to this bug report https://bugs.archlinux.org/task/32025, I was wondering maybe something was happening with the discrete card (especially since cpu usage is low and temps are high).

If I completely disable the discrete card, I am hoping to bypass the high temps. The cards in the bug report seem to only mention AMD discrete gfx chips.

stupidus · 2013-03-19 13:46:32

I just upgraded to 3.8.3 from [core] on my T420. It was using ~18W as compared to ~12W with 3.5.6. Therefore, I went back to 3.5.6.

But I am not sure, if it is the same issue. With 3.6.x it was also running at ~12W until I did a suspend/resume, then it went up to ~24W. I never tried any of the 3.7 kernels. But with 3.8.3 I get the ~17W of power consumption instantaneously and suspend/resume does not change anything.

curson · 2013-03-19 22:36:33

Not sure what's going on here. With 3.8.3-2 things seems to be better, but at the same time, 'powertop' has stopped showing the state in which the GPU is, showing just the various cores and not letting me check if the GPU is in RC6 when due to be. Freshly at boot, the laptop seemed to be running a good 5°-6° warmer than average at startup (around 52°C instead of the usual 47°~48°) and the fan is clearly spinning stronger, but about 5 minutes after boot, I'm sitting in X with Chrome open at the usual idle 47°C... I'm confused.

Maybe I've been getting around with this bug so long that I'm just plain suspicious, and the posts here that suggests that even 3.8.x doesn't fix it, are making me overthink things, but it really seems to me like there's still something off in the behaviour of the fans. Since yesterday, before upgrading, I've been running 3.6.2 and I was checking at every boot for the value of:

/sys/kernel/debug/dri/0/i915_cur_delayinfo (with a simple 'cat')

That value being 650MHz meant a safe boot, being at 1300MHz meant the bug was messing things up... right now, I'm just unsure as it seems that particular entity isn't present anymore in 3.8 and can't be checked manually.

Anyway, now the move has been done, can't downgrade to anything lower than 3.8.x without losing the capability of using bumblebee and bbswitch, and the latter requires a linux>3.8, so I'll have to get the pulse of this and see how it goes from here. This bug has been around for far too long though

KaiSforza · 2013-03-19 23:03:32

Just thought I would come on here and make everyone sad again:

This bug has (albiet very rarely) appeared on 3.9. I am still hopeful that this will be fixed soon, though...but now 3 minor version affected by it...is kind of unacceptable.

From what I've noticed, usually a simple suspend/resume will fix the issue, though.

curson · 2013-03-20 07:39:56

After a bunch of reboots, to test things out, I'd say that 3.8.3-2 seems to work fine for me. At the very least, it works as fine as 3.6.2 was doing. I don't use suspend myself, so I wouldn't know what the recurrence is still with that kind of usage behaviour, but it isn't as bad as I thought it was when I first updated to the 3.8.x series.

Still watching things, and I agree with you KaiSforza, this is starting to get a bit ridiculous...

Schrottfresse · 2013-03-25 08:21:45

For me the bug was gone in 3.8.3, but i wasn't able to use my network interfaces. With 3.8.4 the problem is gone.

stupidus · 2013-03-25 13:38:15

I tried linux-3.8.4 now, and it was again the same behaviour as with 3.8.3 (~18W consumption). Then I had the idea, that it might be a problem with bumblebee, so I turned optimus off in the BIOS. Now it seems to me, that the system is using less power (down to 10.6W), but this value is much more oscillating than with 3.5.6
This means, that it often goes up to 14 - 16 W even though the system is still idle and not doing anything. Especially, after a suspend/resume cycle I observe this behaviour. This is still better than the 20+ W with 3.6.x, but still not as good as 3.5.6 (stable ~11W).

Nanthiel · 2013-03-25 13:41:09

You can read /proc/acpi/bbswitch to see if your nvidia card is on using bumblebee and bbswitch.

ayr0 · 2013-03-25 22:35:49

stupidus wrote:

I tried linux-3.8.4 now, and it was again the same behaviour as with 3.8.3 (~18W consumption). Then I had the idea, that it might be a problem with bumblebee, so I turned optimus off in the BIOS. Now it seems to me, that the system is using less power (down to 10.6W), but this value is much more oscillating than with 3.5.6
This means, that it often goes up to 14 - 16 W even though the system is still idle and not doing anything. Especially, after a suspend/resume cycle I observe this behaviour. This is still better than the 20+ W with 3.6.x, but still not as good as 3.5.6 (stable ~11W).

I am seeing the same behavior. I have about 1hr less battery life with 3.8.4 than with 3.5.6. grrrrr....

Looking at the frequencies from i7z, my processor will sometimes randomly go to full power for a few seconds even though my desktop is idle. This never happened in 3.5.6.
Can't downgrade, because of the upgrade for xorg forced me to upgrade intel gfx driver. The newer intel gfx drivers crash often if you're using an old kernel (xorg will crash randomly several times a day!).

EDIT: This is strange. After a few more suspend-resume cycles, my power usage seems to settle around 9-11W. Idling I consistently get about 8.80W (with wifi on).

Last edited by ayr0 (2013-03-26 14:52:10)

stupidus · 2013-04-05 12:11:23

I tried 3.8.5 now. I'm getting about 15W in idle now, which is still more than with 3.5.6. Also temperatures are about 4K higher than usual (with 3.5.6). I am not really happy with this result, but I am thinking about trying it with this kernel version for a longer time now. This is mainly because I am annoyed by the whole partial update stuff. I rarely use my laptop on battery anyway. But still this bug is really annoying.

BTW suspend/resume does not change anything in terms of power consumption anymore.

EDIT: OK, I had to downgrade again. I couldn't even go one afternoon with the new kernel. The system is running too hot, which makes the fan spin up all the time. The problem is that after a few hours even though the system is mainly idle, power consumption goes up to more than 20W. The only problem with the old kernel is that I cannot use bumblebee anymore because nvidia-dkms depends on linux>=3.8.

Last edited by stupidus (2013-04-05 15:13:02)

cfr · 2013-04-26 21:50:53

I am not sure if I am seeing something related to this or not. Until a couple of days ago, I did not experience it and I have an i3 (not i5 or i7) and no other graphics card. Now, however, the laptop sometimes gets hot for no apparent reason after wake from sleep. Sleeping and rewaking seems to resolve the issue.

I am not at all liking the 3.8.* series - EFI STUB loader gone, overheating... .

chx · 2013-04-29 10:07:57

3.9 is out. What do we know about it?

blackout23 · 2013-04-29 12:44:56

Does this also affect the desktop version of Sandybridge CPUs? I have never had any issues with my 2600K. EFISTUB working like a charm and runs at 28 C° all the time in idle or 60 C° with full load at 4.4 Ghz.

dr_undecided · 2013-05-01 19:27:57

I've stopped using Arch on my laptop because of this issue.
But now I'm again extremely sick of using Windows.
Will I have this issue if I use another Linux distribution? Or is is Arch specific?

KaiSforza · 2013-05-01 19:35:09

This is an issue in the kernel. So far, I haven't had any issues with 3.9{,rc} kernels. I don't really remember how prevalent it was with the 3.8 line of kernels (that was a month and a half ago for me), but it appears to have either been solved or gotten better.

Arch Linux

#401 2013-03-10 09:00:31

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#402 2013-03-11 03:57:04

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#403 2013-03-11 04:09:45

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#404 2013-03-14 13:48:56

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#405 2013-03-14 18:16:20

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#406 2013-03-14 18:29:08

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#407 2013-03-16 10:04:25

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#408 2013-03-18 05:29:43

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#409 2013-03-18 16:11:29

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#410 2013-03-18 16:36:34

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#411 2013-03-18 17:57:08

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#412 2013-03-19 13:46:32

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#413 2013-03-19 22:36:33

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#414 2013-03-19 23:03:32

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#415 2013-03-20 07:39:56

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#416 2013-03-25 08:21:45

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#417 2013-03-25 13:38:15

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#418 2013-03-25 13:41:09

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#419 2013-03-25 22:35:49

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#420 2013-04-05 12:11:23

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#421 2013-04-26 21:50:53

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#422 2013-04-29 10:07:57

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#423 2013-04-29 12:44:56

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#424 2013-05-01 19:27:57

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#425 2013-05-01 19:35:09

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

Board footer