Xref: utzoo comp.lang.c:32048 comp.unix.programmer:78
Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sun-barr!rutgers!mephisto!prism!gt0178a
From: gt0178a@prism.gatech.EDU (BURNS,JIM)
Newsgroups: comp.lang.c,comp.unix.programmer
Subject: Re: shared memory
Message-ID: <13991@hydra.gatech.EDU>
Date: 22 Sep 90 09:18:57 GMT
References: <1990Sep21.183326.13116@DRD.Com>
Followup-To: comp.lang.c
Organization: Georgia Institute of Technology
Lines: 78

in article <1990Sep21.183326.13116@DRD.Com>, mark@DRD.Com (Mark Lawrence) says:

> shmat returns the address.  You can't tell it where you want it.  Then 

Machine dependent.

> /*
>     +-----------------------------------------------------------------
>     | declarations of scope process
> */
> extern int      errno;      /* system call err status */
> extern const char *strerror(int);

I had more compile troubles w/this last line then any other in the program.
I had to substitute:

extern char *sys_errlist[];

and for:

    fprintf(stderr,
         "\n%s: Fatal error from %s, %s\n", who, what, strerror(errno));*/

in CrynDie, I had to substitute:

    fprintf(stderr,
         "\n%s: Fatal error from %s, %s\n", who, what, sys_errlist[errno]);

to get it to compile and load on Dynix, SunOS 4.0.3, A/UX 1.1, & Ultrix
4.0.

>   * The CreateShMem function is to be called only by one process at system
>   * startup.  It will create shared memory for the system.  That process must
>   * then AttachShMem.  All other processes must use the AttachShMem function
>   * only which will get and attach shared memory.
   [...]
>     if ((shmid = shmget(SHMEM_k, sizeof(SHMEM),
>                 IPC_CREAT | SHMEM_PERMISSION)) == -1)
   [...]
>     memset((char *) pshmem, '\0', sizeof(SHMEM));

Just to avoid misunderstandings, these instructions are correct NOT
because of the way he's calling shmget(2), but simply because he's
initializing the shm seg. (Of course, he's also immediately doing a
shmdt(). ) It would only be an error to call shmget(2) w/IPC_CREAT or'd
in w/SHMEM_PERMISSION in the 2nd process if it ALSO had IPC_EXCL or'd in
as well, as the man pages on the above systems state.

I mention this because some systems guarantee that the first shmat() will
zero out the shm seg. From the HP 9000/800 HP-UX Real Time Programmers
Manual:

[example program omitted]

"Note that the fields of the tallytab entries that were not written in
subroutine write_and_read_shm() have values of 0. This is because the
first shmat called (by any process) on that shared memory identifier will
zero out all of the contents of shared memory."

I have not seen this documented in the man pages of the other above
systems, altho' a simple:

main() {
   CreateShMem("main");   /* I commented out the memset in here */
   AttachShMem("main");
   printf("%d %f %d %f\n",Psh->foo,Psh->bar,Psh->gex,Psh->snark);
   printf("%d\n",shmctl(shmid,IPC_RMID,0));
   printf("main exiting\n");
   exit(0);
}

added to the end of the posted subroutines shows all zeroes on A/UX and
Ultrix. Anybody know if this is standard?
-- 
BURNS,JIM
Georgia Institute of Technology, Box 30178, Atlanta Georgia, 30332
uucp:	  ...!{decvax,hplabs,ncar,purdue,rutgers}!gatech!prism!gt0178a
Internet: gt0178a@prism.gatech.edu