Call device lost callback when it is replaced, or when the global is dropped. (#5168)

This fixes two cases where a DeviceLostClosureC might not be consumed
before it is dropped, which is a requirement:

1) When the closure is replaced, this ensures the to-be-dropped closure
is invoked.
2)  When the global is dropped, this ensures that the closure is invoked
before it is dropped.

The first of these two cases is tested in a new test,
DEVICE_LOST_REPLACED_CALLBACK. The second case has a stub,
always-skipped test, DROPPED_GLOBAL_THEN_DEVICE_LOST. The test is
always-skipped because there does not appear to be a way to drop the
global from within a test. Nor is there any other way to reach
Device.prepare_to_die without having first dropping the device.

CHANGELOG.md

@@ -65,6 +65,9 @@ Bottom level categories:
 
 ### Bug Fixes
 
+#### General
+- Device lost callbacks are invoked when replaced and when global is dropped. By @bradwerth in [#5168](https://github.com/gfx-rs/wgpu/pull/5168)
+
 #### DX12
 - Fix `panic!` when dropping `Instance` without `InstanceFlags::VALIDATION`. By @hakolao in [#5134](https://github.com/gfx-rs/wgpu/pull/5134)
 

tests/tests/device.rs

@@ -549,3 +549,69 @@ static DEVICE_DROP_THEN_LOST: GpuTestConfiguration = GpuTestConfiguration::new()
             "Device lost callback should have been called."
         );
     });
+
+#[gpu_test]
+static DEVICE_LOST_REPLACED_CALLBACK: GpuTestConfiguration = GpuTestConfiguration::new()
+    .parameters(TestParameters::default())
+    .run_sync(|ctx| {
+        // This test checks that a device_lost_callback is called when it is
+        // replaced by another callback.
+        let was_called = std::sync::Arc::<std::sync::atomic::AtomicBool>::new(false.into());
+
+        // Set a LoseDeviceCallback on the device.
+        let was_called_clone = was_called.clone();
+        let callback = Box::new(move |reason, _m| {
+            was_called_clone.store(true, std::sync::atomic::Ordering::SeqCst);
+            assert!(
+                matches!(reason, wgt::DeviceLostReason::ReplacedCallback),
+                "Device lost info reason should match DeviceLostReason::ReplacedCallback."
+            );
+        });
+        ctx.device.set_device_lost_callback(callback);
+
+        // Replace the callback.
+        let replacement_callback = Box::new(move |_r, _m| {});
+        ctx.device.set_device_lost_callback(replacement_callback);
+
+        assert!(
+            was_called.load(std::sync::atomic::Ordering::SeqCst),
+            "Device lost callback should have been called."
+        );
+    });
+
+#[gpu_test]
+static DROPPED_GLOBAL_THEN_DEVICE_LOST: GpuTestConfiguration = GpuTestConfiguration::new()
+    .parameters(TestParameters::default().skip(FailureCase::always()))
+    .run_sync(|ctx| {
+        // What we want to do is to drop the Global, forcing a code path that
+        // eventually calls Device.prepare_to_die, without having first dropped
+        // the device. This models what might happen in a user agent that kills
+        // wgpu without providing a more orderly shutdown. In such a case, the
+        // device lost callback should be invoked with the message "Device is
+        // dying."
+        let was_called = std::sync::Arc::<std::sync::atomic::AtomicBool>::new(false.into());
+
+        // Set a LoseDeviceCallback on the device.
+        let was_called_clone = was_called.clone();
+        let callback = Box::new(move |reason, message| {
+            was_called_clone.store(true, std::sync::atomic::Ordering::SeqCst);
+            assert!(
+                matches!(reason, wgt::DeviceLostReason::Dropped),
+                "Device lost info reason should match DeviceLostReason::Dropped."
+            );
+            assert!(
+                message == "Device is dying.",
+                "Device lost info message is \"{}\" and it should be \"Device is dying.\".",
+                message
+            );
+        });
+        ctx.device.set_device_lost_callback(callback);
+
+        // TODO: Drop the Global, somehow.
+
+        // Confirm that the callback was invoked.
+        assert!(
+            was_called.load(std::sync::atomic::Ordering::SeqCst),
+            "Device lost callback should have been called."
+        );
+    });

wgpu-core/src/device/global.rs

@@ -2247,8 +2247,8 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
         }
     }
 
-    // This closure will be called exactly once during "lose the device"
+    // This closure will be called exactly once during "lose the device",
-    // or when the device is dropped, if it was never lost.
+    // or when it is replaced.
     pub fn device_set_device_lost_closure<A: HalApi>(
         &self,
         device_id: DeviceId,
@@ -2258,6 +2258,12 @@ impl<G: GlobalIdentityHandlerFactory> Global<G> {
 
         if let Ok(device) = hub.devices.get(device_id) {
             let mut life_tracker = device.lock_life();
+            if let Some(existing_closure) = life_tracker.device_lost_closure.take() {
+                // It's important to not hold the lock while calling the closure.
+                drop(life_tracker);
+                existing_closure.call(DeviceLostReason::ReplacedCallback, "".to_string());
+                life_tracker = device.lock_life();
+            }
             life_tracker.device_lost_closure = Some(device_lost_closure);
         }
     }

wgpu-core/src/device/resource.rs

@@ -3380,6 +3380,11 @@ impl<A: HalApi> Device<A> {
             current_index,
             self.command_allocator.lock().as_mut().unwrap(),
         );
+        if let Some(device_lost_closure) = life_tracker.device_lost_closure.take() {
+            // It's important to not hold the lock while calling the closure.
+            drop(life_tracker);
+            device_lost_closure.call(DeviceLostReason::Dropped, "Device is dying.".to_string());
+        }
         #[cfg(feature = "trace")]
         {
             *self.trace.lock() = None;

wgpu-types/src/lib.rs

@@ -7099,4 +7099,18 @@ pub enum DeviceLostReason {
     Unknown = 0,
     /// After Device::destroy
     Destroyed = 1,
+    /// After Device::drop
+    ///
+    /// WebGPU does not invoke the device lost callback when the device is
+    /// dropped to prevent garbage collection from being observable. In wgpu,
+    /// we invoke the callback on drop to help with managing memory owned by
+    /// the callback.
+    Dropped = 2,
+    /// After replacing the device_lost_callback
+    ///
+    /// WebGPU does not have a concept of a device lost callback, but wgpu
+    /// does. wgpu guarantees that any supplied callback will be invoked
+    /// exactly once before it is dropped, which helps with managing the
+    /// memory owned by the callback.
+    ReplacedCallback = 3,
 }