# switchroot's mainline kernel — reference material

Vendored for reference, **not** built by this project. Provenance:

- Source: `https://gitlab.com/switchroot/kernel/mainline-kernel`
- Branch: `switch-dev`
- Commit: `5dc55737d3f32324bc8946fa76c2d827947d090b` (resolved 2026-10-19)
- Fetched: 2026-10-18

Pin that SHA. The files below are a **copy**, not a submodule, so they drift:
re-resolve the branch tip and diff before assuming this reference still matches
what switchroot ship. Their tree is also an older mainline than this project's
7.2, so even a correct pin does not make it drop-in.

switchroot have already done mainline bring-up for the Nintendo Switch. Their
tree contains `arch/arm64/boot/dts/nvidia/tegra210-nintendo-switch.dts` (2425
lines) and a `switch-patches/` series that is effectively a map of the whole
job. Everything in this project that was written from scratch — the device tree,
the USB gadget wiring, the display — has a counterpart there, and several of the
things still open here are already solved in their patches.

## Files kept here

| File | What it is |
|---|---|
| `tegra210-nintendo-switch.dts` | their Switch device tree — the reference for display, PCIe, USB, SD |
| `0010-...-jdi-lpm062m326a.patch` | **the panel driver** — 612 lines, `drivers/gpu/drm/panel/panel-jdi-lpm062m326a.c` |
| `0012-...-switch-panel.patch` | the display integration for that panel |
| `0026-...-tegra_xudc-use-dma_alloc_coherent.patch` | a one-hunk fix to the XUDC driver, likely the cause of our USB stalls |

## The full patch list, as a map of the work

```
0001  Tegra210 DFLL support
0002  sdmmc autocal
0003  CPUidle support for Tegra210
0004  Tegra XUSB gadget driver support
0005  Fixes for Tegra soctherm
0006  More soctherm functionality
0007  Tegra210 BPMP driver
0008  DisplayPort support for drm/tegra
0009  iio: light: bh1730fvc (ambient light sensor)
0010  drm: panel: driver for jdi-lpm062m326a        <- the panel
0011  stmfts switch support (touchscreen)
0012  platform integration for switch panel        <- display wiring
0013  of: thermal: allow enabling hwmon
0014  missing SKU IDs
0015  switch driver for PWM fan
0016  input: working Joy-Con rails driver
0017  missing SKU IDs
0018  gpio-tegra: mask IRQs on shutdown
0019  HID: switchcon: Nintendo Switch controller driver
0020  misc: rohm bm92txx driver (USB-PD)
0021  phy: tegra-xusb: ID override support
0022  usb: xhci-tegra: allow extcon to enable host mode
0023  input: touchscreen stmfts: sync mt events
0024  arm64: configs: Nintendo Switch defconfig
0025  Switch DTS
0026  usb: udc: tegra_xudc: use dma_alloc_coherent
```

## What their tree says about our open items

### Display — solved there, and the panel is identified

```dts
		dsi@54300000 {
			status = "okay";
			avdd-dsi-csi-supply = <&vdd_dsi_csi>;

			panel@0 {
				compatible = "jdi,lpm062m326a";
				reg = <0>;

				reset-gpios = <&gpio TEGRA_GPIO(V, 2) GPIO_ACTIVE_HIGH>;
				vdd1-supply = <&pavdd>;
				vdd2-supply = <&navdd>;
				backlight = <&backlight>;
			};
		};
```

So the panel is a **JDI LPM062M326A**, reset is `PV2` and **active high**, and the
supplies are `pavdd`/`navdd`. Their patch 0010 is the driver, 0012 the
integration. This is exactly the piece we could not derive — the timings and the
init sequence.

### PCIe / WiFi — both ports, and the lanes we were missing

```dts
		pci@1,0 {
			phys = <&{/padctl@7009f000/pads/pcie/lanes/pcie-1}>,
			       <&{/padctl@7009f000/pads/pcie/lanes/pcie-2}>,
			       <&{/padctl@7009f000/pads/pcie/lanes/pcie-3}>,
			       <&{/padctl@7009f000/pads/pcie/lanes/pcie-4}>;
			phy-names = "pcie-0", "pcie-1", "pcie-2", "pcie-3";
			status = "okay";
		};

		pci@2,0 {
			phys = <&{/padctl@7009f000/pads/pcie/lanes/pcie-0}>;
			phy-names = "pcie-0";
			status = "okay";
		};
```

**Both ports are enabled**, and `pci@1,0` uses lanes `pcie-1` through `pcie-4`.
Our tree enabled only `pci@2,0` on lane `pcie-0` and left the other lanes out —
which is a plausible reason the endpoint never answered config cycles.

### USB device mode — a different, cleaner approach

```dts
	xudc@700d0000 {
		status = "okay";
		phys = <&{/padctl@7009f000/pads/usb2/lanes/usb2-0}>;
		phy-names = "usb2";
		extcon = <&bm92t 1>;
	};
```

They signal VBUS and role through an **extcon from the bm92txx USB-PD driver**
(patch 0020) rather than the padctl role switch we used. Note `phy-names = "usb2"`
— their driver revision accepts that name, whereas current mainline wants
`"usb2-0"` (see `device-tree/README.md`).

### USB throughput — patch 0026 is probably our answer

```diff
--- a/drivers/usb/gadget/udc/tegra_xudc.c
+++ b/drivers/usb/gadget/udc/tegra_xudc.c
+		dma_alloc_coherent(xudc->dev, XUDC_NR_EPS * ...
```

Our NCM link stalls and manages only ~550 KiB/s with erratic latency. A DMA
allocation fix in the XUDC driver is a very plausible cause, and it is 25 lines.

## Important caveat

Their tree is an **older mainline** than this project's 7.2, so bindings and
driver internals have moved: `xudc@700d0000` vs `usb@700d0000`, `phy-names =
"usb2"` vs `"usb2-0"`, no `usb-role-switch` requirement. Their device tree and
patches are therefore a **reference to port from**, not something to drop in.