Summary: | [gm45] GPU lockup (xorg crash when closing the laptop lid) | ||
---|---|---|---|
Product: | DRI | Reporter: | Geir Ove Myhr <gomyhr> |
Component: | DRM/Intel | Assignee: | Jesse Barnes <jbarnes> |
Status: | CLOSED FIXED | QA Contact: | |
Severity: | normal | ||
Priority: | medium | CC: | 84yelo3, unggnu |
Version: | DRI git | ||
Hardware: | x86-64 (AMD64) | ||
OS: | Linux (All) | ||
Whiteboard: | |||
i915 platform: | i915 features: | ||
Attachments: |
Description
Geir Ove Myhr
2010-03-17 15:41:24 UTC
Created attachment 34169 [details]
dmesg before closing lid
Created attachment 34170 [details]
Xorg.0.log before closing lid
Created attachment 34171 [details]
intel_gpu_dump output before closing lid (gzipped)
Created attachment 34172 [details]
dmesg (with drm.debug=0x02) after closing lid
Unfortunately, the beginning is missing, since the log has been filled with [drm:i915_add_request] lines.
Created attachment 34173 [details]
i915_error_state after lid close
Created attachment 34174 [details]
Xorg.0.log after closing lid
Created attachment 34175 [details]
intel_gpu_dump output after closing lid (gzipped)
Created attachment 34176 [details]
intel_error_decode output on the previously attache i915_error_state (gzipped, lid closed)
Created attachment 34177 [details]
dmesg (with drm.debug=0x02) after opening lid
Created attachment 34178 [details]
dmesg (with drm.debug=0x02) after opening lid
Created attachment 34179 [details]
i915_error_state after lid open
Created attachment 34180 [details]
Xorg.0.log after opening lid
*** Bug 27267 has been marked as a duplicate of this bug. *** The current (hanging) batchbuffer for this bug is (from running intel_error_decode on i915_error_state captured with drm-intel-next or 2.6.34-rcX kernel) 0x0ad96000: 0x09100000: MI_LOAD_SCAN_LINES_INCL 0x0ad96004: 0x00000384: dword 1 0x0ad96008: 0x09100000: MI_LOAD_SCAN_LINES_INCL 0x0ad9600c: 0x00000384: dword 1 0x0ad96010: 0x01800020: MI_WAIT_FOR_EVENT 0x0ad96014: HEAD 0x54f08806: XY_SRC_COPY_BLT (rgb enabled, alpha enabled, src tile 1, dst tile 1) 0x0ad96018: 0x03cc0600: format 8888, dst pitch 1536, clipping disabled 0x0ad9601c: 0x00000000: dst (0,0) 0x0ad96020: 0x038405a0: dst (1440,900) 0x0ad96024: 0x02522000: dst offset 0x02522000 0x0ad96028: 0x00000000: src (0,0) 0x0ad9602c: 0x00000600: src pitch 1536 0x0ad96030: 0x0360e000: src offset 0x0360e000 0x0ad96034: 0x02000000: MI_FLUSH 0x0ad96038: 0x00000000: MI_NOOP 0x0ad9603c: 0x05000000: MI_BATCH_BUFFER_END Most of the 50+ automatically reported Ubuntu bug reports that are duplicates to this fail to record this batchbuffer with intel_gpu_dump. In the limited i915_error_state that is captured with the Ubuntu kernel, one can se IPEHR: 0x01800020 (header of last instruction) in all those reports. This last instruction MI_WAIT_FOR_EVENT (Display Pipe B Scan Line Window Wait Enable) may (or may not) be related to the lines in dmesg output: [ 3188.322798] [drm:gm45_get_vblank_counter], trying to get vblank count for disabled pipe 0 [ 3188.532922] [drm:gm45_get_vblank_counter], trying to get vblank count for disabled pipe 1 This bug has been marked as fixed downstream. The patch that fixed it was [ Jesse Barnes ] * SAUCE: drm/i915: don't change DRM configuration when releasing load detect pipe http://kernel.ubuntu.com/git?p=ubuntu/ubuntu-lucid.git;a=commit;h=0d2907f4bead56cff60f91068b3a3efa7149e702 I haven't seen this being applied upstream in linux-2.6.33.y or linux-2.6 git trees, so I'm not closing this bug yet. Jesse, that patch is still MIA. However, at work I think we may have a move serious issue, namely we need to ensure that the gpu is idle prior to changing connectors lest we have a pending WAIT. Because I know Jesse is this ->||<- close to getting this patch upstream. :) (In reply to comment #17) > Because I know Jesse is this ->||<- close to getting this patch upstream. :) Last message on the intel-gfx mailing list [1] was that this patch is not needed when the patch "drm/i915: cleanup lvds detection function" [2] is applied. So I suppose it is patch [2] that we will see upstream soon. [1]: http://lists.freedesktop.org/archives/intel-gfx/2010-July/007383.html [2]: http://lists.freedesktop.org/archives/intel-gfx/2010-July/007382.html On Sat, 10 Jul 2010 07:47:11 -0700 (PDT) bugzilla-daemon@freedesktop.org wrote: > https://bugs.freedesktop.org/show_bug.cgi?id=27146 > > --- Comment #18 from Geir Ove Myhr <gomyhr@gmail.com> 2010-07-10 07:47:11 PDT --- > (In reply to comment #17) > > Because I know Jesse is this ->||<- close to getting this patch upstream. :) > > Last message on the intel-gfx mailing list [1] was that this patch is not > needed when the patch "drm/i915: cleanup lvds detection function" [2] is > applied. So I suppose it is patch [2] that we will see upstream soon. > > [1]: http://lists.freedesktop.org/archives/intel-gfx/2010-July/007383.html > [2]: http://lists.freedesktop.org/archives/intel-gfx/2010-July/007382.html No, the cleanup function is separate. If you still see issues with current 2.6.35-rc and you need to apply the previous "disable unused" patch, then we have some other issue going on as well. Put another way; is the patch in http://kernel.ubuntu.com/git?p=ubuntu/ubuntu-lucid.git;a=commit;h=0d2907f4bead56cff60f91068b3a3efa7149e702 still required in current (i.e. 2.6.35-rc) kernels? (In reply to comment #20) > Put another way; is the patch in > http://kernel.ubuntu.com/git?p=ubuntu/ubuntu-lucid.git;a=commit;h=0d2907f4bead56cff60f91068b3a3efa7149e702 > still required in current (i.e. 2.6.35-rc) kernels? The Ubuntu 10.10 development version has been without this patch since late May, and we haven't got the storms of bug reports for this issue like we did for 10.04, but then again that started only later in the release cycle when more people started testing. The patch was apparently removed after one person confirmed that the patch was no longer needed for him [1]. We have had one recent bug report with Maverick with the same type of batchbuffer hanging in MI_WAIT_FOR_EVENT after two MI_LOAD_SCAN_LINES_INCL [2,3], but the circumstances seems to be different (watching video) and Chris Wilson has committed a DDX patch [4] which may solve this but it has not been tested yet. I have asked for verification at the downstream bug report and I will report back here and clear the NEEDINFO flag when/if I get some feedback. [1]: https://lists.ubuntu.com/archives/kernel-team/2010-May/010602.html [2]: https://bugs.launchpad.net/ubuntu/+source/xserver-xorg-video-intel/+bug/603064 [3]: http://bugs.freedesktop.org/show_bug.cgi?id=28964 [4]: video: apply the crtc box checks from dri. (http://cgit.freedesktop.org/xorg/driver/xf86-video-intel/commit/?id=272d1c14a39c32ade39b5a8b080a891f2b3d6e8e) Nicolò Chieffo, the original reporter has tested the 10.10 development version which is without this patch (and has a kernel based on 2.6.35-rc4), and the problem is gone. I'm marking this as FIXED. |
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.