H700 port

NextUI → Allwinner H700 (Anbernic RG XX) Port

Developer documentation for the H700 platform — internals, decisions, and test evidence for people working on NextUI itself. For installing and using NextUI on a handheld, see NextUI for Anbernic H700 handhelds. For community Paks, see Porting NextUI Paks to H700.

Goal: NextUI on the Anbernic RG XX line (Allwinner H700) with feature parity with the TrimUI Brick (tg5040) — displaycal, WiFi/BT, first-class sleep — installed on top of the stock Anbernic OS, no reflash, fully reversible.

Lifecycle: h700-rc8 is golden master — the build intended for the upstream merge PR. See 09 for stage gates and 08 for current evidence.

This section is the H700 maintainer knowledge base. Documents 00–07 record durable facts, architecture, decisions, rationale, failure analysis, diagnostics, and lessons for future platform ports. They may say how a fact was established, but they do not own current release status. Documents 08–10 are the operational ledgers for tests, release planning, and per-core coverage.

Documents

Doc	Role	Contents
00-device-facts	Knowledge base	Probed hardware/OS facts, sysfs/ioctl/evdev reference, stock boot chain, known unknowns
01-toolchain-and-build	Knowledge base	dedicated h700-toolchain image, in-tree SDL2, dependency policy, and cross-toolchain lessons
02-boot-and-installer	Knowledge base	dmenu.bin hijack, storage/write policy, boot shim, installer, and launcher responsibilities
03-platform-layer	Knowledge base	Platform API map, device detection, evdev/SDL input split, libmsettings, keymon, and cores
04-video-display	Knowledge base	Mali/fbdev architecture, geometry, rotation and HDMI decision records, DisplayCal, graphics incidents
05-audio	Knowledge base	ALSA architecture, symbol-versioning incident, routing, Bluetooth audio, and sample-rate constraints
06-power-sleep-battery	Knowledge base	Sleep architecture/postmortem, lid semantics, charging behavior, battery facts, and governors
07-wifi-bluetooth	Knowledge base	Stock components, NextUI ownership boundaries, lifecycle scripts, and diagnostics
08-testing-status	Test ledger	Stable test IDs, current results, evidence devices, and model-specific details
09-roadmap	Release policy	Alpha 1/2/3, Beta, RC, and Release gates; deferrals, future work, enhancements, and stretch goals
10-core-game-matrix	Test ledger	Per-system/core/game evidence and emulator defects underlying the GAME test rows

Architecture in one paragraph

One h700 platform serves rg40xx/rg34xx/rg28xx/cube (DEVICE env, detected at boot from the stock dmenu.bin binary). It builds inside the h700-toolchain image (same aarch64/A53 target and TG5040-derived SDK sysroot as the historical tg5040 reuse, forward-compatible glibc) with mali-fbdev SDL2 prebaked in the toolchain image (JohnnyonFlame/SDL-malifbdev-rot) targeting the Mali blob's fbdev EGL winsys. NextUI's shared generic_video.c/generic_wifi.c/generic_bt.c run unchanged. Boot is hijacked by dropping one dmenu.bin file on the stock card's FAT partition; NextUI itself lives entirely on TF2. The stock Ubuntu userland is used aggressively (systemd, timedatectl, dhclient, firmware-provided BlueZ, stock unzip) rather than bundled around.

Key decisions — outcome register

Decision	Planned	Shipped / outcome
Platform	one h700, DEVICE env per device	✅ as planned
Arch / toolchain	64-bit; dedicated h700-toolchain (TG5040 SDK recipe)	✅ dedicated image; sysroot-mismatch pitfalls catalogued in 01
SDL2	malifbdev-rot in h700-toolchain	✅ prebaked in image; early only stages libs
Video	generic_video GLES pipeline on Mali blob	✅ core pipeline works; multi-panel UI scaling + box art pass; panel-matched overlay assets not shipped
Input	SDL joystick route	Deviation: raw evdev remains primary for the built-in pad; the SDL classification fix makes enumeration work, but the built-in SDL ordering differs from external-pad JOY_* mappings and would duplicate/mis-map events. SDL is retained for external pads (03)
Audio linkage	SDK libasound, bundled	Deviation: dlopen'd device libasound (--enable-alsa-shared) after a symbol-versioning bug caused glitchy audio (05)
BT audio	build + ship bluealsa	Deviation: use the stock BlueALSA/ALSA/SBC/BlueZ stack, with an H700-only ALSA option adjustment, initial mixer volume, and post-connect mixer update (05/07)
Sleep	echo mem + tg5040-style wrapper	✅ shared power architecture retained; target-specific service and ALSA sequencing documented in 06
Lid	hallkey → PLAT lid API	✅ sysfs-polled lid integrated with the shared sleep request API (06)
WiFi	NextUI-owned wpa_supplicant	✅ as planned; creds on SD, dhclient + wpa_action renew (07)
Rumble	moto sysfs	✅ as planned; tested-good
LEDs	MAX_LIGHTS 0 — "no RGB on RG XX"	Reversed. RG40XX H/V and RG CubeXX do have RGB, driven by an MCU over UART5 rather than /sys/class/leds, which is why the first probe missed them (03)
Brightness / displaycal	same disp ioctls as tg5040	✅ 1:1 as predicted, tested-good incl. sleep survival
Install	TF1 gets one file; NextUI on TF2; no TF1-only mode	✅ as planned; TF1 writes mountpoint+cmp guarded
Splash	per-panel raw-BMP dd assets	Deviation: fbsplash text renderer — no per-resolution assets needed (02)
RG28XX rotation	SDL-level first, GL hook fallback	✅ SDL/driver level alone is correct; the GL hook double-rotated and broke minarch scaling — removed after hardware testing (04)
HDMI	stretch goal	✅ disp2/fbdev hotplug with layer-geometry repair and ALSA routing (04)

Maintenance model

• Update 00 when hardware, firmware, stock-OS paths, or empirical facts change.
• Update 01–07 when implementation, rationale, diagnostics, or a subsystem lesson changes. Keep incident histories when they explain a non-obvious constraint.
• Record test evidence once in 08 and per-core evidence once in 10.
• Record stage gates, accepted risk, deferrals, enhancements, and future work only in 09. Subsystem documents should link there instead of restating release policy.
• Re-probe stock integration after Anbernic firmware changes and revisit the relevant knowledge-base chapter before treating a new failure as a NextUI regression.