AMD PSP reverse engineering, emulator and vulnerability research https://git.sl0p.foo/amd-psp-re.git/atom/hw/tmr_edge.py?h=main 2026-09-30T00:06:41Z 2026-09-30T00:06:41Z user user@clank 2026-09-30T00:06:41Z urn:sha1:10f3351528ca2e1cc96ce887b36763eb3279c5b0 The previous pass showed a command buffer inside the trusted region is rejected, and the classic failure mode is validating only the start address, which would let a buffer starting outside and extending into the region through so that any output written there would land in trusted memory. Testing the edges with controls at both ends shows a buffer starting below the region and extending into it is rejected, one starting inside and extending past the end is rejected, one starting inside is rejected, and an ordinary buffer and one starting exactly at the region end are both accepted. So the processor validates the whole range rather than merely the start, with correct end-exclusive boundary semantics, and both controls behave so the rejections are the check firing rather than a broken harness. This closes the most promising remaining route to a corruption primitive on the descriptor surface, since the obvious confused-deputy write into trusted memory is properly defended. The objective is not met.