Branch: refs/heads/main
Home: https://github.com/WebKit/WebKit
Commit: 1e4577ba8de434bc8942b01baf6b1d65ba967ab1
https://github.com/WebKit/WebKit/commit/1e4577ba8de434bc8942b01baf6b1d65ba967ab1
Author: Yijia Huang <[email protected]>
Date: 2024-06-08 (Sat, 08 Jun 2024)
Changed paths:
M Source/JavaScriptCore/heap/WeakInlines.h
M Source/JavaScriptCore/runtime/ArrayBuffer.cpp
M Source/JavaScriptCore/runtime/DeferredWorkTimer.cpp
M Source/JavaScriptCore/runtime/DeferredWorkTimer.h
M Source/JavaScriptCore/runtime/JSGlobalObject.cpp
M Source/JavaScriptCore/runtime/JSGlobalObject.h
M Source/JavaScriptCore/runtime/VM.cpp
M Source/JavaScriptCore/runtime/WaiterListManager.cpp
M Source/JavaScriptCore/runtime/WaiterListManager.h
M Source/JavaScriptCore/wasm/WasmStreamingCompiler.cpp
M Source/JavaScriptCore/wasm/js/JSWebAssembly.cpp
Log Message:
-----------
[JSC] Fix Atomics.waitAsync by using weak references in JSC::DeferredWorkTimer
https://bugs.webkit.org/show_bug.cgi?id=272896
rdar://126690921
Reviewed by Yusuke Suzuki.
Previously, we introduced JSC::DeferredWorkTimer by keeping the task related
dependencies and environment alive with strong references. This works fine
until Atomics.waitAsync is introduced. Consider a scenario that an
Atomics.waitAsync
registered in the main thread with infinity timeout is never be resolved in the
runtime. At some point, the corresponding JSGlobalObject is no longer needed.
But the JSGlobalObject cannot be collected due to the previous registered
Atomics.waitAsync's JSPromise is strongly referenced and queued in the
DeferredWorkTimer.
To fix this issue, we can transfer the liveness responsibility from the
DeferredWorkTimer::TicketData to the target's JSGlobalObject. And this is
how it works. Given the graph as shown below:
VM <----------------------------
^ |
| |
JSGlobalObject_0 ----> JSGlobalObject_n -----
^ | ^ |
| | | |
SharedArrayBufferContents ----- | S | S
^ | |
| W | |
Waiter ----> TicketData -----> Target <------|
| |
| W |
------> Dependencies <--------
1. The JSGlobalObject_0 and JSGlobalObject_n may or may not be the same object,
which is dependent on the VM.
2. The SharedArrayBufferContents is dependent on the multiple JSGlobalObjects.
3. The Waiter is dependent on the SharedArrayBufferContents.
4. The Waiter has a TicketData used to schedule a task to DeferredWorkTimer.
5. The TicketData replies on a Target and its corresponding Dependencies to
complete the task.
6. The TicketData has weak references to the Target and Dependencies.
7. The Target's JSGlobalObject_n holds the strong references to the Target and
Dependencies.
Then, we should do:
A. If JSGlobalObject_n is alive, then it should clear those strong references
once the
Waiter done the task.
B. If JSGlobalObject_n is dead, then it should clear the corresponding waiters
first and
then clear the other TicketDatas associating to it.
If no others are referencing JSGlobalObject_n and Target, GC would consider the
strong reference
cycle in Target -> JSGlobalObject_n -> Target as dead status. Then, they are
available
for recycling.
* Source/JavaScriptCore/heap/WeakInlines.h:
(JSC::operator==):
* Source/JavaScriptCore/runtime/DeferredWorkTimer.cpp:
(JSC::DeferredWorkTimer::TicketData::TicketData):
(JSC::DeferredWorkTimer::addPendingWork):
* Source/JavaScriptCore/runtime/DeferredWorkTimer.h:
* Source/JavaScriptCore/runtime/JSGlobalObject.cpp:
(JSC::JSGlobalObject::visitChildrenImpl):
* Source/JavaScriptCore/runtime/JSGlobalObject.h:
(JSC::JSGlobalObject::addObjectsForDeferredTaskTimer):
* Source/JavaScriptCore/wasm/WasmStreamingCompiler.cpp:
(JSC::Wasm::StreamingCompiler::StreamingCompiler):
* Source/JavaScriptCore/wasm/js/JSWebAssembly.cpp:
(JSC::JSWebAssembly::webAssemblyModuleValidateAsync):
(JSC::instantiate):
(JSC::compileAndInstantiate):
Canonical link: https://commits.webkit.org/279852@main
To unsubscribe from these emails, change your notification settings at
https://github.com/WebKit/WebKit/settings/notifications
_______________________________________________
webkit-changes mailing list
[email protected]
https://lists.webkit.org/mailman/listinfo/webkit-changes
