cmp170hx_check

Repo: https://git.qbuissondebon.eu.org/qbuissondebon/cmp170hx_check (SSH push: git@git.ssh.qbuissondebon.eu.org:qbuissondebon/cmp170hx_check.git)

Single-card checker for NVIDIA CMP 170HX cards (PCI IDs 10de:20c2 / 10de:2082) — run it one card at a time: it waits for the card, verifies the PCIe capacitor mod, records PCIe bus health, backs up VBIOS, runs the cmpunlocker verification, offers the opt-in cmpunlocker unlock install (then a cold reboot and a re-run for the final PCIe check), and optionally stresses VRAM. Each card gets its own folder + JSON report named by GPU UUID, so the folders accumulate as you go through your batch (e.g. 23 cards).

Reference harness (auto-cloned into harness/ by preflight):

• 170th-Street — community docs, incl. modifications/pcie-capacitor-mod.md (the pass criteria: LnkSta Width x16 after the mod, x4 when missing)
• cmpunlocker — verify.sh is executed read-only per card; install.sh is only run from the separate, opt-in unlock step (see Safety)

Requirements

• Linux x86-64 (tested target: Ubuntu 26.04), Python 3.8+
• sudo privileges — precondition for --preflight package installation, for privileged checks (PCI rescan, link retrain, sysfs ROM, dmesg, verify.sh), and for the opt-in unlock install
• pciutils (lspci, setpci), git — installed automatically by preflight
• NVIDIA driver with nvidia-smi (nvidia-open, version list in cmpunlocker driver/VERSION — 610.xx / 615.xx) for unlock verification — preflight can install it for you (asks first)
• nvflash — bundled in this repo (nvflash_5.867_linux/x64/nvflash), used directly by the script; no system install needed (see nvflash)
• Optional: vkmark or vulkan-tools for the 5-minute VRAM stress test

Preflight (setup step)

python3 check_cards.py --preflight

On Ubuntu 26.04 this:

1. Checks the sudo precondition — runs sudo -v if needed so credentials are cached; if sudo is unavailable it prints the fix and does not install anything (exit code 3 if blocking tools are still missing)
2. Installs required packages via apt-get update && apt-get install: git, pciutils, vulkan-tools, vkmark
3. Offers to install the NVIDIA driver (ubuntu-drivers install, falling back to the nvidia-driver metapackage) when nvidia-smi is missing — reboot afterwards
4. Clones the two harness repos into harness/
5. Prints a tool status table — the bundled nvflash is detected automatically and its path is shown

Skip installation with --no-install (no sudo needed at all).

nvflash (bundled in this repo)

The Linux build of NVIDIA's nvflash 5.867 is committed to this repo at nvflash_5.867_linux/x64/nvflash (plus aarch64/ppc64/x86 variants) — the script uses it directly, no system-wide installation required.

Binary resolution order:

1. --nvflash /path/to/nvflash (explicit override)
2. repo copy: nvflash_*_linux/<arch>/nvflash, then ./nvflash
3. nvflash on PATH (optional system install)

Preflight reports nvflash OK and prints the resolved path.

python3 check_cards.py --preflight   # shows: nvflash OK / nvflash: <repo>/nvflash_5.867_linux/x64/nvflash

Notes:

• nvflash requires root: the script invokes it via sudo — cache credentials with sudo -v first, or run the script as root.
• Only --save (read-only backup) is used; if it fails while the NVIDIA driver is bound, the script automatically falls back to the sysfs ROM method.
• Do not use the AUR nvflash package (known broken), and ignore Windows .exe builds.

Optional — install it system-wide anyway:

sudo install -m 755 nvflash_5.867_linux/x64/nvflash /usr/local/bin/nvflash
nvflash --version
sudo nvflash --list        # must show the 170HX

Refreshing the bundled copy: download nvflash_5.867_linux.zip from https://www.techpowerup.com/download/nvidia-nvflash/ , verify SHA256 73ab51e121129059f1377b3bf21a8001f6649d64af2471455499c803a9c80301, extract and replace the files (keep the executable bit: chmod +x).

Usage (one card per run)

python3 check_cards.py               # check the connected card
python3 check_cards.py --preflight   # setup: tools, packages, harness clone
python3 check_cards.py --demo        # self-test with fake hardware (no GPU needed)
python3 check_cards.py --unlock      # check + run the cmpunlocker install without asking
python3 check_cards.py --help        # all flags

Flow of a normal run:

1. Waits for a CMP 170HX on the bus — prompts you to insert/reseat the card, Enter triggers a PCI rescan (q quits)
2. Folder cards/<GPU-UUID>/ from the auto-read GPU UUID (nvidia-smi); physical serial probed from nvflash and saved to serial.txt (N/A when the card exposes none) — never prompts
3. VBIOS backup → cards/<GPU-UUID>/bios.rom (nvflash, sysfs ROM fallback). On failure the script reports the precise cause (driver-loaded vs missing rom node) and offers an opt-in driver unload → save → reload retry (Unload NVIDIA driver and retry nvflash --save now? (y/N), default n; --bios-unload runs it without asking)
4. PCIe checks: link speed/width, AER counters, full config-space dump
5. Link retrain test + full bus rescan test (device must survive both)
6. Kernel log scrape (dmesg/journalctl) for this BDF: Xid / AER / NVRM lines
7. nvidia-smi snapshot + harness/cmpunlocker/verify.sh (read-only)
8. Opt-in unlock: cmpunlocker unlock (install) — skipped with a note when memory is already unlocked, otherwise asks Run cmpunlocker install.sh now? (y/N) (default n; --unlock runs it without asking). On yes: git pull --ff-only on the harness clone, then install.sh runs from harness/cmpunlocker/ with live terminal output (full log → cards/<GPU-UUID>/install_unlocker.log; the kernel-module build takes 5–15 min)
9. After a successful install: the VRAM stress is skipped — cold reboot the host (sudo shutdown -h now, then power it back on) and re-run the same command for the final PCIe check, which must report Gen2 x16
10. Last step, optional: Run 5-min Vulkan VRAM stress test now? (y/N) — refused by default, --vulkan always|never to override
11. Result printed + report.json written into the card's folder

Remove the card, plug the next one, repeat — no queue, no indices, folders just pile up in cards/.

Checks & scoring (lenient)

Result	Criteria
FAIL	no device detected, link down/untrained, device lost after rescan, link failed retrain, active uncorrectable AER errors, final post-unlock link other than Gen2 x16 ([ERROR] final PCIe link after unlock is <speed> <width> - expected Gen2 x16)
WARN	link width ≠ x16 (capacitor mod missing/incomplete, pre-unlock runs only), Gen1 only (Gen2 unlock patch not active, pre-unlock runs only), correctable AER errors, kernel errors / Xid, VBIOS backup failed, cmpunlocker verify not unlocked-ok, cmpunlocker install failed, errors during VRAM stress
PASS	everything clean

Exit codes: 0 PASS/WARN · 1 FAIL · 2 no card found · 3 preflight blocking failure.

Output

cards/
└── <GPU-UUID>/
    ├── serial.txt            # physical serial, or N/A
    ├── bios.rom              # VBIOS backup (+ sha256 in report)
    ├── report.json           # full result: link, AER before/after, retrain,
    │                          #   rescan, dmesg, verify, vulkan, PASS/WARN/FAIL
    ├── lspci.txt             # lspci -vvv (before)
    ├── lspci_after.txt       # lspci -vvv (after retrain/rescan)
    ├── config_space.hex      # lspci -nnvvxxxx full hex dump
    ├── dmesg.log             # kernel lines related to this card
    ├── verify_unlocker.log   # cmpunlocker verify.sh output
    └── install_unlocker.log  # full install.sh output (only if the opt-in
                              #   unlock step ran)

Flags

Flag	Meaning
--preflight	setup only: sudo precondition + package install + harness clone, then exit
--demo	run the whole pipeline against fake hardware (self-test)
--nvflash PATH	explicit nvflash binary
--serial XYZ	override physical serial recorded in serial.txt (folder stays UUID-named)
--vulkan ask|always|never	VRAM test policy (default: ask, last step)
--unlock	run the cmpunlocker install.sh step without prompting (default: ask y/N)
--skip-bios	skip VBIOS backup
--bios-unload	unload NVIDIA driver to retry nvflash --save without prompting (console may freeze briefly, driver is reloaded after)
--no-install	never install packages (skips the sudo precondition)
--no-clone	don't clone the harness repos

Safety

This tool reads card state by default. Two steps can change host state, both opt-in:

• Unlock install (opt-in, explicit confirmation) — runs only if you answer y to Run cmpunlocker install.sh now? (y/N) (default n) or pass --unlock. It executes install.sh from the harness clone (harness/cmpunlocker/, refreshed first with git pull --ff-only, no extra installer flags), needs root (sudo credentials are cached first), streams its output live, and saves the full log to cards/<GPU-UUID>/install_unlocker.log. It is skipped entirely when memory is already unlocked, when the harness is not cloned, and in --demo mode. A successful install replaces kernel modules — the step then skips VRAM stress and asks you to cold reboot and re-run.
• VBIOS unload-retry (opt-in, explicit confirmation) — offered only after both backup methods fail with the driver loaded: unloads the NVIDIA modules, retries nvflash --save, then reloads the modules (local console may freeze briefly). Triggered by y or --bios-unload; skipped in --demo mode.
• Everything else remains read-only: verify.sh runs without installing anything, no modules are patched and no firmware is flashed (nvflash is only ever used with --save). The only other writes are local files (reports, logs), PCI link retrain/rescan sysfs triggers, and package installation during --preflight (explicitly gated behind the sudo precondition).