SDL_Init failed with 'kmsdrm not available' because SDL2's KMSDRM backend
cannot resolve the vendor Mali EGL by itself. Export
LD_PRELOAD=/usr/lib/libmali.so in the Setup file_manager.sh and patch it from
the build script. Verified on the device: 351Files starts and runs instead of
returning to ES.
On the device /sys/class/leds is empty; the vendor micro,gamepad driver exposes
the LED as /sys/devices/platform/gamepad/led_ctrl with values off/on/loop/
breath/red/green/blue. Replace the inert udev rule with a sysinit unit that
writes 'loop' as soon as the attribute exists (~0.5s instead of ~11s when ES
opens the input device).
Also restore the WiFi menu section that was lost when the SSH section was
rewritten.
The G90 has no gpio-leds node; the micro,gamepad driver registers red_led,
green_led and blue_led from the DT red/green/blue-gpio properties. It probes at
~0.5s but only lights the LED when userspace opens the input device, which is
EmulationStation at ~11s.
Add a udev rule that writes brightness=1 when blue_led appears, plus a
sysinit.target unit as a fallback. Change the LED name in both files if the
unit uses another colour.
Single static ARMv7 musl binary (3.2 MB), HTTP basic auth, serving / read-write
with upload, delete, search, folder archives, and WebDAV on the same URL.
Enabled at boot via multi-user.target.wants; also useful as a fallback while
SSH key exchange is being debugged.
Pinned to v0.47.0 with the release tarball sha256 checked by the build script.
An OpenSSH 10.6 client died at 'expecting SSH2_MSG_KEX_ECDH_REPLY' when
mlkem768x25519-sha256 was negotiated, before authentication. The same build
completes that KEX under qemu with an OpenSSH 10.6p2 client, so the fault is
specific to the ARMv7 hardware in the PQ path.
Build dropbear with DROPBEAR_MLKEM768=0 and DROPBEAR_SNTRUP761=0; it now
advertises curve25519-sha256 and clients pick it automatically. Rebuilt image
and p1; verified the binary extracted from the image matches the build and
negotiates curve25519 with ssh 10.6p2.
EmulationStation's WiFi screens are blocking calls on the UI thread and the
helper script made it worse:
- batocera-wifi list ran 'connmanctl services <id>' once per AP just to read
Interface=wlan0; 20-40 APs meant 20-40 D-Bus calls while the SSID menu was
built. Now one 'connmanctl services' listing is parsed locally, skipping P2P
and hidden entries.
- No call had a timeout. Every connmanctl invocation is now wrapped in
'timeout 5'; enable/disable are bounded by 25s/10s.
- ES calls 'batocera-wifi enable <ssid> <key>' but only list/scanlist were
implemented, so saving WiFi settings reported success without configuring
anything. enable/disable now delegate to 'batocera-config wifi'.
The remaining stall (ApiSystem::ping() on opening Network Settings, up to 3x2s)
lives in the ES binary and is documented, not patched.
Verified against a stub connmanctl and inside the repacked image.
The stock image ships /usr/bin/351Files, its gptokeyb mapping and
scripts/setup/file_manager.sh, but removes the setup system from
es_systems.cfg and deletes 13 of the 15 scripts its gamelist referenced.
- re-add the setup system (upstream definition, before ports)
- trim the setup gamelist to the two scripts that still exist:
File Manager (351Files) and Copy USB roms to device
- add scripts/enable-setup-menu.sh for cards already flashed, since
/storage/.config/emulationstation/es_systems.cfg overrides the image default
- build script installs system/es/{es_systems.cfg,gamelist.xml}