promotion_lat.c

/* $Id: promotion_lat.c,v 1.8 2020/06/19 14:57:10 kostik Exp kostik $ */

#include <sys/param.h>
#include <sys/mman.h>
#include <machine/cpufunc.h>

#include <err.h>
#include <fcntl.h>
#include <pthread.h>
#include <pthread_np.h>
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>

int
main(int argc, char *argv[])
{
	char *m, *mincore_status;
	uint64_t *lats;
	cpuset_t orig, pin;
	size_t idx, page_sizes[10], shmsz, step;
	int error, n_sps, opt, shmfd, sp_count;
	bool all;

	shmsz = 1024 * 1024 * 1024;
	all = false;
	n_sps = getpagesizes(page_sizes, nitems(page_sizes));
	if (n_sps < 2) {
		fprintf(stderr, "superpages are not supported\n");
		exit(1);
	}

	while ((opt = getopt(argc, argv, "as:")) != -1) {
		switch (opt) {
		case 'a':
			all = true;
			break;
		case 's':
			shmsz = atoll(optarg);
			break;
		default:
			fprintf(stderr, "Unknown option '%c'\n", opt);
			exit(3);
		}
	}
	if (argc != optind) {
		fprintf(stderr, "unknown args\n");
		exit(3);
	}

	if (shmsz % page_sizes[1] != 0) {
		fprintf(stderr, "shmsz must be multiple of %zd\n", page_sizes[1]);
		exit(3);
	}

	shmfd = shm_open(SHM_ANON, O_RDWR | O_CREAT, 0666);
	if (shmfd == -1)
		err(1, "shm_open");
	error = ftruncate(shmfd, shmsz);
	if (error == -1)
		err(1, "ftruncate");

	mincore_status = malloc(shmsz / page_sizes[0]);
	if (mincore_status == NULL)
		errx(1, "malloc mincore_status");
	lats = malloc((shmsz / page_sizes[0]) * sizeof(*lats));
	if (lats == NULL)
		errx(1, "malloc lats");

	m = mmap(NULL, shmsz, PROT_READ | PROT_WRITE,
	    MAP_SHARED | MAP_ALIGNED_SUPER, shmfd, 0);
	if (m == MAP_FAILED)
		err(1, "mmap");

	error = pthread_getaffinity_np(pthread_self(), sizeof(orig), &orig);
	if (error != 0)
		errc(1, error, "getaffinity");
	CPU_ZERO(&pin);
	CPU_SET(0, &pin);
	error = pthread_setaffinity_np(pthread_self(), sizeof(pin), &pin);
	if (error != 0)
		errc(1, error, "setaffinity pin");

	for (idx = 0; idx < shmsz; idx += page_sizes[0]) {
		lats[idx / page_sizes[0]] = -rdtscp();
		m[idx] = 0;
		lats[idx / page_sizes[0]] += rdtscp();
	}

	error = pthread_setaffinity_np(pthread_self(), sizeof(orig), &orig);
	if (error != 0)
		errc(1, error, "setaffinity orig");

	error = mincore(m, shmsz, mincore_status);
	if (error == -1)
		err(1, "mincore");
	for (idx = 0, sp_count = 0; idx < shmsz; idx += page_sizes[1]) {
		if ((mincore_status[idx / page_sizes[0]] & MINCORE_SUPER) != 0)
			sp_count++;
	}

	error = munmap(m, shmsz);
	if (error == -1)
		err(1, "munmap");

	fprintf(stderr, "promoted %d superpages\n", sp_count);
	if (all) {
		for (idx = 0; idx < shmsz / page_sizes[0]; idx++) {
			printf("%jx %jx %c\n", idx, (uintmax_t)lats[idx],
			    (mincore_status[idx] & MINCORE_SUPER) != 0 ?
			    's' : 'n');
		}
	} else {
		step = page_sizes[1] / page_sizes[0];
		for (idx = step - 1; idx < shmsz / page_sizes[0]; idx += step) {
			if ((mincore_status[idx] & MINCORE_SUPER) != 0)
				printf("%jx %jx\n", idx,
				    (uintmax_t)lats[idx]);
		}
	}

	free(mincore_status);
	free(lats);
}