Experimental Physics and
Industrial Control System

"Mark Rivers" <[email protected]> · Fri, 27 Apr 2007 14:37:13 -0500

Hi Eric,

Thanks for helping with this, sorry I misunderstood your question.

The answer is that YES it does fix the problem to put the
epicsThreadSleep(.001) in the start_routine function in osdThread.c.

Without that call to epicsThreadSleep I added print statements in
epicsThreadCreate in that same file so I could see the status returned
by pthread_create().

epicsThreadId epicsThreadCreate(const char *name,
    unsigned int priority, unsigned int stackSize,
    EPICSTHREADFUNC funptr,void *parm)
{
    epicsThreadOSD *pthreadInfo;
    int status;
    sigset_t blockAllSig, oldSig;

    epicsThreadInit();
    assert(pcommonAttr);
    sigfillset(&blockAllSig);
    pthread_sigmask(SIG_SETMASK,&blockAllSig,&oldSig);
    pthreadInfo = init_threadInfo(name,priority,stackSize,funptr,parm);
    if(pthreadInfo==0) return 0;
    pthreadInfo->isEpicsThread = 1;
    setSchedulingPolicy(pthreadInfo,SCHED_FIFO);
    pthreadInfo->isFifoScheduled = 1;
printf("Calling pthread_create\n");
    status = pthread_create(&pthreadInfo->tid,&pthreadInfo->attr,
                start_routine,pthreadInfo);
printf("pthread_create status=%d\n", status);
    if(status==EPERM){
        /* Try again without SCHED_FIFO*/
        free_threadInfo(pthreadInfo);
        pthreadInfo =
init_threadInfo(name,priority,stackSize,funptr,parm);
        if(pthreadInfo==0) return 0;
        pthreadInfo->isEpicsThread = 1;
printf("Calling pthread_create without SCHED_INFO\n");
        status = pthread_create(&pthreadInfo->tid,&pthreadInfo->attr,
                start_routine,pthreadInfo);
printf("pthread_create without SCHED_INFO status=%d\n", status);
    }

Here is the output:

epics> threadTest 1
Calling pthread_create
pthread_create status=0
Sleeping 5 seconds
epics>
epics>
epics> threadTest 0
Calling pthread_create
pthread_create status=11
pthread_create error Resource temporarily unavailable
Segmentation fault (core dumped)

So pthread_create is returning status=11 (EAGAIN) when things mess up.
Any idea what this means and how to fix it?  And why is the error
causing a seg fault, since it did simply return an error.

Mark

rivers@Subaru /corvette/home/epics/devel/threadTest/iocBoot/iocexample
$
> -----Original Message-----
> From: Eric Norum [mailto:[email protected]] 
> Sent: Friday, April 27, 2007 2:10 PM
> To: Mark Rivers
> Cc: [email protected]
> Subject: Re: EPICS thread problem on cygwin-x86
> 
> 
> On Apr 27, 2007, at 1:58 PM, Mark Rivers wrote:
> 
> > Eric,
> >
> > 1) Does putting the epicsThreadSleep(.001); into the thread wrapper
> > function that actually invokes the EPICS thread fix things?
> >
> > No.  I have simplified the program even more, eliminating the
> > epicsEvent.  I just spawn a thread that either returns 
> immediately or
> > sleeps for .001 second.  The wrapper function now sleeps 
> for 5 seconds
> > after spawning the thread.  It still behaves the same, i.e. if the
> > spawned thread sleeps for .001 second all is well, and the calling
> > function sleeps for 5 seconds and prints a message. If it does not  
> > sleep
> > then it crashes.
> 
> I think that I wasn't clear enough about where I meant for the delay  
> to go.
> Here's the code in /src/libCom/osi/os/posix/osdThread.c which 
> I meant  
> when I referred to the 'wrapper' routine:
> 
> static void * start_routine(void *arg)
> {
>      epicsThreadOSD *pthreadInfo = (epicsThreadOSD *)arg;
>      int status;
>      int oldtype;
>      sigset_t blockAllSig;
> 
>      sigfillset(&blockAllSig);
>      pthread_sigmask(SIG_SETMASK,&blockAllSig,NULL);
>      status = pthread_setspecific(getpthreadInfo,arg);
>      checkStatusQuit(status,"pthread_setspecific","start_routine");
>      status = pthread_setcanceltype 
> (PTHREAD_CANCEL_ASYNCHRONOUS,&oldtype);
>      checkStatusQuit(status,"pthread_setcanceltype","start_routine");
>      status = mutexLock(&listLock);
>      checkStatusQuit(status,"pthread_mutex_lock","start_routine");
>      ellAdd(&pthreadList,&pthreadInfo->node);
>      pthreadInfo->isOnThreadList = 1;
>      status = pthread_mutex_unlock(&listLock);
>      checkStatusQuit(status,"pthread_mutex_unlock","start_routine");
> 
>      (*pthreadInfo->createFunc)(pthreadInfo->createArg);
> 
>      epicsExitCallAtThreadExits ();
> 
>      free_threadInfo(pthreadInfo);
>      return(0);
> }
> 
> Does putting the delay just before the free_threadInfo call 
> fix things?
> 
> >
> > 2) Have you tried running this under gdb to see where the crash  
> > occurs?
> >
> > I have found that gdb under Cygwin does not run the threaded EPICS
> > applications at all.  Maybe I'm doing something wrong, but what  
> > works on
> > Linux does not work on Cygwin.  Furthermore, Cygwin does 
> not produce a
> > useful core file.  It produces a file called, in this case
> > "threadTest.exe.stackdump", but it simply contains:
> >
> > $ cat threadTest.exe.stackdump
> > Exception: STATUS_ACCESS_VIOLATION at eip=004A05B6
> > eax=DF0DF046 ebx=004E6F60 ecx=6112CB58 edx=005A2DC0 esi=005A2DC0
> > edi=610E30A0
> > ebp=0022EDD8 esp=0022EDD4
> > program=j:\epics\devel\threadTest\bin\cygwin-x86\threadTest.exe, pid
> > 4028, thread main
> > cs=001B ds=0023 es=0023 fs=003B gs=0000 ss=0023
> > Stack trace:
> > Frame     Function  Args
> > 0022EDD8  004A05B6  (004E6F60, 005A2DC0, 0022EE08, 005A2DC8)
> > 0022EDF8  004918A7  (610E30A0, 004D94F4, 004D9747, 6110A4F0)
> > 0022EE28  0049220E  (004C9438, 0000000A, 00020000, 00401520)
> > 0022EE48  00401593  (00543BD8, 0022EEE8, 00000000, 005407A4)
> > 0022EEF8  00418070  (61157780, FFFFFFFF, 00000000, 004015F3)
> > 0022EF18  0040161E  (00000002, 61157668, 00540090, 77FA88F0)
> > 0022EFD8  61004DD2  (0022EFF0, 00000000, 0022F221, 33323430)
> > 0022FF88  6100594F  (00000000, 00000000, 00000000, 00000000)
> > End of stack trace
> >
> Hmmmmm...
> not too useful.
> 
> -- 
> Eric Norum <[email protected]>
> Advanced Photon Source
> Argonne National Laboratory
> (630) 252-4793
> 
> 
> 

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System