On 9/19/06, Geir Magnusson Jr. <[EMAIL PROTECTED]> wrote:
On Sep 19, 2006, at 2:18 AM, Weldon Washburn wrote:
> All,
> I have noticed endless loop behavior when running gc.LOS. It
> appears to go
> into some sort of endless loop when I try, " build test". Does
> anyone else
> see this problem?
>
> I used MSVC to break into drlvm when it gets stuck. It shows
> basically
> what's been reported before – a bunch of threads in JITed code.
> They keep
> making some system call.
>
> Semis/vm/_smoke.tests/reports/gc.LOS_jit.out shows that somehow
> LOS.java is
> in an infinite loop after it prints all 200 dots. This is rather
> curious.
>
> Looking at gc/LOS.java, there is a "threads[i].join()" where "i"
> goes from 0
> to 199. This thread join happens immediately after a "notifyAll()"
> that is
> intended to tell each of the threads to start running.
>
> I moved the trace(".") to immediately after the synchronized
> statement in
> run(). The test now completes successfully. It might be a bug in the
> implementation of Object.wait() and Object.notifyAll() that
> different HW/SW
> combinations are aggrevating??? Below are the mods that I made:
>
I'd not commit this... we need this to help us find what the problem is.
Exactly. Sorry for not being clearer. I think it might be a bug in
Object.wait() that we need to find.
geir
--
Weldon Washburn
Intel Middleware Products Division
