Experimental Physics and
Industrial Control System

Kate Feng <[email protected]> · Sun, 23 Nov 2008 00:47:25 -0500

I was in a rush for a report previously that I did not get to
reply to all of this.  Please see below.
Till Straumann wrote:

Kate Feng wrote:

Szalata, Zenon M. wrote:

I need a device driver for the CAEN V965 module, which works with
vxWorks 6.x. Here at SLAC we have a device driver for this module
which works with rtems only.
Thanks,
Zen

Is the driver written for EPICS 3.14.x ? If so, it should be
easy to port it to vxWorks.  It is a good idea to write a driver
for multiple platforms.  The "Application Developer's Guide"
provides a good guideline.
1) For the src/Makefile :
Copy the existing Makefile to Makefile.RTEMS.  For the vxworks
Makefile, change "PROD_IOC_RTEMS" to "PROD_IOC_vxWorks"
and take out all the RTEMS related include path.
2) For the source code under src/ :
vxworks access the I/O register via a  pointer to the address.
For example : for the long data access, which is four bytes, it does:
   volatile epicsUInt32 *reg;
    epicsUInt32 value;
 reg = (volatile epicsUInt32 *) io_address ;

My 2 cents:

I am not sure how much does 2 cents cost. :-)

Accessing device registers this way is bad practice.
a) it is not obvious where I/O is done. Makes porting this code e.g., to
    a CPU of different endianness much more cumbersome.

It should not be cumbersome to be implemented for EPICS.  The
endianne  for CA is  already implemented gracefully. It could be
even more cumbersome to port in different ASM languages
among various CPUs into EPICS.

b) may not synchronize I/O properly; modern CPUs may reorder
    instruction execution and need special instructions to enforce
    in-order execution of critical code sequences.

I am sure vxWorks has a smart path to handle this kind of issues.
Did any vxWorks users experience  any reorder problem on
the modern CPUs ?

Both issues are addressed by the RTEMS inline routines in io.h and
byteorder.h

Again this depends on the ASM language of each CPU, which could
be really "cumbersome".

It certainly would be great to have equivalent functionality available
in EPICS/OSI.
For now, you'll have to find out what vxWorks provides in this area
or ultimately code it yourself.

Vxworks provided.   It's up to one's choice.  Of course, I like how it was
done in RTEMS, too.  :-)

However, I suggest you *always* perform register access by means
of a macro or inline routine for sake of portability and ease of
maintenance.

Sometimes, the sake of portability and ease of maintenance is in the
beholder's eyes.  There is no clear cut in each instance. :-)
Ciao,
Kate

-- Till

 *reg = value;
while RTEMS  accesses  the I/O register via the command in_be32(  ) or
in_le32().
You can reference "synAppsRTEMS"  source code at the web site :
http://www.nsls.bnl.gov/facility/expsys/software/EPICS/
The software is meant to be ported for both RTEMS and vxWorks.
Cheers,
Kate

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System