mirror of
https://github.com/Droid-VM/crosvm.git
synced 2026-11-17 06:58:03 +00:00
M4 (VPU_DESIGN.md 7.1). The crate was a probe's library; the camera device needs five things it did not have: - an open that unwinds: Camera::open_with builds through an OpenGuard that releases every handle made so far, in Camera::drop's order, on any failure, and takes an optional StateListener; Camera::open is open_with(.., None), so camera_probe is unchanged; - the device-state callbacks with a context: ACameraDeviceStateCallbacks now points at a Box<StateContext> the camera keeps, and the listener is called with DeviceState::Disconnected / Error(code) from the binder thread that delivers them, which is how the guest learns the host took the camera away; - AImageReader_acquireLatestImage when the NDK has it (the ndk_api! macro grew an optional group resolved with .ok()): next_frame_latest drops the frames a slow consumer left behind, and is next_frame where the symbol is missing; - the two tables the device answers ENUM_FRAMEINTERVALS and S_PARM from, SCALER_AVAILABLE_MIN_FRAME_DURATIONS (YUV_420_888 only) and CONTROL_AE_AVAILABLE_TARGET_FPS_RANGES, in CameraInfo and in camera_probe list; - Plane::as_ptr/len for the interleaved-chroma copy, which a per-plane slice cannot express. libc is a real dependency now and the crate is a workspace member, so it and camera_probe build under cargo on the build host (logs/vpu_wp/scratch-m4). apex_available: com.android.virt, because libdevices, which depends on this since the camera device, is available to that APEX and soong refuses the dependency otherwise. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BkN7vYhuc1w9iVuqsHzSn
61 lines
2.3 KiB
61 lines
2.3 KiB