Comment 179 for bug 541511

I'm trying to backport Daniel Vetters patches (http://cgit.freedesktop.org/~danvet/drm/log/?h=stuff/i8xx_cache_coherency_for_oga) to current lucids kernel (2.6.32-21.32). It seems to be not too complex as lucids kernel has applied drm of kernel 2.6.33.

The experimental result can be found there:
https://launchpad.net/~dnjl/+archive/experimental/+packages?field.name_filter=linux&field.status_filter=published&field.series_filter=lucid

Changes summary:

linux (2.6.32-21.32+i8xxkmsfix2~dnjl1) lucid; urgency=critical

  [ Daniel Baumann ]
  * Revert "SAUCE: i915 KMS -- support disabling KMS for known broken devices"
    - LP: #563277
  * Revert "SAUCE: i915 KMS -- blacklist i830"
    - LP: #542208, #563277
  * Revert "SAUCE: i915 KMS -- blacklist i845g"
    - LP: #541492, #563277
  * Revert "SAUCE: i915 KMS -- blacklist i855"
    - LP: #511001, #541511, #563277
  * x86, lib: Add wbinvd smp helpers (stolen from 2.6.34)
  * Applied patches done by Daniel Vetter to fix above problems:
    - agp/intel-gtt: fix i85x gtt chipset flush
    - agp/intel-gtt: extract mch buffer flush in i830 chipset flush
    - agp/intel-gtt: check cache-coherency on i830 class chipsets
    - drm/i915: add locking around chipset flush
    - agp/intel-gtt: steal the last gtt page
    - agp/intel-gtt: kill previous_size assignments
    - agp/intel-gtt: kill intel_i830_tlbflush
    - agp/intel: make intel-gtt.c into a real source file
    - agp/intel: split out gmch/gtt probe, part 2
    - agp/intel: split out gmch/gtt probe, part 1
    - drm/intel: kill mutli_gmch_chip
    - agp/intel: uncoditionally reconfigure driver on resume
    - agp/intel: split out the GTT support
    - agp/intel: introduce intel-agp.h header file

On my system (Dell Latitude X300 with 855GM) I'm running this kernel and the libdrm and xorg intel driver from xorg-edgers (https://launchpad.net/~xorg-edgers/+archive/ppa). So I'm able to get my system up and login without any kernel/module parameters.
But desktop effects are not working and it's not stable at all. But crashes would not hard freeze the whole system - only X crashs sometimes sporadic. So it's possible to grab the debugging data in sysfs.

Those who are interessted in, please test this. It's not perfect, I know, but its run somewhat.

Does anyone has any hints how to further this?