jetson-customize-pinmux

par nvidia

Configurateur par broche SFIO / direction / état initial pour une carte personnalisée Jetson Orin ou Thor à partir du XLSM de pinmux. Ne PAS utiliser pour une superposition kernel-DT ou…

npx skills add https://github.com/nvidia/skills --skill jetson-customize-pinmux

Customize pinmux (per-pin SFIO / direction / state)

Overview

The Tegra pinmux spreadsheet (.xlsm) is the ground truth for every CVM ball: SoC pin name, supported SFIOs, customer-selected function, direction, and initial state. This skill parses that XLSM, runs a per-pin Q1–Q6 interactive loop, and emits the three BCT DTSIs (pinmux, gpio, padvoltage) in one shot into the overlay tracker at <source.root_path>/Linux_for_Tegra/bootloader/.

Unlike sibling skills jetson-customize-uphy / jetson-customize-pcie / jetson-customize-camera, pinmux has no kernel-DT overlay surface and no ODMDATA edit. The XLSM is the source of truth; the three emitted DTSIs land at flash time via the carrier conf's PINMUX_CONFIG= / GPIOINT_CONFIG= / PMC_CONFIG= references (which /jetson-derive-carrier set up).

Pad classification (silicon-fixed): only BD* and BI* pads have configurable pull / drive / open-drain attributes. LP5XA_*, UPHYDS_*, DP_SINGLE_*, BDMIPI16X_*, BDUSB2_*, OSCI27_* are fixed-function and skip Q4–Q6 (configurable: no).

The bundled scripts/modify_pinmux.py is the workhorse: it parses the XLSM via openpyxl>=3.1, builds the per-carrier pinmap JSON, captures pin edits into a session shim, and (on generate) writes the three DTSIs.

When to invoke

  • The user says "configure pin", "set SFIO", "edit pinmux DTSI", "set pin direction", "set initial state", or asks to repurpose a CVM ball (e.g. flip a pin between GPIO and a peripheral function).
  • A sibling skill (jetson-customize-camera, jetson-customize-pcie, jetson-customize-usb, jetson-customize-mgbe) reports an HSIO pin mismatch via pin_verifier.py and the user wants to fix it.
  • The user pre-derived a custom carrier with /jetson-derive-carrier and now wants to author the pinmux from a freshly-edited .xlsm.

Prerequisites:

  • Active profile selected (target-platform/active_target.yml<profile>.yaml with reference_devkit: AND custom_carrier:).
  • <source.root_path>/Linux_for_Tegra/ exists as a git repo (/jetson-init-source).
  • /jetson-derive-carrier has run — the three pinmux-side BCT DTSIs (PINMUX_CONFIG, GPIOINT_CONFIG, PMC_CONFIG references in the carrier conf) exist in the overlay tracker.
  • A pinmux .xlsm is registered in the active profile at documents.custom_carrier_pinmux_xls (preferred when custom-carrier- specific) or documents.ref_devkit_pinmux_xls (fallback). The bundled modify_pinmux.py requires openpyxl>=3.1.

Procedure

See references/procedure.md for the full step-by-step procedure (Steps 1–8). Summary:

  1. Resolve active target + XLSM. Validate active profile, custom_carrier:, overlay-tracker prerequisites; resolve the pinmux .xlsm path from documents.custom_carrier_pinmux_xlsdocuments.ref_devkit_pinmux_xls → single XLSM under documents.root_path → user prompt.
  2. Probe. Run modify_pinmux.py probe to parse the XLSM into the per-skill scratch <KB>/pinmap/<custom-carrier>.json plus session.json shim.
  3. Lookup. Resolve a free-form user query (CVM ball, Verilog name, signal, DT pin) via modify_pinmux.py lookup; surface supported SFIO list, defaults, and configurable: yes/no.
  4. Set-pin (HARD GATE — Q1–Q6 via AskUserQuestion). Q1–Q3 (sfio / direction / initial_state) always asked; Q4–Q6 (pull / drive_type / open_drain) only when configurable: yes. tristate and e_input are derived from direction, never asked.
  5. Generate. modify_pinmux.py generate --out-dir <source.root_path>/Linux_for_Tegra/bootloader/ (root, not bootloader/generic/BCT/ — derive-carrier .dts forks live there, do not colocate). Emits: tegra<soc>-mb1-bct-{pinmux,gpio,padvoltage}-<carrier-key>.dtsi. <carrier-key> comes from the carrier conf's PINMUX_CONFIG= reference, NOT the kebab-cased carrier name.
  6. Commit (single batched commit per workflow rule). All three DTSIs are one logical edit → one customization commit. Run the commit-preview gate before each commit.
  7. Run-state sidecar + session shim. Write the user-facing <profile-stem>.jetson-customize-pinmux.json sidecar and the transient session.json shim under <workspace>/target-platform/.
  8. Summary. Emit the standard one-line + table summary.

Gotchas

  • No kernel-DT overlay; no OVERLAY_DTB_FILE edit; no render_conf.py hand-off. This skill ends at the three BCT DTSIs — the carrier conf already references them via PINMUX_CONFIG= / GPIOINT_CONFIG= / PMC_CONFIG= (set up by /jetson-derive-carrier).
  • Re-point the .dts wrapper's #include after generate. /jetson-derive-carrier forks the .dts wrappers at bootloader/generic/BCT/, but their #include lines may still pull the upstream devkit .dtsi (e.g. …-p3834-xxxx-p4071-0000.dtsi). After generate writes the new <CARRIER_KEY>.dtsi to bootloader/ root, edit each wrapper's #include to the new filename — by bare basename (#include "tegra<soc>-mb1-bct-pinmux-<CARRIER_KEY>.dtsi"), not ../../… filesystem-relative. The BCT build's cpp -I bootloader/ resolves bare basenames; that's the convention every other BCT include in the tree follows. Roll the wrapper edits into the same customization commit as the three DTSIs. See references/procedure.md Step 5 ("Sanity-check the carrier .dts wrapper").
  • Q4–Q6 gated on configurable: yes. Asking pull / drive_type / open_drain on a fixed-function pad (LP5XA_*, UPHYDS_*, BDMIPI16X_*, etc.) is silently dropped by the script and confuses the user. lookup prints configurable: yes/no — always check it before prompting Q4–Q6.
  • tristate and e_input are derived, never asked. unused → tristate=ENABLE; input / bidirectional → enable-input=ENABLE. Exposing them as separate prompts produces inconsistent DTSIs.
  • sfio=gpio requires a parseable gpio=GPIOn_PD.NN entry in the pinmap row's sfio list. Pins without one are GPIO-incapable silicon; set-pin rejects the call. Surface the rejection — don't silently fall back to a non-GPIO SFIO.
  • Marker idempotency. Every per-pin edit carries // custom-bsp: pinmux on the closing brace; gpio default-state entries carry the same marker as a trailing comment. Re-running generate must detect and update — never duplicate.
  • modify_pinmux.py is unchanged from the original framework — it reads its own session.json shim under --kb-dir. The shim is regenerated each run from the active profile + the user-facing sidecar. Do not hand-edit the shim; it's transient.
  • Multiple pinmux DTSI variants per Thor module SKU. Some carrier pins live in a different DTSI variant than the one the carrier conf references. modify_pinmux.py commit (legacy patch-in-place flow) tolerates missing per-pin blocks via pinmux.warnings[] rather than failing. Surface the warning; point at the alternate DTSI variant.
  • Don't touch the upstream BSP at <bsp_image.root_path>. All edits land in <source.root_path>/Linux_for_Tegra/bootloader/ under the pristine + customization commit pattern.

Available Scripts

ScriptPurposeArguments
scripts/modify_pinmux.pyXLSM parser + per-pin DTSI generator. Invoked via run_script() from Steps 3-6 with the subcommand of the current phase.probe | lookup | set-pin | apply | generate | commit [...] (see --help)
scripts/generate_dtsi.pyRenders pinmux/GPIO/padvoltage DTSI fragments from the bundled session state. Called by modify_pinmux.py generate.--session <path> --out-dir <dir>

Invoke from the skill body as a subprocess via run_script():

# run_script: probe the carrier pinmux XLSM and write a session state
scripts/modify_pinmux.py probe --xlsm carrier.xlsm --session .pinmux-session.json

# run_script: render DTSI fragments from the final session state
scripts/modify_pinmux.py generate --session .pinmux-session.json --out-dir bsp_sources/pinmux/

References

Plus de skills de nvidia

compileiq-debug
nvidia
Utilisez quand quelque chose ne va pas : Search() bloque, toutes les évaluations retournent INVALID_SCORE, les scores ne s'améliorent pas, chaque configuration retourne le même nombre, erreurs ptxas…
create-github-pr
nvidia
Créer des pull requests GitHub en utilisant l'interface en ligne de commande gh. Utiliser lorsque l'utilisateur souhaite créer une nouvelle PR, soumettre du code pour révision, ou ouvrir une pull request. Mots-clés de déclenchement -…
nemoclaw-maintainer-cross-issue-sweep
nvidia
Analyse les autres problèmes ouverts pour trouver ceux qu’une PR donnée pourrait également corriger ou casser accidentellement. Génère des opportunités de correctifs adjacents et des risques de contradiction avec fichier:ligne…
fhir-basics
nvidia
Apprend aux agents comment fonctionnent les API FHIR R4, quelles ressources sont disponibles, comment les interroger avec des paramètres de recherche, et comment analyser correctement tous les formats de réponse…
compileiq-validate-result
nvidia
Utiliser APRÈS qu'une recherche soit terminée et AVANT de réclamer un accélérateur ou d'expédier un ACF. Charge le CSV dump_results, extrait les K meilleurs candidats (mono-objectif)…
changelog-audit
nvidia
Auditer le CHANGELOG.md de Warp avant une publication : récupérer les entrées perdues, trier par impact utilisateur, affiner le langage des entrées, ajuster les retours à la ligne et (en mode branche de publication) mettre à jour la comparaison…
maintain-dynamic-plugins
nvidia
Maintenir les chargeurs de plugins dynamiques NeMo Relay, les manifestes, les SDK natifs Rust, le protocole worker gRPC, le SDK worker Python, la documentation, les tests et la couverture du workflow de publication
dgx-diagnose
nvidia
Diagnostiquer les problèmes courants du DGX Station GB300 — plantages CUDA, ciblage incorrect du GPU, bugs de conteneur vLLM/SGLang, problèmes d'état MIG, erreurs NVLink/Fabric Manager,…