Re: [rvm-research] Stopping a single thread

[Ian R. <ian...@ma...>]

Athul Acharya wrote:
> Hi,
>
> I originally asked this a few weeks ago, but since then I've run into
> some issues, and I'm hoping someone on this list can clarify things
> for me :-)
>
> On Dec 1, 2007 11:59 AM, Ian Rogers <ian...@ma...> wrote:
>   
>> Athul Acharya wrote:
>>     
>>> This is along the lines of what I was looking for - to suspend a
>>> thread from inside the VM temporarily, and I think I found it in the
>>> form of VM_Thread.suspend() (right?).
>>>       
>
>   
>> Yep, you can use the java.lang.Thread interface to this of course :-)
>>     
>
> As far as I can tell, using the java.lang.Thread interface would call
> VMThread.suspend() (right?), which immediately calls
> VM_Thread.suspend() anyway.  Also as far as I can tell,
> VMThread.suspend() is the /only/ thing that calls VM_Thread.suspend(),
> and nothing within the rvm calls VMThread.suspend() (right?).  The
> reason I ask is because I suspect VM_Thread.suspend() doesn't
> actuallly work as advertised.
>   

The suspend/resume mechanism is also used by on stack replacement (OSR)
to suspend a thread and resume it when the method is recompiled.

> Assuming the VM_Thread is a green thread, VM_Thread.suspend() calls
> VM_GreenThread.suspendInternal() to change a flag to let the process
> dispatch code know that it needs to be suspended.  Additionally, if it
> sees that it is the current thread on this processor, it goes ahead
> and yield()s.  I don't see anywhere that suspend() will signal the
> thread if it's on a different processor, or wait to return until the
> thread is actually suspended, i.e. no guarantee the thread is
> suspended when the call to suspend() returns.  This may or may not be
> intended behavior, I'm not sure.
>   

You're right iirc that suspended threads aren't placed on any queues,
they should support interruption and other mechanisms.

> Additionally, the function that checks the suspend flag is
> VM_GreenThread.suspendIfPending().  This function clears the suspend
> flag if it's set and returns true, and ostensibly any caller that sees
> it return a true should dequeue that thread. Eclipse shows that the
> only place that calls suspendIfPending() is
> VM_GreenProcessor.dispatch(), and this function does _not_ dequeue
> when suspendIfPending() returns true. So I'm not sure a thread that's
> been suspend()ed ever actually gets suspended.  Please let me know if
> I'm missing something here :-)
>   

Is the thread already dequeued as part of the dispatch?

> All that being said, is there a function somewhere that /will/
> guarantee the thread in question is suspended when it returns?  Where
> do mutator threads get suspended for GC?
>   

I'm not sure what you mean by your first question, you can't arbitrarily
just suspend a thread as it needs to run to a yield point - this is why
the actual suspension occurs in dispatch. Iirc the mutator threads run
into a barrier in VM_HandShake for stop-the-world GC.

Regards,
Ian

> Thanks,
>
> Athul Acharya
> Purdue University