M5Stack Tab5 (m5stack-tab5, chip rev v1.3): ST7123 display init succeeds but panel stays black #2186

Summary

M5Stack Tab5 (m5stack-tab5 board variant, Rev < 3 chip) boots and runs correctly — WiFi, audio (I2S/ES8388), touch, camera, and backlight all confirmed working — but the ST7123 MIPI-DSI display never actually renders anything. The backlight turns on, the driver logs a fully successful init sequence, but the panel stays black.

Hardware

• M5Stack Tab5, ESP32-P4 chip revision v1.3 (confirmed via esptool chip-id)
• Board variant built: m5stack-tab5 (not -p4x), per the board README's own Rev<3 guidance
• Confirmed the physical unit/panel is not faulty: M5Stack's own M5Tab5-UserDemo factory firmware renders correctly on this exact unit

What was tried

1. ESP-IDF v6.0.2 (recommended in the board README): boots but hard-crashes at esp_startup_start_app app_startup.c:83 (res == pdTRUE) — traced to the SDIO transport buffer pool for the onboard ESP32-C6 (WiFi co-processor) exhausting internal RAM. Fixed by setting CONFIG_ESP_HOSTED_MEMPOOL_PREFER_SPIRAM=y (this option already exists in the bundled espressif__esp_hosted component's Kconfig, just defaults to off). After this fix, boot succeeds cleanly and the device is fully functional except for the display.
2. Also tried ESP-IDF v5.5.5 to sidestep the above — no crash, but this hit the known display bug tracked in M5Stack Tab5 (ESP32-P4) DSI issues on ESP-IDF v5.5.X (IDFGH-17043) espressif/esp-idf#18083 (blank/garbled screen on v5.5.x). Went back to v6.0.2 + the mempool fix instead, since that's a clean fix rather than a version dodge.
3. Compared main/boards/m5stack/tab5/config.h's st7123_vendor_specific_init_default[] byte-for-byte against M5Stack's own working init table in M5Tab5-UserDemo — identical, command-for-command, byte-for-byte.
4. Compared the DPI video timing config (h_size/v_size/porches/pulse widths, dpi_clock_freq_mhz, lane_bit_rate_mbps) against the same UserDemo reference — identical.
5. UserDemo sets .flags = { .use_dma2d = true } on esp_lcd_dpi_panel_config_t; that field doesn't exist in this project's pinned ESP-IDF v6.0.2 (esp_lcd_mipi_dsi.h only has disable_lp), so it's not portable as-is — UserDemo is evidently built against a different/newer esp_lcd API.
6. Tried dpi_config.num_fbs = 2 (a fix that resolved a similar-sounding ESPHome ST7123 issue) — no change.
7. Tried dpi_config.flags.disable_lp = 1 (disabling low-power mode during DPI video transmission, since continuous LP/HS switching is a plausible cause of "driver reports success, panel shows nothing") — no change.
8. Full power-cycle (unplug/replug USB, not just soft reset) tried at multiple points, since some DSI PHY/LDO state doesn't reinitialize on soft reset — no change.

Current state

Everything is initialized without error at the log level (ST7123 Display initialized with resolution 720x1280, Touch panel create success, Backlight: Set brightness to 75), but the physical panel never displays anything — confirmed on real hardware, backlight on, screen black, across every tested ESP-IDF version and every config permutation tried above.

Given the init bytes and timing are verified identical to a known-working reference build, the remaining bug is most likely in the vendored esp_lcd_st7123.c driver's actual init/reset/DBI-transaction sequencing logic (order of operations, timing between esp_lcd_panel_reset and esp_lcd_panel_init, or something in how DBI commands are transmitted relative to DPI video start) rather than anything tunable from the board config file.

Environment

• ESP-IDF v6.0.2
• xiaozhi-esp32 main branch, board m5stack/tab5, variant m5stack-tab5
• macOS (Apple Silicon shouldn't be relevant here, host toolchain only)