Experimental Physics and
Industrial Control System

Eric Norum <[email protected]> · Fri, 19 Mar 2010 07:01:33 -0700

Thanks.
I'm surprised that this problem survived so long.   I wonder what has changed with the SRQ setup in the scope.  

Marty and I asked Tektronix several times for information about theSRQ setup protocol.  We always received the same answer, "Tektronix provides Labview VISA drivers".    It would be greatly appreciated if someone with one of these scopes, and EPICS, and the Labview driver, and a network traffic sniffer could compare the first few packets between the host and the scope.  Then perhaps we could fix the SRQ handling properly.

On Mar 19, 2010, at 4:34 AM, Benjamin Franksen wrote:

> On Wednesday 17 March 2010, Benjamin Franksen wrote:
>> On Wednesday 17 March 2010, Benjamin Franksen wrote:
>>> I'm trying to talk to the mentioned scope via asyn/stream and get
>>> mysterious crashes the first time the IOC tries to communicate over
>>> the link. Below is a trace of the IOC shell up to the crash. I am using
>>> the latest and greatest versions of everything except EPICS base:
>>> base-3.14.8.2, asyn-4.11a (supposedly the last version that works with
>>> base-3.14.8), StreamDevice-2- snapshot20081007. The crash happens on
>>> the latest ubuntu version as well as on a debian stable system. I
>>> should point out that with the exact same setup I can talk fine with
>>> e.g. a GPIB controller.
>> 
>> Ok I did dome more research. Götz suggested valgrind and indeed I got
>> interesting results:
>> 
>> (1) When I run the IOC under valgrind it doesn't crash, it even talks to
>> the device. Nice.
>> 
>> (2) Valgrind complains about StreamDevice doing "Mismatched free() /
>> delete / delete []". See below for a transcript.
>> 
>> My current theory is that these mismatched calls cause the memory free
>> list to become corrupted. Note that VXI-11 is based on RPC which heavily
>> uses malloc/free internally during runtime, so there is some probability
>> that a corrupted free list will lead to a crash inside the RPC library.
> 
> I had to abandon this theory. I managed to fix the mismatched delete calls 
> (so that valgrind no longer complained) but it did not fix anything. No 
> wonder, the calls where valgrind wants a 'delete[]' instead of just 'delete' 
> are for an array of chars. No destructor for type char, so this should not 
> matter at all and indeed it doesn't.
> 
> Still, IOC works under valgrind, but crashes without.
> 
> I tried electric-fence but this tool just makes any soft IOC crash almost 
> immediately.
> 
> Then I removed the call to vxiCreateIrqChannel from vxiConnectPort (in file 
> drvVxi11.c). This finally let the IOC run without crashing.
> 
> So, it seems, streamDevice is innocent, after all. There must be something 
> wrong (no idea what, exactly) with the vxi-11 SRQ implementation in asyn.
> 
> Cheers
> Ben
> 

-- 
Eric Norum
[email protected]

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System