[Valgrind-developers] Proposed VALGRIND_TRY_OPTS and division patches

[D. J. B. <aut...@bo...>]

Attached are the latest versions of valgrind patches stemming from the
paper "KyberSlash: Exploiting secret-dependent division timings in Kyber
implementations". The paper will appear at CHES 2025 next month; CHES is
a highly ranked conference with many software engineers in the audience,
so the talk would be a good moment to report the latest valgrind status:

    https://ches.iacr.org/2025/program.php
    http://jianying.space/conference-ranking.html

These two patches apply to 112f1080b7c21e37dfce0a2e589d0dc7aa115afa
cleanly. I ran "ulimit -n 8192; make regtest" on Zen 2 under Debian 12
before and after the patches. Before the patches, there were 947 tests
with the following failures:

    memcheck/tests/x86-linux/scalar          (stderr)
    memcheck/tests/x86-linux/scalar_exit_group (stderr)
    memcheck/tests/x86-linux/scalar_supp     (stderr)
    memcheck/tests/x86-linux/shm             (stderr)
    none/tests/x86-linux/sigcontext          (stdout)

After the patches: 26 new successes; no new failures.

I've also attached patch-overview.txt with an updated description of the
rationale for, and status of, the patches. I believe all feedback has
been handled. Happy to address any further questions.

---D. J. Bernstein