On 7/14/16 4:57 PM, Ivan Gerasimov wrote:
Hey!
I've updated the webrev in place at
http://cr.openjdk.java.net/~igerasim/8160892/01/webrev/
Thumbs up!
Dan
On 15.07.2016 0:31, Daniel D. Daugherty wrote:
On 7/14/16 12:46 PM, Ivan Gerasimov wrote:
Thank you David for looking into this!
Here's the webrev updated in accordance with your and Daniel's
suggestions:
http://cr.openjdk.java.net/~igerasim/8160892/01/webrev/
src/os/windows/vm/os_windows.cpp
L3900: bool registered = false;
Please consider adding a comment above this variable:
// We only attempt to register threads until a process exiting
// thread manages to set the process_exiting flag. Any threads
// that come through here after the process_exiting flag is set
// are unregistered and will be caught in the SuspendThread()
// infinite loop below.
Sure. Added this comment.
My request is different than David's. I prefer the comment up
here by the 'registered' variable because that variable is
key for getting into the SuspendThread() infinite loop. The
comment also occurs before all the code paths that you have
trace out with the different Ept types and flags states.
I'm really hoping that this is the last tweak we need to make to
clearly comment what we're doing to alleviate this OS race.
The number of occurrences of the underlying bug keep going down
with every iteration so we're getting close.
Yes. This is my hope too.
With kind regards,
Ivan
