In #intel-gfx, ajax confirmed that this is probably a bug in the 915(GM) part
of the driver. Unfortunately Intel has not yet released the gen3 documentation,
so for non-Intel employees it will be hard to fix this issue. Comparing
register dumps of a working and non-working Linux kernel would be one option,
but unfortunately as seen in bug 26294 I could not find a working one, although
I think it is a regression.

In the meantime ajax asked me if vesa works (boot with `nomodeset`) and I will
try to get back with this information as soon as possible.