Experimental Physics and
Industrial Control System

Jameson Graef Rollins <[email protected]> · Thu, 14 May 2015 10:53:55 -0700

On Wed, May 13 2015, Michael Davidsaver <[email protected]> wrote:
> In my experience, when pthread_mutex_unlock fails w/ EINVAL it is either
> a sign of memory corruption, or that the lock has already been free'd
> (w/ pthread_mutex_destory).
>
> Given that the process doesn't crash I'm guessing its the later.

The process doesn't crash, but the main thread becomes unresponsive.

> Is your application creating, or more importantly destroying, casPVI
> instances when this error appears?  A race during destruction would
> probably explain things.

Certainly not at the level I'm working.  I'm using the pcaspy interface
more or less as prescribed, where all of this kind of stuff is happening
under the hood.

I think I mentioned that this failure seems to be associated with the
severing of subscriptions, but only occasionally.  A race condition
associated with destruction of a casPVI sounds exactly right.  I wonder
if there's something I can do to trigger this directly.

Xiaoqiang, can you comment on what's going on under the hood in pcaspy?
Can I help diagnose further, so we can come up with a targeted fix?

jamie.

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System