Add WorkerClient for WebCore to access WebKit from worker threads, and use it to create a RemoteRenderingBackendProxy per-thread

[mattwoodrow]

280e6cd

Add WorkerClient for WebCore to access WebKit from worker threads, and use it to create a RemoteRenderingBackendProxy per-thread
https://bugs.webkit.org/show_bug.cgi?id=244828

Reviewed by Kimmo Kinnunen.

Creates a new interface class WorkerClient (which implements GraphicsClient), for accessing WebKit graphics APIs from a worker thread. Adds an implementation
of this in WebWorkerClient.

Allocates an instance of (Web)WorkerClient for every dedicated and shared worker, and makes it available via WorkerGlobalScope.

WebWorkerClient creates a dedicated GPUProcessConnection and RemoteRenderingBackendProx for each worker thread/instance, and uses this to create
RemoteImageBufferProxys that are safe to use on that worker thread.

* Source/WebCore/Headers.cmake:
* Source/WebCore/page/Chrome.cpp:
(WebCore::Chrome::createWorkerClient):
* Source/WebCore/page/Chrome.h:
* Source/WebCore/page/ChromeClient.h:
(WebCore::ChromeClient::createWorkerClient):
* Source/WebCore/page/WorkerClient.h: Added.
* Source/WebCore/workers/DedicatedWorkerGlobalScope.cpp:
(WebCore::DedicatedWorkerGlobalScope::create):
(WebCore::DedicatedWorkerGlobalScope::DedicatedWorkerGlobalScope):
* Source/WebCore/workers/DedicatedWorkerGlobalScope.h:
* Source/WebCore/workers/DedicatedWorkerThread.cpp:
(WebCore::DedicatedWorkerThread::createWorkerGlobalScope):
* Source/WebCore/workers/WorkerGlobalScope.cpp:
(WebCore::WorkerGlobalScope::WorkerGlobalScope):
* Source/WebCore/workers/WorkerGlobalScope.h:
(WebCore::WorkerGlobalScope::workerClient):
* Source/WebCore/workers/WorkerMessagingProxy.cpp:
(WebCore::WorkerMessagingProxy::startWorkerGlobalScope):
* Source/WebCore/workers/WorkerThread.h:
(WebCore::WorkerThread::setWorkerClient):
(WebCore::WorkerThread::workerClient):
* Source/WebCore/workers/service/ServiceWorkerGlobalScope.cpp:
(WebCore::ServiceWorkerGlobalScope::ServiceWorkerGlobalScope):
* Source/WebCore/workers/shared/SharedWorkerGlobalScope.cpp:
(WebCore::SharedWorkerGlobalScope::SharedWorkerGlobalScope):
* Source/WebCore/workers/shared/SharedWorkerGlobalScope.h:
* Source/WebCore/workers/shared/context/SharedWorkerThread.cpp:
(WebCore::SharedWorkerThread::createWorkerGlobalScope):
* Source/WebCore/workers/shared/context/SharedWorkerThreadProxy.cpp:
(WebCore::SharedWorkerThreadProxy::SharedWorkerThreadProxy):
* Source/WebKit/Sources.txt:
* Source/WebKit/WebKit.xcodeproj/project.pbxproj:
* Source/WebKit/WebProcess/WebCoreSupport/WebChromeClient.cpp:
(WebKit::WebChromeClient::createWorkerClient):
* Source/WebKit/WebProcess/WebCoreSupport/WebChromeClient.h:
* Source/WebKit/WebProcess/WebCoreSupport/WebWorkerClient.cpp: Added.
(WebKit::WebWorkerClient::WebWorkerClient):
(WebKit::WebWorkerClient::ensureRenderingBackend const):
(WebKit::WebWorkerClient::clone):
(WebKit::WebWorkerClient::displayID const):
(WebKit::WebWorkerClient::createImageBuffer const):
* Source/WebKit/WebProcess/WebCoreSupport/WebWorkerClient.h: Added.

Canonical link: https://commits.webkit.org/257507@main

7088983

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 🧪 win
✅ 🧪 bindings	✅ 🛠 ios-sim	✅ 🛠 mac-AS-debug	✅ 🛠 gtk	✅ 🛠 wincairo
✅ 🧪 webkitperl	❌ 🧪 ios-wk2	✅ 🧪 api-mac	✅ 🧪 gtk-wk2
	✅ 🧪 api-ios	❌ 🧪 mac-wk1	✅ 🧪 api-gtk
	✅ 🛠 tv	❌ 🧪 mac-wk2
	✅ 🛠 tv-sim	❌ 🧪 mac-AS-debug-wk2
✅ 🛠 🧪 merge	✅ 🛠 watch	✅ 🧪 mac-wk2-stress
	✅ 🛠 watch-sim

[mattwoodrow]

Full branch for extra context: https://github.com/mattwoodrow/WebKit/commits/eng/remote-rendering-workers

[webkit-early-warning-system]

EWS run on previous version of this PR (hash f71351e)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	❌ 🛠 ios	❌ 🛠 mac	❌ 🛠 wpe	❌ 🛠 🧪 win
✅ 🧪 bindings	❌ 🛠 ios-sim	❌ 🛠 mac-debug	❌ 🛠 gtk	❌ 🛠 wincairo
✅ 🧪 webkitperl	❌ 🧪 ios-wk2	❌ 🛠 mac-AS-debug	❌ 🧪 gtk-wk2
	❌ 🧪 api-ios	❌ 🧪 api-mac	❌ 🧪 api-gtk
	❌ 🛠 tv	❌ 🧪 mac-wk1
	❌ 🛠 tv-sim	❌ 🧪 mac-wk2
	❌ 🛠 watch	❌ 🧪 mac-AS-debug-wk2
	❌ 🛠 watch-sim	❌ 🧪 mac-wk2-stress

[kkinnunen-apple]

I think this work could be simplified to first pursue the changes needed with GraphicsClient.

It seems that we are lacking a per-worker factory for polymorphic object instantiations.
We have per-page factory ChromeClient that creates objects that depend on global state.
We need per-worker factory of similar sort, WorkerClient.

I would start with:

1. Extract what ImageBuffer/WebGL needs from ChromeClient to GraphicsClient, make it work with current code.

2a. Investigate what is the object that workers use to configure the polymorphic objects via the global state. If there is none, add this interface

alternatively

2b. Add logic for ChromeClient to hold per-thread GraphicsClients

Small detail:
I suppose the ScriptExecutionContext would be the place to add the initial polymorphic query to get the GraphicsClient?

E.g. there's ScriptExecutionContext::notificationClient(), and similar to this, there could be one for graphics?

Things that initially do not make sense to me:

GPUProcess updated to handle multiple instances of GPUConnectionToWebProcess per web process

This does not seem to be needed. To me, it appears as if we just need a RemoteRenderingBackendProxy per worker. Then we just need to make sure that the closure/crash support works if it doesn't work with the main connection.

SerialDispatchQueue -- We already have WorkQueue, which is a serial dispatch queue. So the code about that doesn't make sense initially. Note: the implementation seems to be missing from this patch.

[mattwoodrow]

Things that initially do not make sense to me:

GPUProcess updated to handle multiple instances of GPUConnectionToWebProcess per web process

This does not seem to be needed. To me, it appears as if we just need a RemoteRenderingBackendProxy per worker. Then we just need to make sure that the closure/crash support works if it doesn't work with the main connection.

We'd also need to add support to IPC::Connection for sending sync messages from the worker thread, and make sure responses are handled without touching/blocking the main thread. That's pretty complicated, with all the complexity we already have for handling unrelated messages while blocking for a sync message.

Instead, I added support for creating new IPC::Connections that are bound to the worker thread (and can send sync messages from that thread), but don't support the re-entrant message handling (for simplicity, and we don't need that functionality for the current use case).

This does require creating a new GPUConnectionToWebProcess though, bound to the IPC::Connection for the worker.

This is all in #4043 (which is a dependency here, though GitHub doesn't let me express that in a useful way).

SerialDispatchQueue -- We already have WorkQueue, which is a serial dispatch queue. So the code about that doesn't make sense initially. Note: the implementation seems to be missing from this patch.

The implementation is in #4043.

The functionality we're trying to represent here is 'the main thread, or a worker', that we can dispatch tasks to, assert that we're currently running in the expected place, and hold on to a reference to (could be weak, or strong).

WorkQueue is a specific implementation, that neither the main thread or workers are. The closest existing thing that I can find is FunctionDispatcher (which isn't implemented by WorkerOrWorkletThread, but can be), but that doesn't offer refcounting, or assertions.

[mattwoodrow]

SerialDispatchQueue -- We already have WorkQueue, which is a serial dispatch queue. So the code about that doesn't make sense initially. Note: the implementation seems to be missing from this patch.

Note that WorkQueue could implement the SerialDispatchQueue interface, but ConcurrentWorkQueue can't really, so it's a bit tricky.

[webkit-early-warning-system]

EWS run on previous version of this PR (hash 73a4c71)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	❌ 🛠 wpe	❌ 🛠 🧪 win
✅ 🧪 bindings	✅ 🛠 ios-sim	✅ 🛠 mac-debug	❌ 🛠 gtk	❌ 🛠 wincairo
✅ 🧪 webkitperl	❌ 🧪 ios-wk2	✅ 🛠 mac-AS-debug	❌ 🧪 gtk-wk2
	❌ 🧪 api-ios	🧪 api-mac	❌ 🧪 api-gtk
✅ 🛠 🧪 jsc	✅ 🛠 tv	✅ 🧪 mac-wk1	✅ 🛠 jsc-armv7
	✅ 🛠 tv-sim	❌ 🧪 mac-wk2	✅ 🧪 jsc-armv7-tests
	✅ 🛠 watch	❌ 🧪 mac-AS-debug-wk2	✅ 🛠 jsc-mips
	✅ 🛠 watch-sim	✅ 🧪 mac-wk2-stress	✅ 🧪 jsc-mips-tests

[mattwoodrow]

Please note, only the top two commits are truly part of this PR. The others are just there because GitHub (and EWS) don't support dependent branches.

[webkit-early-warning-system]

EWS run on previous version of this PR (hash 03e85a2)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
❌ 🧪 style	❌ 🛠 ios	❌ 🛠 mac	❌ 🛠 wpe	❌ 🛠 🧪 win
❌ 🧪 bindings	❌ 🛠 ios-sim	❌ 🛠 mac-debug	❌ 🛠 gtk	❌ 🛠 wincairo
❌ 🧪 webkitperl	❌ 🧪 ios-wk2	❌ 🛠 mac-AS-debug	❌ 🧪 gtk-wk2
	❌ 🧪 api-ios	❌ 🧪 api-mac	❌ 🧪 api-gtk
❌ 🛠 🧪 jsc	❌ 🛠 tv	❌ 🧪 mac-wk1	❌ 🛠 jsc-armv7
	❌ 🛠 tv-sim	❌ 🧪 mac-wk2	❌ 🧪 jsc-armv7-tests
	❌ 🛠 watch	❌ 🧪 mac-AS-debug-wk2	❌ 🛠 jsc-mips
	❌ 🛠 watch-sim	❌ 🧪 mac-wk2-stress	❌ 🧪 jsc-mips-tests

[webkit-early-warning-system]

EWS run on previous version of this PR (hash 2519ee1)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
❌ 🧪 style	❌ 🛠 ios	❌ 🛠 mac	❌ 🛠 wpe	❌ 🛠 🧪 win
❌ 🧪 bindings	❌ 🛠 ios-sim	⏳ 🛠 mac-debug	❌ 🛠 gtk	⏳ 🛠 wincairo
❌ 🧪 webkitperl	❌ 🧪 ios-wk2	❌ 🛠 mac-AS-debug	❌ 🧪 gtk-wk2
	❌ 🧪 api-ios	❌ 🧪 api-mac	❌ 🧪 api-gtk
⏳ 🛠 🧪 jsc	❌ 🛠 tv	❌ 🧪 mac-wk1	❌ 🛠 jsc-armv7
	❌ 🛠 tv-sim	❌ 🧪 mac-wk2	❌ 🧪 jsc-armv7-tests
	❌ 🛠 watch	❌ 🧪 mac-AS-debug-wk2	❌ 🛠 jsc-mips
	❌ 🛠 watch-sim	❌ 🧪 mac-wk2-stress	❌ 🧪 jsc-mips-tests

[webkit-early-warning-system]

EWS run on previous version of this PR (hash cb76e02)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	❌ 🛠 ios	❌ 🛠 mac	❌ 🛠 wpe	❌ 🛠 🧪 win
✅ 🧪 bindings	❌ 🛠 ios-sim	⏳ 🛠 mac-debug	❌ 🛠 gtk	⏳ 🛠 wincairo
✅ 🧪 webkitperl	❌ 🧪 ios-wk2	❌ 🛠 mac-AS-debug	❌ 🧪 gtk-wk2
	❌ 🧪 api-ios	❌ 🧪 api-mac	❌ 🧪 api-gtk
⏳ 🛠 🧪 jsc	❌ 🛠 tv	❌ 🧪 mac-wk1	✅ 🛠 jsc-armv7
	❌ 🛠 tv-sim	❌ 🧪 mac-wk2	✅ 🧪 jsc-armv7-tests
	❌ 🛠 watch	❌ 🧪 mac-AS-debug-wk2	✅ 🛠 jsc-mips
	❌ 🛠 watch-sim	❌ 🧪 mac-wk2-stress	✅ 🧪 jsc-mips-tests

[webkit-early-warning-system]

EWS run on previous version of this PR (hash bd97b32)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	❌ 🛠 wpe	❌ 🛠 🧪 win
✅ 🧪 bindings	✅ 🛠 ios-sim	✅ 🛠 mac-debug	❌ 🛠 gtk	❌ 🛠 wincairo
✅ 🧪 webkitperl	❌ 🧪 ios-wk2	✅ 🛠 mac-AS-debug	❌ 🧪 gtk-wk2
	✅ 🧪 api-ios	✅ 🧪 api-mac	❌ 🧪 api-gtk
✅ 🛠 🧪 jsc	✅ 🛠 tv	✅ 🧪 mac-wk1	✅ 🛠 jsc-armv7
	✅ 🛠 tv-sim	✅ 🧪 mac-wk2	✅ 🧪 jsc-armv7-tests
	✅ 🛠 watch	❌ 🧪 mac-AS-debug-wk2	✅ 🛠 jsc-mips
	✅ 🛠 watch-sim	✅ 🧪 mac-wk2-stress	❌ 🧪 jsc-mips-tests

[webkit-early-warning-system]

EWS run on previous version of this PR (hash 7844c4a)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 🧪 win
✅ 🧪 bindings	✅ 🛠 ios-sim	✅ 🛠 mac-debug	✅ 🛠 gtk	✅ 🛠 wincairo
✅ 🧪 webkitperl	✅ 🧪 ios-wk2	✅ 🛠 mac-AS-debug	✅ 🧪 gtk-wk2
	✅ 🧪 api-ios	🧪 api-mac	✅ 🧪 api-gtk
	✅ 🛠 tv	✅ 🧪 mac-wk1
	✅ 🛠 tv-sim	✅ 🧪 mac-wk2
	✅ 🛠 watch	❌ 🧪 mac-AS-debug-wk2
	✅ 🛠 watch-sim	✅ 🧪 mac-wk2-stress

[webkit-early-warning-system]

EWS run on previous version of this PR (hash 4904912)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 🧪 win
✅ 🧪 bindings	✅ 🛠 ios-sim	✅ 🛠 mac-debug	✅ 🛠 gtk	⏳ 🛠 wincairo
✅ 🧪 webkitperl	🧪 ios-wk2	✅ 🛠 mac-AS-debug	✅ 🧪 gtk-wk2
	✅ 🧪 api-ios	🧪 api-mac	🧪 api-gtk
	✅ 🛠 tv	🧪 mac-wk1
	✅ 🛠 tv-sim	✅ 🧪 mac-wk2
	✅ 🛠 watch	🧪 mac-AS-debug-wk2
	✅ 🛠 watch-sim	✅ 🧪 mac-wk2-stress

[shallawa]

There are many #if ENABLE(GPU_PROCESS) in this header file and in the source file. Should not the GPUP case be moved to a new class named RemoteWebWorkerClient which inherits from WebWorkerClient ? WebChromeClient::createWorkerClient() will make the decision from the beginning so we do not have to deal with two different code paths in the same file.

[shallawa]

Why is this always true? And what is the point in asserting it hasOneRef() and then create a new RefPtr of the ImageBuffer which will increment the ref count?

[mattwoodrow]

As above, the naming might not be great here.

It asserts that there's only one ref, because it's a private virtual and the only caller is the static which also asserts that it owns the only ref.

The default behaviour is that if the static owns the only ref to the ImageBuffer, then we can allow it to be passed to a new thread as-is since there won't be any other cross-thread references.

[webkit-early-warning-system]

EWS run on previous version of this PR (hash 9d2f63b)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 🧪 win
✅ 🧪 bindings	✅ 🛠 ios-sim	✅ 🛠 mac-debug	✅ 🛠 gtk	✅ 🛠 wincairo
✅ 🧪 webkitperl	✅ 🧪 ios-wk2	✅ 🛠 mac-AS-debug	✅ 🧪 gtk-wk2
	✅ 🧪 api-ios	✅ 🧪 api-mac	✅ 🧪 api-gtk
	✅ 🛠 tv	✅ 🧪 mac-wk1
	✅ 🛠 tv-sim	✅ 🧪 mac-wk2
	✅ 🛠 watch	✅ 🧪 mac-AS-debug-wk2
	✅ 🛠 watch-sim	✅ 🧪 mac-wk2-stress

[webkit-early-warning-system]

EWS run on previous version of this PR (hash 3aba2e8)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 🧪 win
✅ 🧪 bindings	✅ 🛠 ios-sim	✅ 🛠 mac-debug	✅ 🛠 gtk	✅ 🛠 wincairo
✅ 🧪 webkitperl	✅ 🧪 ios-wk2	✅ 🛠 mac-AS-debug	✅ 🧪 gtk-wk2
	✅ 🧪 api-ios	✅ 🧪 api-mac	✅ 🧪 api-gtk
	✅ 🛠 tv	✅ 🧪 mac-wk1
	✅ 🛠 tv-sim	✅ 🧪 mac-wk2
	❌ 🛠 watch	✅ 🧪 mac-AS-debug-wk2
	✅ 🛠 watch-sim	✅ 🧪 mac-wk2-stress

[webkit-early-warning-system]

EWS run on previous version of this PR (hash 5d21d44)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	❌ 🛠 🧪 win
✅ 🧪 bindings	✅ 🛠 ios-sim	✅ 🛠 mac-debug	❌ 🛠 gtk	❌ 🛠 wincairo
✅ 🧪 webkitperl	✅ 🧪 ios-wk2	✅ 🛠 mac-AS-debug	❌ 🧪 gtk-wk2
	✅ 🧪 api-ios	✅ 🧪 api-mac	✅ 🧪 api-gtk
	✅ 🛠 tv	✅ 🧪 mac-wk1
	❌ 🛠 tv-sim	✅ 🧪 mac-wk2
	✅ 🛠 watch	✅ 🧪 mac-AS-debug-wk2
	✅ 🛠 watch-sim	✅ 🧪 mac-wk2-stress

[webkit-early-warning-system]

EWS run on current version of this PR (hash 7088983)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 🧪 win
✅ 🧪 bindings	✅ 🛠 ios-sim	✅ 🛠 mac-AS-debug	✅ 🛠 gtk	✅ 🛠 wincairo
✅ 🧪 webkitperl	❌ 🧪 ios-wk2	✅ 🧪 api-mac	✅ 🧪 gtk-wk2
	✅ 🧪 api-ios	❌ 🧪 mac-wk1	✅ 🧪 api-gtk
	✅ 🛠 tv	❌ 🧪 mac-wk2
	✅ 🛠 tv-sim	❌ 🧪 mac-AS-debug-wk2
✅ 🛠 🧪 merge	✅ 🛠 watch	✅ 🧪 mac-wk2-stress
	✅ 🛠 watch-sim

[kkinnunen-apple]

To solve the most of the ifdefing problems, you can obtain the data needed by RemoteRenderingBackendCreationParameters expanded to the constructor args.

[webkit-commit-queue]

Committed 257507@main (280e6cd): https://commits.webkit.org/257507@main

Reviewed commits have been landed. Closing PR #4085 and removing active labels.