/* forkcost.c - fork cost vs touched memory; COW and zero-fill fault costs. C89.
   usage: forkcost MB [reps]   */
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/time.h>
#include <sys/wait.h>
#include <sys/mman.h>

static double now(void) { struct timeval tv; gettimeofday(&tv, 0); return tv.tv_sec + tv.tv_usec / 1e6; }
static int cmp(const void *a, const void *b) { double x = *(const double *)a, y = *(const double *)b; return x < y ? -1 : x > y; }

int main(int argc, char **argv)
{
    long mb = argc > 1 ? atol(argv[1]) : 0, reps = argc > 2 ? atol(argv[2]) : 20;
    long bytes = mb << 20, pg = sysconf(_SC_PAGESIZE), i, npg = bytes / pg;
    char *buf = 0, *mm;
    double t0, t1, times[1000], med;
    pid_t pid; int st;

    if (reps > 1000) reps = 1000;
    if (bytes) {
        buf = malloc(bytes);
        if (!buf) { perror("malloc"); return 1; }
        t0 = now(); for (i = 0; i < npg; i++) buf[i * pg] = 1; t1 = now();
        printf("zero-fill touch  %ld pages  %.1f us/page\n", npg, (t1 - t0) * 1e6 / npg);
        /* second pass: no faults, the loop's own cost */
        t0 = now(); for (i = 0; i < npg; i++) buf[i * pg] = 2; t1 = now();
        printf("retouch          %ld pages  %.2f us/page\n", npg, (t1 - t0) * 1e6 / npg);
        mm = mmap(0, bytes, PROT_READ|PROT_WRITE, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
        if (mm != MAP_FAILED) {
            t0 = now(); for (i = 0; i < npg; i++) mm[i * pg] = 1; t1 = now();
            printf("mmap-anon touch  %ld pages  %.1f us/page\n", npg, (t1 - t0) * 1e6 / npg);
            munmap(mm, bytes);
        } else perror("mmap");
    }
    for (i = 0; i < reps; i++) {
        t0 = now();
        pid = fork();
        if (pid == 0) _exit(0);
        t1 = now();
        waitpid(pid, &st, 0);
        times[i] = t1 - t0;
    }
    qsort(times, reps, sizeof(double), cmp);
    med = times[reps / 2];
    printf("fork             %ld MB touched  min %.0f us  median %.0f us\n", mb, times[0] * 1e6, med * 1e6);
    if (bytes) {
        /* child writes every inherited page: COW faults */
        int fd[2]; double cow;
        pipe(fd);
        pid = fork();
        if (pid == 0) {
            t0 = now(); for (i = 0; i < npg; i++) buf[i * pg] = 3; t1 = now();
            cow = (t1 - t0) * 1e6 / npg; write(fd[1], &cow, sizeof cow); _exit(0);
        }
        read(fd[0], &cow, sizeof cow); waitpid(pid, &st, 0);
        printf("child COW write  %ld pages  %.1f us/page\n", npg, cow);
    }
    return 0;
}