So it seems there definitely is a regression on 4.17 on this issue (the patches are not required as the lines were already there in 4.17). This time it isn't a kernel panic, but an "invalid opcode" error caused by dm_update_crtcs_state that was called by amdgpu_dm_atomic_check. The issue is not 100% reproducible, but still means going to standby with an AMD GPU and DC is still dangerous and may result in losses of unsaved work.