Kernel 3.6.2 Power Regressions (Sandy Bridge)

KaiSforza · 2013-07-17 20:16:01

dr_undecided wrote:

Sweet!
What kernel version will it be included in? When do you estimate it will reach the core repository?

It's not even in the official tree yet. It's only been proposed, though I expect it to be put into either a later 3.11 rc or 3.12.

eazy · 2013-07-24 09:46:17

Tried the latest 3.9.9 with the pm_async workaround, still 10 °C hotter than kernel 3.5.6 -- 60 °C idle vs 50 °C [Sony Vaio VPCSA.]

From powertop it seems that while there's a regression on the i915 driver which the recent commit should have fixed, there's another huge power regression which started some time after 3.5.6 and hasn't been tackled yet.

As I seem to understand it there are two major regressions:

1) High power usage after suspend/resume cycle -- mainly due to i915 driver (pm_async workaround)
2) High power usage after boot -- due to ?? (no workaround yet, apart from downgrading to 3.5.6)

#1 has been fixed by latest commit, #2 is still at large.

Anyone else can confirm this diagnosis?

Last edited by eazy (2013-07-24 09:50:30)

Druedain · 2013-07-25 20:33:22

It's definitely better with 3.10.2. I'm using stable kernel from official repo. With 3.9.9 after suspend cpu was mostly overheating. For now I'm working without any problem. Don't know if it's fixed, but it's still better.

firewalker · 2013-07-25 20:36:37

Yes, it seems to be ok with 3.10.2.

trsk · 2013-07-26 00:42:46

Are you two referring only to the suspend regression or also to the general higher power consumption after boot?

firewalker · 2013-07-26 08:16:03

I was referring to the suspend regression. But it just happened again with 3.10.2. I guess the patch is not included.

firewalker · 2013-07-26 08:18:56

I have also noticed some image degradation after resume with 3.10.2.

firewalker · 2013-07-26 08:46:40

No, the patch wasn't present. Just embedded it. Compiling now.

www123 · 2013-07-27 20:33:06

Dell Inspiron 7720 (17r SE) i7 3630qm CPU idles much cooler in Windows 7/8 than Arch +10 degrees C (kernel 3.9.9-1-ARCH)
Is there any solution to this?

Last edited by www123 (2013-07-27 20:34:45)

firewalker · 2013-07-27 22:19:26

With the patch applied to 3.10.2 everything is ok for me.

trsk · 2013-07-27 22:25:54

@firewalker: So there is no graphics corruption and no further problems either? How is your power consumption now (compared to 3.5.6)?

firewalker · 2013-07-27 22:41:29

No, no corruption, no overheating after suspend.

As far as the new power scene, Intel PSTATE. It is better when in powersave mode than the old powersave (acpi). The overall impression is that Intel PSTATE regulates power consumption in a better way.

trsk · 2013-07-27 22:50:15

Thanks. Are you using laptop-mode-tools or similar? Could you also post some exemplary numbers when idling with no foreground or background programms running, with brightness set to lowest, enabled WLAN, disabled BT, disabled Ethernet, no attached devices (USB, displays etc.), with a charged battery? After boot and after resuming from suspend?

If you're using smapi, you could use the following:

cat /sys/devices/platform/smapi/BAT0/{power_avg,power_now}

donniezazen · 2013-07-27 23:05:45

www123 wrote:

Dell Inspiron 7720 (17r SE) i7 3630qm CPU idles much cooler in Windows 7/8 than Arch +10 degrees C (kernel 3.9.9-1-ARCH)
Is there any solution to this?

https://wiki.ubuntu.com/Kernel/PowerMan … vingTweaks
https://wiki.archlinux.org/index.php/TLP
https://aur.archlinux.org/packages/thermald-git/

cfr · 2013-07-28 00:09:16

trsk wrote:

Thanks. Are you using laptop-mode-tools or similar? Could you also post some exemplary numbers when idling with no foreground or background programms running, with brightness set to lowest, enabled WLAN, disabled BT, disabled Ethernet, no attached devices (USB, displays etc.), with a charged battery? After boot and after resuming from suspend?
If you're using smapi, you could use the following:
cat /sys/devices/platform/smapi/BAT0/{power_avg,power_now}

Why don't you just test it yourself? Nothing anybody posts here can guarantee it will perform the same way for you in any case. If it doesn't pan out, downgrade back to your working kernel.

What would this information usefully tell you in any case? My laptop, for example, is essentially unusable with brightness at the lowest setting - might as well save the power and put it to sleep since you can't read anything anyway. I certainly wouldn't be able to get numbers off the thing in that state.

This just looks like you are trying to get other people to do your testing for you. If you want to know, test it. Just like if you want to know which DE or WM you should use, install some and decide which you like.

trsk · 2013-07-28 00:14:23

Sorry, I didn't mean to appear lazy, it's just that I don't have access to my X220 currently. Also, I wanted to specify the most important parameters in some way so that we can compare the results amongst each other. Disabling most of the things and turning down the brightness has the advantage of making small differences between wattages relatively larger.

Last edited by trsk (2013-07-28 00:14:45)

WonderWoofy · 2013-07-28 00:17:00

You should just wait until you get your machine back then. It sounds like whatever poor soul tries to provide this information for you is then going to be bombarded with a billion other test requests...

KaiSforza · 2013-07-28 01:42:43

Running 3.11-rc2, and still getting hit by this. Have not tried the patch yet, but I still haven't been able to reliably reproduce yet, so I don't know if I can ever say if it's working.

mar04 · 2013-07-29 06:24:41

The patch is now in kernel 3.11-rc3, but sadly not in 3.10.4. We will have to wait a little longer till it lands in [core].

Last edited by mar04 (2013-07-29 06:33:42)

KaiSforza · 2013-07-29 06:41:09

If I don't see it again for the rest of time I'll be happy.

firewalker · 2013-07-31 11:23:39

With the patch to the 3.10.2 there is definitely a display degradation.

http://i.imgur.com/WgRpBLr.png

After some mouse movement the display is ok (with minor artifacts that eventualy goes away).

And some dmesg errors:

[ 7073.391125] [drm:i915_hangcheck_hung] *ERROR* Hangcheck timer elapsed... GPU hung
[ 7073.391130] [drm] capturing error event; look for more information in /sys/kernel/debug/dri/0/i915_error_state
[ 7073.402402] [drm:kick_ring] *ERROR* Kicking stuck semaphore on render ring
[13137.145779] [drm] Wrong MCH_SSKPD value: 0x16040307
[13137.145780] [drm] This can cause pipe underruns and display issues.
[13137.145780] [drm] Please upgrade your BIOS to fix this.
[13138.285473] [drm] Enabling RC6 states: RC6 on, RC6p on, RC6pp on

Thermally, everything is ok.

mar04 · 2013-07-31 17:56:57

firewalker wrote:

With the patch to the 3.10.2 there is definitely a display degradation.

Might not be related to this issue, but there were some follow up patches: http://lists.freedesktop.org/archives/i … 30858.html

KaiSforza · 2013-07-31 18:05:38

Alright, I think we should keep this just to the cpu/gpu scaling issue for 3.6-3.1?.x (?). If you've got graphical issues, I'm just going to recommend starting a new thread.

Last edited by KaiSforza (2013-07-31 18:05:49)

cfr · 2013-07-31 21:56:43

KaiSforza wrote:

Alright, I think we should keep this just to the cpu/gpu scaling issue for 3.6-3.1?.x (?). If you've got graphical issues, I'm just going to recommend starting a new thread.

There are already threads on the graphical issue (at least two). There are also at least two apparently distinct graphical issues. So it would be best if people posted to those threads rather than contaminating this one or starting new ones, I think.

Thaodan · 2013-07-31 23:26:55

mar04 wrote:

The patch is now in kernel 3.11-rc3, but sadly not in 3.10.4. We will have to wait a little longer till it lands in [core].

See https://bugs.archlinux.org/task/36107

Arch Linux

#551 2013-07-17 20:16:01

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#552 2013-07-24 09:46:17

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#553 2013-07-25 20:33:22

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#554 2013-07-25 20:36:37

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#555 2013-07-26 00:42:46

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#556 2013-07-26 08:16:03

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#557 2013-07-26 08:18:56

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#558 2013-07-26 08:46:40

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#559 2013-07-27 20:33:06

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#560 2013-07-27 22:19:26

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#561 2013-07-27 22:25:54

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#562 2013-07-27 22:41:29

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#563 2013-07-27 22:50:15

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#564 2013-07-27 23:05:45

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#565 2013-07-28 00:09:16

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#566 2013-07-28 00:14:23

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#567 2013-07-28 00:17:00

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#568 2013-07-28 01:42:43

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#569 2013-07-29 06:24:41

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#570 2013-07-29 06:41:09

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#571 2013-07-31 11:23:39

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#572 2013-07-31 17:56:57

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#573 2013-07-31 18:05:38

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#574 2013-07-31 21:56:43

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

#575 2013-07-31 23:26:55

Re: Kernel 3.6.2 Power Regressions (Sandy Bridge)

Board footer