Skip to content

Stress/gate v2 harness - #36

Merged
YUZHEthefool merged 3 commits into
mainfrom
stress/gate-v2-harness
Sep 1, 2026
Merged

Stress/gate v2 harness#36
YUZHEthefool merged 3 commits into
mainfrom
stress/gate-v2-harness

Conversation

@YUZHEthefool

@YUZHEthefool YUZHEthefool commented Sep 1, 2026

Copy link
Copy Markdown
Member

Problem and outcome

Related issue or audit finding

Related: #

Design and scope

Kernel safety and invariants

  • Trust boundaries and attacker-controlled inputs:
  • unsafe / MMIO / page-table / usercopy justification:
  • IRQ, preemption, lock ordering, and SMP behavior:
  • Fallible allocation, quotas, and rollback/commit ordering:
  • ABI/layout/syscall compatibility:
  • Fail-closed and error-path behavior:

Validation evidence

make fmt-check
make clippy
make lint
make build
make boot-check
make test

Additional focused checks:

Compatibility, performance, and rollout

Checklist

  • The change is focused, bisectable, and free of unrelated refactoring.
  • I added a regression test or explained why an automated test is not practical.
  • I ran the relevant formatting, lint, build, runtime, ABI, SMP, storage, or fuzz gates listed above.
  • New unsafe blocks have local SAFETY reasoning and a minimized unsafe scope.
  • I preserved IRQ/lock-order rules, bounded attacker-controlled work, and reserve-before-commit accounting.
  • I updated architecture, safety, testing, roadmap, or user documentation when behavior changed.
  • New fixtures/artifacts are minimal, reproducible, redistributable, and contain no confidential data.
  • Breaking behavior, residual risk, deferred validation, and untested hardware are called out explicitly.

@YUZHEthefool
fix(stress): repair the stress-v2 harness and add the V2 guest workload
62f8646 
The monthly stress gate has never passed end-to-end; its green history is
last-day-of-month skips (2 real runs, both failed). This commit lands the
verified harness repairs and the missing guest:

- scripts/stress_test.sh: split the 'local duration=$1 deadline=...' set -u
  crash in wait_soak_window, and stop QEMU's writable 'fat:rw:' ESP from
  mutating BOOTX64.EFI in the source tree (per-boot throwaway ESP copy +
  stale NvVars drop) - the corruption produced a false bad-bootloader
  diagnosis this round.
- userspace/stress_runner.c: new NILIX_STRESS_V2 guest at the path the
  Makefile already expects - NILSTR2 config parser with independent
  SHA-256, marker/sequencing state machine, and the memory/cpu/smp/
  process/combined profiles; block fail-closes until fsync exists.
  Config/header handshake proven on a booted kernel (digest matches the
  host's make-config output).
- docs/stress-gate-status.md: the honest gate record - failure history,
  reconstructed V2 contract, kernel gaps K1-K4, open decisions.
- docs/quality-gates.md: mark the stress soak gate NOT GREEN and link the
  status doc.

Kernel-side gaps (K1-K4) remain open and are tracked as ST-K1..ST-K4 in
plan v15.60; do not treat this gate as passing until a profile produces a
validated PASS+HEARTBEAT from a booted run.
@YUZHEthefool
chore(toolchain): pin rust-analyzer alongside the nightly channel
f1cae74 
The VS Code extension's bundled rust-analyzer refuses to index
nightly-2025-12-08 ('toolchain too old'); listing the component makes the
extension launch the version-matched rustup rust-analyzer instead.
@YUZHEthefool
feat(stress/diag): ST-K3 Phase D - tag all 8 sys_mmap ENOMEM sites an…
0ce7007 
…d emit guest cgroup diagnostics

Kernel (debug builds only, R159-17 gating style; no addresses logged):
- E1 max_map_count, E2 cgroup_charge (error variant bound and named),
  E3 phase1_admission, E4 prot_none_commit (annotated-unreachable - the
  tag exists to falsify the annotation), E5 frame_alloc, E6 map_page,
  E7 phase2_track_reserve (heap-admission class, distinct from E5),
  E8 phase3_commit (no-rollback hazard: a hit here also wedges every
  later fork via the stuck PENDING entry).

Guest (userspace/stress_runner.c):
- emit_mmap_diag() prints a NILIX_STK3_DIAG line (run-cgroup + root
  stats via syscall 516, tolerated-failure, rc values reported) just
  before the FAIL marker; the prefix is deliberately NOT
  NILIX_STRESS_V2_* because stress_protocol.py's line filter and
  fullmatch FAIL regex (detail must be a bare integer) would reject
  free-form detail text.
- report_mmap/shared_mmap/memory_mmap/combined_mmap fail sites now
  route through fail_with_stats(); errno is captured at argument time
  so the stats syscalls cannot clobber it, and the helper never calls
  fail() itself (would consume the failure_emitted latch).

Design: docs/review/design/st-k3-mmap-enomem-design.md (plan v15.60,
ST-K3 Phase D). Diagnosis run + family selection follow separately.
@YUZHEthefool
YUZHEthefool merged commit aedf6f4 into main Sep 1, 2026
6 checks passed
@github-actions github-actions Bot added documentation Improvements or additions to documentation area: process/ABI Processes, syscalls, usercopy, signals, ELF, and Linux ABI area: userspace Ring-3 programs, libc compatibility, and userspace support area: tests Unit, hosted, runtime, QEMU, SMP, stress, and conformance tests area: tooling/CI Build system, scripts, hooks, automation, and repository tooling dependencies Dependency or toolchain updates size: XL At least 1000 review lines, excluding generated/binary artifacts labels Sep 1, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

area: process/ABI Processes, syscalls, usercopy, signals, ELF, and Linux ABI area: tests Unit, hosted, runtime, QEMU, SMP, stress, and conformance tests area: tooling/CI Build system, scripts, hooks, automation, and repository tooling area: userspace Ring-3 programs, libc compatibility, and userspace support dependencies Dependency or toolchain updates documentation Improvements or additions to documentation size: XL At least 1000 review lines, excluding generated/binary artifacts

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@YUZHEthefool