At present, using the store buffer off-main-thread would be extremely problematic, so I would think this should be runtimeFromMainThread. This was last touched by f836042326f9: Bug 898886 - Improve threadsafe assertions when accessing runtimes and zones.

These methods can be called off thread during parsing and other future operations, so using runtimeFromMainThread() would be incorrect. The store buffer shouldn't be used off the main thread, so the intent is that these post() barriers never trigger because nothing the non-main threads ever do will use nursery things. All the writes done off thread are to fresh zones which don't contain nursery things, and we don't allocate from the nursery off the main thread (cx->hasNursery()). This might need tweaking in the future since the first condition will be relaxed when we start serializing objects off thread, but for now this should hold. The attached patch beefs up assertions in the store buffer to make sure the buffer is being used in a threadsafe way.

Comment on attachment 801882 [details] [diff] [review] beef up asserts Review of attachment 801882 [details] [diff] [review]: ----------------------------------------------------------------- Excellent! Thank you. ::: js/src/gc/StoreBuffer.h @@ +443,5 @@ > > /* Insert an entry into the generic buffer. */ > template <typename T> > void putGeneric(const T &t) { > + JS_ASSERT(CurrentThreadCanAccessRuntime(runtime)); Could you move these up to MonoTypeBuffer::put and GenericBuffer::put? Right below |JS_ASSERT(!owner->inParallelSection());|, or replacing it, if the new assertion subsumes it. You should be able to access the runtime there as |owner->runtime|.