webgpu: Implement device lost (#32354)

* device lost promise should be init at creation of device object

* device lost impl

* lock for device poll

workaround for wgpu deadlocks

* expect

* Less lost reason reasoning in script

components/script/dom/bindings/codegen/Bindings.conf

@@ -169,7 +169,7 @@ DOMInterfaces = {
 },
 
 'GPUDevice': {
-    'inRealms': ['PopErrorScope', 'GetLost', 'CreateComputePipelineAsync', 'CreateRenderPipelineAsync'],
+    'inRealms': ['PopErrorScope', 'CreateComputePipelineAsync', 'CreateRenderPipelineAsync'],
 }
 
 }

components/script/dom/globalscope.rs

@@ -56,8 +56,9 @@ use script_traits::{
 };
 use servo_url::{ImmutableOrigin, MutableOrigin, ServoUrl};
 use uuid::Uuid;
-use webgpu::WebGPUDevice;
+use webgpu::{DeviceLostReason, WebGPUDevice};
 
+use super::bindings::codegen::Bindings::WebGPUBinding::GPUDeviceLostReason;
 use super::bindings::trace::HashMapTracedValues;
 use crate::dom::bindings::cell::{DomRefCell, RefMut};
 use crate::dom::bindings::codegen::Bindings::BroadcastChannelBinding::BroadcastChannelMethods;
@@ -3093,16 +3094,25 @@ impl GlobalScope {
             .insert(device.id(), Dom::from_ref(device));
     }
 
-    pub fn remove_gpu_device(&self, device: WebGPUDevice) {
+    pub fn gpu_device_lost(&self, device: WebGPUDevice, reason: DeviceLostReason, msg: String) {
-        let _ = self.gpu_devices.borrow_mut().remove(&device);
+        let reason = match reason {
+            DeviceLostReason::Unknown => GPUDeviceLostReason::Unknown,
+            DeviceLostReason::Destroyed => GPUDeviceLostReason::Destroyed,
+        };
+        let _ac = enter_realm(&*self);
+        self.gpu_devices
+            .borrow_mut()
+            .remove(&device)
+            .expect("GPUDevice should still exists")
+            .lose(reason, msg);
     }
 
     pub fn handle_uncaptured_gpu_error(&self, device: WebGPUDevice, error: webgpu::Error) {
-        self.gpu_devices
+        if let Some(gpu_device) = self.gpu_devices.borrow().get(&device) {
-            .borrow()
+            gpu_device.fire_uncaptured_error(error);
-            .get(&device)
+        } else {
-            .expect("GPUDevice not found")
+            warn!("Recived error for lost GPUDevice!")
-            .fire_uncaptured_error(error);
+        }
     }
 
     pub fn handle_gamepad_event(&self, gamepad_event: GamepadEvent) {

components/script/dom/gpudevice.rs

@@ -84,8 +84,9 @@ pub struct GPUDevice {
     #[no_trace]
     device: webgpu::WebGPUDevice,
     default_queue: Dom<GPUQueue>,
+    /// <https://gpuweb.github.io/gpuweb/#dom-gpudevice-lost>
     #[ignore_malloc_size_of = "promises are hard"]
-    lost_promise: DomRefCell<Option<Rc<Promise>>>,
+    lost_promise: DomRefCell<Rc<Promise>>,
     valid: Cell<bool>,
 }
 
@@ -100,6 +101,7 @@ impl GPUDevice {
         device: webgpu::WebGPUDevice,
         queue: &GPUQueue,
         label: String,
+        lost_promise: Rc<Promise>,
     ) -> Self {
         Self {
             eventtarget: EventTarget::new_inherited(),
@@ -111,7 +113,7 @@ impl GPUDevice {
             label: DomRefCell::new(USVString::from(label)),
             device,
             default_queue: Dom::from_ref(queue),
-            lost_promise: DomRefCell::new(None),
+            lost_promise: DomRefCell::new(lost_promise),
             valid: Cell::new(true),
         }
     }
@@ -131,9 +133,18 @@ impl GPUDevice {
         let queue = GPUQueue::new(global, channel.clone(), queue);
         let limits = GPUSupportedLimits::new(global, limits);
         let features = GPUSupportedFeatures::Constructor(global, None, features).unwrap();
+        let lost_promise = Promise::new(global);
         let device = reflect_dom_object(
             Box::new(GPUDevice::new_inherited(
-                channel, adapter, extensions, &features, &limits, device, &queue, label,
+                channel,
+                adapter,
+                extensions,
+                &features,
+                &limits,
+                device,
+                &queue,
+                label,
+                lost_promise,
             )),
             global,
         );
@@ -206,18 +217,11 @@ impl GPUDevice {
     }
 
     /// <https://gpuweb.github.io/gpuweb/#lose-the-device>
-    pub fn lose(&self, reason: GPUDeviceLostReason) {
+    pub fn lose(&self, reason: GPUDeviceLostReason, msg: String) {
-        if let Some(ref lost_promise) = *self.lost_promise.borrow() {
+        let ref lost_promise = *self.lost_promise.borrow();
-            let global = &self.global();
+        let global = &self.global();
-            let msg = match reason {
+        let lost = GPUDeviceLostInfo::new(global, msg.into(), reason);
-                GPUDeviceLostReason::Unknown => "Unknown reason for your device loss.",
+        lost_promise.resolve_native(&*lost);
-                GPUDeviceLostReason::Destroyed => {
-                    "Device self-destruction sequence activated successfully!"
-                },
-            };
-            let lost = GPUDeviceLostInfo::new(global, msg.into(), reason);
-            lost_promise.resolve_native(&*lost);
-        }
     }
 }
 
@@ -248,10 +252,8 @@ impl GPUDeviceMethods for GPUDevice {
     }
 
     /// <https://gpuweb.github.io/gpuweb/#dom-gpudevice-lost>
-    fn GetLost(&self, comp: InRealm) -> Fallible<Rc<Promise>> {
+    fn Lost(&self) -> Rc<Promise> {
-        let promise = Promise::new_in_current_realm(comp);
+        self.lost_promise.borrow().clone()
-        *self.lost_promise.borrow_mut() = Some(promise.clone());
-        Ok(promise)
     }
 
     /// <https://gpuweb.github.io/gpuweb/#dom-gpudevice-createbuffer>
@@ -1000,8 +1002,6 @@ impl GPUDeviceMethods for GPUDevice {
         if self.valid.get() {
             self.valid.set(false);
 
-            self.lose(GPUDeviceLostReason::Destroyed);
-
             if let Err(e) = self
                 .channel
                 .0

components/script/dom/webidls/WebGPU.webidl

@@ -1075,14 +1075,13 @@ enum GPUDeviceLostReason {
     "destroyed",
 };
 
-[Exposed=(Window, DedicatedWorker), Pref="dom.webgpu.enabled"]
+[Exposed=(Window, Worker), Pref="dom.webgpu.enabled"]
 interface GPUDeviceLostInfo {
     readonly attribute GPUDeviceLostReason reason;
     readonly attribute DOMString message;
 };
 
 partial interface GPUDevice {
-    [Throws]
     readonly attribute Promise<GPUDeviceLostInfo> lost;
 };
 

components/script/script_thread.rs

@@ -2424,12 +2424,14 @@ impl ScriptThread {
             WebGPUMsg::FreeTexture(id) => self.gpu_id_hub.lock().kill_texture_id(id),
             WebGPUMsg::FreeTextureView(id) => self.gpu_id_hub.lock().kill_texture_view_id(id),
             WebGPUMsg::Exit => *self.webgpu_port.borrow_mut() = None,
-            WebGPUMsg::CleanDevice {
+            WebGPUMsg::DeviceLost {
                 pipeline_id,
                 device,
+                reason,
+                msg,
             } => {
                 let global = self.documents.borrow().find_global(pipeline_id).unwrap();
-                global.remove_gpu_device(device);
+                global.gpu_device_lost(device, reason, msg);
             },
             WebGPUMsg::UncapturedError {
                 device,

components/webgpu/poll_thread.rs

@@ -7,7 +7,7 @@
 //! This is roughly based on <https://github.com/LucentFlux/wgpu-async/blob/1322c7e3fcdfc1865a472c7bbbf0e2e06dcf4da8/src/wgpu_future.rs>
 
 use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
-use std::sync::Arc;
+use std::sync::{Arc, Mutex, MutexGuard};
 use std::thread::JoinHandle;
 
 use log::warn;
@@ -40,14 +40,25 @@ pub(crate) struct Poller {
     is_done: Arc<AtomicBool>,
     /// Handle to the WGPU poller thread (to be used for unparking the thread)
     handle: Option<JoinHandle<()>>,
+    /// Lock for device maintain calls (in poll_all_devices and queue_submit)
+    ///
+    /// This is workaround for wgpu deadlocks: https://github.com/gfx-rs/wgpu/issues/5572
+    lock: Arc<Mutex<()>>,
 }
 
 #[inline]
-fn poll_all_devices(global: &Arc<Global>, more_work: &mut bool, force_wait: bool) {
+fn poll_all_devices(
+    global: &Arc<Global>,
+    more_work: &mut bool,
+    force_wait: bool,
+    lock: &Mutex<()>,
+) {
+    let _guard = lock.lock().unwrap();
     match global.poll_all_devices(force_wait) {
         Ok(all_queue_empty) => *more_work = !all_queue_empty,
         Err(e) => warn!("Poller thread got `{e}` on poll_all_devices."),
     }
+    // drop guard
 }
 
 impl Poller {
@@ -56,9 +67,11 @@ impl Poller {
         let is_done = Arc::new(AtomicBool::new(false));
         let work = work_count.clone();
         let done = is_done.clone();
+        let lock = Arc::new(Mutex::new(()));
         Self {
             work_count,
             is_done,
+            lock: Arc::clone(&lock),
             handle: Some(
                 std::thread::Builder::new()
                     .name("WGPU poller".into())
@@ -69,9 +82,9 @@ impl Poller {
                             // so every `ẁake` (even spurious) will do at least one poll.
                             // this is mostly useful for stuff that is deferred
                             // to maintain calls in wgpu (device resource destruction)
-                            poll_all_devices(&global, &mut more_work, false);
+                            poll_all_devices(&global, &mut more_work, false, &lock);
                             while more_work || work.load(Ordering::Acquire) != 0 {
-                                poll_all_devices(&global, &mut more_work, true);
+                                poll_all_devices(&global, &mut more_work, true, &lock);
                             }
                             std::thread::park(); //TODO: should we use timeout here
                         }
@@ -101,6 +114,11 @@ impl Poller {
             .thread()
             .unpark();
     }
+
+    /// Lock for device maintain calls (in poll_all_devices and queue_submit)
+    pub(crate) fn lock(&self) -> MutexGuard<()> {
+        self.lock.lock().unwrap()
+    }
 }
 
 impl Drop for Poller {

components/webgpu/script_messages.rs

@@ -15,6 +15,13 @@ use crate::wgc::id::{
     ShaderModuleId, StagingBufferId, SurfaceId, TextureId, TextureViewId,
 };
 
+/// <https://gpuweb.github.io/gpuweb/#enumdef-gpudevicelostreason>
+#[derive(Clone, Copy, Debug, Deserialize, Serialize)]
+pub enum DeviceLostReason {
+    Unknown,
+    Destroyed,
+}
+
 #[derive(Clone, Debug, Deserialize, Serialize)]
 pub enum WebGPUMsg {
     FreeAdapter(AdapterId),
@@ -34,14 +41,16 @@ pub enum WebGPUMsg {
     FreeRenderBundle(RenderBundleId),
     FreeStagingBuffer(StagingBufferId),
     FreeQuerySet(QuerySetId),
-    CleanDevice {
-        device: WebGPUDevice,
-        pipeline_id: PipelineId,
-    },
     UncapturedError {
         device: WebGPUDevice,
         pipeline_id: PipelineId,
         error: Error,
     },
+    DeviceLost {
+        device: WebGPUDevice,
+        pipeline_id: PipelineId,
+        reason: DeviceLostReason,
+        msg: String,
+    },
     Exit,
 }

components/webgpu/wgpu_thread.rs

@@ -20,7 +20,7 @@ use webrender_api::{DirtyRect, DocumentId};
 use webrender_traits::{WebrenderExternalImageRegistry, WebrenderImageHandlerType};
 use wgc::command::{ImageCopyBuffer, ImageCopyTexture};
 use wgc::device::queue::SubmittedWorkDoneClosure;
-use wgc::device::{DeviceDescriptor, HostMap, ImplicitPipelineIds};
+use wgc::device::{DeviceDescriptor, DeviceLostClosure, HostMap, ImplicitPipelineIds};
 use wgc::id::DeviceId;
 use wgc::instance::parse_backends_from_comma_list;
 use wgc::pipeline::ShaderModuleDescriptor;
@@ -68,7 +68,7 @@ pub(crate) struct WGPU {
     script_sender: IpcSender<WebGPUMsg>,
     global: Arc<wgc::global::Global>,
     adapters: Vec<WebGPUAdapter>,
-    devices: HashMap<DeviceId, DeviceScope>,
+    devices: Arc<Mutex<HashMap<DeviceId, DeviceScope>>>,
     // Track invalid adapters https://gpuweb.github.io/gpuweb/#invalid
     _invalid_adapters: Vec<WebGPUAdapter>,
     //TODO: Remove this (https://github.com/gfx-rs/wgpu/issues/867)
@@ -114,7 +114,7 @@ impl WGPU {
             script_sender,
             global,
             adapters: Vec::new(),
-            devices: HashMap::new(),
+            devices: Arc::new(Mutex::new(HashMap::new())),
             _invalid_adapters: Vec::new(),
             error_command_encoders: RefCell::new(HashMap::new()),
             webrender_api: Arc::new(Mutex::new(webrender_api_sender.create_api())),
@@ -525,6 +525,8 @@ impl WGPU {
                     WebGPURequest::DestroyDevice(device) => {
                         let global = &self.global;
                         gfx_select!(device => global.device_destroy(device));
+                        // Wake poller thread to trigger DeviceLostClosure
+                        self.poller.wake();
                     },
                     WebGPURequest::DestroySwapChain {
                         external_id,
@@ -580,15 +582,8 @@ impl WGPU {
                         };
                     },
                     WebGPURequest::DropDevice(device_id) => {
-                        let device = WebGPUDevice(device_id);
-                        let pipeline_id = self.devices.remove(&device_id).unwrap().pipeline_id;
-                        if let Err(e) = self.script_sender.send(WebGPUMsg::CleanDevice {
-                            device,
-                            pipeline_id,
-                        }) {
-                            warn!("Unable to send CleanDevice({:?}) ({:?})", device_id, e);
-                        }
                         let global = &self.global;
+                        // device_drop also calls device lost callback
                         gfx_select!(device_id => global.device_drop(device_id));
                         if let Err(e) = self.script_sender.send(WebGPUMsg::FreeDevice(device_id)) {
                             warn!("Unable to send FreeDevice({:?}) ({:?})", device_id, e);
@@ -686,8 +681,44 @@ impl WGPU {
                         };
                         let device = WebGPUDevice(device_id);
                         let queue = WebGPUQueue(queue_id);
-                        self.devices
+                        {
-                            .insert(device_id, DeviceScope::new(device_id, pipeline_id));
+                            self.devices
+                                .lock()
+                                .unwrap()
+                                .insert(device_id, DeviceScope::new(device_id, pipeline_id));
+                        }
+                        let script_sender = self.script_sender.clone();
+                        let devices = Arc::clone(&self.devices);
+                        let callback =
+                            DeviceLostClosure::from_rust(Box::from(move |reason, msg| {
+                                let _ = devices.lock().unwrap().remove(&device_id);
+                                let reason = match reason {
+                                    wgt::DeviceLostReason::Unknown => {
+                                        Some(crate::DeviceLostReason::Unknown)
+                                    },
+                                    wgt::DeviceLostReason::Destroyed => {
+                                        Some(crate::DeviceLostReason::Destroyed)
+                                    },
+                                    wgt::DeviceLostReason::Dropped => None,
+                                    wgt::DeviceLostReason::ReplacedCallback => {
+                                        panic!("DeviceLost callback should only be set once")
+                                    },
+                                    wgt::DeviceLostReason::DeviceInvalid => {
+                                        Some(crate::DeviceLostReason::Unknown)
+                                    },
+                                };
+                                if let Some(reason) = reason {
+                                    if let Err(e) = script_sender.send(WebGPUMsg::DeviceLost {
+                                        device: WebGPUDevice(device_id),
+                                        pipeline_id,
+                                        reason,
+                                        msg,
+                                    }) {
+                                        warn!("Failed to send WebGPUMsg::DeviceLost: {e}");
+                                    }
+                                }
+                            }));
+                        gfx_select!(device_id => global.device_set_device_lost_closure(device_id, callback));
                         if let Err(e) = sender.send(Some(Ok(WebGPUResponse::RequestDevice {
                             device_id: device,
                             queue_id: queue,
@@ -744,6 +775,7 @@ impl WGPU {
                                 "Invalid command buffer submitted",
                             )))
                         } else {
+                            let _guard = self.poller.lock();
                             gfx_select!(queue_id => global.queue_submit(queue_id, &command_buffers))
                                 .map_err(Error::from_error)
                         };
@@ -951,6 +983,7 @@ impl WGPU {
                         data,
                     } => {
                         let global = &self.global;
+                        let _guard = self.poller.lock();
                         //TODO: Report result to content process
                         let result = gfx_select!(queue_id => global.queue_write_texture(
                             queue_id,
@@ -959,6 +992,7 @@ impl WGPU {
                             &data_layout,
                             &size
                         ));
+                        drop(_guard);
                         self.maybe_dispatch_wgpu_error(queue_id.transmute(), result.err());
                     },
                     WebGPURequest::QueueOnSubmittedWorkDone { sender, queue_id } => {
@@ -1070,18 +1104,18 @@ impl WGPU {
                     },
                     WebGPURequest::PushErrorScope { device_id, filter } => {
                         // <https://www.w3.org/TR/webgpu/#dom-gpudevice-pusherrorscope>
-                        let device_scope = self
+                        let mut devices = self.devices.lock().unwrap();
-                            .devices
+                        if let Some(device_scope) = devices.get_mut(&device_id) {
-                            .get_mut(&device_id)
+                            device_scope.error_scope_stack.push(ErrorScope::new(filter));
-                            .expect("Using invalid device");
+                        } // else device is lost
-                        device_scope.error_scope_stack.push(ErrorScope::new(filter));
                     },
                     WebGPURequest::DispatchError { device_id, error } => {
                         self.dispatch_error(device_id, error);
                     },
                     WebGPURequest::PopErrorScope { device_id, sender } => {
                         // <https://www.w3.org/TR/webgpu/#dom-gpudevice-poperrorscope>
-                        if let Some(device_scope) = self.devices.get_mut(&device_id) {
+                        if let Some(device_scope) = self.devices.lock().unwrap().get_mut(&device_id)
+                        {
                             if let Some(error_scope) = device_scope.error_scope_stack.pop() {
                                 if let Err(e) =
                                     sender.send(Some(Ok(WebGPUResponse::PoppedErrorScope(Ok(
@@ -1133,7 +1167,7 @@ impl WGPU {
 
     /// <https://www.w3.org/TR/webgpu/#abstract-opdef-dispatch-error>
     fn dispatch_error(&mut self, device_id: id::DeviceId, error: Error) {
-        if let Some(device_scope) = self.devices.get_mut(&device_id) {
+        if let Some(device_scope) = self.devices.lock().unwrap().get_mut(&device_id) {
             if let Some(error_scope) = device_scope
                 .error_scope_stack
                 .iter_mut()