646411 - TI: Crash [@ JSString::length]

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Christian Holler (:decoder)]

The following code crashes on TI tip (run with -n -m -a), tested on 64 bit:

__proto__ = Function();
eval("\
var MS = 16;\
addNewTestCase(new Date(1899,11,31,16,0,0), \"new Date(1899,11,31,16,0,0)\", typeof UTC_DAY == 'undefined');\
function addNewTestCase( DateCase, DateString, ResultArray ) {\
        ResultArray[MS];\
}\
");


Backtrace (looks like near-null pointer deref):


0x0000000000412c64 in JSString::length (this=0x1) at ./jsstr.h:244
244             return d.lengthAndFlags >> LENGTH_SHIFT;
(gdb) bt
#0  0x0000000000412c64 in JSString::length (this=0x1) at ./jsstr.h:244
#1  0x000000000077a2e7 in js::mjit::stubs::GetElem (f=@0x7fff19787e60) at ./methodjit/StubCalls.cpp:444
#2  0x00007f0c292a555d in ?? ()
#3  0x00007f0c292a4ac0 in ?? ()
#4  0x0000000001252b10 in ?? ()
#5  0x00007fff197883b0 in ?? ()
#6  0x0000000000000000 in ?? ()

Comment 1

[Brian Hackett [Laid off!]]

We got confused by the opcode fusion in Compiler::jsop_typeof, and gave the value pushed by the following '==' a type set containing only strings, so that when we propagated types to addNewTestCase in the call IC it thought the argument was a string, and passed a torn value to GetElem.

This fix does not specify type sets for values pushed by fused opcodes.

http://hg.mozilla.org/projects/jaegermonkey/rev/df80ae4a8776

Comment 2

[Christian Holler (:decoder)]

A testcase for this bug was automatically identified at js/src/jit-test/tests/jaeger/bug646411.js.