Experimental Physics and
Industrial Control System

"Jeff Hill" <[email protected]> · Tue, 22 Mar 2005 09:09:17 -0700

Hello All,

Sorry about the delay, I have been tied up with travel to the APS
and the SNS over the last two weeks.

This message is relevant to a CA gateway failure reported by Rolf
Keitel (Mantis Entry 113). The symptom observed was a corrupt
value written through the CA gateway when it is under a heavy
client induced write through load. The bug has been isolated to
the portable CA server. An incoming payload for a CA put was byte
swapped twice when an IO operation was restarted due to heavy
client loads.

I have committed a fix to CVS. This fix should appear first in
EPICS R3.14.7. 

It will be important to *not* enable writes through your gateway
until you receive a patch for this bug.

Thanks to Rolf Keitel, Ralph Lange, and Ken Evans for their
assistance with this matter.

Sorry about any inconvenience that this has caused.

Jeff

> -----Original Message-----
> From: Kenneth Evans, Jr. [mailto:[email protected]]
> Sent: Monday, March 14, 2005 1:02 PM
> To: 'tech_talk'
> Subject: RE: dm and edm problems with CA gateway
> 
> I thought Jeff would respond to this topic, but since he has
> not, let me say
> he was at ANL last week for another reason, and we traced the
> problem to the
> fact that CAS is swapping the byte order each time the request
> is sent.
> This causes it to be garbage every other time when it is (1)
> postponed on
> (2) machines that do not have network byte order (e.g. Linux).
> 
> That it has not happened more often is because there is usually
> no need to
> postpone write requests.  A slider on a fast machine can cause
> it to happen,
> however.
> 
> It will be fixed in base at some point.
> 
> 	-Ken

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System