I wonder if I/we should just jump straight to the final step of resolving the generalised VM case (as well as the generalised software rendering case): bug 1118903
It might perform worse than this, and it might take a lot more effort than this, but it's something we have to do anyway. Once that's in place we could make mesa-kms fail to load if PRIME is unsupported and then Mir will fall back to the generalised software driver (which doesn't exist yet).
I wonder if I/we should just jump straight to the final step of resolving the generalised VM case (as well as the generalised software rendering case): bug 1118903
It might perform worse than this, and it might take a lot more effort than this, but it's something we have to do anyway. Once that's in place we could make mesa-kms fail to load if PRIME is unsupported and then Mir will fall back to the generalised software driver (which doesn't exist yet).