Re: [Sablevm-developer] question about _svmf_stop_the_world()

[Prof. E. M. G. <eti...@uq...>]

Archie Cobbs wrote:
> Presumably you could have simplified things by just waiting for all
> threads to stop (instead of just waiting for non-JNI threads) before
> GC'ing, but then you might have to wait an arbitrarily long time,
> because a JNI thread could be doing something slow, sleeping, etc.

1) Non-JNI threads have at least one GC check point per loop iteration (this is 
a simplification of the reality which includes method entry, etc.)
2) Nothing prevents a JNI thread to go into a infinite loop, or not to call-back 
the VM, or to sleep indefinitely.

So, it is reasonable to assume a non-JNI thread will soon reach a GC point, and 
it would be an error to wait for JNI threads to return or call back the VM 
before proceeding with GC.

In fact, if you read the JNI spec carefully, you'll notice how direct access to 
a reference is never given to native code.  You may rightfully deduce that this 
was made specifically to allow concurrent disruptive GC (e.g. moving collectors 
like copying GC).

If you enjoy learning about SableVM's stop-the-world protocol, go back to 
reading the monitor enter/exit related code which is closely related...

Etienne

-- 
Etienne M. Gagnon, Ph.D.             http://www.info.uqam.ca/~egagnon/
SableVM:                                       http://www.sablevm.org/
SableCC:                                       http://www.sablecc.org/