EPICSRTEMS at NSLS

1
EPICS/RTEMS at NSLS

• S. Kate Feng
• EPICS Meeting at DLS, UK
• June 19, 2003

1
2
Projects that use EPICS at NSLS

• DUV-FEL accelerator (only one beamline)
• I began to implement EPICS/VxWorks in 2000
• 400 nm SASE achieved in February, 2002
• 266 nm HGHG achieved in October, 2002
• Currently in operation for users
• NSLS beamlines (currently gt 10)
• Peter Siddons and Ivan So began to implement
  EPICS/VxWorks in 2000
• Peter Siddons proposed to use RTEMS for beamline
  applications

2
3
EPICS/RTEMS at NSLS

• I began the project in October, 2002
• Development platform is Linux RedHat 7.3
• EPICS 3.14 IOC test platform is RTEMS
• Evaluated cexp vs. esh for RTEMS dynamic loading
• cexp is ideal for run-time dynamic loading plus
  other nice features
• esh would be a good debugging tool (e.g.
  breakpoint and single step) if it is fully
  developed as promised. All the run-time modules
  to be loaded have to be pre-mounted into the IMFS
  file system via a makefile before running

3
4
Sample features as of June 10, 2003
4
5
RTEMS IOC configuration at NSLS

• RTEMS IOC configuration
• CPU MVME2306
• RTEMS gcc-3.2.2, newlib-1.11.0, rtems-4.6.0pre3
• RTEMS image download bootp/tftp
• Shell rtems-cexp for dynamic loading (T.S.,
  SSRL)
• Remote login telnetd (Till Straumann, SSRL,
  SLAC)
• Portability RTEMS devLib.c (T.S., SSRL), epics
  threads, epicsMutex, and so on (Eric Norum, APS)
• EPICS 3.14.0beta2 (Marty Kraimer, APS, and Jeff
  Hill LANL)

5
6
EPICS 3.14 OSI for the DUV-FEL project

• Radiation monitor IOC (needed for operation)
• Tested on the RTEMS and Tornado2.02 platform
• Used epicsTime thread instead of routines from
  drvTS.c
• drvXVme293.c driver for the MVME293 counter
• devWFXvme293.c radiation counts at various
  locations
• devWF30secTotal.c average radiation dose every
  30 seconds at various locations
• devWFHourTotal.c total radiation dose hourly at
  various locations
• devWFWorkdayTotal.c total radiation dose during
  an eight-hour workday at various locations

6
7
EPICS/RTEMS for NSLS beamlines

• Added OSI over the beamline device/driver
  routines from the R3.13.x release and tested
  successfully on the RTEMS platform for the bitIO,
  scaler and OMS58 motor controller
• stdApp (50)
• motorApp
• Collaborating with Mark River, Time Mooney and
  others from APS to install and test the
  up-to-date beamline device/driver routines

7
8
What did I modify in EPICS 3.14?

• I modified the registerRecordDeviceDriver.pl file
  so that I could continue the scheme of dynamic
  loading to load the Record/Device/Driver for
  stdApp and motorApp in the startup script
• Instead of registerRecordDeviceDriver(pdbase) for
  both the stdApp and motorApp, I used
• std_registerRecordDeviceDriver(pdbase)
• motor_registerRecordDeviceDriver(pdbase)

8
9
Excerpt of the RTEMS startup script
cexpModuleLoad("std.obj") cexpModuleLoad("motor.ob
j) vmeUniverseInit() devAvme9440Config(1,0x8000,0
x80) VSCSetup(1, 0x20000000, 200) oms58Setup(1,
8, 0xa000, 190, 7,0.2) dbLoadDatabase("../dbd/std.
dbd") std_registerRecordDeviceDriver(pdbbase) dbLo
adDatabase("../dbd/motor.dbd") motor_registerRecor
dDeviceDriver(pdbbase) dbLoadRecords("../db/Acroma
g_16IO.db","PKate,AB1") dbLoadRecords("../db/Js
caler.db","PKate,Sscaler1,C0") dbLoadTemplate(
"../db/motor.substitutions") dbLoadRecords("../db/
scanParms.db","PKate,Qm10,SCANREC(P)scan1,POS
(Q).VAL,RDBK(Q).RBV") iocInit()
9
10
Next step

• To install and test RTEMS NFS (Till Straumann,
  SSRL) so that we can use EPICS save_restore (Tim
  Mooney, APS)
• To find a different VME/CPU board to run RTEMS in
  the event that Motorola discontinues the MVME230x
  board

10
11
RTEMS-specific issues that I encountered

• Most of the RTEMS-specific issues are
  transparent over the EPICS OSI. The
  RTEMS-specific issues I dealt with were very
  minor. For example
• Use strtok instead of strtok_r for strict ANSI C
• I/O Ops. in_8(), in_be16(), ....out_be32() for
  PPC
• The NTP client for RTEMS reads only the GMT time.
  An easy workaround is to change the value of
  rtems_bsdnet_timeoffset in the osdTime.cpp file
  to offset the local time.

11
12
Summary

• While RTEMS is lacking in source level debugging
  tools, the existing EPICS OSI/RTEMS/CEXP packages
  are sufficient for us to port the IOC
  applications to RTEMS.
• As of today, RTEMS meets the requirements of our
  real-time IOC applications. The challenge to its
  wider use lies in its limited range of BSPs.

12