fix: resolve SHM hang, DRM device mismatch, and duplicate VAAPI context

BUG-2 (HIGH): SHM Buffer event caused permanent hang
  In the ZwlrScreencopyFrameV1 dispatcher, receiving a SHM Buffer event
  left in_flight_surface stuck at AllocQueued forever, preventing
  queue_alloc_frame() from requesting new frames.
  Fix: treat Buffer as a metadata offer (v3 protocol), wait for
  BufferDone to decide failure, and add AllocQueued state guard to
  LinuxDmabuf handler.

BUG-3 (MEDIUM): Portal backend picked wrong GPU on multi-GPU systems
  state_portal.rs hardcoded /dev/dri/renderD128 then renderD129, which
  selects the wrong GPU when PipeWire uses a different device.
  Fix: extract find_drm_render_nodes() as shared utility; defer DRM
  device selection to first PipeWire frame; test each candidate with
  av_hwframe_transfer_data to find the GPU that can actually import
  the DMA-BUF frame.

BUG-4 (LOW): VAAPI device context created twice unnecessarily
  try_finalize_output() created an AvHwDevCtx stored in EverythingButFmt,
  but negotiate_format() discarded it (_hw_device_ctx) and EncState::new
  created a new one.
  Fix: thread the existing hw_device_ctx through negotiate_format() and
  create_encoder() to EncState::new() which reuses it when provided.

src/avhw.rs

@@ -1,4 +1,6 @@
 use std::ffi::CString;
+use std::mem;
+use std::os::fd::AsRawFd;
 use std::path::Path;
 use std::ptr;
 
@@ -7,6 +9,7 @@ use ffmpeg_next as ff;
 use ffmpeg_next::ffi;
 use ffmpeg_next::packet::Mut as _;
 
+use crate::cap_portal::PwDmaBufFrame;
 use crate::transform::{transpose_if_transform_transposed, Transform};
 
 // ---------------------------------------------------------------------------
@@ -123,6 +126,74 @@ impl Drop for AvHwFrameCtx {
     }
 }
 
+/// Test whether `drm_device` can import the PipeWire DMA-BUF frame via VAAPI.
+pub fn test_dma_buf_import(drm_device: &Path, frame: &PwDmaBufFrame) -> Result<()> {
+    let hw_dev = AvHwDevCtx::new_vaapi(drm_device)?;
+    let frames = AvHwFrameCtx::for_capture(
+        &hw_dev,
+        frame.width,
+        frame.height,
+        ff::format::Pixel::RGBZ,
+    )?;
+
+    // SAFETY: AVDRMFrameDescriptor is a C POD struct. Zero-initialization is the
+    // expected FFmpeg setup before filling the fields used below.
+    let mut desc: ffi::AVDRMFrameDescriptor = unsafe { mem::zeroed() };
+    desc.nb_objects = 1;
+    desc.objects[0].fd = frame.fd.as_raw_fd();
+    desc.objects[0].size = 0;
+    desc.objects[0].format_modifier = frame.modifier;
+    desc.nb_layers = 1;
+    desc.layers[0].format = frame.format;
+    desc.layers[0].nb_planes = 1;
+    desc.layers[0].planes[0].object_index = 0;
+    desc.layers[0].planes[0].offset = frame.offset as isize;
+    desc.layers[0].planes[0].pitch = frame.stride as isize;
+
+    let desc_box = Box::new(desc);
+    let mut raw_frame = ff::frame::Video::empty();
+    // SAFETY: raw_frame owns a valid AVFrame. data[0] is used by FFmpeg's
+    // DRM_PRIME frame convention to point at an AVDRMFrameDescriptor. The Box is
+    // recovered before every return path below.
+    unsafe {
+        let raw_ptr = raw_frame.as_mut_ptr();
+        (*raw_ptr).data[0] = Box::into_raw(desc_box) as *mut u8;
+        (*raw_ptr).format = ffi::AVPixelFormat::AV_PIX_FMT_DRM_PRIME as i32;
+        (*raw_ptr).width = frame.width as i32;
+        (*raw_ptr).height = frame.height as i32;
+    }
+
+    let mut hw_frame = ff::frame::Video::empty();
+    // SAFETY: frames is an initialized AVHWFramesContext and hw_frame is a valid
+    // writable AVFrame wrapper.
+    let ret = unsafe { ffi::av_hwframe_get_buffer(frames.as_ptr(), hw_frame.as_mut_ptr(), 0) };
+    if ret < 0 {
+        // SAFETY: data[0] still contains the Box pointer installed above.
+        unsafe {
+            let _ = Box::from_raw((*raw_frame.as_ptr()).data[0] as *mut ffi::AVDRMFrameDescriptor);
+            (*raw_frame.as_mut_ptr()).data[0] = ptr::null_mut();
+        }
+        bail!("av_hwframe_get_buffer failed: error {ret}");
+    }
+
+    // SAFETY: hw_frame is a valid VAAPI frame allocated from `frames`; raw_frame
+    // is a DRM_PRIME source frame whose descriptor describes `frame`'s DMA-BUF.
+    let ret = unsafe { ffi::av_hwframe_transfer_data(hw_frame.as_mut_ptr(), raw_frame.as_ptr(), 0) };
+
+    // SAFETY: data[0] still contains the Box pointer installed above. Recover it
+    // before checking the transfer result so all paths clean up the descriptor.
+    unsafe {
+        let _ = Box::from_raw((*raw_frame.as_ptr()).data[0] as *mut ffi::AVDRMFrameDescriptor);
+        (*raw_frame.as_mut_ptr()).data[0] = ptr::null_mut();
+    }
+
+    if ret < 0 {
+        bail!("av_hwframe_transfer_data failed: error {ret}");
+    }
+
+    Ok(())
+}
+
 // ---------------------------------------------------------------------------
 // EncState
 // ---------------------------------------------------------------------------
@@ -139,8 +210,8 @@ pub struct EncState {
 
 unsafe impl Send for EncState {}
 
-#[allow(clippy::too_many_arguments)]
 impl EncState {
+    #[allow(clippy::too_many_arguments)]
     pub fn new(
         drm_device: &Path,
         output_path: &Path,
@@ -152,12 +223,16 @@ impl EncState {
         gop_size: u32,
         fps: u32,
         transform: Transform,
+        existing_hw_ctx: Option<AvHwDevCtx>,
     ) -> Result<Self> {
         tracing::info!(
             "EncState::new: {width}x{height} enc={enc_width}x{enc_height} transform={transform:?}"
         );
-        // 1. VAAPI device
+        // 1. VAAPI device — reuse existing context if provided
-        let hw_device_ctx = AvHwDevCtx::new_vaapi(drm_device)?;
+        let hw_device_ctx = match existing_hw_ctx {
+            Some(ctx) => ctx,
+            None => AvHwDevCtx::new_vaapi(drm_device)?,
+        };
 
         // 2. Frame context for capture (XRGB/RGBZ)
         let frames_rgb =
@@ -482,6 +557,7 @@ pub fn create_encoder(
     transform: Transform,
     bitrate: Option<u64>,
     gop_size: Option<u32>,
+    existing_hw_ctx: Option<AvHwDevCtx>,
 ) -> Result<EncState> {
     let (enc_w, enc_h) =
         transpose_if_transform_transposed(transform, width as i32, height as i32);
@@ -500,6 +576,7 @@ pub fn create_encoder(
         actual_gop_size,
         fps,
         transform,
+        existing_hw_ctx,
     )
 }
 

src/state.rs

@@ -188,23 +188,29 @@ pub struct State<S: CaptureSource> {
 // Helpers
 // ---------------------------------------------------------------------------
 
+/// Scan /dev/dri for all available DRM render nodes (renderD*), sorted by node number.
+pub(crate) fn find_drm_render_nodes() -> Vec<PathBuf> {
+    let Ok(entries) = std::fs::read_dir("/dev/dri") else {
+        return Vec::new();
+    };
+
+    let mut nodes: Vec<(u32, PathBuf)> = entries
+        .filter_map(Result::ok)
+        .filter_map(|entry| {
+            let path = entry.path();
+            let name = path.file_name()?.to_str()?;
+            let number = name.strip_prefix("renderD")?.parse::<u32>().ok()?;
+            std::fs::metadata(&path).ok()?;
+            Some((number, path))
+        })
+        .collect();
+    nodes.sort_by_key(|(number, _)| *number);
+    nodes.into_iter().map(|(_, path)| path).collect()
+}
+
 /// Scan /dev/dri for the first available DRM render node (renderD*).
 fn find_drm_render_node() -> Option<PathBuf> {
-    std::fs::read_dir("/dev/dri")
+    find_drm_render_nodes().into_iter().next()
-        .ok()?
-        .filter_map(|e| e.ok())
-        .filter(|e| {
-            e.file_name()
-                .to_str()
-                .map(|s| s.starts_with("renderD"))
-                .unwrap_or(false)
-        })
-        .filter_map(|e| {
-            let path = e.path();
-            std::fs::metadata(&path).ok()?;
-            Some(path)
-        })
-        .min_by_key(|e| e.to_path_buf())
 }
 
 impl<S: CaptureSource> State<S> {
@@ -584,18 +590,18 @@ impl<S: CaptureSource> State<S> {
             EncConstructionStage::EverythingButFmt {
                 output_info,
                 output,
-                hw_device_ctx: _hw_device_ctx,
+                hw_device_ctx,
                 cap,
                 screencopy_manager,
                 dmabuf,
-            } => (output_info, output, cap, screencopy_manager, dmabuf),
+            } => (output_info, output, hw_device_ctx, cap, screencopy_manager, dmabuf),
             other => {
                 tracing::warn!("negotiate_format: not in EverythingButFmt stage");
                 self.stage = other;
                 return;
             }
         };
-        let (output_info, output, cap, screencopy_manager, dmabuf) = stage_data;
+        let (output_info, output, hw_device_ctx, cap, screencopy_manager, dmabuf) = stage_data;
         let drm_path = self.resolve_drm_path();
         let fps = self.args.fps;
         let bitrate = self.args.bitrate.unwrap_or_else(|| {
@@ -610,6 +616,7 @@ impl<S: CaptureSource> State<S> {
             output_info.transform,
             self.args.bitrate,
             self.args.gop_size,
+            Some(hw_device_ctx),
         ) {
             Ok(enc) => enc,
             Err(e) => {
@@ -1228,11 +1235,13 @@ impl Dispatch<ZwlrScreencopyFrameV1, ()> for State<CapWlrScreencopy> {
         _qhandle: &QueueHandle<State<CapWlrScreencopy>>,
     ) {
         match event {
+            // SHM buffer offer — in v3 the compositor enumerates supported buffer
+            // types (buffer and/or linux_dmabuf) before buffer_done.  We only
+            // support DMA-BUF, so just log and wait for linux_dmabuf / buffer_done.
             ScreencopyFrameEvent::Buffer { .. } => {
-                tracing::warn!(
+                tracing::debug!(
-                    "Received SHM Buffer event — only DMA-BUF capture is supported. Ignoring."
+                    "Received SHM Buffer offer — only DMA-BUF capture is supported"
                 );
-                return;
             }
             ScreencopyFrameEvent::LinuxDmabuf {
                 format,
@@ -1240,6 +1249,12 @@ impl Dispatch<ZwlrScreencopyFrameV1, ()> for State<CapWlrScreencopy> {
                 height,
             } => {
                 tracing::debug!("Screencopy LinuxDmabuf: format={format}, {width}x{height}");
+
+                if !matches!(state.in_flight_surface, InFlightSurface::AllocQueued) {
+                    tracing::warn!("Received LinuxDmabuf while no frame allocation was queued");
+                    return;
+                }
+
                 if matches!(state.stage, EncConstructionStage::EverythingButFmt { .. }) {
                     state.negotiate_format(format, width, height);
                     if state.errored {
@@ -1251,6 +1266,20 @@ impl Dispatch<ZwlrScreencopyFrameV1, ()> for State<CapWlrScreencopy> {
                 }
                 state.on_frame_allocd((), format, width, height);
             }
+            // v3 terminal event: all buffer offers have been enumerated.
+            // If still AllocQueued, the compositor never sent linux_dmabuf —
+            // DMA-BUF screencopy is unsupported, so we must error out.
+            ScreencopyFrameEvent::BufferDone => {
+                if matches!(state.in_flight_surface, InFlightSurface::AllocQueued) {
+                    tracing::error!(
+                        "Compositor did not offer DMA-BUF screencopy (only SHM); \
+                         DMA-BUF capture is required"
+                    );
+                    state.in_flight_surface = InFlightSurface::None;
+                    proxy.destroy();
+                    state.errored = true;
+                }
+            }
             ScreencopyFrameEvent::Ready {
                 tv_sec_hi,
                 tv_sec_lo,

src/state_portal.rs

@@ -1,5 +1,6 @@
 // 采集门户状态模块 —— 通过 PipeWire/DMA-BUF 进行屏幕采集并编码
 use std::mem;
+use std::os::fd::AsRawFd;
 use std::path::PathBuf;
 
 use anyhow::{bail, Result};
@@ -39,8 +40,8 @@ pub struct StatePortal {
     errored: bool,
     /// 是否为第一帧（首帧跳过帧率限制）
     first_frame: bool,
-    /// DRM 渲染设备路径（如 /dev/dri/renderD128）
+    /// DRM 渲染设备路径（如 /dev/dri/renderD128）；None 表示首帧自动检测
-    drm_device: PathBuf,
+    drm_device: Option<PathBuf>,
     /// 第一帧的时间戳（纳秒），用于计算相对 PTS
     first_pts_ns: Option<i64>,
 }
@@ -51,7 +52,11 @@ impl StatePortal {
     /// 初始化 DRM 设备路径和 PipeWire 采集端点，编码器延迟到第一帧到达时创建。
     pub fn new(args: Args) -> Result<Self> {
         let drm_device = resolve_drm_device(&args)?;
-        tracing::info!("Using DRM device: {}", drm_device.display());
+        if let Some(ref drm_device) = drm_device {
+            tracing::info!("Using DRM device: {}", drm_device.display());
+        } else {
+            tracing::info!("DRM device auto-detection enabled");
+        }
 
         let cap = CapPortal::new(&args)?;
 
@@ -92,8 +97,9 @@ impl StatePortal {
                             frame.modifier
                         );
 
+                        let drm_path = self.resolve_drm_device_for_frame(&frame)?;
                         let enc = avhw::create_encoder(
-                            &self.drm_device,
+                            &drm_path,
                             self.args.output.as_ref(),
                             frame.width,
                             frame.height,
@@ -101,6 +107,7 @@ impl StatePortal {
                             Transform::Normal,
                             self.args.bitrate,
                             self.args.gop_size,
+                            None,
                         )?;
 
                         self.enc = Some(enc);
@@ -128,6 +135,44 @@ impl StatePortal {
         Ok(true)
     }
 
+    fn resolve_drm_device_for_frame(&mut self, frame: &PwDmaBufFrame) -> Result<PathBuf> {
+        if let Some(ref drm) = self.drm_device {
+            return Ok(drm.clone());
+        }
+
+        let candidates = crate::state::find_drm_render_nodes();
+        if candidates.is_empty() {
+            bail!("No DRM render device found. Specify --drm-device.");
+        }
+
+        let mut failures = Vec::new();
+        for candidate in &candidates {
+            match crate::avhw::test_dma_buf_import(candidate, frame) {
+                Ok(()) => {
+                    tracing::info!(
+                        "Auto-selected DRM device: {} (can import PipeWire DMA-BUF)",
+                        candidate.display()
+                    );
+                    self.drm_device = Some(candidate.clone());
+                    return Ok(candidate.clone());
+                }
+                Err(err) => {
+                    tracing::debug!(
+                        "DRM device {} cannot import frame: {err:#}",
+                        candidate.display()
+                    );
+                    failures.push(format!("{}: {err:#}", candidate.display()));
+                }
+            }
+        }
+
+        bail!(
+            "No DRM render device can import the PipeWire DMA-BUF frame. \
+             Specify --drm-device. Tried: {}",
+            failures.join("; ")
+        )
+    }
+
     /// 处理单帧 DMA-BUF 数据
     ///
     /// 完整的帧处理流水线：
@@ -311,25 +356,14 @@ fn build_drm_descriptor(frame: &PwDmaBufFrame) -> ffi::AVDRMFrameDescriptor {
     desc
 }
 
-use std::os::fd::AsRawFd;
-
 /// 解析 DRM 渲染设备路径
 ///
-/// 优先使用命令行指定的设备路径，否则依次尝试
+/// 仅使用命令行指定的设备路径；未指定则在首帧到达时自动检测。
-/// `/dev/dri/renderD128` 和 `/dev/dri/renderD129`。
+fn resolve_drm_device(args: &Args) -> Result<Option<PathBuf>> {
-fn resolve_drm_device(args: &Args) -> Result<PathBuf> {
     if let Some(ref drm) = args.drm_device {
-        return Ok(PathBuf::from(drm));
+        return Ok(Some(PathBuf::from(drm)));
     }
-
+    Ok(None)
-    for render in &["/dev/dri/renderD128", "/dev/dri/renderD129"] {
-        let path = PathBuf::from(render);
-        if path.exists() {
-            return Ok(path);
-        }
-    }
-
-    bail!("No DRM render device found. Specify --drm-device.")
 }
 
 #[cfg(test)]