[not-matthias]

TLDR: Argument and environment sizes move the initial guest stack pointer and change simulated cache misses. Reserve a fixed startup region to keep the stack pointer stable (Linux).

• Store pointer tables and original strings in a fixed 16 MiB region; reject overflow rather than shifting the stack. Preserve auxiliary data, including AT_RANDOM.
• Account for the region separately from the runtime stack limit.
• A frozen workload executable shows variation in all 47 workloads before the fix and exact equality across all 11 checked counters afterward. Changed-work and child-failure controls still behave as expected.
• Verified argument/environment preservation with none, Callgrind and Memcheck; exercised stack growth and stable mmap addresses.
• Local 12-case startup regression passes with the fix and fails on the rebuilt parent: a 15-byte argument moves the initial SP by 16 bytes. The full none suite retains the same 38 pre-existing stdout/stderr failure records, with no new failures.

This removes startup stack-address variation, not input-dependent work, randomness or scheduling nondeterminism.

Before/after variance matrix — 47 workloads

Same x86-64 Linux host and byte-identical workload executable; parent e1e4bdf3 versus stack-layout fix a5dc7701. Both Valgrind builds report 3.26.0.codspeed9; the parent was rebuilt cleanly. Each build runs 18 controlled process states × three repetitions: 54 samples per workload, 2,538 samples total. The states vary ignored environment values/counts, argument contents/counts and directory contents.

CV is population standard deviation / mean × 100 across these controlled states, not walltime variance or a speedup. D1mr/D1mw are L1 data read/write misses; DLmr/DLmw are last-level data read/write misses. Small miss counts can produce large relative CVs. The after column is the maximum CV across all 11 checked counters (Ir, Dr, Dw, I1mr, D1mr, D1mw, ILmr, DLmr, DLmw, Ct, Cl): every value is exactly invariant, not merely rounded to zero.

Benchmark	Before D1mr CV	Before D1mw CV	Before DLmr CV	Before DLmw CV	After max CV
expr_eval_recursive	0.16%	0.00%	0.16%	0.00%	0.00%
expr_accumulate_recursive	0.16%	3.22%	0.16%	3.22%	0.00%
insertion_sort_64	6.28%	3.70%	6.28%	3.70%	0.00%
merge_sort_recursive_64	4.20%	3.10%	4.20%	3.10%	0.00%
dense_stack_checksum_4k	1.25%	14.39%	1.25%	14.39%	0.00%
strided_static_table_sum	1.23%	14.39%	1.23%	14.39%	0.00%
heap_slice_sum_control	0.32%	14.39%	0.32%	14.39%	0.00%
register_arithmetic_control	34.02%	18.33%	34.02%	18.33%	0.00%
sparse_64b_d08	23.39%	2.63%	23.39%	2.63%	0.00%
sparse_80b_d08	23.39%	2.34%	23.39%	2.34%	0.00%
dense_64b_d08	23.39%	2.63%	23.39%	2.63%	0.00%
dense_80b_d08	23.39%	2.34%	23.39%	2.34%	0.00%
aligned_64b_d08	23.39%	1.53%	23.39%	1.53%	0.00%
aligned_80b_d08	23.39%	1.78%	23.39%	1.78%	0.00%
sparse_64b_d32	23.39%	0.72%	23.39%	0.72%	0.00%
sparse_80b_d32	23.39%	0.63%	23.39%	0.63%	0.00%
dense_64b_d32	23.39%	0.72%	23.39%	0.72%	0.00%
dense_80b_d32	23.39%	0.63%	23.39%	0.63%	0.00%
aligned_64b_d32	23.39%	0.42%	23.39%	0.42%	0.00%
aligned_80b_d32	23.39%	0.49%	23.39%	0.49%	0.00%
sparse_64b_d64	23.39%	0.37%	23.39%	0.37%	0.00%
sparse_80b_d64	23.39%	0.32%	23.39%	0.32%	0.00%
dense_64b_d64	23.39%	0.37%	23.39%	0.37%	0.00%
dense_80b_d64	23.39%	0.32%	23.39%	0.32%	0.00%
aligned_64b_d64	23.39%	0.21%	23.39%	0.21%	0.00%
aligned_80b_d64	23.39%	0.25%	23.39%	0.25%	0.00%
sparse_784b_d08	23.39%	3.98%	23.39%	3.98%	0.00%
sparse_800b_d08	23.39%	3.98%	23.39%	3.98%	0.00%
dense_784b_d08	23.39%	0.39%	23.39%	0.39%	0.00%
aligned_784b_d08	23.39%	1.78%	23.39%	1.78%	0.00%
spread_784b_d08	23.39%	11.61%	23.39%	11.61%	0.00%
spread_800b_d08	23.39%	3.03%	23.39%	3.03%	0.00%
aligned_spread_784b_d08	23.39%	1.38%	23.39%	1.38%	0.00%
sparse_784b_d32	23.39%	1.14%	23.39%	1.14%	0.00%
sparse_800b_d32	23.39%	1.14%	23.39%	1.14%	0.00%
dense_784b_d32	23.39%	0.10%	23.39%	0.10%	0.00%
aligned_784b_d32	11.00%	0.49%	23.39%	0.49%	0.00%
spread_784b_d32	23.39%	13.98%	23.39%	13.98%	0.00%
spread_800b_d32	23.39%	0.84%	23.39%	0.84%	0.00%
aligned_spread_784b_d32	11.00%	0.37%	23.39%	0.37%	0.00%
sparse_784b_d64	23.39%	0.58%	23.39%	0.58%	0.00%
sparse_800b_d64	23.39%	0.58%	23.39%	0.58%	0.00%
dense_784b_d64	27.16%	1.41%	23.39%	0.05%	0.00%
aligned_784b_d64	1.16%	0.21%	23.39%	0.25%	0.00%
spread_784b_d64	23.39%	14.46%	23.39%	14.46%	0.00%
spread_800b_d64	23.39%	0.43%	23.39%	0.43%	0.00%
aligned_spread_784b_d64	1.16%	0.17%	23.39%	0.19%	0.00%

Workload executable SHA-256: cd9d44401bb329cb33d54463ecea22d6ebb1b6517d67dd948f1804b91aa921cc.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The PR appears safe to merge; no actionable new issue was established.

Summary

Keeps the Linux initial stack pointer at a fixed offset from the stack top.

• Linux clients now start with a fixed stack pointer.

The supplied previous threads have no Comment numbers, so they have no numbered entries in previousFindings. Both concerns are addressed by the current code.

The previous reviewed commit is not an ancestor of the head, so this review used the full PR diff. No tests were executed.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Choose runtime stack size] --> B[Add 8 MiB startup allowance]
  B --> C[Place initial SP below startup region]
  C --> D[Write argument and environment tables above SP]
  C --> E[Place strings near stack top]
  C --> F[Reserve runtime stack below SP]

Loading

_{Reviews (4) · Last reviewed commit: "fix(initimg): stabilize Linux startup st..." · Reviewed by Greptile}

[codspeed]

Merging this PR will not alter performance

✅ 84 untouched benchmarks
⏩ 60 skipped benchmarks¹

---

_{Comparing cod-3934-define-fixed-buffer-for-environment-variables-inside-process (921efb9) with master (eadd52e)}

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. ↩

[GuillaumeLagrange]

olgtm, let's wait for uv bench variance results before merging to check if we need additionnal changes here