I can confirm that reverting to ffa305d0fc926418e4dff432381ead8907dc18d9 makes
it work, before it would fail before reaching the dm login page (I'm using
sddm), and now it reaches the login page. I also did a manual unpatching of
56d8dd6a9c03680700e0b0043cb56e0af7e3e3de when on the latest commit, which gave
me the same result.

The conflict looks to me like it's trying to initialize the device multiple
times - I did some debugging, and it reaches the amdgpu_device_initialize
function several times with the same virtual address. I'm not sure if this is
intentional or not, but it does seem a bit odd to me. Maybe there should be a
saved global state of vamgr's that corresponds to the virtual address? Or is it
just not supposed to initialize the same device multiple times?