Kernel 3.6.2 Power Regressions (Sandy Bridge)

Nanthiel · 2013-01-21 11:23:38

LordChaos73 wrote:

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.

Do you use any other measure of power saving? That's a pretty good result. Will check out your link.

LordChaos73 · 2013-01-21 11:37:53

Nanthiel wrote:

LordChaos73 wrote:
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.
Do you use any other measure of power saving? That's a pretty good result. Will check out your link.

I've used PowerTop to determine what could be improved; run

# powertop --html

on your laptop and check the html file.

I've put most of these recommendations in the following file so it gets picked up by systemd:

[ericd@vaiopro13 Scripts]$ cat /etc/tmpfiles.d/power-savings.conf 
w /sys/class/scsi_host/host0/link_power_management_policy - - - - min_power
w /proc/sys/kernel/nmi_watchdog - - - - 0
w /sys/bus/usb/devices/1-1.1/power/control - - - - auto
w /sys/bus/usb/devices/1-1.2/power/control - - - - auto
w /sys/bus/usb/devices/3-1.1/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:02.0/power/control - - - - auto
w /sys/bus/pci/devices/0000:05:00.0/power/control - - - - auto
w /sys/bus/pci/devices/0000:03:00.0/power/control - - - - auto
w /sys/bus/pci/devices/0000:02:00.0/power/control - - - - auto
w /sys/bus/pci/devices/0000:01:00.0/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:16.0/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:1a.0/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:1c.0/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:1c.1/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:1c.2/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:1c.3/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:1d.0/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:1f.0/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:1f.2/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:01.0/power/control - - - - auto
w /sys/bus/pci/devices/0000:00:1b.0/power/control - - - - auto

EDIT: just installed 3.7.3 from the official repo, the discrete card is still turned on after the laptop wakes from resume. Can't comment on the Intel GPU bug, yet.

Last edited by LordChaos73 (2013-01-21 12:34:09)

hendry · 2013-01-22 01:36:53

3.7.3 temperature and suspend is fine.

However wifi cuts off for me from time to time. https://bugs.archlinux.org/task/33503

I had a very similar wifi experience running 3.8rc4. *sigh* another annoying regression that seems to have survived a complete kernel cycle.

KaiSforza · 2013-01-22 04:31:20

hendry wrote:

3.7.3 temperature and suspend is fine.
However wifi cuts off for me from time to time. https://bugs.archlinux.org/task/33503
I had a very similar wifi experience running 3.8rc4. *sigh* another annoying regression that seems to have survived a complete kernel cycle.

File a kernel bug report. Have you found any bug reports on https://bugzilla.kernel.org with the same issue? I didn't see anything pop out at me in a quick search.

hendry · 2013-01-22 04:43:43

KaiSforza wrote:

File a kernel bug report. Have you found any bug reports on https://bugzilla.kernel.org with the same issue? I didn't see anything pop out at me in a quick search.

Would you believe me that it tooks me 10 minutes to *find* my own bug comment on Bugzilla?

https://bugzilla.kernel.org/show_bug.cgi?id=48921#c11

Do you think I should create a new one? Not sure how things work on https://bugzilla.kernel.org

killsto · 2013-01-23 20:20:31

I'm not at my laptop now, but I did not seem to have this issue when I was on 3.6.? (whatever the latest release in core was). But as soon as I upgraded to 3.7 yesterday I experienced battery life of a little over an hour, when I usually get about 3.5-4 hours. At first I thought it was bumblebee, but I reinstalled both bumblebee and bbswitch from the aur. I have since downgraded back to 3.6 and the battery seems to be back to normal. However, I do not suspend much so maybe that's why I didn't experience it before?

How long is it viable to stay at a kernel release that is not in core? Will any packages depend on 3.7? I don't see how they could since it's so recent.

Last edited by killsto (2013-01-23 20:21:23)

chx · 2013-01-23 20:31:50

What about using linux-lts? It's 3.0 currently but I hear it'll be 3.4 soon. Sounds a great to avoid these problems. Is there anything in 3.5 or 3.6 that we need? Edit: nevermind, I mixed up http://www.linux.com/news/featured-blog … 4-released with linux-lts.

As for dependencies, who knows. glibc 2.17 (Dec 25, 2012) requires Linux 2.6.16 (Mar 20, 2006). Linux 3.6 was released Sep 30, 2012. Presuming the same time frame, we have some time until early 2019 to begin to worry...

Last edited by chx (2013-01-23 20:46:22)

KaiSforza · 2013-01-23 20:43:10

killsto wrote:

How long is it viable to stay at a kernel release that is not in core? Will any packages depend on 3.7? I don't see how they could since it's so recent.

You've got a few options, really.
1) You can use the LTS kernel, which has binary modules for virtualbox, etc. (You won't have to build any packages that require on a specific kernel)

2) You can build one of the stable kernels from kernel.org (3.2.x or 3.4.x) yourself to get some of the new features. If you use

virtualbox, tp_smapi, tpacpi-bat, acpi_call, or anything that requires a specific kernel version will have to be rebuilt for whatever kernel you choose if you go with option 2. (It sounds bad, but it's really not)

killsto · 2013-01-23 21:00:28

KaiSforza wrote:

killsto wrote:
How long is it viable to stay at a kernel release that is not in core? Will any packages depend on 3.7? I don't see how they could since it's so recent.
You've got a few options, really.
1) You can use the LTS kernel, which has binary modules for virtualbox, etc. (You won't have to build any packages that require on a specific kernel)
2) You can build one of the stable kernels from kernel.org (3.2.x or 3.4.x) yourself to get some of the new features. If you use
virtualbox, tp_smapi, tpacpi-bat, acpi_call, or anything that requires a specific kernel version will have to be rebuilt for whatever kernel you choose if you go with option 2. (It sounds bad, but it's really not)

Virtualbox, bumbleebee, etc seem to need to be rebuilt every kernel update anyway. I even had to rebuild dkms-bbswitch for some reason on 3.7. That reminds me, (completely off topic) the other problem I had was getting VMware Workstation to find 3.7's headers.

So basically I can just stay on an older kernel with no issues as long as I don't need newer features.

Lassulus · 2013-01-24 00:46:07

I created a package in the AUR by the name of linux-sandyfix.
https://aur.archlinux.org/packages/linux-sandyfix/

I use the stock kernel with this little patch http://ix.io/45T
With this patch I don't experience the "overheating after suspend" problem anymore

To boot this kernel you must edit your Bootmanager accordingly (Boot into linux-sandyfix instead of linux)

KaiSforza · 2013-01-24 02:29:04

Lassulus wrote:

I created a package in the AUR by the name of linux-sandyfix.
https://aur.archlinux.org/packages/linux-sandyfix/
I use the stock kernel with this little patch http://ix.io/45T
With this patch I don't experience the "overheating after suspend" problem anymore
To boot this kernel you must edit your Bootmanager accordingly (Boot into linux-sandyfix instead of linux)

Thanks! I've been publishing mainline pkgbuilds, but having a nice stable pkgbuild will probably be good for most people. (Just be warned, the AUR doesn't like split packages very much so the description is always the last one in the packages- list. Still okay to put it there, though.)

Deth · 2013-01-24 21:46:10

My 2 cents for one of my laptops:

-All 3.7.*-arch kernels I tried burned my laptop.
-All 3.7.*-ck kernels has that issue too. Both "usual" and optimized for Sandy Bridge from here
-The last 3.0.60-lts kernel is not affected, but it lacks my cardreader module (solved anyway) and has broken virtualbox-host-modules package (vboxnetadp and vboxnetflt could not be loaded). I did try it yesterday though, I see today`s update for virtualbox modules, but didn`t try it yet.
-Now I`m on a 3.6.11, the latest kernel I find that doesn`t burn my hands

KaiSforza · 2013-01-26 04:20:13

3.8-rc5 has been pushed to my repository.
http://sakuscans.com/pacmanpkg/x86_64/00README

the sad clown · 2013-01-26 04:50:44

KaiSforza wrote:

3.8-rc5 has been pushed to my repository.
http://sakuscans.com/pacmanpkg/x86_64/00README

Very nice. I'll try it out.

the sad clown · 2013-01-26 06:51:42

KaiSforza wrote:

3.8-rc5 has been pushed to my repository.
http://sakuscans.com/pacmanpkg/x86_64/00README

Unfortunately, I get an error when I try it:

error: failed to update wgiokas (no servers configured for repository)

koz · 2013-01-26 07:58:46

the sad clown wrote:

KaiSforza wrote:
3.8-rc5 has been pushed to my repository.
http://sakuscans.com/pacmanpkg/x86_64/00README
Unfortunately, I get an error when I try it:
error: failed to update wgiokas (no servers configured for repository)

# William Giokas's (aka KaiSforza's) packages
[wgiokas]
SigLevel = Optional
http://sakuscans.com/pacmanpkg/$arch

Missing " Server = " on the last line.

Server = http://sakuscans.com/pacmanpkg/$arch

the sad clown · 2013-01-26 08:00:16

I just tried linux-mainline 3.8-rc4 from the AUR, but gave up on it before I could see if it resolved the gpu issue since I lost my wireless network interface when booting from it. Not sure why that happened.

the sad clown · 2013-01-26 08:02:09

koz wrote:

the sad clown wrote:
KaiSforza wrote:
3.8-rc5 has been pushed to my repository.
http://sakuscans.com/pacmanpkg/x86_64/00README
Unfortunately, I get an error when I try it:
error: failed to update wgiokas (no servers configured for repository)
# William Giokas's (aka KaiSforza's) packages
[wgiokas]
SigLevel = Optional
http://sakuscans.com/pacmanpkg/$arch
Missing " Server = " on the last line.
Server = http://sakuscans.com/pacmanpkg/$arch

Ah, duh, I should have looked more carefully before copying it over. Thanks. I'll give it a go again.

the sad clown · 2013-01-26 08:20:27

No go. rc5 has the same problem that AUR rc4 has, no wireless network interface to speak of.

ip link only shows lo and enp4s0, my wired interface.

edit: I'll give it a go again when a new rc becomes available and report back here.

Last edited by the sad clown (2013-01-26 08:22:57)

Nanthiel · 2013-01-26 13:17:43

I just noticed that after going from linux-ck 3.7.3 to the official kernel from the repo (3.7.4) the power situtation got worse!

Before, if I rebooted the computer after powering it on, the consumption went to 12 W idle (from 25--30 W). Now a reboot doesn't help anymore.

KaiSforza · 2013-01-26 18:09:35

the sad clown wrote:

No go. rc5 has the same problem that AUR rc4 has, no wireless network interface to speak of.
ip link only shows lo and enp4s0, my wired interface.
edit: I'll give it a go again when a new rc becomes available and report back here.

Thanks for pointing that out, fixed the readme.

Also, what wireless hardware do you have?

the sad clown · 2013-01-26 19:25:13

I should have mentioned that from the start. I've got Atheros AR9285, which uses the ath9k driver. I noticed that in the AUR build, the ath packages are present in src, but not in lib when installed. Strangely, it was the only wireless driver not present.

the sad clown · 2013-01-26 19:31:39

Just installed your kernel and again, no ath. Here:

# ls /lib/modules/3.8.0-1-rc5-mainline-dirty/kernel/drivers/net/wireless/
total 292K
 16K adm8211.ko.gz       4.0K b43legacy    4.0K mwifiex           4.0K rtl818x
 48K airo.ko.gz          4.0K brcm80211     28K mwl8k.ko.gz       4.0K rtlwifi
4.0K airo_cs.ko.gz       4.0K hostap       4.0K orinoco           4.0K ti
 24K at76c50x-usb.ko.gz  4.0K ipw2x00      4.0K p54                16K wl3501_cs.ko.gz
 24K atmel.ko.gz         4.0K iwlegacy     4.0K prism54            16K zd1201.ko.gz
4.0K atmel_cs.ko.gz      4.0K iwlwifi       16K ray_cs.ko.gz      4.0K zd1211rw
4.0K atmel_pci.ko.gz     4.0K libertas      20K rndis_wlan.ko.gz
4.0K b43                 4.0K libertas_tf  4.0K rt2x00

Versus the standard repo kernel:

# ls /lib/modules/3.7.4-1-ARCH/kernel/drivers/net/wireless/
total 316K
 16K adm8211.ko.gz       4.0K b43        4.0K libertas_tf       4.0K rt2x00
 48K airo.ko.gz          4.0K b43legacy  4.0K mwifiex           4.0K rtl818x
4.0K airo_cs.ko.gz       4.0K brcm80211   28K mwl8k.ko.gz       4.0K rtlwifi
 24K at76c50x-usb.ko.gz  4.0K hostap     4.0K orinoco           4.0K ti
4.0K ath                 4.0K ipw2x00    4.0K p54                16K wl3501_cs.ko.gz
 24K atmel.ko.gz         4.0K iwlegacy   4.0K prism54            16K zd1201.ko.gz
4.0K atmel_cs.ko.gz      4.0K iwlwifi     28K ray_cs.ko.gz      4.0K zd1211rw
4.0K atmel_pci.ko.gz     4.0K libertas    28K rndis_wlan.ko.gz

Note, ath is the only one missing. The AUR package does the same thing, but I can't figure out why.

KaiSforza · 2013-01-26 20:44:01

Ah. looks like a config issue. I'll fix that config and build it soon.

EDIT: Fixed. 3.8.rc5-g949db15-3 is up.

Last edited by KaiSforza (2013-01-26 22:12:22)

Hspasta · 2013-01-27 00:41:23

Serious thanks for such quick builds for the rc kernels.

Let's hope this fixes my GPU hang problem I've had issued with the last few 3.8rc's. At least the power regression seems to be fine for me.

Edit: Power regression still fixed, but GPU hang still there

Last edited by Hspasta (2013-01-27 00:45:40)

Arch Linux

#301 2013-01-21 11:23:38

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#302 2013-01-21 11:37:53

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#303 2013-01-22 01:36:53

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#304 2013-01-22 04:31:20

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#305 2013-01-22 04:43:43

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#306 2013-01-23 20:20:31

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#307 2013-01-23 20:31:50

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#308 2013-01-23 20:43:10

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#309 2013-01-23 21:00:28

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#310 2013-01-24 00:46:07

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#311 2013-01-24 02:29:04

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#312 2013-01-24 21:46:10

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#313 2013-01-26 04:20:13

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#314 2013-01-26 04:50:44

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#315 2013-01-26 06:51:42

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#316 2013-01-26 07:58:46

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#317 2013-01-26 08:00:16

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#318 2013-01-26 08:02:09

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#319 2013-01-26 08:20:27

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#320 2013-01-26 13:17:43

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#321 2013-01-26 18:09:35

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#322 2013-01-26 19:25:13

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#323 2013-01-26 19:31:39

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#324 2013-01-26 20:44:01

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#325 2013-01-27 00:41:23

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

Board footer