Board support: Waveshare ESP32-P4-WIFI6-Touch-LCD-4.3 (esp32p4 target, MIPI-DSI display, touch, speaker)

[bitcoin3us]

What

First ESP32-P4 board for MicroPythonOS, following the plan in #307. Eight commits, on top of current main:

1. build: add an esp32p4 target — ./scripts/build_mpos.sh esp32p4: ESP32_GENERIC_P4 / C6_WIFI variant (Wi-Fi 6 and BLE through the ESP32-C6 co-processor over SDIO, esp_hosted), 32 MB flash with 4 MiB app partitions, RISC-V frozen builds (rv32imc), PSRAM at 200 MHz and the 256 KB L2 cache with 128-byte lines (the MIPI-DSI panel is scanned out of PSRAM continuously). Also fixes c_mpos/src/quirc_decode.c's FreeRTOS stub guard, which keyed on __xtensa__.
2. boards: detect the Waveshare ESP32-P4-WIFI6-Touch-LCD-4.3 — the P4 gets its own probe branch in detect_board() (GT911 at 0x5D/0x14 on SDA 7 / SCL 8): the xtensa boards' probe pins map to unrelated peripherals on the P4 and hang the chip. main.py also stops cleanly when a board file registers no LVGL display instead of panicking in init_rootscreen().
3. boards: ... display and touch — drivers/display/st7701_dsi: ST7701 over lcd_bus.DSIBus (video mode; hardware reset, Waveshare's vendor register table over DCS, no window addressing) and the board file: 2-lane DSI at 500 Mbps, DPI 30 MHz, RGB565, two PSRAM frame buffers, backlight on GPIO26 (PWM, active low), GT911 polled (its RST/INT are not wired), drivers/indev/gt911.py fixed for interrupt_pin=None.
4. boards: ... speaker output — ES8311 codec + the board amplifier (I2S MCLK 13 / BCLK 12 / LRCK 10 / data 9, PA enable GPIO53 active high), sequenced with the DAC soft-mute like the other ES8311 boards. The microphones sit behind a separate ES7210 ADC that has no driver yet.
5. boards: ... in landscape — the UI runs 800x480 like the other boards. The panel is natively portrait and the ST7701 cannot rotate on the video interface, so the P4's PPA does it (DSIBus(rotation=270), same rotation on the LVGL display). LCD_ROTATION in the board file picks which way is up; 270 is upright in the enclosure.
6. c_mpos: hosted module — hosted.host_version() / hosted.coprocessor_version() report the esp_hosted versions on both sides of the SDIO link, and hosted.ota_begin/ota_write/ota_end/ota_activate wrap esp_hosted's co-processor update calls. Compiled only where CONFIG_ESP_HOSTED_ENABLED is set (an empty file elsewhere; desktop builds don't include it). See "Wi-Fi co-processor firmware" below for why it matters on this board.
7. boards: warm speaker output — now that WAVStream: opt-in warm output so back-to-back clips don't click (ES8311) #301 is in, the speaker uses warm_ms=30000 like the 3.5" board, so back-to-back clips don't click on every I2S restart.
8. boards: ... Wi-Fi 30 s — the board file raises WifiService.CONNECT_TIMEOUT_S (added in WifiService: cancel a pending reconnect (#309), let a board lengthen the connection wait #311) to 30 s, because this board's C6 can need up to 30 s to join on channel 6 (see below). Until WifiService: cancel a pending reconnect (#309), let a board lengthen the connection wait #311 is merged nothing reads the attribute and the wait stays 13 s.

Depends on

• ESP32-P4: DSIBus (MIPI-DSI video mode, PPA rotation) and two builder fixes lvgl_micropython#16 (DSIBus, PPA rotation, and the builder fix for frozen native code on RISC-V — without it the first @micropython.native call panics on the P4). The lvgl_micropython submodule pointer will be bumped here once that is merged, so this stays a draft until then.
• Recommended alongside: WifiService: cancel a pending reconnect (#309), let a board lengthen the connection wait #311. It makes WifiService cancel a pending reconnect before scanning or connecting; without that, a P4 that loses Wi-Fi never recovers, because the stuck retry loop makes every scan come back empty. It also adds the CONNECT_TIMEOUT_S setting that commit 8 uses.

Verified on the board

• Boots to the launcher (no panics, ~30 MB free), Wi-Fi scan through the C6 (14 networks).
• Display: DCS reads answer (RDDPM 0x9C), LVGL at 800x480, forced full-screen refreshes 29–33 ms (render + PPA rotation + vsync), no DMA underruns; picture upright in the enclosure.
• Touch: GT911 product 911, firmware 0x1060, 480x800.
• Speaker: a 1 s WAV plays to completion with the amplifier and codec sequenced correctly. With warm output, three back-to-back clips open the output once and never close it in between; it closes once, 30 s after the last clip, with the amplifier off and the DAC muted.
• Rebased onto main with WAVStream: opt-in warm output so back-to-back clips don't click (ES8311) #301, Waveshare ESP32-S3-Touch-LCD-3.5: LCD SPI clock 80 MHz #303, Re-pin micropython-camera-API to db42861 (restores sccb_i2c_port; fixes "No camera found" on the 3.5) #304, c_mpos: jpegdec, a native JPEG decoder module (esp_new_jpeg) for video playback #305 (jpegdec and the size trims also apply to the P4 build) and Waveshare ESP32-S3-Touch-LCD-3.5: reliable panel init on cold boot #306: boots cleanly, jpegdec imports, 617 KB of app-partition headroom.
• Wi-Fi wait (commit 8, with WifiService: cancel a pending reconnect (#309), let a board lengthen the connection wait #311 and build: apply lvgl_micropython's USER_C_MODULES and RISC-V native patches #312): WifiService.CONNECT_TIMEOUT_S reads 30 after boot, the board joins its saved network by itself, and a connection attempt to a channel-6 AP waits the full 30 s instead of 13 s. Image 616 KB below the partition size.

Wi-Fi co-processor firmware

The board's ESP32-C6 ships with esp_hosted firmware older than 1.4.7 (it does not answer the co-processor version request, which 1.4.7 already supports), while this build's host side is esp_hosted 2.7.0. esp_hosted warns about that mismatch at boot ("upgrade co-proc to avoid RPC timeouts"), but the message is lost while USB re-enumerates, hence the hosted module. On the test board the C6 was updated to 2.7.0 over SDIO: esp_hosted's slave project built for esp32c6, the image written into the P4's unused ota_1 with esptool, then streamed to the C6 with hosted.ota_begin/ota_write/ota_end (under 10 s; the old firmware switches and restarts at ota_end, 2.x needs ota_activate). A documented update path for users is a follow-up.

Separately, this board's C6 joins slowly or not at all on 2.4 GHz channels 6–9. Against a test AP 1 m away:

• channels 1 and 11 join in ~3 s;
• channel 6 joined in 6 of 12 attempts, after 14–30 s (none in the latest run);
• channels 7–9 did not join at all;
• the other way round (Mikey1774's test in Fixed macOS casing because make.py expects macOS instead of macos #7), with the C6 as a soft-AP, another ESP32 saw it on channels 6–9 in at most 2 of 12 scans per channel, against 6 of 6 on channels 1 and 11;
• results differ between sessions.

It reproduces with Waveshare's own 04_wifistation example and with plain ESP-IDF. No clear improvement came from changing the co-processor firmware, the SDIO clock or stopping the display, nor from moving the PSRAM clock source from 400 to 320 MHz (CONFIG_SPIRAM_SPEED_80M, which also slows a full-screen refresh to ~53 ms, so the build stays at 200 MHz). Reported as waveshareteam/ESP32-P4-WIFI6-Touch-LCD-4.3#7. Commit 8 gives the board 30 s to connect, so the slow channel-6 joins succeed when they happen.

Not yet

ES7210 microphones, the TF card slot (slot 0 needs the P4's on-chip LDO channel 4 switched on, which machine.SDCard cannot do), the MIPI-CSI camera.

Flashing

USB-Serial-JTAG, so plain esptool --chip esp32p4 --before default_reset works (no TinyUSB caveats): full image at 0x0, app-only at 0x20000. Factory partition layout differs (the factory image is worth backing up first).

🤖 Generated with Claude Code

With thanks to the scientists and engineers who did the hard, unglamorous work that got us here.