Bug 134945 – Need ability to fuzz exception throwing

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Filip Pizlo]

Patch forthcoming.

[Filip Pizlo]

Created attachment 234957 [details]
almost done

[Filip Pizlo]

This causes a rare crash in v8-earley-boyer.  I will try to fix it.

[Filip Pizlo]

(In reply to comment #2)
> This causes a rare crash in v8-earley-boyer.  I will try to fix it.

Here's the repro:

[pizlo@dethklok OpenSource] DYLD_FRAMEWORK_PATH=WebKitBuild/Release/ lldb -- WebKitBuild/Release/jsc Source/JavaScriptCore/tests/exceptionFuzz/earley-boyer.js --enableConcurrentJIT=false --enableExceptionFuzz=true  --fireExceptionFuzzAt=13582
Current executable set to 'WebKitBuild/Release/jsc' (x86_64).
(lldb) r
Process 47764 launched: '/Volumes/Data/pizlo/quartary/OpenSource/WebKitBuild/Release/jsc' (x86_64)
JSC EXCEPTION FUZZ: Throwing fuzz exception.
Process 47764 stopped
* thread #1: tid = 0x3d48178, 0x000000010022b8bd JavaScriptCore`JSC::UnwindFunctor::operator()(JSC::StackVisitor&) [inlined] JSC::WriteBarrierBase<JSC::SymbolTable>::get() const at WriteBarrier.h:92, queue = 'com.apple.main-thread, stop reason = EXC_BAD_ACCESS (code=EXC_I386_GPFLT)
    frame #0: 0x000000010022b8bd JavaScriptCore`JSC::UnwindFunctor::operator()(JSC::StackVisitor&) [inlined] JSC::WriteBarrierBase<JSC::SymbolTable>::get() const at WriteBarrier.h:92
   89  	    T* get() const
   90  	    {
   91  	        // Copy m_cell to a local to avoid multiple-read issues. (See <http://webkit.org/b/110854>)
-> 92  	        JSCell* cell = m_cell;
   93  	        if (cell)
   94  	            validateCell(cell);
   95  	        return reinterpret_cast<T*>(static_cast<void*>(cell));
(lldb) bt
[0x0000000000000000 - 0x0000000000000198)
[0x0000000000000198 - 0x0000000000000336)
[0x0000000000000336 - 0x000000000000035c)
[0x000000000000035c - 0x000000000000632b)
* thread #1: tid = 0x3d48178, 0x000000010022b8bd JavaScriptCore`JSC::UnwindFunctor::operator()(JSC::StackVisitor&) [inlined] JSC::WriteBarrierBase<JSC::SymbolTable>::get() const at WriteBarrier.h:92, queue = 'com.apple.main-thread, stop reason = EXC_BAD_ACCESS (code=EXC_I386_GPFLT)
    frame #0: 0x000000010022b8bd JavaScriptCore`JSC::UnwindFunctor::operator()(JSC::StackVisitor&) [inlined] JSC::WriteBarrierBase<JSC::SymbolTable>::get() const at WriteBarrier.h:92
    frame #1: 0x000000010022b8bd JavaScriptCore`JSC::UnwindFunctor::operator()(JSC::StackVisitor&) [inlined] JSC::JSSymbolTableObject::symbolTable() const at JSSymbolTableObject.h:43
    frame #2: 0x000000010022b8bd JavaScriptCore`JSC::UnwindFunctor::operator()(JSC::StackVisitor&) [inlined] JSC::JSActivation::tearOff(this=0xffff000000000002) at JSActivation.h:151
    frame #3: 0x000000010022b8bd JavaScriptCore`JSC::UnwindFunctor::operator()(JSC::StackVisitor&) [inlined] JSC::unwindCallFrame(JSC::StackVisitor&) + 312 at Interpreter.cpp:461
    frame #4: 0x000000010022b785 JavaScriptCore`JSC::UnwindFunctor::operator(this=0x00007fff5fbfc8c8, visitor=0x00007fff5fbfc910)(JSC::StackVisitor&) + 117 at Interpreter.cpp:661
    frame #5: 0x000000010022876b JavaScriptCore`JSC::Interpreter::unwind(JSC::ExecState*&, JSC::JSValue&) [inlined] void JSC::StackVisitor::visit<JSC::UnwindFunctor>(functor=0x00007fff5fbfc9a0, startFrame=<unavailable>) + 42 at StackVisitor.h:123
    frame #6: 0x0000000100228741 JavaScriptCore`JSC::Interpreter::unwind(JSC::ExecState*&, JSC::JSValue&) [inlined] void JSC::ExecState::iterate<JSC::UnwindFunctor>(functor=0x00007fff5fbfc9a0, this=<unavailable>) at CallFrame.h:308
    frame #7: 0x0000000100228741 JavaScriptCore`JSC::Interpreter::unwind(this=<unavailable>, callFrame=0x00007fff5fbfc9a0, exceptionValue=0x00007fff5fbfc998) + 513 at Interpreter.cpp:734
    frame #8: 0x0000000100245db8 JavaScriptCore`JSC::genericUnwind(vm=0x000000010086e000, callFrame=0x00007fff5fbfcab0, exceptionValue=JSValue at 0x00007fff5fbfc998) + 72 at JITExceptions.cpp:51
    frame #9: 0x0000538cb201172f
    frame #10: 0x0000538cb20462be
    frame #11: 0x0000538cb201da1e
    frame #12: 0x0000538cb201db3e
    frame #13: 0x0000538cb204625e
    frame #14: 0x0000538cb201a4de
    frame #15: 0x0000538cb20461fe
    frame #16: 0x0000538cb204625e
    frame #17: 0x0000538cb20187fe
    frame #18: 0x0000538cb204625e
    frame #19: 0x0000538cb201a4de
    frame #20: 0x0000538cb20461fe
    frame #21: 0x0000538cb204625e
    frame #22: 0x0000538cb201a4de
    frame #23: 0x0000538cb20461fe
    frame #24: 0x0000538cb20462be
    frame #25: 0x0000538cb201da1e
    frame #26: 0x0000538cb201da1e
    frame #27: 0x0000538cb201db3e
    frame #28: 0x0000538cb20462be
    frame #29: 0x0000538cb201da1e
    frame #30: 0x0000538cb201db3e
    frame #31: 0x0000538cb2018224
    frame #32: 0x0000538cb201a4de
    frame #33: 0x0000538cb2011353
    frame #34: 0x0000538cb2018224
    frame #35: 0x0000538cb201a4de
    frame #36: 0x0000538cb201131c
    frame #37: 0x0000538cb2018224
    frame #38: 0x0000538cb201a4de
    frame #39: 0x0000000100336f20 JavaScriptCore`llint_entry + 22744
    frame #40: 0x0000000100336f20 JavaScriptCore`llint_entry + 22744
    frame #41: 0x0000000100336eb5 JavaScriptCore`llint_entry + 22637
    frame #42: 0x0000000100336eb5 JavaScriptCore`llint_entry + 22637
    frame #43: 0x0000000100336f20 JavaScriptCore`llint_entry + 22744
    frame #44: 0x0000000100336eb5 JavaScriptCore`llint_entry + 22637
    frame #45: 0x0000000100336eb5 JavaScriptCore`llint_entry + 22637
    frame #46: 0x0000000100331423 JavaScriptCore`callToJavaScript + 311

[Filip Pizlo]

Looks like this particular failure was caused by exception fuzzing incorrectly causing an exception to be thrown from operationOptimize.  In an ideal world, we would fix this by removing the exception check from operationOptimize.  But there's the possibility that this exception check was picking up a stale exception that we hadn't ever checked - this is a really dumb bug we have sometimes.  Then, we'd crash instead of throwing.

[Filip Pizlo]

Created attachment 234974 [details]
the patch

[Sam Weinig]

Comment on attachment 234974 [details]
the patch

View in context: https://bugs.webkit.org/attachment.cgi?id=234974&action=review

> Source/JavaScriptCore/jit/JITOperations.cpp:1812
> +    ExecState* exec = static_cast<ExecState*>(__builtin_frame_address(1));

It's probably worth noting why you are doing these shenanigans for future sanity.

[Geoffrey Garen]

Comment on attachment 234974 [details]
the patch

View in context: https://bugs.webkit.org/attachment.cgi?id=234974&action=review

> Source/JavaScriptCore/interpreter/Interpreter.cpp:461
> +        if (activation && activation.isCell())
>              jsCast<JSActivation*>(activation)->tearOff(*scope->vm());

When is the activation present but not a cell?

[Filip Pizlo]

(In reply to comment #7)
> (From update of attachment 234974 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=234974&action=review
> 
> > Source/JavaScriptCore/interpreter/Interpreter.cpp:461
> > +        if (activation && activation.isCell())
> >              jsCast<JSActivation*>(activation)->tearOff(*scope->vm());
> 
> When is the activation present but not a cell?

If we throw an exception between op_enter and op_init_lazy_reg, then both the activation and the arguments registers will be undefined.

This can definitely happen due to my fuzzer.

It's also possible if we have a pending exception at the time that we execute a function's prologue.  It's kind of funny, but it could probably happen if we have other bugs elsewhere.

[Radar WebKit Bug Importer]

<rdar://problem/17722027>

[Filip Pizlo]

Landed in http://trac.webkit.org/changeset/171213

[Csaba Osztrogonác]

(In reply to comment #10)
> Landed in http://trac.webkit.org/changeset/171213

it made 3 jsc tests fail everywhere

[Filip Pizlo]

(In reply to comment #11)
> (In reply to comment #10)
> > Landed in http://trac.webkit.org/changeset/171213
> 
> it made 3 jsc tests fail everywhere

Investigating.

[Filip Pizlo]

(In reply to comment #12)
> (In reply to comment #11)
> > (In reply to comment #10)
> > > Landed in http://trac.webkit.org/changeset/171213
> > 
> > it made 3 jsc tests fail everywhere
> 
> Investigating.

Will land a fix shortly.

[Filip Pizlo]

(In reply to comment #13)
> (In reply to comment #12)
> > (In reply to comment #11)
> > > (In reply to comment #10)
> > > > Landed in http://trac.webkit.org/changeset/171213
> > > 
> > > it made 3 jsc tests fail everywhere
> > 
> > Investigating.
> 
> Will land a fix shortly.

Should be fixed in http://trac.webkit.org/changeset/171216.

[Filip Pizlo]

(In reply to comment #14)
> (In reply to comment #13)
> > (In reply to comment #12)
> > > (In reply to comment #11)
> > > > (In reply to comment #10)
> > > > > Landed in http://trac.webkit.org/changeset/171213
> > > > 
> > > > it made 3 jsc tests fail everywhere
> > > 
> > > Investigating.
> > 
> > Will land a fix shortly.
> 
> Should be fixed in http://trac.webkit.org/changeset/171216.

Make that http://trac.webkit.org/changeset/171217.

[Csaba Osztrogonác]

And it broke the CLOOP build:
Undefined symbols for architecture x86_64:
  "__ZN3JSC27numberOfExceptionFuzzChecksEv", referenced from:
      __Z7jscmainiPPc in jsc.o

[Filip Pizlo]

(In reply to comment #16)
> And it broke the CLOOP build:
> Undefined symbols for architecture x86_64:
>   "__ZN3JSC27numberOfExceptionFuzzChecksEv", referenced from:
>       __Z7jscmainiPPc in jsc.o

Fixed in http://trac.webkit.org/changeset/171228