Experimental Physics and
Industrial Control System

Michael Davidsaver <[email protected]> · Tue, 18 Nov 2014 15:12:37 -0500

On 11/18/2014 11:37 AM, Michael Davidsaver wrote:
> ...
> From a quick look I would guess that at least dbBkptExit(),
> dbLockExit(), and errlogExitHandler() can cause similar problems.  So
> I'll look at making these conditional as well.

Ok, so I think I've backed out enough of the DB cleanup to prevent
crashing due to shutdown races.

I think the remaining atexit hooks for softIoc shouldn't crash.  What is
done is:

> atExit exitDatabase((nil))

For a real IOC (as opposed to a unittest) only doCloseLinks() runs to
cleanup CA_LINK.  For records with extended device support it also
disables "I/O Intr" scans and call del_record().

> atExit ca_client_exit_handler((nil))

Frees a thread local variable.

> atExit dbCaExit((nil))

Stops the CA link worker.

> atExit cacExitHandler((nil))

Frees a thread local variable.

> atExit twdShutdown((nil))

Stops the watchdog thread.

> atExit errlogExitHandler((nil))

Stops the errlog thread

> atExit ClockTime_Shutdown((nil))

Stops the ClockTimeSync() thread.

None of these should free any global locks so that concurrent operations
using these locks shouldn't operate on free'd memory.

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System