How to make your EPICS driver operating system independent

How to make your EPICS driver operating system independent

W. Eric Norum

Introduction

The following table shows the changes made to some existing drivers to allow them to be used on systems other than vxWorks. Note that it is a very preliminary start at a set of conversion instructions. There's still no `cookbook' document to guide you, but many drivers can be converted without doing much more than applying the translations shown. A technique that I've found works well is to first change all the #include statements that mention vxWorks header file to their OSI equivalents and then enter a compile-edit cycle until all compile errors have been eliminated.

Conversions

vxWorks	OSI
#include <vxWorks.h>
#include <stdlib.h>	#include <epicsStdlib.h>
#include <stdio.h>	#include <epicsStdioRedirect.h>
#include <iosLib.h>
#include <taskLib.h>	#include <epicsThread.h>
#include <semLib.h>	#include <epicsMutex.h>
#include <memLib.h>
#include <rebootLib.h>	#include <epicsExit.h>
#include <intLib.h>	#include <epicsInterrupt.h>
#include <lstLib.h>
#include <vme.h>	#include <devLib.h>
#include <sysLib.h>
#include <iv.h>
#include <epicsExport.h>
#include <iocsh.h>
ERROR	-1
OK	0
SEM_ID LinkLock; if ((LinkLock = semBCreate(SEM_Q_PRIORITY, SEM_FULL)) == NULL) semTake(LinkLock, WAIT_FOREVER); semGive(LinkLock);	epicsMutexId LinkLock; if ((LinkLock = epicsMutexCreate()) == NULL) epicsMutexLock(pLink->LinkLock); epicsMutexUnlock(pLink->LinkLock);
taskDelay(1)	To generate a `short' delay: epicsThreadSleep(0.001) This works because the operating system specific layer ensures a delay of at least one system clock tick for epicsThreadSleep arguments greater than 0.
taskDelay(30)	Much old code assumes a 60 Hz vxWorks system clock. On such systems the equivalent for the command shown would be: epicsThreadSleep(0.5) Of course, lots of other old code assumes a 50 Hz or 100 Hz system clock rate...
taskSuspend(0)	epicsThreadSuspendSelf()
intConnect(INUM_TO_IVEC(IrqVector), ErIrqHandler, pLink)	devConnectInterruptVME(IrqVector, ErIrqHandler, pLink)
sysIntEnable(IrqLevel)	devEnableInterruptLevelVME(IrqLevel)
sysBusToLocalAdrs(VME_AM_SUP_SHORT_IO, (char*)CardAddress, (char**)&ErLink[Card].pEr)	devRegisterAddress("apsEr", atVMEA16, CardAddress, 0x40, (void*)&ErLink[Card].pEr)
vxMemProbe(pReg, READ, sizeof(short), &value)	devReadProbe(sizeof(short), pReg, &value)
vxMemProbe(pReg,WRITE,sizeof(short), &value)	devWriteProbe(sizeof(short), pReg, &value)
key = intLock()	key = epicsInterruptLock()
intUnlock(key)	epicsInterrupUnlock(key)
rebootHookAdd((FUNCPTR)ErRebootFunc)	epicsAtExit(ErRebootFunc, NULL)
#include <fast_lock.h> FAST_LOCK lock; FASTLOCKINIT(&mzconf[card].lock); FASTLOCKFREE(&mzconf[card].lock); FASTLOCK(&mzconf[card].lock); FASTUNLOCK(&mzconf[card].lock);	You'll have to use your judgement here. If the section of code being protected is quick (less than a few microseconds) it's reasonable to simply disable interrupts while the code is active. In this case you can remove the 'lock' variable and the init/free operations and then add a local 'key' variable and replace the lock/unlock operations with epicsInterruptLock/Unlock operations. If the section of code being protected is longer you'll have to convert the FASTLOCK to an EPICS mutex.
taskSpawn(...)	epicsThreadCreate(...)
You should consider rewriting the driver to use the "worker thread" environment provided by the ASYN package. This is likely to result in a shorter, simpler, and more robust driver.

Some other changes

• Some R3.13 record support database definitions explicitly mention all fields rather than including dbCommon.dbd. This can cause problems with FAST_LOCK. The solution is to add include "dbCommon.dbd" at the beginning of the file in question and to remove all entries for the fields provided by dbCommon.dbd.
• You should set up IOC shell command registrations for all `configure' functions and for any other commands you may need to call from the IOC shell. Here's an example of a fairly complex configuration command:

 /*
  * IOC shell command registration
  */
 #include <iocsh.h>
 static const iocshArg vtr10012ConfigArg0 = { "card",iocshArgInt};
 static const iocshArg vtr10012ConfigArg1 = { "VME A16 offset",iocshArgInt};
 static const iocshArg vtr10012ConfigArg2 = { "VME memory offset",iocshArgInt};
 static const iocshArg vtr10012ConfigArg3 = { "interrupt vector",iocshArgInt};
 static const iocshArg vtr10012ConfigArg4 = { "interrupt level",iocshArgInt};
 static const iocshArg vtr10012ConfigArg5 = { "use DMA",iocshArgInt};
 static const iocshArg vtr10012ConfigArg6 = { "nchannels",iocshArgInt};
 static const iocshArg vtr10012ConfigArg7 = { "kilosamplesPerChan",iocshArgInt};
 static const iocshArg *vtr10012ConfigArgs[] = {
     &vtr10012ConfigArg0, &vtr10012ConfigArg1, &vtr10012ConfigArg2,
     &vtr10012ConfigArg3, &vtr10012ConfigArg4, &vtr10012ConfigArg5,
     &vtr10012ConfigArg6,&vtr10012ConfigArg7};
 static const iocshFuncDef vtr10012ConfigFuncDef =
                       {"vtr10012Config",8,vtr10012ConfigArgs};
 static void vtr10012ConfigCallFunc(const iocshArgBuf *args)
 {
     vtr10012Config(args[0].ival, args[1].ival, args[2].ival,
                  args[3].ival, args[4].ival, args[5].ival,
                  args[6].ival, args[7].ival);
 }
 
 /*
  * This routine is called before multitasking has started, so there's
  * no race condition in the test/set of firstTime.
  */
 static void
 drvVtr10012RegisterCommands(void)
 {
     static int firstTime = 1;
     if (firstTime) {
         iocshRegister(&vtr10012ConfigFuncDef,vtr10012ConfigCallFunc);
         firstTime = 0;
     }
 }
 epicsExportRegistrar(drvVtr10012RegisterCommands);

You must also provide a corresponding registrar statement in a .dbd file:

registrar(drvVtr10012RegisterCommands)

• If your driver provides functions that are referred to by name from database files you must:
  • #include <registryFunction.h>
  • Add an epicsExportFuncion() statement for each such function.
  • Add a corresponding function statement in a .dbd file.

Additional Information

The procedure for converting a support module from R3.13 to R3.14 is described in the Converting R3.13 Apps to R3.14 document.

Acknowledgments

• Peter Denison (Diamond) for getting me started on this document and for useful suggestions.
• Steve Shoaf and Nick DiMonte (ANL) for some early conversions.

---

Eric Norum 2006-08-17