RFR 8160892: VM warning: WaitForMultipleObjects timed out

[David Holmes]

Hi Ivan,

On 7/07/2016 2:55 AM, Ivan Gerasimov wrote:
> Hello!
>
> When fixing JDK-8069048 a potential race was overlooked:
> 1) Thread A wants to call _endthreadex() and registers itself,
> 2) Thread B wants to call exit(), takes its turn and raises the flag
> `process_exiting`,
> 3) Thread A continues, and gets captured in the loop at 4020, in
> SuspendThread(GetCurrentThread()),
> 4) Now, the thread B will have to wait for all the registered threads,
> including the thread A, which will never wake up.
>
> Would you please help review the fix?
>
> BUGURL: https://bugs.openjdk.java.net/browse/JDK-8160892
> WEBREV: http://cr.openjdk.java.net/~igerasim/8160892/00/webrev/

Nit: can we change 'registered_itself" to just "registered" please.

Can you explain under what conditions a thread will now reach the 
self-suspension code. Is that only if an error occurred such that we 
were unable to register our handle for the process-exiting thread to 
wait on? If so some commentary on that block seems appropriate - perhaps 
more appropriate there than back up at the place where it failed to get 
the handle (as Dan requested).

Given we keep missing conditions I'm only cautiously optimistic about 
this. And I'd like to understand how we still sometimes end up exiting 
with an "error code" that seems to be the value of an exception! :(

Thanks,
David

> With kind regards,
> Ivan
>