fix: hangs on ARM64 due to atomics

[not-matthias]

This reverts commit e0e62db which was previously thought to be an issue in the stress-ng utility, but turned out to be a valgrind bug.

[greptile-apps]

Greptile Summary

This PR fixes intermittent ARM64 hangs in callgrind CI by extending the Cortex-A72 (implementer 0x41, part 0xd08) detection in /proc/cpuinfo to auto-enable the fallback LL/SC implementation, alongside the existing Cavium/ThunderX check. Under Valgrind instrumentation, the LDAXR..STXR window spans two translated superblocks with hundreds of host instructions, so the hardware exclusive monitor cannot be relied on for forward progress.

• coregrind/m_machine.c: Adds Cortex-A72 identification to VG_(parse_cpuinfo), setting arm64_requires_fallback_LLSC when both 0x41 (ARM Ltd implementer) and 0xd08 (A72 part) appear in /proc/cpuinfo; the comment correctly notes that a false positive only enables the safe fallback.
• bench/testdata/llsc_tzconvert_bench.c (Git LFS): New benchmark fixture exercising the localtime()/__tz_convert() call path that triggered the livelock; compiled from source in CI.
• bench/generate_config.py / .github/workflows/codspeed.yml: Adds stress-ng and the new fixture to the benchmark suite with a CI compile step.

Confidence Score: 5/5

Safe to merge. The fix is a targeted extension of an existing, well-tested detection path, and the fallback it enables is the conservative direction.

The Cortex-A72 detection is correctly guarded by requiring both the implementer and part strings, matches the same proven pattern used for Cavium, and the comment explicitly documents that a false positive is safe. The new benchmark fixture is compiled from source in CI after an LFS-aware checkout, so there is no risk of a stale binary. No existing logic is removed or weakened.

No files require special attention.

Important Files Changed

Filename	Overview
coregrind/m_machine.c	Extends parse_cpuinfo to detect Cortex-A72 (0x41/0xd08) alongside Cavium; false-positive-safe and well-commented.
bench/testdata/llsc_tzconvert_bench.c	New Git LFS file; CI checks out with lfs:true so the compile step receives the actual C source.
bench/generate_config.py	Adds stress-ng and llsc_tzconvert_bench commands; follows existing relative-path convention.
.github/workflows/codspeed.yml	Adds compile step for new C fixture; source and libc are both available unconditionally.
.gitignore	Correctly ignores compiled fixture binary and Python bytecode cache.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[VG_machine_get_hwcaps] --> B[VG_parse_cpuinfo]
    B --> C[Read /proc/cpuinfo]
    C --> D{Cavium?\nimplementer 0x43}
    D -->|yes| E[arm64_requires_fallback_LLSC = True]
    D -->|no| F{ARM Ltd?\nimplementer 0x41}
    F -->|no| G[return True - no fallback]
    F -->|yes| H{Cortex-A72?\npart 0xd08}
    H -->|no| G
    H -->|yes| E
    E --> I[return True - fallback enabled]
    I --> J[Guest LDAXR/STXR via host CAS]
    G --> K[Guest LDAXR/STXR via hardware monitor]

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A[VG_machine_get_hwcaps] --> B[VG_parse_cpuinfo]
    B --> C[Read /proc/cpuinfo]
    C --> D{Cavium?\nimplementer 0x43}
    D -->|yes| E[arm64_requires_fallback_LLSC = True]
    D -->|no| F{ARM Ltd?\nimplementer 0x41}
    F -->|no| G[return True - no fallback]
    F -->|yes| H{Cortex-A72?\npart 0xd08}
    H -->|no| G
    H -->|yes| E
    E --> I[return True - fallback enabled]
    I --> J[Guest LDAXR/STXR via host CAS]
    G --> K[Guest LDAXR/STXR via hardware monitor]

Loading

_{Reviews (5): Last reviewed commit: "test(bench): add llsc_tzconvert_bench fi..." | Re-trigger Greptile}

[codspeed-hq]

Merging this PR will not alter performance

⚡ 11 improved benchmarks
❌ 2 regressed benchmarks
✅ 57 untouched benchmarks
🆕 14 new benchmarks
⏩ 60 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	test_valgrind[valgrind-3.25.1, python3 testdata/test.py, no-inline]	4.4 s	337.4 s	-98.68%
❌	test_valgrind[valgrind-3.25.1, python3 testdata/test.py, full-with-inline]	7.1 s	8.4 s	-15.27%
⚡	test_valgrind[valgrind.codspeed, stress-ng --cpu 4 --cpu-ops 10, full-no-inline]	20 s	2.9 s	×6.8
⚡	test_valgrind[valgrind.codspeed, stress-ng --cpu 4 --cpu-ops 10, full-with-inline-with-cycle-estimation]	14.6 s	3.5 s	×4.2
⚡	test_valgrind[valgrind.codspeed, stress-ng --cpu 1 --cpu-ops 10, no-inline]	4.4 s	3.7 s	+17.19%
⚡	test_valgrind[valgrind.codspeed, stress-ng --cpu 4 --cpu-ops 10, inline]	2.3 s	2 s	+15.98%
⚡	test_valgrind[valgrind.codspeed, stress-ng --cpu 1 --cpu-ops 10, cycle-estimation]	4.8 s	4.2 s	+15.75%
⚡	test_valgrind[valgrind.codspeed, stress-ng --cpu 4 --cpu-ops 10, cycle-estimation]	2.3 s	2 s	+15.48%
⚡	test_valgrind[valgrind.codspeed, stress-ng --cpu 4 --cpu-ops 10, no-inline]	2.1 s	1.8 s	+15.14%
⚡	test_valgrind[valgrind-3.26.0, python3 testdata/test.py, full-with-inline]	8.3 s	7.2 s	+14.85%
⚡	test_valgrind[valgrind.codspeed, stress-ng --cpu 4 --cpu-ops 10, full-with-inline]	3.6 s	3.2 s	+14.11%
⚡	test_valgrind[valgrind.codspeed, stress-ng --cpu 1 --cpu-ops 10, full-with-inline]	8.8 s	7.8 s	+13.06%
⚡	test_valgrind[valgrind.codspeed, stress-ng --cpu 1 --cpu-ops 10, full-with-inline-with-cycle-estimation]	9.5 s	8.5 s	+12.15%
🆕	test_valgrind[valgrind-3.25.1, testdata/llsc_tzconvert_bench 5000, full-no-inline]	N/A	1 s	N/A
🆕	test_valgrind[valgrind-3.25.1, testdata/llsc_tzconvert_bench 5000, full-with-inline]	N/A	1.2 s	N/A
🆕	test_valgrind[valgrind-3.25.1, testdata/llsc_tzconvert_bench 5000, inline]	N/A	887.8 ms	N/A
🆕	test_valgrind[valgrind-3.25.1, testdata/llsc_tzconvert_bench 5000, no-inline]	N/A	711.1 ms	N/A
🆕	test_valgrind[valgrind.codspeed, testdata/llsc_tzconvert_bench 5000, cycle-estimation]	N/A	797.5 ms	N/A
🆕	test_valgrind[valgrind.codspeed, testdata/llsc_tzconvert_bench 5000, full-no-inline]	N/A	1 s	N/A
🆕	test_valgrind[valgrind.codspeed, testdata/llsc_tzconvert_bench 5000, full-with-inline-with-cycle-estimation]	N/A	1.3 s	N/A
...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing cod-3103-valgrind-atomics-have-incorrect-time-cause-hangs (b738cff) with master (8a82521)²}

Footnotes

1. 60 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

2. No successful run was found on master (b738cff) during the generation of this report, so 8a82521 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩