Add step to lower ASLR bits to 28 when running libc++ sanitizer builds.

[EricWF]

The libc++ sanitizer bots have been broken for a while. This appears to be related to ASLR, since it only occurs when vm.mmap_rnd_bits > 28.

This change attempts to address this issue by lowing the ASLR bits to 28 (though allegedly LLVM 21 supports ASLR up to 32 bits?).

[llvmbot]

@llvm/pr-subscribers-libcxx

@llvm/pr-subscribers-github-workflow

Author: Eric (EricWF)

Changes

The libc++ sanitizer bots have been broken for a while. This appears to be related to ASLR, since it only occurs when vm.mmap_rnd_bits > 28.

This change attempts to address this issue by lowing the ASLR bits to 28 (though allegedly LLVM 21 supports ASLR up to 32 bits?).

---

Full diff: https://github.com/llvm/llvm-project/pull/146273.diff

1 Files Affected:

• (modified) .github/workflows/libcxx-build-and-test.yaml (+4)

diff --git a/.github/workflows/libcxx-build-and-test.yaml b/.github/workflows/libcxx-build-and-test.yaml
index f0bdf6c0b5899..da8f8848c1e25 100644
--- a/.github/workflows/libcxx-build-and-test.yaml
+++ b/.github/workflows/libcxx-build-and-test.yaml
@@ -166,6 +166,10 @@ jobs:
     container: ghcr.io/llvm/libcxx-linux-builder:2b57ebb50b6d418e70382e655feaa619b558e254
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
+
+      - name: Reduce ASLR for Sanitizers
+        if: ${{ matrix.config == 'generic-tsan' || matrix.config  == 'generic-msan' }}
+        run: sudo sysctl vm.mmap_rnd_bits=28; sudo sysctl vm.mmap_rnd_bits;
       - name: ${{ matrix.config }}
         run: libcxx/utils/ci/run-buildbot ${{ matrix.config }}
         env:

[EricWF]

OK, so this doesn't work.

It turns out that half of the clusters are running on ubuntu-22.04, and the other ones are on ubuntu-24, the latter have 32 bits of randomization, the former 28.

I don't believe I can downgrade the clusters on Ubuntu 24, and the clusters still on version 22 will likely be auto-upgraded soon, so that's unfortunate.

I've tried setting the sysctl settings on the cluster using the instructions here, but they don't work.

[EricWF]

The fix ended up being making the changes on the node.