Kernel 3.6.2 Power Regressions (Sandy Bridge)

flockyrocky · 2012-10-25 02:57:58

I uninstalled laptop-mode-tools and installed TLP from AUR yesterday, and until now it works for me. Maybe thinkpad users can have a try.

donniezazen · 2012-10-25 03:41:22

@flockryrocky I haven't found much difference between laptop-mode-tools and tlp. TLP doesn't seem to handle PCI power saving.

Anthony25 · 2012-10-25 07:38:37

After a suspend, I have the same problem as many of you : ondemand stays at the highest frequency.

I have to switch on the powersave governor before the suspend, and after resume I switch again on ondemand. It doesn't work like it did with the Kernel 3.5.8, but it switch between the freq mini and freq maxi (not by levels).

l3nkz · 2012-10-25 07:44:45

flockyrocky wrote:

I uninstalled laptop-mode-tools and installed TLP from AUR yesterday, and until now it works for me. Maybe thinkpad users can have a try.

Did this fix the graphic card problem or just the problem with the frequency scaling of the cpu?

EDIT
In the official kernel bugtracker a patch was published! See the following link.

EDIT2
I finally managed to try the patch mentioned in the above url and after the first short suspend everything seems to be ok again.
I use the mainline kernel from the aur (link) with the following patch. I had to customize the original patch from above to fit the new source base, but it's working!
I'll go on testing, but this seems to fix the problem!!

Last edited by l3nkz (2012-10-25 10:32:13)

shadyabhi · 2012-10-25 09:50:10

vheissu wrote:

shadyabhi wrote:
3.6.3-1 fixed it for me.
Lenovo Thinkpad X220 here. Now, it stays at arounf 43-50 C during normal usage.
sry, can’t confirm that, it’s still not working with my X220 and ondemand governor after suspend.

After a reboot, it was all fine but then I rebooted again, the problem is still there.. Now, I am wondering how it got fine this last time.

So, the problem still persists.

Anthony25 · 2012-10-25 12:15:48

l3nkz wrote:

flockyrocky wrote:
I uninstalled laptop-mode-tools and installed TLP from AUR yesterday, and until now it works for me. Maybe thinkpad users can have a try.
Did this fix the graphic card problem or just the problem with the frequency scaling of the cpu?
EDIT
In the official kernel bugtracker a patch was published! See the following link.
EDIT2
I finally managed to try the patch mentioned in the above url and after the first short suspend everything seems to be ok again.
I use the mainline kernel from the aur (link) with the following patch. I had to customize the original patch from above to fit the new source base, but it's working!
I'll go on testing, but this seems to fix the problem!!

Cool !

So the ondemand governors works like with the Kernel 3.5.8 (in changing the frequency by levels) or the patch just fix the suspend problem ?

l3nkz · 2012-10-25 13:05:53

Anthony25 wrote:

Cool !
So the ondemand governors works like with the Kernel 3.5.8 (in changing the frequency by levels) or the patch just fix the suspend problem ?

It fixed both problems in my case!

My Laptop (T420s with i5@2.5GHz) is consuming 7-8W when idling around at the moment! And after 10 suspend/resume cycles without any problem this solution probably fixes the problems finally.

KaiSforza · 2012-10-25 14:47:30

l3nkz wrote:

flockyrocky wrote:
I uninstalled laptop-mode-tools and installed TLP from AUR yesterday, and until now it works for me. Maybe thinkpad users can have a try.
Did this fix the graphic card problem or just the problem with the frequency scaling of the cpu?
EDIT
In the official kernel bugtracker a patch was published! See the following link.
EDIT2
I finally managed to try the patch mentioned in the above url and after the first short suspend everything seems to be ok again.
I use the mainline kernel from the aur (link) with the following patch. I had to customize the original patch from above to fit the new source base, but it's working!
I'll go on testing, but this seems to fix the problem!!

I was just about to report that an unpatched 3.7.0-1-mainline has been working for me for 12 hours and 2 suspend resume cycles (one short, sub 5 minutes, one long, overnight).

pyropirat · 2012-10-25 23:31:01

on my laptop ist randomly, most of the time its working, but somtimes on resuming or restarting the speed steping stops working

system atm: patched mainline-kernel 3.7 rc2
Kerneloptions: i915.i915_enable_rc6=7 i915.i915_enable_fbc=1 i915.lvds_downclock=1 i915.semaphores=1 i915.modeset=1

flockyrocky · 2012-10-26 07:08:52

After several days testing I found TLP does make it better, but didn't resolve this issue.The chance to run into problem is reduced for me. Anyone else tried it?
@l3nkz
I just mean frequency issue. I turned off my Nvidia card because I almost never use it.

l3nkz · 2012-10-26 07:19:22

pyropirat wrote:

on my laptop ist randomly, most of the time its working, but somtimes on resuming or restarting the speed steping stops working
system atm: patched mainline-kernel 3.7 rc2
Kerneloptions: i915.i915_enable_rc6=7 i915.i915_enable_fbc=1 i915.lvds_downclock=1 i915.semaphores=1 i915.modeset=1

I set the i915_enable_rc6 to 1 because i read here that deepest rc6 is buggy. Can you try setting this option to -1 or 1?

My actual kernelparameters are the following:

$ cat /etc/default/grub | grep GRUB_CMDLINE_LINUX                                                      [0]
GRUB_CMDLINE_LINUX_DEFAULT="quiet"
GRUB_CMDLINE_LINUX="cryptdevice=/dev/sdb2:system:allow-discards i915.i915_enable_rc6=-1 i915.i915_enable_fbc=1 i915.lvds_downclock=1 drm.vblankoffdelay=1"

By the way, I've been running my previously posted setting for 1 day without any problems. Resume/Suspend is working again as well as the cpu governer.

flockyrocky wrote:

@l3nkz
I just mean frequency issue. I turned off my Nvidia card because I almost never use it.

EDIT Ah ok thanks. This is fixed too for me! But I never had frequency issues after rebooting. They occured together with the gpu activity issues! And they both are gone now.

Last edited by l3nkz (2012-10-26 07:27:33)

eduedix · 2012-10-28 10:09:30

gaah im bored of this bug, my fan never goes silent, the laptop is sooo hot. are you sure that wireless driver is fine?

shadyabhi · 2012-10-28 14:49:09

I don't get it. top says that my cpus are almost at 0-1% but when I fired up iz7.it's saying one of the threads for two cores at 100%. http://i.imgur.com/qcP02.png

WonderWoofy · 2012-10-28 15:07:02

@shadyabhi, those are c-states. They are various levels of power saving modes that the processor can drop into. The higher the number, the lower power state it is. So I think that it is actually telling you that your processor is staying in c1 for most of the time. It should be dropping to c6 if you are not doing anything. I would suggest trying powertop to really get a better idea of what is going on. You can view the processor c-states as well as gpu rc6 states in powertop as well.

donniezazen · 2012-10-28 19:21:32

3.6.4 has been released. I haven't checked if there is anything relevant in the kernel for us.

l3nkz · 2012-10-28 19:57:28

So guys, I want to give a new status report:

I've been running my patched linux-mainline kernel for about 2 days with a lot of resume and supend cycles (long as well as short ones) without any problems.

$uptime                                                                                                                             [0]
20:51:17 up 2 days, 11:11,  1 user,  load average: 0.05, 0.15, 0.13

CPU governor and GPU RC6 states are working like a charm!

Has anybody else who has/had the problem tried the linux-mainline kernel and the patch I posted?

Cheers

1126 · 2012-10-29 15:49:52

l3nkz wrote:

So guys, I want to give a new status report:
I've been running my patched linux-mainline kernel for about 2 days with a lot of resume and supend cycles (long as well as short ones) without any problems.
$uptime                                                                                                                             [0]
20:51:17 up 2 days, 11:11,  1 user,  load average: 0.05, 0.15, 0.13
CPU governor and GPU RC6 states are working like a charm!
Has anybody else who has/had the problem tried the linux-mainline kernel and the patch I posted?

Cheers

I did. It helped a lot by reducing heat and increasing battery lifetime.

Last edited by 1126 (2012-10-29 18:05:53)

eduedix · 2012-10-29 18:14:58

can you explain in short how you patch&install the kernel?

l3nkz · 2012-10-30 10:58:10

eduedix wrote:

can you explain in short how you patch&install the kernel?

Yes of course! This is quite easy:

1. Download the needed files
- get the tarball from here

wget https://aur.archlinux.org/packages/li/linux-mainline/linux-mainline.tar.gz

- extract the tarball and change into the directory
- get the patch from here

wget http://pastebin.com/raw.php\?i\=tQcAzLcs -O sandybridge.patch

2. Edit PKGBUILD
- add the patch to the source array that it lookes like this

source=("http://www.kernel.org/pub/linux/kernel/v3.x/testing/${_srcname}.tar.xz"
        #"http://www.kernel.org/pub/linux/kernel/v3.x/patch-${pkgver}.xz"
        # the main kernel config files
        'config' 'config.x86_64'
        # standard config files for mkinitcpio ramdisk
        'linux.preset'
        'change-default-console-loglevel.patch'
        'sandybridge.patch')

- add the md5sum for the patch to the md5sums array that it lookes like this

md5sums=('885ce2287e76128dbf0e276b19c80f63'
         '443f285744cb96401f9a631f542b9248'
         '2793e97a9e98c501afd3e18d2e3b9762'
         'eb14dcfd80c00852ef81ded6e826826a'
         '9d3c56a4b999c8bfbd4018089a62f662'
         '33a7aafb7673c5a2d66e3e77ac6fb066')

Where 33a7aafb7673c5a2d66e3e77ac6fb066 is the result of md5sum sandybridge.patch
- add the patch command to the build() routine like this

build() {
  cd "${srcdir}/${_srcname}"

  # set DEFAULT_CONSOLE_LOGLEVEL to 4 (same value as the 'quiet' kernel param)
  # remove this when a Kconfig knob is made available by upstream
  # (relevant patch sent upstream: https://lkml.org/lkml/2011/7/26/227)
  patch -Np1 -i "${srcdir}/change-default-console-loglevel.patch"

  # patch for the sandybridge suspend/resume bug
  patch -Np1 -i "${srcdir}/sandybridge.patch"

  if [ "${CARCH}" = "x86_64" ]; then
    cat "${srcdir}/config.x86_64" > ./.config
  else
    cat "${srcdir}/config" > ./.config
  fi

3. Compile the Kernel
run the following command as normal user in the folder where your PKGBUILD is located

makepkg

4. Install the new Kernel
install the compiled kernel with the following command

pacman -U linux-mainline.....tar.xz

where linux-mainline.....tar.xz is the result of the makepkg command.

And you're done

Good luck and have fun!

Cheers

eduedix · 2012-10-30 16:20:13

Thank you very much

hendry · 2012-10-30 16:25:05

yes, very nice tutorial there l3nkz. I didn't know about the mainline AUR package.

My machine is now 46C instead of almost 80C and it can be considered a laptop again

be nice to know when this patch is going to land in 3.7. I wonder why you didn't patch the linux package?

I like the linux-mainline package since it seems to have far less patches than the Arch linux one. Wonder what the need was to change the loglevel though.

thanks again

eduedix · 2012-10-30 18:16:44

well one issue here, after i closed my lid the laptop went on suspend. and then i guess kernel panicked: very fast toggle of green power/suspend led, laptop didnt resume. had to shut it down by press&holding power key. BEWARE.

hendry · 2012-10-31 06:25:01

eduedix, your panic is probably a different issue entirely. The patch trivially adds a microsleep AFAICT.

eduedix · 2012-10-31 06:48:34

well yeah, but there is definitely another issue on this kernel. l3nkz, is yours X220,too?

pyknite · 2012-10-31 19:27:10

Thx l3nkz, your patch seem to be working on my x220... At least for the high temperature, didn't try the rest for the moment.

Arch Linux

#76 2012-10-25 02:57:58

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#77 2012-10-25 03:41:22

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#78 2012-10-25 07:38:37

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#79 2012-10-25 07:44:45

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#80 2012-10-25 09:50:10

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#81 2012-10-25 12:15:48

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#82 2012-10-25 13:05:53

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#83 2012-10-25 14:47:30

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#84 2012-10-25 23:31:01

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#85 2012-10-26 07:08:52

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#86 2012-10-26 07:19:22

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#87 2012-10-28 10:09:30

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#88 2012-10-28 14:49:09

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#89 2012-10-28 15:07:02

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#90 2012-10-28 19:21:32

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#91 2012-10-28 19:57:28

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#92 2012-10-29 15:49:52

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#93 2012-10-29 18:14:58

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#94 2012-10-30 10:58:10

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#95 2012-10-30 16:20:13

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#96 2012-10-30 16:25:05

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#97 2012-10-30 18:16:44

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#98 2012-10-31 06:25:01

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#99 2012-10-31 06:48:34

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#100 2012-10-31 19:27:10

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

Board footer