unom/punktfunk
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases 20 Wiki Activity

fix(presenter): resize crash — never vkDeviceWaitIdle while the pump decodes

Browse Source 
vkDeviceWaitIdle's external-sync rule claims EVERY queue on the device;
with Vulkan Video the session pump concurrently submits FFmpeg decode
work to the shared device's video queue, so the resize path's wait-idle
(and the video-image/staging rebuilds') raced it — observed as a crash
on window resize mid-stream (software/VAAPI never touched the device
from the pump, which is why this only appeared now).

Mid-session quiescing is now fence-only ( waits the single
in-flight fence, which covers every command buffer WE submitted), and
the replaced swapchain + its per-image semaphores/overlay targets are
parked in a retire list — the presentation engine may still hold the old
swapchain's final semaphore wait, which no fence covers — and destroyed
after the next present on the successor completes a fence cycle. The one
legitimate device-wide idle left is teardown, and the run loop now JOINS
the pump thread first (SessionHandle carries the JoinHandle; quick — the
pump notices stop within its 20 ms receive timeout).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Enrico Bühler
2026-07-07 22:20:15 +02:00
parent 349d16382e
commit c299a41a67
3 changed files with 113 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files
crates/pf-presenter/src/run.rs
+21 -4
View File
@@ -170,6 +170,17 @@ impl StreamState {
}
}
/// Stop the pump and JOIN its thread — required before any device-wide idle or
/// teardown (the pump submits decode work to the shared device). Quick: the pump
/// notices `stop` within its 20 ms receive timeout, and on a normal end it's
/// already returning.
fn shutdown(mut self) {
self.handle.stop.store(true, Ordering::SeqCst);
if let Some(t) = self.handle.thread.take() {
let _ = t.join();
}
}
/// Deliberate user exit (chord / window close): release capture, close with
/// QUIT_CLOSE_CODE so the host tears down instead of lingering, stop the pump.
/// The pump then emits `Ended(None)` — the loop's normal end path picks it up.
@@ -530,7 +541,9 @@ fn run_inner(mut opts: SessionOpts, mut mode: ModeCtl) -> Result<Option<Outcome>
}
ModeCtl::Browse(_) => {
tracing::warn!(%msg, "connect failed — back to the library");
stream = None;
if let Some(st) = stream.take() {
st.shutdown();
}
mouse.set_relative_mouse_mode(&window, false);
mouse.show_cursor(true);
if let Some(o) = overlay.as_mut() {
@@ -550,7 +563,9 @@ fn run_inner(mut opts: SessionOpts, mut mode: ModeCtl) -> Result<Option<Outcome>
ModeCtl::Single(_) => break 'main Some(Outcome::Ended(reason)),
ModeCtl::Browse(_) => {
window.set_title(&opts.window_title).ok();
stream = None;
if let Some(st) = stream.take() {
st.shutdown();
}
if let Some(o) = overlay.as_mut() {
o.session_phase(SessionPhase::Ended(reason.as_deref()));
}
@@ -736,8 +751,10 @@ fn run_inner(mut opts: SessionOpts, mut mode: ModeCtl) -> Result<Option<Outcome>
}
};
if let Some(st) = &stream {
st.handle.stop.store(true, Ordering::SeqCst);
// Join the pump BEFORE the device-wide idle: its decode submissions on the shared
// device would race vkDeviceWaitIdle otherwise.
if let Some(st) = stream.take() {
st.shutdown();
}
// Overlay resources live on the presenter's device: quiesce the queue first, drop
// the overlay (its Drop destroys the Skia surfaces), THEN the presenter tears down.