Experimental Physics and
Industrial Control System

"Jeff Hill" <[email protected]> · Mon, 16 Aug 2010 17:46:41 -0600

> + this->iiuExistenceCount = this->circuitList.count();

Examining this change a bit closer I see that it will introduce a race condition when circuits are being created and destroyed at close to the same instant in time. See revision 11786 of cac.cpp which fixes mantis 334 if you are interested in what this code does. After running the regression tests, I pushed in a fix (which is removing above mentioned change and restoring the increment of iiuExistenceCount in cac::findOrCreateVirtCircuit.

Jeff
______________________________________________________
Jeffrey O. Hill           Email        [email protected]
LANL MS H820              Voice        505 665 1831
Los Alamos NM 87545 USA   FAX          505 665 5107

Message content: TSPA

> -----Original Message-----
> From: Ralph Lange [mailto:[email protected]]
> Sent: Monday, August 16, 2010 10:03 AM
> To: Andrew Johnson
> Cc: Jeff Hill; EPICS Core Talk
> Subject: Fixed: Segfaults in 3.14 branch since merging ca-over-tcp
> 
>   Hi Andrew,
> 
> it looks as I have found the reason for the segfaults, which I think is
> a bug in CAC.
> 
> In the cac::~cac() destructor, a private member
> "this->iiuExistenceCount" is used to determine if cac should wait for
> TCP threads to shutdown. While this counter gets correctly decreased in
> cac::destroyIIU(), it never gets increased or set (except initialized to
> 0 in the cpp constructor).
> 
> So cac::~cac() was never waiting for the TCP threads to shutdown, and
> happily kept tearing down the timer/mutex infrastructure. If the threads
> were not fast enough, they were hitting on destroyed semaphores when
> cancelling their watchdog timers - that was causing the exceptions and
> segfaults.
> 
> This one-line fix will take care of the problem:
> 
> 
> === modified file 'src/ca/cac.cpp'
> --- src/ca/cac.cpp      2010-04-15 21:06:16 +0000
> +++ src/ca/cac.cpp      2010-08-16 15:24:13 +0000
> @@ -287,6 +287,7 @@
>               //
>               // shutdown all tcp circuits
>               //
> +            this->iiuExistenceCount = this->circuitList.count();
>               tsDLIter < tcpiiu > iter = this->circuitList.firstIter ();
>               while ( iter.valid() ) {
>                   // this causes a clean shutdown to occur
> 
> 
> Maybe it actually takes care of several shutdown related issues...
> 
> Yay!
> Ralph

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System