03:01imirkin: skeggsb: so with danvet's patch, we run into a BUG_ON because you can't have GEM objects that aren't a multiple of PAGE_SIZE
03:01imirkin: skeggsb: i don't see how this is working now though -- we have no such provisions that i can see around ensuring that the *size* is a multiple of PAGE_SIZE
03:04imirkin: we have a nouveau_bo_fixup_align() which takes care of it ...
03:05imirkin: but that is not called on the size that's passed to drm_gem_object_init...
03:17imirkin: btw, all this "do relocs" stuff seems overly complicated... it's saving a copy_from_user in an error case
03:39imirkin: oh i see. so most of the time, buffers don't move, and relocations don't have to be applied, so it saves a memcpy there... seems like a weak argument.
20:58imirkin: danvet: i think i lied
20:59imirkin: i think the base kernel is actually bogus too
20:59imirkin: i didn't actually test starting X :)
20:59imirkin: i just tested my thing, and i had already disabled X, and i didn't need X to test
20:59imirkin: tagr: 019cbd4a4feb3aa3a917d78e7110e3011bbff6d5
20:59danvet: imirkin, ah, makes at least more sense
20:59imirkin: i think this kills nouveau_gem_new
21:00imirkin: it used to do size alignment before calling drm_gem_object_init, but now it does after
21:01imirkin: danvet: will retest after i fix up the baseline
21:42danvet: imirkin, htx