Merge r180649 - ASan does not like JSC::MachineThreads::tryCopyOtherT…

…hreadStack.

<https://webkit.org/b/141672>

Reviewed by Alexey Proskuryakov.

ASan does not like the fact that we memcpy the stack for GC scans.  So,
we're working around this by using our own memcpy (asanUnsafeMemcpy)
implementation that we can tell ASan to ignore.

Source/JavaScriptCore:

* heap/MachineStackMarker.cpp:
(JSC::asanUnsafeMemcpy):

Tools:

Also removed the previous added directive to ignore *tryCopyOtherThreadStack*
which isn't effective for working around this issue.

* asan/webkit-asan-ignore.txt:

Source/JavaScriptCore/ChangeLog

@@ -1,3 +1,17 @@
+2015-02-25  Mark Lam  <mark.lam@apple.com>
+
+        ASan does not like JSC::MachineThreads::tryCopyOtherThreadStack.
+        <https://webkit.org/b/141672>
+
+        Reviewed by Alexey Proskuryakov.
+
+        ASan does not like the fact that we memcpy the stack for GC scans.  So,
+        we're working around this by using our own memcpy (asanUnsafeMemcpy)
+        implementation that we can tell ASan to ignore.
+
+        * heap/MachineStackMarker.cpp:
+        (JSC::asanUnsafeMemcpy):
+
 2015-02-15  Sam Weinig  <sam@webkit.org>
 
         Add experimental <attachment> element support

Source/JavaScriptCore/heap/MachineStackMarker.cpp

@@ -443,6 +443,26 @@ static std::pair<void*, size_t> otherThreadStack(void* stackBase, const Platform
     return std::make_pair(begin, static_cast<char*>(end) - static_cast<char*>(begin));
 }
 
+#if ASAN_ENABLED
+void asanUnsafeMemcpy(void* dst, const void* src, size_t);
+void asanUnsafeMemcpy(void* dst, const void* src, size_t size)
+{
+    size_t dstAsSize = reinterpret_cast<size_t>(dst);
+    size_t srcAsSize = reinterpret_cast<size_t>(src);
+    RELEASE_ASSERT(dstAsSize == WTF::roundUpToMultipleOf<sizeof(intptr_t)>(dstAsSize));
+    RELEASE_ASSERT(srcAsSize == WTF::roundUpToMultipleOf<sizeof(intptr_t)>(srcAsSize));
+    RELEASE_ASSERT(size == WTF::roundUpToMultipleOf<sizeof(intptr_t)>(size));
+
+    intptr_t* dstPtr = reinterpret_cast<intptr_t*>(dst);
+    const intptr_t* srcPtr = reinterpret_cast<const intptr_t*>(src);
+    size /= sizeof(intptr_t);
+    while (size--)
+        *dstPtr++ = *srcPtr++;
+}
+
+#define memcpy asanUnsafeMemcpy
+#endif
+
 // This function must not call malloc(), free(), or any other function that might
 // acquire a lock. Since 'thread' is suspended, trying to acquire a lock
 // will deadlock if 'thread' holds that lock.

Tools/ChangeLog

@@ -1,3 +1,28 @@
+2015-02-25  Mark Lam  <mark.lam@apple.com>
+
+        ASan does not like JSC::MachineThreads::tryCopyOtherThreadStack.
+        <https://webkit.org/b/141672>
+
+        Reviewed by Alexey Proskuryakov.
+
+        ASan does not like the fact that we memcpy the stack for GC scans.  So,
+        we're working around this by using our own memcpy (asanUnsafeMemcpy)
+        implementation that we can tell ASan to ignore.
+
+        Also removed the previous added directive to ignore *tryCopyOtherThreadStack*
+        which isn't effective for working around this issue. 
+
+        * asan/webkit-asan-ignore.txt:
+
+2015-02-17  Dana Burkart  <dburkart@apple.com>
+
+        ASan does not like JSC::MachineThreads::tryCopyOtherThreadStack
+        https://bugs.webkit.org/show_bug.cgi?id=141672
+
+        Reviewed by David Kilzer.
+
+        * asan/webkit-asan-ignore.txt:
+
 2015-02-17  Carlos Garcia Campos  <cgarcia@igalia.com>
 
         Unreviewed. Fix GTK+ make distcheck.

Tools/asan/webkit-asan-ignore.txt

@@ -4,3 +4,4 @@ fun:*DFG*prepareOSREntry*
 # FIXME (rdar://problem/19379214): Register::jsValue() only needs to be blacklisted when
 # called from prepareOSREntry(), but there is currently no way to express this in a blacklist.
 fun:*JSC*Register*jsValue*
+fun:*asanUnsafeMemcpy*