12:49 fdobridge: <m​arysaka> I got a bit lost but this seems wrong https://gitlab.freedesktop.org/nouveau/mesa/-/blob/nvk/main/src/nouveau/winsys/nouveau_bo.c#L53
12:50 fdobridge: <k​arolherbst🐧🦀> in the gallium driver there was `NOUVEAU_BO_CONTIG` to not do that
12:50 fdobridge: <m​arysaka> that seems to make everything have ``NOUVEAU_GEM_DOMAIN_MAPPABLE``
12:50 fdobridge: <k​arolherbst🐧🦀> huh?
12:51 fdobridge: <m​arysaka> ``NOUVEAU_GEM_TILE_NONCONTIG = 8``
12:51 fdobridge: <m​arysaka> ``NOUVEAU_GEM_DOMAIN_MAPPABLE = 1 << 3``
12:51 fdobridge: <m​arysaka> and it's set on the same field here
12:51 fdobridge: <k​arolherbst🐧🦀> "fun"
12:51 fdobridge: <m​arysaka> I'm not to familiar with this flag but shouldn't this be set in tile_mode?
12:52 fdobridge: <k​arolherbst🐧🦀> yeah.. let me check
12:54 fdobridge: <k​arolherbst🐧🦀> ahh yeah.. sounds aobut right
12:54 fdobridge: <k​arolherbst🐧🦀> do you want to fix it up?
12:55 fdobridge: <m​arysaka> sure I can do MR if you want
12:55 fdobridge: <k​arolherbst🐧🦀> yeah, would be perfect
13:00 fdobridge: <m​arysaka> looking at libdrm side it's actually in tile_flags oops
13:01 fdobridge: <k​arolherbst🐧🦀> but now I'm a little bit worried.. let's see..
13:02 fdobridge: <k​arolherbst🐧🦀> the gallium driver allocates bos for images without `NOUVEAU_GEM_TILE_NONCONTIG` set, so either we just drop that flag or figure out how to deal with that for images...
13:02 fdobridge: <k​arolherbst🐧🦀> maybe set it only if tile_mode is 0? no idea
13:03 fdobridge: <k​arolherbst🐧🦀> uhm.. actually only for cursors and scanout buffers
13:08 fdobridge: <m​arysaka> I will remove it for now
13:09 fdobridge: <m​arysaka> Was mostly trying to get mapping of BO working on some python scripts I have around
13:13 fdobridge: <k​arolherbst🐧🦀> yeah... soo.. mapping BOs is currently a mess
13:13 fdobridge: <k​arolherbst🐧🦀> we can't really map VRAM for reading
13:14 fdobridge: <k​arolherbst🐧🦀> we can write to it tho
13:14 fdobridge: <k​arolherbst🐧🦀> there is a little trick tho. You can get the actual VRAM bo content if you manage to tell the kernel driver to evict all VRAM to sysmem (like e.g. for preparing system suspend or runtime suspend)