aboutsummaryrefslogtreecommitdiffstats
path: root/hw/auc_len_test.py (follow)
Commit message (Collapse)AuthorAge
* seventy-fourth pass: test the proposed root cause on hardware and falsify ituser2026-09-30
The previous pass proposed that the wedge is caused by an unguarded length subtraction underflowing. Dispatcher analysis shows the worker's argument is the request's argument area, so the field in question is the host shared buffer at a known offset and the hypothesis was directly testable. The test tries predicted-safe values first, since a wedge costs a power cycle, and the very first one falsifies the hypothesis: a valid non-underflowing length wedges the part exactly as zero does. The unguarded subtraction remains a real code defect, since the guard tests one field while the arithmetic uses another, but it does not explain this hang and the finding is corrected so it cannot be cited as a memory-corruption primitive. A better unexplored candidate is the call the worker makes before the subtraction, which is a more likely place for a blocking wait. The card was recovered and verified healthy. The previous pass's closest-yet claim is withdrawn, since the alignment between a memory-safety defect and the hardware fault was coincidental and one experiment settled it. What stands is a reproducible host-triggered denial of service with unknown root cause, plus the separate well-verified defect family that is not host-reachable.