Changeset 163428 in webkit for trunk/Source/JavaScriptCore/llint/LowLevelInterpreter.cpp

Timestamp:

Feb 4, 2014, 9:32:21 PM (11 years ago)

Author:

[email protected]

Message:

The stack limit computation does not work for Windows.
<​https://webkit.org/b/128226>

Reviewed by Geoffrey Garen.

Source/JavaScriptCore:

• llint/LowLevelInterpreter.cpp:

(JSC::CLoopRegister::CLoopRegister):
(JSC::CLoop::execute):

• Suppressed some compiler warnings for the C loop build.
• runtime/VM.cpp:

(JSC::VM::updateStackLimitWithReservedZoneSize):

• Use the new StackBounds::recursionLimit() to compute the stack limit the right way.

Source/WTF:

The current stack limit computation assumes that the stack is allocated
in high memory. On Windows, the stack seems to be allocated very near
address 0, and is smaller in size than our Options::maxPerThreadStackUsage().
This combination triggers a bug where StackBounds::recursionLimit()
underflows and computes a limit address that is ridiculously high in memory.
The net effect is that the Windows port thinks its out of stack space all
the time.

We now check for potential underflows in StackBounds and return an
appropriate limit address.

• wtf/StackBounds.h:

(WTF::StackBounds::recursionLimit):

File:

• trunk/Source/JavaScriptCore/llint/LowLevelInterpreter.cpp (modified) (2 diffs)

trunk/Source/JavaScriptCore/llint/LowLevelInterpreter.cpp

@@ -154 +154 @@
 
 struct CLoopRegister {
+    CLoopRegister() { i = static_cast<intptr_t>(0xbadbeef0baddbeef); }
     union {
         intptr_t i;
@@ -480 +481 @@
     #undef SIGN_BIT32
 
+    return JSValue(); // to suppress a compiler warning.
 } // Interpreter::llintCLoopExecute()