A device with /dev/mali0 and no KGSL starts with the bridge on, so it draws
without a trip to Setup. An Adreno keeps Turnip and its automatic driver
route whatever the setting says, other GPUs keep the switch off until it is
turned on, and a choice made in Setup always wins. The GPU checks are read
once per process.
Setup has a "DXVK" row under the bridge's safe mode: automatic (the default, as before) or
Proton's own (SessionPrefs.bridgeProtonDxvk). With Proton's own, nothing is chosen and an earlier
automatic choice goes back to the Proton's copy.
Automatic choices also go back when the session is not a bridge session (the bridge turned off),
even when nothing is known of the Vulkan; a queued automatic swap is dropped with them, the
player's own queued swaps stay. Turning the bridge off, or choosing Proton's own, does it right
away in the background instead of at the next session. A failed detection in a bridge session
still keeps the last choice.
When nothing could be read (no apps-vulkan file and VulkanInfo failing), every automatic choice was
undone: a 1.1 phone went back to a DXVK that cannot start, and the next session flipped it again.
Now an automatic choice is undone only when the Vulkan is known and wants none.
Each Proton's swap or undo is on its own: a failure (a missing saved copy, an I/O error) is said in
the result and the remaining Protons still get theirs. When the Proton is the same but its saved copy
is gone, the choice stays and is said, instead of being forgotten with DXVK 1.10.3 still installed.
apps-vulkan: the version counts only with promotion=on; otherwise, and in files from servers before
the field (which wrote a promoted version even where it was not trusted), driver_api does. A file
whose driver_api is not the driver's own version (a driver update, a server rolled back) is left out.
Proton's DXVK 3 refuses a device below Vulkan 1.3, without shaderInt64 or
with less than 256 bytes of push constants, so on a Mali with a 1.1 driver
(OnePlus Ace 2V, Mali-G710, r32p1) every D3D9-11 game exits. On a bridge
session the forced-package step now picks the newest DXVK that starts on
what games see: 1.10.3 below Vulkan 1.3, 2.6.2 on 1.3 without shaderInt64
or 256 bytes of push constants, Proton's own otherwise.
What games see comes from <bridge cache dir>/apps-vulkan (a bridge with
Vulkan promotion writes it after its self-test) or else from the phone's
driver (vkinfo now reports shader_int64 and max_push_constants). These
choices are marked automatic: a later session with another answer swaps or
restores them, and the player's own choice in Components is never touched.
With BL_VKBRIDGE_CEF_ANGLE=vulkan the CEF GPU process stopped crashing but
Chromium still composited in software (gpu_compositing: disabled_software,
Steam "Disabling GPU acceleration: Disabled/RuntimeDetect", VizCompositorThread
at a full core). The GPU report shows why: "eglCreateWindowSurface failed with
error EGL_BAD_CONFIG". ANGLE's Vulkan backend on X11 gives every EGL config the
root window's visual; Chromium's X11 surface asks for a config whose
EGL_NATIVE_VISUAL_ID is the ARGB visual whenever a compositing manager owns
_NET_WM_CM_S0 (gamescope does), finds none and passes a null config.
bl_webhelper_layout() now adds +Vulkan +VulkanFromANGLE +DefaultANGLEVulkan
for ANGLE on Vulkan, so Skia composites through an xcb swapchain on ANGLE's
Vulkan device and no EGL window surface is needed. BL_VKBRIDGE_CEF_VULKAN=0
leaves it off; unset BL_VKBRIDGE_CEF_ANGLE still means Steam's stock layout.
Steam drops -cef-use-angle: on the bridge the CEF GPU process reports
ANGLE_OPENGL (ANGLE on Zink) with -cef-use-angle=vulkan on the command line.
An explicit BL_VKBRIDGE_CEF_ANGLE now also becomes --use-gl=angle
--use-angle=<backend> on steamwebhelper's own argv (stock process layout kept);
unset, nothing changes.
FrontEndActivity read vkBridge and vkBridgeSafe only when the Performance page
opened, so after an app restart (every APK update) Setup showed
"Mali bridge: off" while sessions used the bridge. refresh() now reads them.
VkBridgeComponent logs how long the server took to listen.
The bridge check tested the server's socket once. It now polls every 50 ms
(BL_VKBRIDGE_WAIT_MS, default 10000) on the first pass, logs how long the
socket took, stops waiting when the session is told to stop, and names what
is missing (icd, library, socket) when it falls back.
A second row under the GPU check, shown only when the bridge is on: "Safe mode" keeps the
bridge's proven workarounds wherever its startup probe has not verified this GPU and driver;
off is performance mode, native paths wherever the probe passed. It reaches the server as
<cache dir>/mode ("safe" or "performance", written atomically before the server starts or is
reused, so the in-app server's JNI start stays as it is) and the guest as VKBRIDGE_MODE, which
/root/.vkbridge-env can still override. A server without modes ignores the file.
Android empties an app's cacheDir when storage runs low; it took the bridge's pipeline
caches, the probe profile, stuck-pipelines and the env file from under a running server.
The directory is now files/vkbridge-pipelines (--cache-dir and the in-app start alike), and
an existing cache/vkbridge-pipelines is moved there once. Nothing trims filesDir, so the
server's own limits bound it (64 MB per pipeline cache file, a total limit in newer servers).
Steam's web helper now runs its GPU work inside the browser process
(--in-process-gpu --no-zygote, #566). On the Mali bridge that browser stops
answering after its first window, Steam restarts it every 10 s and then quits
before the login page. Bridge sessions keep the stock process layout
(BL_WEBHELPER_ARGS=stock); BL_VKBRIDGE_WEBHELPER_ARGS overrides it for tests.
Adreno sessions are unchanged.
On a GPU without KGSL, Setup offers an experimental Mali bridge switch
(off by default; ignored on Adreno). With it on, the session starts the
bridge server, points the Vulkan loader at the bridge ICD, names Mali's
device as the stand-in render node, runs the bridge self-test into the
session log, and caps vkd3d-proton's feature level at 12_0. Flatpak
sandboxes get the bridge ICD and socket. The DRM preload falls back to
flock in a bridge session when kcmp is refused, and a clock/thermal log
is kept beside the bridge logs.
A non-QTI gralloc handle (Arm/MediaTek gralloc on Mali) has no layout
the compositor can read, so the pool import guessed a linear dma-buf and
failed. On drivers other than Qualcomm's the compositor device enables
VK_ANDROID_external_memory_android_hardware_buffer and
VK_EXT_queue_family_foreign and imports such buffers as AHardwareBuffers;
the linear dma-buf stays as the last resort. Adreno/Turnip keeps the
same device and the dma-buf path.
The components job checks out external/MaliBridge and builds the guest
ICD, its manifest and the self-test as the vkbridge recipe; a submodule
counts in a recipe's identity by its pinned commit.
The server is built with the NDK into libvkbridge_server.so (a program
the app execs from its native library directory); the Linux ICD, its
manifest and the self-test go into the runtime assets with the other
glibc preloads, locally and in CI.
MaliBridge (https://github.com/ASEbrahim/MaliBridge) is a Vulkan
command bridge that lets the Linux side draw on a non-Adreno GPU (Mali)
through the phone's own Vulkan driver. Pinned at v0.5.3.1.
Steam's native QAM sliders now control Android brightness and media
volume, with device changes reflected back in Steam. Holding Android
Back for 700 ms sends a real held QAM button for Steam shortcuts. Single
and double taps keep the existing menu/QAM order and swap setting; the
double-tap window remains 500 ms.
- Discover native brightness through a guest backlight descriptor and
use Android's Modify system settings permission. Retain the latest
pending request until permission is granted or the session stops, with a
notification action for hidden sessions.
- Give DirectAudio and fallback AAudio sinks Android media volume/mute
callbacks, keeping software gain at unity. Recover when initial volume
state is absent or invalid, brightness acknowledgements differ or never
arrive, or the Display bridge is not ready yet.
- Forward other buttons while QAM is held in Steam Deck and Steam
Controller modes. Consume release after a hold and clear pending input
on overlays, focus loss, pause, disconnect, and teardown. The session
menu retains its compact Back choice without an explanatory paragraph.
Validation:
- Local ARM64 release build passed. The latest timing change passed 20
targeted input tests, including release at 600 ms remaining a tap. The
preceding full feature check passed 61 Android input/device-control
tests, with permission cases on API 28 and 33.
- 66 Linux Python/Node/C tests passed for device controls, audio
recovery, Steam compatibility, and file dialogs. Both audio sink modules
rebuilt with Android NDK 27.2.
- Release-signed integration build 14d12cf0 is installed on the unrooted
AYN Thor, carrying the Android/input changes through 749ade4a plus the
existing portrait changes. Device checks confirmed the configured
QAM/menu tap order, Back dismissal, removal of the inline menu
description, and held QAM overlapping X with clean release. Hands-on
testing confirmed this installed build works.
- Earlier Thor testing covered brightness permission and pending
requests, brightness synchronization in both directions, native volume
changes, hardware volume key readback, and both sink backends. Failure
recovery is covered by the automated tests above.
The subsequent queued-notification fix in 4d7ceb23 is covered by the
Linux regression suite and has not replaced the Thor install.
Bluetooth/wired output switching and other device models remain
untested. CI and CodeRabbit monitoring continue against the pushed head.
A tag that matches no longer vouches for the build: FEX and FEXServer must
also load (ld.so --list), otherwise the session reinstalls it and x86
programs use the Steam or system FEX meanwhile.
The pin moved there from droiddeck-steam-x64, so the client settings sheet
and its test read it from the new script. The x86-64 client notes now name
the updater and Components' FEX watcher.
A failed catalog install is retried after an hour or when the session names
another build, a missing updater falls back to the FEX already here, and the
archive download gives up on a stalled connection.
The Sentry GDExtension was turned off for every Godot game that ships it,
because its crash handler takes the .NET runtime down under FEX. GDScript
games such as Pathogenic call SentrySDK from their scripts, so without the
extension their autoloads fail to parse and the game quits. Only Godot .NET
builds have it turned off now.
The launch view lives under the app's files folder, so games saw an
install path like Z:\data\user\0\com.droiddeck.launcher\files\xbox-runtime
\views\launch-<n>\droiddeck-xbox-<id>. Forza Horizon checks that length and
quits with "Installation path is too long". The view is now mapped to a
free drive letter in the prefix (X: first) for the length of the launch.