open(/dev/dri/card0): No such file or directory — no radeon DRM device on this machine; trigger is hardware-gated.

--- arithmetic simulation (same guest) ---
VGT_STRMOUT_BUFFER_SIZE_0 = 0x40000000 DWs
kernel track->vgt_strmout_size = raw*4 (u32) = 0x00000000
kernel check: offset (0) > bo_size (4096)? NO -> CS ACCEPTED

GPU executes the register verbatim: streamout bound = raw*4 = 4294967296 bytes = 4 GiB of write space past the 4 KiB BO.

<<8 truncation: reg raw = 0x01000000
  kernel-tracked offset = (u32)(raw<<8) = 0x0000000000
  GPU base offset       = (u64)raw<<8    = 0x0100000000
  -> every CB/DB/htile/streamout BO-size check validates the
     WRONG offset; the GPU addresses 4294967296 bytes away from where
     the kernel believes.