i915 DRM problems in lucid.

[Andy Whitcroft]

On Sat, Mar 27, 2010 at 05:15:09PM +0100, Robert Hooker wrote:
> Hello, I've found a lot of bug fixes for i915 that seem needed in the
> lucid kernel and I thought I'd list them here for consideration since
> I'm not having any luck in IRC.
> 
> 
> * GPU hangs after DPMS events
> *Quite* a lot of reports of this bug, the fix was included in
> 2.6.32-15 but was dropped during the 2.6.33 DRM merge that happened in
> 2.6.32-16. Due to the number of people it affects it seems pretty
> urgent to bring this back in. Symptoms include hangs after lid close,
> idle screen blanking, and suspend/resume. dmesg and old xorg/gdm logs
> will show batchbuffer I/O errors when it crashed, EIR is always
> 00000000 and it eventually ends up with an assert like this happening
> - X: ../../src/i830_batchbuffer.h:79: intel_batch_emit_dword:
> Assertion `pI830->batch_ptr != ((void *)0)' failed.
> 
> Bug: https://launchpad.net/bugs/535640
> Fix: http://kernel.ubuntu.com/git?p=ubuntu/ubuntu-lucid.git;a=commit;h=7b56712ff524ee55e38afaee3954d125f56a6070

Ok, this one is restored in the 2.6.32-19.28 kernel.  Thanks for the
heads up.

> * Framebuffer compression causing rendering to stop on 915 and 945
> mobile devices after resume.
> This one was brought up on the list previously, but it is pretty nasty
> and affects all mobile 915-945 users. I submitted my patch upstream
> and they applied it and I will send it to stable once it's in linus'
> tree but that will probably be awhile with how rarely the intel tree
> is pulled in lately.
> 
> Bug: https://bugs.edge.launchpad.net/ubuntu/+source/linux/+bug/492392
> Fix: https://patchwork.kernel.org/patch/86999/

This one was also pulled in for the 2.6.32-19.28 kernel.

> * i915 module getting loaded before agpgart
> This has been an issue since karmic at least, and is also happening
> with radeon. Intel has fixed this in 2.6.34 but it hasn't been
> backported to stable yet. This especially seems like a backport worth
> considering due to how fast boots are now. Side note about this:
> upstream might even actually require AGP modules to be built into the
> kernel at some point, the consensus seems to be module AGP=bad.
> 
> Bug: https://bugs.launchpad.net/bugs/542251
> Fix: http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=1f7a6e372e9cb4d749f34c0738d832e6cadb4071

I thought we had a work around for this.  We have a couple of commits
relating to the MODULE_IMPORT macro which Tim added which should trigger
preloading of the appropriate AGP module for i915 at least.

    UBUNTU: SAUCE: Use MODULE_IMPORT macro to tie intel_agp to i915

I suspect we lost part of tha in the drm33 update.  Will investigate.

> * Large number of reports requiring lid quirks
> Upstream has decided to drop the requirement of having a lid open
> status to enable the LVDS completely because a large number of quirks
> were getting sent, a lot of manufacturers screw this up because
> Windows/OSX don't require it.
> 
> Fix: http://git.kernel.org/?p=linux/kernel/git/anholt/drm-intel.git;a=commit;h=6e6c822868f113dabe3c33bdd91e883cc28fa11b

This one was also pulled into -19.28 but is cause regressions in the
other direction so we will need to work out how to deal wit this one.
Surbhi is looking at the fallout and we will use that to form a plan.

-apw