Experimental Physics and
Industrial Control System

[email protected] (Jeff Hill) · Wed, 8 Dec 1999 19:36:52 -0700

>
>      I have a particular IOC (an MV162) that has been giving me loads of
> grief lately.  It keeps giving me "Access Fault"'s, even when I haven't
> started any programs on the sequencer.  The databases get loaded, and the
> object files, but sometime after the iocInit, there's an access fault.
> The latest one happened after the startup script had finished executing,
> but before I had (manually) started running any programs on the sequencer.
> The full description was:
>

This occurs when the CA UDP daemon receives a particularly badly formulated
request message. We have seen this on rare occasions at our site.

In one situation we saw this when there was a thinwire Ethernet network in
an office that was invalid (it was a tree structure which isn't allowed
with thinwire). The problem vanished when this wiring problem was fixed.

Recently we saw this problem when a new 100 Mb switch was installed. The
problem
was not resolved until we powered down our IOCs which have 10/100 Mb
auto-sensing
Ethernet interface cards installed.

Even more recently I have seen situations where the CA UDP daemon has
received
invalid messages during regression testing.

Just a few days ago I received a message about a CA UDP daemon at KEK
receiving
invalid messages.

I am not convinced at this time that this isn't a bug in CA so I am
investigating the problem. It is also possible that if many frames on the
network are damaged, and a large number of bytes are damaged in each frame,
that eventually an invalid UDP frame will pass the Ethernet/UDP CRC checksum
validations and be passed off to the CA UDP daemon. If this were the case
then you should see errors when you type "ifShow" and "ipstatShow" on
vxWorks.

If other sites have seen situations where the CA UDP daemon is crashing or
there are messages about CA corrupt UDP requests then I would like to hear
about the frequency of occurrence and any surrounding circumstances that
appear to be correlated with the problem.

The R3.13.1 branch in CVS contains a patch which makes the CA server much
more
robust against poorly formulated request message. APS is preparing
a release of this version.

Jeff

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System