Ferris
Retired Dev
Joined: 13 Jan 2003
Posts: 426
Location: N. Virginia (USA)
|
 Posted: Tue Aug 26, 2003 6:56 pm Post subject: semctl(semid,offset,SETVAL,ctl) doesn't on sparc-64? |
 |
|
Does anyone know the status of the "semctl" call on sparc-64? Or, in
somewhat longer form:
The LAM-MPI (message passing for multiprocessors) package likes to
use shared memory & semaphores in some of its modes of operations. On
my Sparc20-SMP, this works fine, but on U2, U60, the atomic setvalue
calls "semctl(id,offest,SETVAL,..)" don't set values.
I am including a little test case extracted from the LAM-MPI code, and
I'd appreciate it if someone with a similar system could check this so I
can tell if this is a local problem or something more general in the library.
First, here're what the configurations look like:
- SS20-SMP, kernel=2.4.21-sparc-r1 #1 SMP,
gcc=sys-devel/gcc-3.2.3-r1, glibc= sys-libs/glibc-2.3.1-r4
- U2-SMP, kernel=2.4.21-sparc-r1 #1 SMP,
gcc=sys-devel/gcc-3.2.3-r1, glibc=sys-libs/glibc-2.3.1-r4
- U60-SMP --- Do you see a pattern here?
Second, you can use the ipcs command (ipcs -s -i <semaphore-id>) to
verify what the program is doing.
Here's the test case:
| Code: |
/*
* sysv-semaphore test.
* This program should allocate three semaphores,
* set their values to something,
* read those values back and print them,
* destroy the semaphores,
* and exits.
*
* On my SS20-2.4.21-SMP, it does just that.
*
* On Ultra2, Ultra60-2.4.21-SMP, for me it sets no value
* into the semaphore.
*
* HOWEVER, the SETALL call =does= set in values!
* */
#include <stdio.h>
#include <sys/sem.h>
/*
* Documentation says to define this because headers don't
* */
static union {
int val;
struct semid_ds *buf;
unsigned short *array;
} semctl_arg;
int
main(int argc, char ** argv) {
int semid;
char * line;
int v, i;
int stat;
short all_at_once[] = {32, 23, 1};
line = (char*) malloc(256);
/*
* Grab a semaphore
* */
semid = semget(IPC_PRIVATE, 3, 0777 | IPC_CREAT);
if(semid < 0) {
fprintf(stderr,"OOPS! Can't get semaphore! <%d>\n", semid);
return(semid);
}
/* See "man semctl" -- we are setting the semaphore values one
* at a time to 1, 32, 23 respectively */
printf("Play with id=%d (%x)\n", semid, semid);
semctl_arg.val = 1;
stat = semctl(semid, 0, SETVAL, semctl_arg );
printf("<set 0 to val=%d\n", semctl_arg.val);
semctl_arg.val = 23;
stat |= semctl(semid, 2, SETVAL, semctl_arg );
printf("<set 2 to val=%d\n", semctl_arg.val);
semctl_arg.val = 32;
stat |= semctl(semid, 1, SETVAL, semctl_arg );
printf("<set 1 to val=%d\n", semctl_arg.val);
printf("semid=%d has values set, ORed stat=%d; CR to go on\n", semid, stat);
gets(line);
for(i=0; i < 3; ++i) {
v = semctl(semid, i,GETVAL);
printf("%d(%d) has value = %d\n", semid,i, v);
}
printf("Those are what %d has; CR again to batch (SETALL)\n", semid);
gets(line);
semctl_arg.array = all_at_once;
stat = semctl(semid,0, SETALL, semctl_arg);
printf("For set-all (32,23,1), stat=%d; CR to read them\n", stat);
gets(line);
for(i=0; i < 3; ++i) {
v = semctl(semid, i,GETVAL);
printf("%d(%d) has value = %d\n", semid,i, v);
}
printf("That's values again from %d; CR to end\n", semid);
semctl(semid,0,IPC_RMID);
return 0;
}
|
What should happen looks like this on the SS20:
Play with id=5144577 (4e8001)
<set 0 to val=1
<set 2 to val=23
<set 1 to val=32
semid=5144577 has values set, ORed stat=0; CR to go on
5144577(0) has value = 1
5144577(1) has value = 32
5144577(2) has value = 23
Those are what 5144577 has; CR again to batch (SETALL)
For set-all (32,23,1), stat=0; CR to read them
5144577(0) has value = 32
5144577(1) has value = 23
5144577(2) has value = 1
That's values again from 5144577; CR to end
What happens (and is wrong) on the U2 & U60 is this:
fmccor@antaresia:DASSF/DaSSF-LAM/app [454]% sem
Play with id=917504 (e0000)
<set 0 to val=1
<set 2 to val=23
<set 1 to val=32
semid=917504 has values set, ORed stat=0; CR to go on
917504(0) has value = 0
917504(1) has value = 0
917504(2) has value = 0
Those are what 917504 has; CR again to batch (SETALL)
For set-all (32,23,1), stat=0; CR to read them
917504(0) has value = 32
917504(1) has value = 23
917504(2) has value = 1
That's values again from 917504; CR to end
(Assume here that the program was compiled as "gcc -o sem sem.c"
but optimisation levels don't matter in my tests at least.)
For the impatient, here is a much smaller test:
| Code: | #include <sys/sem.h>
main() {
int s, v=43;
s=semget(IPC_PRIVATE,1,0777|IPC_CREAT);
semctl(s,0,SETVAL,&v);
printf("val=%d\n",semctl(s,0,GETVAL));
return(0);
}
|
Sorry for the very long post, but I don't see where the problem is, yet. 
Regards, |
|
Gentoo on Sparc
