harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Weldon Washburn" <weldon...@gmail.com>
Subject Re: [drlvm] some pending shutdown bugs
Date Fri, 20 Apr 2007 17:33:52 GMT

I can't get to the shutdown bugs right now.

If you do look at the bugs, please tell us if Rana's patches look like the
right thing to do.  I suspect some/all of the shutdown problems are related
to race conditions that I plan to look at once the ugly threading memory
leak is somewhat under control


PS --- I have been "shutdown" myself for the last 36 hours as my laptop had
to be replaced (including the hard disk).  Soon I will be browsing for
the svn client, jdk client.exe...

On 4/18/07, Tim Ellison <t.p.ellison@gmail.com> wrote:
> Rana Dasgupta wrote:
> > Hi,
> > I went through a set of old and new shutdown bugs. Right now they are
> > all assigned to Weldon.
> >
> >  3582: This is a hang problem for gc.Manythreads that I have off and
> > on on multicore machines. It happens due to mutex corruption when non
> > responding daemon threads ( during shutdown ) are terminated. There is
> > a patch for this which weakens the locks for the shutdown thread so
> > that it does not hang. This has been discussed elsewhere.
> > 3272: Similar problem. Test attached. It intentionally causes
> > deadlock on some daemon threads. Linux uses deferred cancellation and
> > pthread_mutex_lock() { where the deadlocked threads are } is not a
> > cancellation point. However, this does not block the shutdown thread
> > in any way, and shutdown happens cleanly. We should close it.
> > 3151 and 2187: dups . The changes in shudown handling of daemon
> > threads has fixed both bugs, they don't repro.
> > 3314: This is a tools bug, but related to the same shutdown hang
> > issue. My tools environment is messed up and I could not do much with
> > this. But it may be fixed by 3582 and the recent shutdown changes.
> If Weldon isn't able to pick up some of these then I can help look at
> them.
> Regards,
> Tim

Weldon Washburn

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message