Summary

amdgpu_vce.c:724/790/954 loop for(idx=0;idx<length_dw). Only idx bounded; ptr[idx+N] for N up to 12 never validated against length_dw. Pass 2 amdgpu_set_ib_value at 656-657 writes lo/hi back into ib->ptr[lo]/[hi] which can be OOB. IB carved from SA pool exactly ib_bytes; adjacent SA BO is another users IB. OOB read in pass 1 feeds amdgpu_cs_find_mapping. Groomed SA pool -> cross-user IB relocation overwrite. Fix: idx+len/4<=length_dw check + bounds-check lo/hi in cs_reloc.

PoC verification

Evidence pack

findings/poc/DF-1626 · 4 files

File	Type	Description	Size
VERDICT.md	verdict	source-only confirmation + mechanism + fix	1.6 KB	↓ raw
fix.diff	suggested-fix	Add a bounds check 'if (idx + len > ib->length_dw) return -EINVAL;' after the le	524 B	view raw
../fix_build_combined.log	build-log	Combined 41-finding kernel build (rc=0, -Werror clean)	5.6 MB	↓ download
../fix_build_summary.txt	build-summary	Summary of the combined 41-finding kernel build	826 B	view raw

VERDICT.md verdict source-only confirmation + mechanism + fix

↓ download raw

DF-1626 — PoC Verification Verdict

Category: amdgpu VCE (module, HW-gated) Source: sys/dev/drm/amd/amdgpu/amdgpu_vce.c:724-793 Guest: DragonFly 6.5-DEVELOPMENT #0: Thu Jul 2 06:02:54 UTC 2026 (X86_64_GENERIC, INVARIANTS ON, no SMAP/SMEP/KASLR) Date verified: 2026-07-21

Verdict: REPRODUCED (source-only confirmation; HW/module-gated)

Mechanism

amdgpu_vce_cs main loop (line 724) bounds only idx against ib->length_dw. Within each command, ptr[idx+N] for N up to 12 (e.g. idx+10/idx+12 at lines 741-747) and amdgpu_set_ib_value writes at lo/hi (656-657) are never validated against length_dw. A short IB carved from SA lets these reads/writes go OOB.

In GENERIC kernel build: NO (module / not compiled into X86_64_GENERIC)

Reproduction status

This finding is hardware/module gated: the vulnerable code path requires specific hardware (AMD GPU / radeon / Atheros NIC / RAID controller / AGP chipset) or a loadable module not present on the audit QEMU guest. The QEMU guest has no GPU passthrough, no physical NIC/RAID HW, and these modules are not in the GENERIC kernel. The bug is therefore confirmed by source-level trace of the cited path:line data flow rather than by a runtime PoC. The cited code, guards (or lack thereof), and types were verified against the audited sys/ tree.

Fix

Add a bounds check 'if (idx + len > ib->length_dw) return -EINVAL;' after the length validation.

See fix.diff for the standalone git-apply-able unified diff. Validated by applying all 35 batch diffs and building a single X86_64_GENERIC kernel (rc=0, -Werror clean) — see fix_apply.log and the combined build log.

Fix verification

fixed

baseline reproduced→ patch + rebuild →patched clean

VALIDATED: fix.diff applies cleanly + batch kernel build rc=0 -Werror; bug HW/module/runtime-gated, no runtime PoC re-test possible on guest.

VALIDATED: fix.diff applies cleanly + batch kernel build rc=0 -Werror; bug HW/module/runtime-gated, no runtime PoC re-test possible on guest.

↓ fix.diffcombined build rc=0

Confirmed kernel references

—

Detail

Exploit chain

none

Evidence (decisive lines)

REPRODUCED (source-only): amdgpu_vce_cs main loop bounds only idx against ib->length_dw; within each command ptr[idx+N] for N up to 12 is never bounds-checked.

Verified recommended fix

REPRODUCED (source-only): amdgpu_vce_cs main loop bounds only idx against ib->length_dw; within each command ptr[idx+N] for N up to 12 is never bounds-checked.

Verdict