Changeset 213238 in webkit for trunk/Source/JavaScriptCore/runtime/JSLock.h

Timestamp:

Mar 1, 2017, 12:15:08 PM (8 years ago)

Author:

[email protected]

Message:

[Re-landing] Change JSLock to stash PlatformThread instead of std::thread::id.
​https://bugs.webkit.org/show_bug.cgi?id=168996

Reviewed by Filip Pizlo and Saam Barati.

PlatformThread is more useful because it allows us to:

1. find the MachineThreads::Thread which is associated with it.
2. suspend / resume threads.
3. send a signal to a thread.

We can't do those with std::thread::id. We will need one or more of these
capabilities to implement non-polling VM traps later.

Update: Since we don't have a canonical "uninitialized" value for PlatformThread,
we now have a JSLock::m_hasOwnerThread flag that is set to true if and only the
m_ownerThread value is valid. JSLock::currentThreadIsHoldingLock() now checks
JSLock::m_hasOwnerThread before doing the thread identity comparison.

• JavaScriptCore.xcodeproj/project.pbxproj:
• heap/MachineStackMarker.cpp:

(JSC::MachineThreads::Thread::createForCurrentThread):
(JSC::MachineThreads::machineThreadForCurrentThread):
(JSC::MachineThreads::removeThread):
(JSC::MachineThreads::Thread::suspend):
(JSC::MachineThreads::tryCopyOtherThreadStacks):
(JSC::getCurrentPlatformThread): Deleted.

• heap/MachineStackMarker.h:
• runtime/JSCellInlines.h:

(JSC::JSCell::classInfo):

• runtime/JSLock.cpp:

(JSC::JSLock::JSLock):
(JSC::JSLock::lock):
(JSC::JSLock::unlock):
(JSC::JSLock::currentThreadIsHoldingLock): Deleted.

• runtime/JSLock.h:

(JSC::JSLock::ownerThread):
(JSC::JSLock::currentThreadIsHoldingLock):

• runtime/PlatformThread.h: Added.

(JSC::currentPlatformThread):

• runtime/VM.cpp:

(JSC::VM::~VM):

• runtime/VM.h:

(JSC::VM::ownerThread):

• runtime/Watchdog.cpp:

(JSC::Watchdog::setTimeLimit):
(JSC::Watchdog::shouldTerminate):
(JSC::Watchdog::startTimer):
(JSC::Watchdog::stopTimer):

• tools/JSDollarVMPrototype.cpp:

(JSC::JSDollarVMPrototype::currentThreadOwnsJSLock):

• tools/VMInspector.cpp:

File:

• trunk/Source/JavaScriptCore/runtime/JSLock.h (modified) (3 diffs)

trunk/Source/JavaScriptCore/runtime/JSLock.h

@@ -21 +21 @@
 #pragma once
 
+#include "PlatformThread.h"
 #include <mutex>
-#include <thread>
 #include <wtf/Assertions.h>
 #include <wtf/Lock.h>
@@ -94 +94 @@
     VM* vm() { return m_vm; }
 
-    std::thread::id ownerThread() const { return m_ownerThreadID; }
-    JS_EXPORT_PRIVATE bool currentThreadIsHoldingLock();
+    std::optional<PlatformThread> ownerThread() const
+    {
+        if (m_hasOwnerThread)
+            return m_ownerThread;
+        return { };
+    }
+    bool currentThreadIsHoldingLock() { return m_hasOwnerThread && m_ownerThread == currentPlatformThread(); }
 
     void willDestroyVM(VM*);
@@ -127 +132 @@
 
     Lock m_lock;
-    std::thread::id m_ownerThreadID;
+    // We cannot make m_ownerThread an optional (instead of pairing it with an explicit
+    // m_hasOwnerThread) because currentThreadIsHoldingLock() may be called from a
+    // different thread, and an optional is vulnerable to races.
+    // See https://bugs.webkit.org/show_bug.cgi?id=169042#c6
+    bool m_hasOwnerThread { false };
+    PlatformThread m_ownerThread;
     intptr_t m_lockCount;
     unsigned m_lockDropDepth;