Comment 20 for bug 814325

Thanks for testing, hopefully the regression really is gone reverting the 3 patches. If you could find an way to reproduce the problem more easily, may be come and go many times in sequence from screen saver/blank would be great, problem is I think you'll get bored easily always having to check the corrupted screen many times in a row :)

The three remaining reverted patches in r2 were:
drm/i915: Fix gen6 (SNB) missed BLT ring interrupts.
drm/i915: disable PCH ports if needed when disabling a CRTC
drm/i915: fix user irq miss in BSD ring on g4x

We should identify now which one is introducing the problem. I took a deep look again at them, and compared now against your reported hardware (lspci). You have an "ironlake" vga. The first and last patches touches code paths only related to sandybridge or gm45/g45 hardware, which shouldn't affect/change your case. Hoping this isn't a timing/code shuffle issue, now I strongly believe we could have a problem with the remaining patch:

drm/i915: disable PCH ports if needed when disabling a CRTC

This was a backport for hardware enablement, and also it is missing a following fix to this change which is upstream but wasn't backported to natty: "drm/i915: DP_PIPE_ENABLED must check transcoder on CPT".

First lets try to double check and really see if "drm/i915: disable PCH ports if needed when disabling a CRTC" introduces your regression, now I built a new kernel only with it reverted, please try the package http://people.canonical.com/~herton/lp814325/r3/

If this one works, then next step is including also "drm/i915: DP_PIPE_ENABLED must check transcoder on CPT" on natty, and doing another test to see if issue is gone.