---
title: "PMIC Inspect"
description: "Inspector for the V2N SoM's on-module power chips — ACT88760, DA9292 and TPS628640 — run as a Linux/Yocto user-space app on the Cortex-A55 cluster over BRD_I2C. It is read-only unless you pass --write together with one action flag."
source: https://docs.alplab.ai/sdk/examples/v2n/v2n-pmic-inspect
---

# `v2n-pmic-inspect`

Inspector for the V2N SoM's on-module power chips — ACT88760, DA9292 and TPS628640 — run as a Linux/Yocto user-space app on the Cortex-A55 cluster over `BRD_I2C`. It is **read-only unless you pass `--write` together with one action flag**.

Source: [`examples/v2n/v2n-pmic-inspect/`](https://github.com/alplabai/alp-sdk/tree/main/examples/v2n/v2n-pmic-inspect). New in alp-sdk v0.17.0.

:::danger[The write actions change live power rails]
`--set-mv`, `--enable` / `--disable`, `--fix-gpio4-polarity` and `--deepx-rail-sequence` reprogram PMICs that power the SoC, its memory and (on V2N-M1) the DEEPX NPU. A wrong setpoint can take a board off-rail. The drivers enforce the SoM's metadata-generated voltage windows and refuse out-of-window or critical-rail requests, but **nothing here has been run on silicon yet** — treat every write as a bench operation on a unit you can afford to power-cycle, and stay read-only otherwise.
:::

## What it reads

| Address | Chip | What is shown |
|---|---|---|
| `0x25` / `0x26` | ACT88760 (`act8760`) | Buck1..7 and LDO1..6: net, enable, mV, POK, OV/ILIM; GPIO1..11: net, MODE byte, polarity, MUX, push-pull, level, IRQ mask |
| `0x1E` | DA9292 (`da9292`) | identity (DEV_ID / REV_ID / CFG_REV); CH1 / CH2: enable, active mV, VSTEP / VSEL, PG, UV / OV / OC; latched EVENT_00/01 (peeked, not cleared) |
| `0x44` / `0x48` / `0x4F` | TPS628640 | The DEEPX-side bucks on V2N-M1: VOUT, CONTROL. Skipped when absent — they answer only once the DEEPX rail has been enabled |
| `0x4D` | TPS628640 | The LPDDR4X 0.6 V buck, an assembly option; skipped when absent |

Rail net names, guard windows, critical flags, the expected DA9292 identity and the expected ACT88760 MODE4 byte all come from the generated `<alp/chips/v2n_power_tree.h>`, which the SDK generates from the SoM's power-tree metadata. The app hard-codes no window: it hands the generated tables to the drivers, and the drivers enforce them.

## Usage

```sh
v2n-pmic-inspect                      # text dump, family v2n-m1 (a superset for reading)
v2n-pmic-inspect --family v2n --json  # machine-readable
v2n-pmic-inspect --include-clear-on-read
```

| Exit code | Meaning |
|---|---|
| `0` | Silicon matches the metadata. |
| `1` | Silicon and metadata disagree: a DA9292 identity byte differs, ACT88760 MODE4 is not `0x08`, a power chip is missing, or a present rail's live voltage reads outside its window. |
| `2` | Usage error, or `/dev/i2c-8` could not be opened. |
| `3` | A write action was refused or failed. |

**Clear-on-read registers** are never read unless you pass `--include-clear-on-read`: ACT88760 ADD1 `0x00` (VSYS latches), `0x04` (GPIO1..8 toggles) and `0x2B` (GPIO9..11 toggles — so without the flag the GPIO9..11 level and IRQ-mask columns show `?`), and TPS628640 `0x05` (STATUS). `act8760_init()` reads `0x00` as its presence probe, so the VSYS latches are consumed even without the flag. The DA9292 EVENT registers are write-1-to-clear, so they are always only peeked.

**Kernel-owned addresses** are checked first with an `I2C_SLAVE`-based probe; an address a kernel driver holds (`EBUSY`) is reported as `kernel-owned` and never touched — the same check [`v2n-brd-i2c-bringup`](https://docs.alplab.ai/sdk/examples/v2n/v2n-brd-i2c-bringup) does.

## Write actions

Every write needs `--write`, **exactly one** action flag, and an explicit `--family v2n|v2n-m1` (which selects the guard tables). The app also refuses to write when silicon disagrees with the metadata — except for the GPIO4 fix below.

| Action | What it does |
|---|---|
| `--set-mv <rail> <mV>` | Programs a setpoint. The driver rounds down to the chip's grid in the rail's live range and refuses (`ALP_ERR_OUT_OF_RANGE`) unless the encoded value is inside the rail's window. |
| `--enable <rail>` / `--disable <rail>` | Changes the enable bit. Disabling a critical rail — every ACT88760 rail, DA9292 CH1, TPS628640 `0x4D` — always returns `ALP_ERR_NOSUPPORT`. The app refuses `--disable` of any DA9292 / TPS628640 rail: DEEPX rails only go down in order, through the `--deepx-rail-sequence` shutdown. |
| `--fix-gpio4-polarity` | Acts only when ACT88760 MODE4 reads exactly `0x88` (an inverted OTP that holds the GD32 supervisor in reset); writes `0x08` and reads it back. Prints a no-op message if MODE4 already reads `0x08`, and refuses any other value. **The fix is volatile — repeat it after every power cycle.** |
| `--deepx-rail-sequence` | V2N-M1 only. Runs the DA9292 CH2 DEEPX-rail sequence through the Linux GPIO character device. `M1_RESET` is not touched. |

`<rail>` is an id (`buck1`..`buck7`, `ldo1`..`ldo6`, `da9292.ch1`, `da9292.ch2`, a TPS628640 net) or the rail's net name as the app prints it in the read-only dump. Buck5 / Buck6 have no confirmed net name — use their `buck5` / `buck6` ids.

### `--deepx-rail-sequence` safety rules

On a normally booted board **U-Boot has already run this sequence**; this action is for bench bring-up only.

- **It refuses when DA9292 CH2 is already enabled** — that state means the boot owner already ran the sequence, and claiming the enable line would drop the DEEPX core enable under a live DEEPX.
- **It refuses when either DEEPX control line is held by another consumer**, or is already an output, and unless the kernel names the lines as expected.
- **The process stays running while the rail is up.** If the sequence succeeds it keeps holding the lines, because the RZ pin controller returns a released line to input. `SIGINT` / `SIGTERM` / `SIGHUP` power the rail down in order. **Killing the process uncleanly (`SIGKILL`) releases the core-enable line, which then floats under a live CH2.**
- **CM33-boot mode is blocked.** In A55-boot mode U-Boot owns the DEEPX sequence and Linux is the sole `BRD_I2C` master. The Cortex-M33 must not master `BRD_I2C`, so no CM33 image may run the sequence. The boot-mode strap level is not recorded yet, so the app prints it but cannot gate on it.

## `board.yaml`

```yaml
som:
  sku: E1M-V2N101

preset: e1m-x-evk

cores:
  a55_cluster:
    app: ./src
    peripherals:
      - i2c
      - gpio

chips:
  - act8760
  - da9292
  - tps628640

diagnostics:
  log_level: info
```

No `meta-alp-sdk` recipe packages this app yet, so `tan build` blocks the A55 slice with a `yocto-recipe-missing` warning. Build it with the Yocto SDK.

## Build (Yocto SDK)

```sh
# The SDK that includes meta-alp-sdk ships libalp_sdk.so + libalp_chips.a
# with the act8760, da9292 and tps628640 drivers on by default:
. /opt/poky/<ver>/environment-setup-aarch64-poky-linux

cmake -B build -S . -DCMAKE_TOOLCHAIN_FILE=$OECORE_NATIVE_SYSROOT/usr/share/cmake/OEToolchainConfig.cmake
cmake --build build
```

Copy `build/v2n-pmic-inspect` to the target and run it as a user that can open `/dev/i2c-8` (and `/dev/gpiochip*` for the DEEPX action).

## See also

- [`v2n-brd-i2c-bringup`](https://docs.alplab.ai/sdk/examples/v2n/v2n-brd-i2c-bringup) — the read-only scan of the whole `BRD_I2C` bus
- [`v2n-power-monitor`](https://docs.alplab.ai/sdk/examples/v2n/v2n-power-monitor) — live per-rail current on the EVK
- [E1M-X V2N-M1 — DEEPX bring-up](https://docs.alplab.ai/e1m/e1m-x-v2n-m1#deepx-bring-up)
- [Chip catalogue](https://docs.alplab.ai/sdk/chips)
