Thanks. Judging from the Oops, it looks like that NULL pointer is passed to ttm->pages in radeon_gart_bind() that is called in radeon_ttm_backend_bind(). It's likely the bad side-effect of the recent refactoring of TTM stuff. It tried to change the SG stuff but radeon has still the access to the old page array. Could you open a bug report on the upstream tracker, gitlab.freedesktop.org Issues?