On Mon Aug 3, 2026 at 4:37 PM +08, Peter Zijlstra wrote:
> I'm not sure this Changelog is convincing me. To me this just looks like
> more weight / bloat without benefit.
The overhead is negligible; like other TRACE_EVENTS this is an if
statement when disabled.

Knowledge of when an object is considered
end-of-usable-life by the refcount implementation is one of the
first steps in UAF debugging. Alternatives to obtain this
information (GDB, watchpoints, manual instrumentation)
require live analysis.

When debugging from static information of remote origin,
like syzbot, this may not be practical. Live analysis requires bug
reproduction and manual effort.

Bug reproduction, especially in rare timing-sensitive UAFs that
thus require many iterations, may require significant
time/hardware resources.

This tracepoint eliminates significant effort/time cost, and
removes a hardware barrier for a common debugging scenario at a
negligible performance cost.

Reply via email to