Skip to content- Sources: project
- Summary: VUSec describes Branch Target Reuse as a class in which the branch predictor retains targets from code that has since been rewritten, so memory freed and reused by a just-in-time compiler exposes the earlier targets. The demonstrated exploit reaches kernel memory through unprivileged classic BPF, the filter layer used by Linux socket filtering, seccomp, Docker, and Chrome. VUSec states that no CPU it tested, across Intel, AMD, and Arm, keeps the branch predictor in sync with self-modified code. Two CVEs cover the upstreamed x86 mitigation, CVE-2026-64507 for an IBPB flush on BPF JIT allocation and CVE-2026-64508 for hardening against JIT spraying. The page gives no affected or fixed kernel version ranges. It is dated 2026-09-25 and records a modification on 2026-09-29 to its mitigation section, and it carries no Hacker News thread, so the only source here is the project page itself.
- Why it matters: Patching is the mitigation rather than configuration, and the shipped fixes sit at different layers: Linux issues an IBPB on all cores when a classic BPF program reuses a previously executed cBPF or eBPF region, Oracle randomizes JIT code-cache locations in GraalVM, which is a language runtime rather than a kernel path, and Mozilla has not shipped an IBPB mitigation, so Firefox users rely on site isolation that is not yet complete.
- Follow-up: Track a Mozilla IBPB mitigation, and whether JIT-bearing runtimes other than classic BPF are shown exploitable through the same class.