659839 - IonMonkey: Implement OOM Ballast

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[David Anderson [:dvander] - inactive, e-mail if emergency]

As of bug 659782 the allocator is treated as infallible for small, fix-sized structures. The ballast isn't actually implemented though, so at some point we'll have to modify IonAllocPolicy::ensureBallast to reserve space in the arena.

Comment 1

[Nicholas Nethercote [inactive]]

A tip from TM:  be careful where you put the ballast.  TM has some ballast (for nanojit) in every compartment, and you can have a lot of compartments, so that's just wasted space.  It's much better if you can put it in the JSRuntime, say.  Though I guess you then have locking considerations.  Hmm.  Ballast kinda sucks.

Comment 2

[David Anderson [:dvander] - inactive, e-mail if emergency]

Would the threaddata be okay in terms of memory use? The compiler won't re-enter and the ballast could live and die with the compiler, if need be.

Comment 3

[Nicholas Nethercote [inactive]]

(In reply to comment #2)
> Would the threaddata be okay in terms of memory use? The compiler won't
> re-enter and the ballast could live and die with the compiler, if need be.

Sounds ideal.  Do you mean ThreadData in jscntxt.h?  I wonder if the tracer ballast could be moved in there as well.

Comment 4

[David Anderson [:dvander] - inactive, e-mail if emergency]

This got fixed a while ago.