fischer@utower.UUCP (Axel Fischer) (05/26/89)
In article <30067@conexch.UUCP> sandy@conexch.UUCP (Sandford Zelkovitz) writes: >Below, is my ported version of lload.c for Xenix. The original code, >by Lars Pensj from Myab Gothenburg, Sweden, was distributed in >alt.sources. Hmmm, I don't think that this one works right. First of all on my Xenix system you must exclude the line #include <stropts.h> I don't have it and I don't need it. Furtheron poll.h is in sys/poll.h, that means you must change the line: #include <poll.h> to #include <sys/poll.h> What for a Xenix are you using ? Ok, after compiling I have started it. And what has happend ? The three numbers should show me the load average but I've waited several minutes and the numbers have gone from at the beginning: Verbose (1 3) 0.066114 0.013311 0.004442 to (1 3) 3.493572 1.354233 0.514 And I was the only user online. Have I done something wrong ? And what is the meaning of the two numbers in brackets ? (1 3) Maybe the Program runs only with SCO Xenix 286 well ? Please clarify my questions. -Axel -- Domain: fischer%utower@tmpmbx.UUCP / fischer%utower%tmpmbx@db@tui6.BITNET USA: ...!uunet!pyramid!tmpmbx!utower!fischer => Beam me up, Scotty - there is no intelligent life down here. <= => No, it's life Jim - but not as we know it! <=
sandy@conexch.UUCP (Sandford Zelkovitz) (05/27/89)
Below, is my ported version of lload.c for Xenix. The original code,
by Lars Pensj from Myab Gothenburg, Sweden, was distributed in
alt.sources.
--------------------------- cut for lload.c ----------------------------
/*
* Load Average deamon.
*
* The load average is updated every constant time interval, and the result
* written to a file as 3 double values.
*
* The load average is for the last 1, 5 and 15 minutes (3 values).
*
* A second argument (-v) will set the program in a verbose mode, writing
* the load average to the standard output and not to the file.
*
* Preferably start this process from the inittab. It needs special
* priviledges to read from /dev/kmem.
*
* The following processes are regarded as "runnning":
* A process that has the SRUN status (simply running).
* A process that is being created (SIDLE).
* A process that is being swapped out (SXBRK).
* A process that waits for disk I/O.
*
* A process is regarded as waiting for disk I/O if it is SSLEEP and
* has wchan set to a buf in the buffer pool.
*
* The sleep is implemented using poll on a stream device, not the
* more usual sleep() call. Why ?
* Because you do not want to wake up simultaneosly with other programs
* doing sleep(), which might give wrong load average.
* Of course, if you do not have the stream pipe device, use the normal
* sleep().
*/
/* For Xenix: compile as follows: cc -Ox lload.c -lm -lx -o lload */
/* Modified for Xenix by Sanford Zelkovitz XBBS 714-898-8634 */
#include <fcntl.h>
#ifdef M_XENIX
#include <sys/a.out.h>
#else
#include <nlist.h>
#endif
#include <stdio.h>
#include <stropts.h>
#include <poll.h>
#include <math.h>
#include <sys/types.h>
#include <sys/param.h>
#include <sys/buf.h>
#ifdef M_XENIX
#include <sys/page.h>
#else
#include <sys/immu.h>
#include <sys/region.h>
#endif
#include <sys/var.h>
#include <sys/proc.h>
/* #define DEBUG */ /* Will append all values in a debug file */
#define LOADAV "/etc/loadav" /* Where to write the load avarge */
#define STREAM_PIPE "/dev/spx" /* Used for polling with timeout */
/*
* You may or may not need the '_' in the following names.
*/
#define VAR_NAME "_v"
#ifdef M_I386
#define BUF_NAME "_pbuf"
#else
#define BUF_NAME "_buf"
#endif
#define PROC_NAME "_proc"
struct nlist nl[] = {
{VAR_NAME},
{BUF_NAME},
{PROC_NAME},
{0},
};
int loadfile; /* file descr to result file */
struct proc *p;
struct var v;
int kmem;
struct nlist *v_p, *proc_p, *buf_p;
int size;
int first_buf, last_buf;
int sleeping = 1; /* Poll frequency in seconds */
int verbose = 0;
double av[3] = { 0.0, 0.0, 0.0 }; /* The loadaverage */
double apa[3]; /* Holding constants */
main(argc, argv)
char **argv;
{
int i, n, n_run, n_disk;
#ifdef DEBUG
int debug_fd;
char buff[100];
debug_fd = open("/tmp/loadavdebug", O_CREAT|O_WRONLY);
#endif
if (argc == 2 && strcmp(argv[1], "-v") == 0) {
verbose = 1;
printf("Verbose\n");
}
kmem = open("/dev/kmem", 0);
if (kmem == -1) {
perror("/dev/kmem");
exit(1);
}
if (!verbose) {
loadfile = open(LOADAV, 1|O_CREAT,0664);
if (loadfile == -1) {
fprintf(stderr, "%s:", argv[0]);
perror(LOADAV);
exit(1);
}
}
#ifdef M_XENIX
if (nlist("/xenix", nl) == -1) {
#else
if (nlist("/unix", nl) == -1) {
#endif
perror("nlist");
exit(1);
}
for (i=0; nl[i].n_name; i++) {
#ifdef M_XENIX
if(nl[i].n_name[0] == '\0')
break;
#endif
#ifdef DEBUG
fprintf(stderr, "nl[%d] = %s\n", i, nl[i].n_name);
#endif
if (nl[i].n_value == 0) {
fprintf(stderr, "Could not get address for %s\n", nl[i].n_name);
exit(1);
}
if (strcmp(nl[i].n_name, VAR_NAME) == 0)
v_p = &nl[i];
if (strcmp(nl[i].n_name, PROC_NAME) == 0)
proc_p = &nl[i];
if (strcmp(nl[i].n_name, BUF_NAME) == 0)
buf_p = &nl[i];
}
/*
* Setup the constants used for computing load average.
*/
apa[0] = exp(-sleeping/60.0);
apa[1] = exp(-sleeping/300.0);
apa[2] = exp(-sleeping/900.0);
/*
* Start looping
*/
while(1) {
/*
* Read the 'v' structure every time. It says how
* many procs are used.
*/
if (lseek(kmem, v_p->n_value, 0) == -1) {
perror("lseek v");
exit(1);
}
if (read(kmem, &v, sizeof v) == -1) {
perror("read v");
exit(1);
}
size = (struct proc *)v.ve_proc - (struct proc *)proc_p->n_value;
first_buf = buf_p->n_value;
last_buf = first_buf + v.v_buf * sizeof (struct buf);
if (lseek(kmem, proc_p->n_value, 0) == -1) {
perror("lseek proc");
exit(1);
}
p = (struct proc *)malloc(size * sizeof (struct proc));
if (p == 0) {
fprintf(stderr, "Could not malloc %d bytes\n",
size * sizeof (struct proc));
exit(1);
}
n = read(kmem, p, size * sizeof (struct proc));
if (n != size * sizeof (struct proc)) {
if (n == -1) {
perror("read procs");
exit(1);
}
fprintf(stderr, "Could only read %d (%d) procs\n",
n, size);
size = n / sizeof (struct proc);
}
n_run = 0;
n_disk = 0;
for (i=0; i<size; i++) {
if (p[i].p_stat == SRUN || p[i].p_stat == SIDL ||
p[i].p_stat == SXBRK)
n_run++;
else if (p[i].p_stat == SSLEEP &&
(unsigned int)p[i].p_wchan >= first_buf &&
(unsigned int)p[i].p_wchan < last_buf)
n_disk++;
}
/*
* Update the load average using a decay filter.
*/
for (i = 0; i < 3; i++)
av[i] = apa[i] * av[i] + (n_run + n_disk) * (1.0 - apa[i]);
if (!verbose) {
if (lseek(loadfile, 0L, 0) == -1) {
fprintf(stderr, "Couldn't seek in %s\n", LOADAV);
exit(1);
}
if (write(loadfile, (char *)av, sizeof av) != sizeof av) {
perror(argv[0]);
exit(1);
}
} else
printf("(%d %d) %f %f %f\n", n_run, n_disk,
av[0], av[1], av[2]);
#ifdef DEBUG
sprintf(buff, "(%d %d) %4.2f\n", n_run, n_disk,
av[0]);
write(debug_fd, buff, strlen(buff));
#endif
(void)nap(sleeping * 1000);
free(p);
}
}
#ifndef M_XENIX
/*
* Use a stream pipe to implement a sleep.
* We have a stream pipe for ourselves, so we know noone will write
* on it.
*/
nap(milli) {
static int fd = 0;
static struct pollfd pollfd;
if (fd == 0) {
fd = open(STREAM_PIPE, 0);
if (fd == -1) {
perror(STREAM_PIPE);
exit(1);
}
pollfd.fd = fd;
pollfd.events = POLLIN;
}
if (poll(&pollfd, 1, milli) == -1) {
perror("nap: poll");
exit(1);
}
if (pollfd.revents != 0) {
fprintf(stderr, "nap: poll: got something\n");
exit(1);
}
}
#endifccea3@rivm.UUCP (Adri Verhoef) (05/30/89)
I tried to compile the Load Average Daemon on UNIQ (UNIX) System V.3.0, but at first it wouldn't let me. I removed two includes: /* #include <sys/immu.h> NON-EXISTENT */ /* #include <sys/region.h> SYNTAX ERROR (pte_t) */ Then I compiled: "cc lload.c -lm -o lload". I ran "lload -v". It stopped after ten seconds, saying Could not get address for _buf I compiled with -DM_I386, although the machine is a microVAX, as to use _pbuf instead of _buf, and I got lseek v: Invalid argument The belonging call is: lseek(kmem, v_p->n_value, 0) Printing v_p->n_value yields -2147211432. Any ideas? Thanks.
br@laura.UUCP (Bodo Rueskamp) (05/30/89)
In article <30067@conexch.UUCP> sandy@conexch.UUCP (Sandford Zelkovitz) writes: >Below, is my ported version of lload.c for Xenix. The original code, >by Lars Pensj from Myab Gothenburg, Sweden, was distributed in >alt.sources. Diffs for Xenix 286 [ sizeof(int) != sizeof(long) !! ]: *** lload.c Tue May 30 12:34:15 1989 --- lload.c.orig Tue May 30 12:34:20 1989 *************** *** 33,40 /* For Xenix: compile as follows: cc -Ox lload.c -lm -lx -o lload */ - /* For Xenix 286: compile as follows: - cc -M2e -i -Ox lload.c -lm -lx -o lload */ /* Modified for Xenix by Sanford Zelkovitz XBBS 714-898-8634 */ /* Modified for Xenix 286 by Bodo Rueskamp, <br@unido.uucp> */ --- 33,38 ----- /* For Xenix: compile as follows: cc -Ox lload.c -lm -lx -o lload */ /* Modified for Xenix by Sanford Zelkovitz XBBS 714-898-8634 */ *************** *** 37,43 cc -M2e -i -Ox lload.c -lm -lx -o lload */ /* Modified for Xenix by Sanford Zelkovitz XBBS 714-898-8634 */ - /* Modified for Xenix 286 by Bodo Rueskamp, <br@unido.uucp> */ #include <fcntl.h> #ifdef M_XENIX --- 35,40 ----- /* For Xenix: compile as follows: cc -Ox lload.c -lm -lx -o lload */ /* Modified for Xenix by Sanford Zelkovitz XBBS 714-898-8634 */ #include <fcntl.h> #ifdef M_XENIX *************** *** 46,52 #include <nlist.h> #endif #include <stdio.h> - #ifndef M_XENIX #include <stropts.h> #include <poll.h> #endif --- 43,48 ----- #include <nlist.h> #endif #include <stdio.h> #include <stropts.h> #include <poll.h> #include <math.h> *************** *** 49,55 #ifndef M_XENIX #include <stropts.h> #include <poll.h> - #endif #include <math.h> #include <sys/types.h> #include <sys/param.h> --- 45,50 ----- #include <stdio.h> #include <stropts.h> #include <poll.h> #include <math.h> #include <sys/types.h> #include <sys/param.h> *************** *** 169,175 * Read the 'v' structure every time. It says how * many procs are used. */ ! if (lseek(kmem, (long) v_p->n_value, 0) == -1) { perror("lseek v"); exit(1); } --- 164,170 ----- * Read the 'v' structure every time. It says how * many procs are used. */ ! if (lseek(kmem, v_p->n_value, 0) == -1) { perror("lseek v"); exit(1); } *************** *** 180,186 size = (struct proc *)v.ve_proc - (struct proc *)proc_p->n_value; first_buf = buf_p->n_value; last_buf = first_buf + v.v_buf * sizeof (struct buf); ! if (lseek(kmem, (long) proc_p->n_value, 0) == -1) { perror("lseek proc"); exit(1); } --- 175,181 ----- size = (struct proc *)v.ve_proc - (struct proc *)proc_p->n_value; first_buf = buf_p->n_value; last_buf = first_buf + v.v_buf * sizeof (struct buf); ! if (lseek(kmem, proc_p->n_value, 0) == -1) { perror("lseek proc"); exit(1); } *************** *** 233,239 av[0]); write(debug_fd, buff, strlen(buff)); #endif ! (void)nap(sleeping * 1000L); free(p); } } --- 228,234 ----- av[0]); write(debug_fd, buff, strlen(buff)); #endif ! (void)nap(sleeping * 1000); free(p); } }
kevin@iisat.UUCP (Kevin Davies) (05/30/89)
In article <1349@rivm05.UUCP>, ccea3@rivm.UUCP (Adri Verhoef) writes: > I compiled with -DM_I386, although the machine is a microVAX, > as to use _pbuf instead of _buf, and I got > lseek v: Invalid argument > The belonging call is: > lseek(kmem, v_p->n_value, 0) > Printing v_p->n_value yields -2147211432. > > Any ideas? Thanks. There's a couple of lseek's in the program that did this. The problem is that n_value in the nlist structure is defined as unsigned int, but the lseek argument has to be a long (since in this realm int != long). so, lseek(kmem, (long) v_p->n_value,0) this should do the trick... -- Kevin Davies International Information Service (IIS) UUCP: {uunet,utai,watmath}!dalcs!iisat!kevin Bitnet/Uucp: kevin@iisat.uucp Arpanet: kevin%iisat.uucp@uunet.uu.net
sandy@conexch.UUCP (Sandford Zelkovitz) (06/01/89)
In article <96@utower.UUCP>, fischer@utower.UUCP (Axel Fischer) writes: > In article <30067@conexch.UUCP> sandy@conexch.UUCP (Sandford Zelkovitz) writes: > >Below, is my ported version of lload.c for Xenix. The original code, > >by Lars Pensj from Myab Gothenburg, Sweden, was distributed in > >alt.sources. > > Hmmm, I don't think that this one works right. > First of all on my Xenix system you must exclude the line > #include <stropts.h> > I don't have it and I don't need it. > Furtheron poll.h is in sys/poll.h, that means you must change the line: > #include <poll.h> > to > #include <sys/poll.h> > What for a Xenix are you using ? > -Axel > I am running Xenix386 2.3.1 with the 2.3 development system. Yes, I do have /usr/include/poll.h and /usr/include/stropts.h!!!!! /usr/include/poll.h just says to include <sys/poll.h> and /usr/include/stropts.h just says include <sys/stropts.h>. I really don't know why you are missing these files; however, if I were you, I would call SCO ASAP. Sandy Sanford <sandy> Zelkovitz XBBS 714-898-8634 uunet!zardoz!alphacm!sandy att!hermix!alphacm!sandy
csch@netcs.UUCP (Clemens Schrimpe) (06/01/89)
Ha! Thanks to SCO's freaky developers, the load-average stuff is already
included in SCO 2.3.1!
You'll find the label '_avenrun' in /xenix, which is somewhat declared as
unsigned short avenrun[3].
As far, as I have found out (try looking into the 'w' command), they
use a strange floating-point notation for each value. The integer part
is in the higher 8 bits, while the fractional part is in the lower 8 bits.
(That's how 'w' will diplay 'em)
Anyhow - this is how one of our utilities gets the load average:
(simple collection of various subroutines)
#include <sys/a.out.h>
#include <fcntl.h>
print_average ()
{
int fd;
struct nlist nl;
unsigned short loadav[3];
if ((fd = open ("/dev/kmem", O_RDONLY)) == -1)
return;
strcpy(nl.n_name, "_avenrun");
nlist ("/xenix", &nl);
if (nl.n_type == 0 || nl.n_value == 0)
return;
if (lseek (fd, nl.n_value, 0) == -1)
return;
if (read (fd, loadav, sizeof (loadav)) != sizeof (loadav))
return 0;
printf (" %2.2f %2.2f %2.2f\n", (double)loadav[0]/256.,
(double)loadav[1]/256., (double)loadav[2]/256.);
}
Any corrections from SCO welcome - but guys: better correct your 'w' too
then :-)
greeting the whole SCO community on the net,
Clemens Schrimpe, netCS GmbH Informationstechnik Berlin
PS:
By looking more deeply into the kernel's tables you'll also find a fully
working select() and other fancy stuff in Xenix 2.3.1! (thnx 2 sco)
(even though the PTY's have a minor bug, which will prevent the select()
from working with them - but this is a fast fix ...)
--
UUCP: csch@netcs BITNET: csch@db0tui6.BITNET
ARPA/NSF: csch@garp.mit.edu PSI: PSI%26245300033047::CSCH
PHONE: +49-30-24 42 37 FAX: +49-30-24 38 00
BTX: 0303325016-0003 TELEX: 186672 net d