evidence · 6.5 KiB

README.md

Committed at docs/evidence/mainline-uart-progress-memory-printk/nohz-off-host/README.md. This is what grounds the requirements that cite it.

Open raw source

Same-image MemoryPrintk nohz-off controller

Source/fixture proof on 2026-10-04 UTC adds the typed --uart-progress-memory-printk-nohz-off child only. It requires minimal mode and all same-image/progress/Memory/no-stimulus/MemoryPrintk selectors; polling, point, clock and shutdown selectors or invalid types/modes reject before preparation, output creation or UART open. Base MemoryPrintk still rejects nohz additions unless this exact child is selected. Arbitrary/bare/empty/alternate/duplicate nohz/nohz_full and other argument edits fail exact reconstruction.

Actual qualification requires same-derivation realized dev/config and reviewed MemoryPrintk source/Image/archive gates, plus built-in NO_HZ_COMMON, NO_HZ_FULL, HIGH_RES_TIMERS, RISCV_TIMER, RISCV_SBI and HZ=250. It verifies unchanged config and Image hashes and one linked NUL-terminated nohz= setup. Unset legacy CONFIG_NO_HZ does not reject the actual supported image. Only trailing nohz=off is added to qualified MemoryPrintk, giving the planned 416→425-byte literal.

The observation parser is unchanged: exact fresh received args/backend, strict UMK framing/state rules and the narrowly qualified Readline prompt prefix. Candidate capture remains zero-input on success/error/finally, NOT_REQUESTED receipt, RX NOT_TESTED and 180-second passive observation. Only the independent ordered fresh normal-return boundary and protected postflight permit helper writes. No candidate reboot/retry, build, card/profile or source/kernel change.

Narrow proof with TMPDIR=$HOME/tmp:

python3 tests/test_mainline_uart_progress_memory_printk_nohz_off_controller.py
python3 tests/test_mainline_uart_progress_memory_printk_controller.py
python3 tests/test_mainline_uart_progress_memory_poll_idle_controller.py
python3 tests/test_mainline_uart_progress_memory_no_stimulus_controller.py
openspec validate the-board-runs-a-mainline-kernel --strict

13 new tests PASS (0.274s), 23 prior MemoryPrintk PASS, 14 polling PASS and 15 no-stimulus PASS. Strict validation/whitespace PASS. The new fixtures exercise actual configuration/Image checks, each missing/wrong/duplicate gate, hash/setup mismatch, exact sole transform and original transport rejection, pre-open typed conflicts, real rolling pump/split CRLF/strict records, unknown/read-error/overflow zero input and protected recovery separation. Full mocked transport executes the qualifier and all five loads/CRCs/printed exact arguments; no bytes follow bootm. The first development run exposed fixture-only argument-prefix and copied-name errors; final fixtures correct those without changing artifact/runtime semantics.

Actual existing-artifact proof for task 5m.2 passed on the first executed run. The safe receipt and exact executed qualifier record UTC timestamps, controller b99d972e, source/config/Image/archive/manifest identities and proof boundaries. Source peer review independently repeated all 13 focused fixtures PASS; root reviewed and repeated the affected UART pattern 143 and shell-PID1 14 tests PASS before landing the code.

TMPDIR="$HOME/tmp" python3 "$HOME/tmp/k230-mainline-uart-memory-printk-nohz-host-qualification/qualify.py" \
  --bundle /nix/store/p2kdar89q7dhwdajjy0mrhmxxzcgsalw-k230-mainline-drm-trial-boot-files \
  --dev /nix/store/24hbalyljs6gn6fzkkl24znv8a0x6jdl-linux-riscv64-unknown-linux-gnu-7.3.0-rc5-dev \
  --normal-report "$HOME/tmp/k230-mainline-uart-memory-printk-board/normal-report.json"

Execution requires the existing successful frozen artifact receipt 1c59f856, explicit exact bundle/dev and fresh outputs in the protected directory. Python 3.14 native Zstd, nix-store/nix offline queries, fdtget and sha256sum are host dependencies. The committed script is byte-identical to its executed private copy and retains that protected directory/worktree contract. It reads immutable artifacts and copies protected baseline/manifest privately; originals remain unchanged. No implicit realization or UART is performed.

The actual selected source worker is the reviewed 30e8eb1e variant under 0l4mgw9; Image SHA c2c9663b and config SHA 52e7470b match previous actual MemoryPrintk proof. Config has the required built-in gates/HZ=250 and unset legacy NO_HZ; unique linked nohz= offset is 19425910. Actual archive executables/common loader, kernel/dev, DT/chosen args, five load hashes/CRCs and manifest equal the previous positive proof. Bundle SHA256SUMS passes. Selected bootargs/transport are exactly baseline plus sole trailing nohz=off; five original mode transports remain equal to frozen fcd1020f. Unchanged preparation/observation/backend AST is compared only within its applicable function domain; the intentionally extended transport is checked through actual original values and sole token.

Private normal report, selected helper, manifest and transport stay outside the repository. The report is a historical host-only anchor, not a new recovery or preflight. Registration absence is asserted in the prepared pre/post helper; this host run does not claim a newly performed board assertion. No wrapper readback/copy or flash is added. No initial actual-artifact failure occurred. No UART/board or physical nohz-off trial was performed for this checkpoint. Task 5m.3 still needs NEW protected normal recovery, reviewed host gate and one reserved operator capture. Received token and linked support do not prove runtime tick policy, hardware timer IRQs, output-call return or cause. Ordinary root/touch and 5b.5 remain UNVERIFIED.

After NEW protected normal recovery and reviewed host gates, the sole operator may use fresh protected log/result paths:

python3 tools/mainline-drm-initrd-shell-trial.py --mode minimal \
  --same-image-shell-pid1 --uart-progress --uart-progress-memory \
  --uart-progress-memory-no-stimulus --uart-progress-memory-printk \
  --uart-progress-memory-printk-nohz-off \
  --bundle /nix/store/p2kdar89q7dhwdajjy0mrhmxxzcgsalw-k230-mainline-drm-trial-boot-files \
  --manifest PRIVATE_MANIFEST --normal-report PRIVATE_NORMAL \
  --log PRIVATE_LOG --result PRIVATE_RESULT

Coordinator independent review passed. The integration controller bytes match the executed qualifier source. A separate protected scratch execution of the byte-identical qualifier passed against the same actual artifacts; 143 UART controller tests and 14 shell-PID1 regression tests also passed. Independent host receipt preserves this additional host proof. It does not replace the still-pending NEW protected normal recovery or the physical nohz-off comparison.