Starting with Linux 3.7.3 the power usage of my laptop does not drop as low as before when the display is switched off (~2W difference on AC). With the display being on, power usage is normal. I bisected from v3.7.2 (good) to v3.7.3 (bad) and found the responsible commit cf787fa284598caea41c63d10cff91ff3ecb9671 I'm currently running Linux 3.9 and the problem is still there. Hardware info: 00:02.0 VGA compatible controller: Intel Corporation Mobile GM965/GL960 Integrated Graphics Controller (primary) (rev 0c) 00:02.1 Display controller: Intel Corporation Mobile GM965/GL960 Integrated Graphics Controller (secondary) (rev 0c) Please tell me what additional information you need for resolving this bug. Thanks!
Xorg.0.log?
Created attachment 100591 [details] Xorg.0.log running linux 3.9 (logs are the same with 3.7.2)
Thanks, can you try Option "AccelMethod" "SNA" as that actually turns the pipe off as well as the display?
SNA does not change power uasge. $ grep -i sna /var/log/Xorg.0.log [ 26.667] (**) intel(0): Option "AccelMethod" "sna" [ 27.327] (II) intel(0): SNA initialized with Broadwater/Crestline backend I already tried SNA in the last time but 3D performance (tested with Neverball) is bad with xf86-video-intel 2.21.6. Version 2.21.5 had even better performance with SNA than with UXA but the versions before also decrease the performance with SNA. Except all this SNA has minor rendering problems and UXA is default so I sticked with that.
"Except all this SNA has minor rendering problems" example? Ok, that rules out a theory I had that maybe the crtc->active was not being cleared with DPMS off.
"in the last time" was no good english, I meant "recently". In Chromium the text on the tabs does flicker, when the page is loading or when hovering with the cursor, the "url overlays" are sometimes missing pieces of the boarder (see attachment) and sometimes the titlebars in Gnome get wrong colors which only goes away when reloading gnome-shell (a black X in the close button instead of a white one, for inactive windows the whole titlebar does not fit in brightness). There are probably more issues I'm not aware of. And to complete all my issues, since Linux 3.9 I get these messages all the time, if framebuffer compression is enabled (i915_enable_fbc=1): [ 2115.042671] [drm] not enough stolen space for compressed buffer (need 4096000 bytes), disabling [ 2115.042677] [drm] hint: you may be able to increase stolen memory size in the BIOS to avoid this [ 2121.041058] [drm] not enough stolen space for compressed buffer (need 4096000 bytes), disabling [ 2121.041065] [drm] hint: you may be able to increase stolen memory size in the BIOS to avoid this [ 0.635741] agpgart-intel 0000:00:00.0: detected 8192K stolen memory There is no option to change this in the BIOS.
Created attachment 100621 [details] rendering issue in Chromium with SNA
Created attachment 100631 [details] the close button always looks a bit different; SNA on the left and UXA on the right
The close button is intentional, by the toolkit's own request.(The only issue is that UXA doesn't have a faster less precise mode and so always ignores the application's request in that regard.) The chromium issue looks to be the gen4 GPU bug -- something is very wrong inside the GPU regarding stale VUEs. UXA doesn't trip over this, as it uses a much, much larger hammer to workaround an even more critical gen4 GPU bug. For SNA, I found a smaller hammer which gets past the first problem of sending more than one rectangle to processed by the GPU at one time, but then exposes us to the fact that the GPU sometimes misrenders when it has multiple rectangles. It's a pita.
Sorry for the long delay in locking at this bug. The SNA issues are unrelated (and already tracked I think), so let's just track the regression introduced in commit cf787fa284598caea41c63d10cff91ff3ecb9671 Author: Chris Wilson <chris@chris-wilson.co.uk> Date: Mon Jan 7 10:11:40 2013 +0000 drm/i915: Treat crtc->mode.clock == 0 as disabled Can you please install intel-gpu-tools (git is available at http://cgit.freedesktop.org/xorg/app/intel-gpu-tools/ ) and grap the output of intel_reg_dumper for both a working configuration and a broken one.
Created attachment 104861 [details] intel_reg_dump with Linux 3.7.2 (low power)
Created attachment 104871 [details] intel_reg_dump with Linux 3.9.6 (higher power)
I'm running Intel GM45 and UXA (not SNA), and I got the same syslog output. 2013-06-20T17:26:07.827440-04:00 mike-laptop kernel: [ 1320.857209] [drm] not enough stolen space for compressed buffer (need 4194304 bytes), disabling 2013-06-20T17:26:07.827462-04:00 mike-laptop kernel: [ 1320.857214] [drm] hint: you may be able to increase stolen memory size in the BIOS to avoid this During boot, however, stolen memory size is sufficiently large. 2013-06-20T12:45:04.526539-04:00 mike-laptop kernel: [ 0.724618] agpgart-intel 0000:00:00.0: detected 32768K stolen memory 2013-06-20T12:45:04.526545-04:00 mike-laptop kernel: [ 0.724749] agpgart-intel 0000:00:00.0: AGP aperture is 256M @ 0xd0000000 mike@mike-laptop% uname -a Linux mike-laptop 3.9.6-13.g8ead728-desktop #1 SMP PREEMPT Thu Jun 13 19:23:10 UTC 2013 (8ead728) x86_64 x86_64 x86_64 GNU/Linux Any ideas what's happening? Thanks
We seem to disable fbc here. Can you please boot with drm.debug=0xe added to your kernel cmdline, reproduce the issue and then attach the complete dmesg?
Created attachment 105891 [details] Thinkpad X200 with drm.debug on
(In reply to comment #14) > We seem to disable fbc here. Can you please boot with drm.debug=0xe added to > your kernel cmdline, reproduce the issue and then attach the complete dmesg? Hi. I've attached the log with drm.debug=0xe. Hope this can help. BTW, I don't think I explicitly specified i915_enable_fbc boot parameter in grub. I think it's trying to turn that on by default.
fbc/watermark issues - one for Ville. Also can you please retest on latest drm-intel-nightly - we've improved the watermark quite a bit recently.
Timeout. Please reopen if the problem persists with recent kernels.