893679 - Crash at weird memory address with testcase involving ParallelArray

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Gary Kwong [:gkw] [:nth10sd] (NOT official MoCo now)]

Attachment: stack

s = newGlobal();
function ff(code) {
    try {
        evalcx(code, s)
    } catch (e) {}
}
ff("\
    Object.defineProperty(this,\"v2\",{\
        get:function(){\
            f1({})\
        }\
    });\
    o2 = new Object;\
    f2 = (function(j){\
        if(j) {}\
    });\
    f1 = f2;\
    for(v of v2) {}\
");
ff("\
    ParallelArray([89], function(x0){\
        if(x0%86==71) {\
            f1(o2.p)\
        }\
    })\
")

crashes js opt shell (threadsafe, deterministic) on m-c changeset 18467a85acf6 with --baseline-eager at a weird memory address.

Not sure what's going on here, setting s-s just-in-case.

Comment 1

[Christian Holler (:decoder)]

JSBugMon: Cannot process bug: Unable to automatically reproduce, please track manually.

Comment 2

[Shu-yu Guo [:shu]]

I can reproduce, looking at it.

Comment 3

[Shu-yu Guo [:shu]]

Attachment: fix

Comment 4

[Brian Hackett [Laid off!]]

Comment on attachment 775653 [details] [diff] [review]
fix

Review of attachment 775653 [details] [diff] [review]:
-----------------------------------------------------------------

This doesn't seem like the right fix.  property->empty() means that no object with that type has that property, but even if !property->empty() there could be some objects with the type which have the property, and which could generate an undefined value when read from which would normally be reflected via a Monitor call.

I think that, as long as you're not able to call TypeScript::Monitor from within the ParallelGetPropertyIC VM call, you always need a barrier when generating ParallelGetPropertyIC instructions.

Comment 5

[Brian Hackett [Laid off!]]

Oh, that should read 'there could be some objects with the type which *don't* have the property'

Comment 6

[Shu-yu Guo [:shu]]

Attachment: fix v2

Yes, you were right about the previous patch, that fix was bogus.

Comment 7

[Brian Hackett [Laid off!]]

Comment on attachment 775710 [details] [diff] [review]
fix v2

Review of attachment 775710 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/ion/IonBuilder.cpp
@@ +7956,5 @@
> +            barrier = true;
> +        break;
> +      default:
> +        MOZ_ASSUME_UNREACHABLE("No such execution mode");
> +    }

This code is pretty verbose.  How about:

if (info().executionMode() == ParallelExecution &&
    !types->hasType(types::Type::UndefinedType()))
{
    barrier = true;
}

(The NULL check is not necessary on |types|)

Comment 8

[Shu-yu Guo [:shu]]

(In reply to Brian Hackett (:bhackett) from comment #7)
> Comment on attachment 775710 [details] [diff] [review]
> fix v2
> 
> Review of attachment 775710 [details] [diff] [review]:
> -----------------------------------------------------------------
> 
> ::: js/src/ion/IonBuilder.cpp
> @@ +7956,5 @@
> > +            barrier = true;
> > +        break;
> > +      default:
> > +        MOZ_ASSUME_UNREACHABLE("No such execution mode");
> > +    }
> 
> This code is pretty verbose.  How about:
> 
> if (info().executionMode() == ParallelExecution &&
>     !types->hasType(types::Type::UndefinedType()))
> {
>     barrier = true;
> }
> 
> (The NULL check is not necessary on |types|)

Fine with me. The verbosity was to keeping with other sites where we case on execution mode, to be extra explicit about what we're doing.

Comment 9

[Gary Kwong [:gkw] [:nth10sd] (NOT official MoCo now)]

autoBisect shows this is probably related to the following changeset:

The first bad revision is:
changeset:   http://hg.mozilla.org/mozilla-central/rev/0c576cf51a80
user:        Shu-yu Guo
date:        Wed Jul 03 09:47:28 2013 -0700
summary:     Bug 886102 - Ignore idempotency for parallel ICs. (r=nmatsakis)

Comment 10

[Shu-yu Guo [:shu]]

Attachment: fix for SM(e) orange

r=bhackett or irc.

Comment 11

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/2a9ef042a6d0
https://hg.mozilla.org/mozilla-central/rev/6adcce906728