At Sat, 8 Dec 2012 13:01:29 +0100, Daniel Vetter wrote: > > On Fri, Dec 07, 2012 at 07:07:32PM +0100, Daniel Vetter wrote: > > On Fri, Dec 7, 2012 at 7:05 PM, Daniel Vetter <daniel.vetter at ffwll.ch> wrote: > > >> I tried to move the w/a to enable code, but also didn't help. > > >> > > >> So... your guess appears really feasible. > > >> Then this is a side-effect of the new modeset code by optimizing the > > >> enable/disable step? > > > > > > Nope, the commit you've cited is way past the modeset-rework merge. > > > That really just moved the enable/disable points of the w/a bit around > > > to the supposedly correct places. Also, if this w/a fails you're > > > supposed to get a black screen with stuck cpu-side pipe, not a display > > > shifted down a few lines. So something strange seems to be going on. > > > > Can you please check whether this is a dual-link lvds machine? Some > > other w/a would apply in this case. > > Also please retest with latest drm-intel-nightly and the following patch: > > https://patchwork.kernel.org/patch/1852411/ Tried that, but didn't help, unfortunately. > Otherwise we might be hitting fallout from know issues. Looks so... Takashi
