Merge r224539 - AccessCase::generateImpl() should exclude the result …

…register when restoring registers after a call.

https://bugs.webkit.org/show_bug.cgi?id=179355
<rdar://problem/35263053>

Reviewed by Saam Barati.

JSTests:

* stress/regress-179355.js: Added.

Source/JavaScriptCore:

In the Transition case in AccessCase::generateImpl(), we were restoring registers
using restoreLiveRegistersFromStackForCall() without excluding the scratchGPR
where we previously stashed the reallocated butterfly.  If the generated code is
under heavy register pressure, scratchGPR could have been from the set of preserved
registers, and hence, would be restored by restoreLiveRegistersFromStackForCall().
As a result, the restoration would trash the butterfly result we stored there.
This patch fixes the issue by excluding the scratchGPR in the restoration.

* bytecode/AccessCase.cpp:
(JSC::AccessCase::generateImpl):

JSTests/ChangeLog

@@ -1,3 +1,13 @@
+2017-11-07  Mark Lam  <mark.lam@apple.com>
+
+        AccessCase::generateImpl() should exclude the result register when restoring registers after a call.
+        https://bugs.webkit.org/show_bug.cgi?id=179355
+        <rdar://problem/35263053>
+
+        Reviewed by Saam Barati.
+
+        * stress/regress-179355.js: Added.
+
 2017-11-02  Filip Pizlo  <fpizlo@apple.com>
 
         AI does not correctly model the clobber case of ArithClz32

JSTests/stress/regress-179355.js

@@ -0,0 +1,25 @@
+var arr0 = [1,2,3,4];
+var arr1 = new Array(1000);
+
+Array.prototype.__defineGetter__(1, function() {
+    [].concat(arr1); //generate to invalid JIT code here?
+});
+
+Array.prototype.__defineGetter__(Symbol.isConcatSpreadable, (function() {
+    for(var i=0;i<10000;i++) {
+        if(i==0)
+            arr1[i];
+        this.x = 1.1;
+        arr1.legnth = 1;
+    }
+}));
+
+var exception;
+try {
+    arr1[1].toString();
+} catch (e) {
+    exception = e;
+}
+
+if (exception != "RangeError: Maximum call stack size exceeded.")
+    throw "FAILED";

Source/JavaScriptCore/ChangeLog

@@ -1,3 +1,22 @@
+2017-11-07  Mark Lam  <mark.lam@apple.com>
+
+        AccessCase::generateImpl() should exclude the result register when restoring registers after a call.
+        https://bugs.webkit.org/show_bug.cgi?id=179355
+        <rdar://problem/35263053>
+
+        Reviewed by Saam Barati.
+
+        In the Transition case in AccessCase::generateImpl(), we were restoring registers
+        using restoreLiveRegistersFromStackForCall() without excluding the scratchGPR
+        where we previously stashed the reallocated butterfly.  If the generated code is
+        under heavy register pressure, scratchGPR could have been from the set of preserved
+        registers, and hence, would be restored by restoreLiveRegistersFromStackForCall().
+        As a result, the restoration would trash the butterfly result we stored there.
+        This patch fixes the issue by excluding the scratchGPR in the restoration.
+
+        * bytecode/AccessCase.cpp:
+        (JSC::AccessCase::generateImpl):
+
 2017-11-03  Michael Saboff  <msaboff@apple.com>
 
         The Abstract Interpreter needs to change similar to clobberize() in r224366

Source/JavaScriptCore/bytecode/AccessCase.cpp

@@ -945,7 +945,9 @@ void AccessCase::generateImpl(AccessGenerationState& state)
                 state.emitExplicitExceptionHandler();
 
                 noException.link(&jit);
-                state.restoreLiveRegistersFromStackForCall(spillState);
+                RegisterSet resultRegisterToExclude;
+                resultRegisterToExclude.set(scratchGPR);
+                state.restoreLiveRegistersFromStackForCall(spillState, resultRegisterToExclude);
             }
         }