IOW, I'm inclined to just do the revert. The "fix" clearly breaks things, and now you're adding random parts of the function back rather than just calling the "fill_fix()" function like things used to. Why?
Its not the final patch. I'm trying to get feedback on what broke exactly.
The commit message in dfe63bb0ad98 doesn't support any of these hacks
- it just seems to say that drm_fb_helper_fill_fix() should also have
been called from setcolreg().
So why don't we just revert the commit and instead add that drm_fb_helper_fill_fix() to setcolreg()?
Fine by me. The real problem is the nouveau driver is using the framebuffer device (fb_fillrect) before set_par is ever called. Before calling set_par the framebuffer state is not defined.