[linux-block.git] / samples / bpf / map_perf_test_user.c

// SPDX-License-Identifier: GPL-2.0-only
/* Copyright (c) 2016 Facebook
 */
#define _GNU_SOURCE
#include <sched.h>
#include <stdio.h>
#include <sys/types.h>
#include <asm/unistd.h>
#include <unistd.h>
#include <assert.h>
#include <sys/wait.h>
#include <stdlib.h>
#include <signal.h>
#include <string.h>
#include <time.h>
#include <arpa/inet.h>
#include <errno.h>

#include <bpf/bpf.h>
#include <bpf/libbpf.h>

#define TEST_BIT(t) (1U << (t))
#define MAX_NR_CPUS 1024

static __u64 time_get_ns(void)
{
	struct timespec ts;

	clock_gettime(CLOCK_MONOTONIC, &ts);
	return ts.tv_sec * 1000000000ull + ts.tv_nsec;
}

enum test_type {
	HASH_PREALLOC,
	PERCPU_HASH_PREALLOC,
	HASH_KMALLOC,
	PERCPU_HASH_KMALLOC,
	LRU_HASH_PREALLOC,
	NOCOMMON_LRU_HASH_PREALLOC,
	LPM_KMALLOC,
	HASH_LOOKUP,
	ARRAY_LOOKUP,
	INNER_LRU_HASH_PREALLOC,
	LRU_HASH_LOOKUP,
	NR_TESTS,
};

const char *test_map_names[NR_TESTS] = {
	[HASH_PREALLOC] = "hash_map",
	[PERCPU_HASH_PREALLOC] = "percpu_hash_map",
	[HASH_KMALLOC] = "hash_map_alloc",
	[PERCPU_HASH_KMALLOC] = "percpu_hash_map_alloc",
	[LRU_HASH_PREALLOC] = "lru_hash_map",
	[NOCOMMON_LRU_HASH_PREALLOC] = "nocommon_lru_hash_map",
	[LPM_KMALLOC] = "lpm_trie_map_alloc",
	[HASH_LOOKUP] = "hash_map",
	[ARRAY_LOOKUP] = "array_map",
	[INNER_LRU_HASH_PREALLOC] = "inner_lru_hash_map",
	[LRU_HASH_LOOKUP] = "lru_hash_lookup_map",
};

enum map_idx {
	array_of_lru_hashs_idx,
	hash_map_alloc_idx,
	lru_hash_lookup_idx,
	NR_IDXES,
};

static int map_fd[NR_IDXES];

static int test_flags = ~0;
static uint32_t num_map_entries;
static uint32_t inner_lru_hash_size;
static int lru_hash_lookup_test_entries = 32;
static uint32_t max_cnt = 10000;

static int check_test_flags(enum test_type t)
{
	return test_flags & TEST_BIT(t);
}

static void test_hash_prealloc(int cpu)
{
	__u64 start_time;
	int i;

	start_time = time_get_ns();
	for (i = 0; i < max_cnt; i++)
		syscall(__NR_getuid);
	printf("%d:hash_map_perf pre-alloc %lld events per sec\n",
	       cpu, max_cnt * 1000000000ll / (time_get_ns() - start_time));
}

static int pre_test_lru_hash_lookup(int tasks)
{
	int fd = map_fd[lru_hash_lookup_idx];
	uint32_t key;
	long val = 1;
	int ret;

	if (num_map_entries > lru_hash_lookup_test_entries)
		lru_hash_lookup_test_entries = num_map_entries;

	/* Populate the lru_hash_map for LRU_HASH_LOOKUP perf test.
	 *
	 * It is fine that the user requests for a map with
	 * num_map_entries < 32 and some of the later lru hash lookup
	 * may return not found.  For LRU map, we are not interested
	 * in such small map performance.
	 */
	for (key = 0; key < lru_hash_lookup_test_entries; key++) {
		ret = bpf_map_update_elem(fd, &key, &val, BPF_NOEXIST);
		if (ret)
			return ret;
	}

	return 0;
}

static void do_test_lru(enum test_type test, int cpu)
{
	static int inner_lru_map_fds[MAX_NR_CPUS];

	struct sockaddr_in6 in6 = { .sin6_family = AF_INET6 };
	const char *test_name;
	__u64 start_time;
	int i, ret;

	if (test == INNER_LRU_HASH_PREALLOC && cpu) {
		/* If CPU is not 0, create inner_lru hash map and insert the fd
		 * value into the array_of_lru_hash map. In case of CPU 0,
		 * 'inner_lru_hash_map' was statically inserted on the map init
		 */
		int outer_fd = map_fd[array_of_lru_hashs_idx];
		unsigned int mycpu, mynode;
		LIBBPF_OPTS(bpf_map_create_opts, opts,
			.map_flags = BPF_F_NUMA_NODE,
		);

		assert(cpu < MAX_NR_CPUS);

		ret = syscall(__NR_getcpu, &mycpu, &mynode, NULL);
		assert(!ret);

		opts.numa_node = mynode;
		inner_lru_map_fds[cpu] =
			bpf_map_create(BPF_MAP_TYPE_LRU_HASH,
				       test_map_names[INNER_LRU_HASH_PREALLOC],
				       sizeof(uint32_t),
				       sizeof(long),
				       inner_lru_hash_size, &opts);
		if (inner_lru_map_fds[cpu] == -1) {
			printf("cannot create BPF_MAP_TYPE_LRU_HASH %s(%d)\n",
			       strerror(errno), errno);
			exit(1);
		}

		ret = bpf_map_update_elem(outer_fd, &cpu,
					  &inner_lru_map_fds[cpu],
					  BPF_ANY);
		if (ret) {
			printf("cannot update ARRAY_OF_LRU_HASHS with key:%u. %s(%d)\n",
			       cpu, strerror(errno), errno);
			exit(1);
		}
	}

	in6.sin6_addr.s6_addr16[0] = 0xdead;
	in6.sin6_addr.s6_addr16[1] = 0xbeef;

	if (test == LRU_HASH_PREALLOC) {
		test_name = "lru_hash_map_perf";
		in6.sin6_addr.s6_addr16[2] = 0;
	} else if (test == NOCOMMON_LRU_HASH_PREALLOC) {
		test_name = "nocommon_lru_hash_map_perf";
		in6.sin6_addr.s6_addr16[2] = 1;
	} else if (test == INNER_LRU_HASH_PREALLOC) {
		test_name = "inner_lru_hash_map_perf";
		in6.sin6_addr.s6_addr16[2] = 2;
	} else if (test == LRU_HASH_LOOKUP) {
		test_name = "lru_hash_lookup_perf";
		in6.sin6_addr.s6_addr16[2] = 3;
		in6.sin6_addr.s6_addr32[3] = 0;
	} else {
		assert(0);
	}

	start_time = time_get_ns();
	for (i = 0; i < max_cnt; i++) {
		ret = connect(-1, (const struct sockaddr *)&in6, sizeof(in6));
		assert(ret == -1 && errno == EBADF);
		if (in6.sin6_addr.s6_addr32[3] <
		    lru_hash_lookup_test_entries - 32)
			in6.sin6_addr.s6_addr32[3] += 32;
		else
			in6.sin6_addr.s6_addr32[3] = 0;
	}
	printf("%d:%s pre-alloc %lld events per sec\n",
	       cpu, test_name,
	       max_cnt * 1000000000ll / (time_get_ns() - start_time));
}

static void test_lru_hash_prealloc(int cpu)
{
	do_test_lru(LRU_HASH_PREALLOC, cpu);
}

static void test_nocommon_lru_hash_prealloc(int cpu)
{
	do_test_lru(NOCOMMON_LRU_HASH_PREALLOC, cpu);
}

static void test_inner_lru_hash_prealloc(int cpu)
{
	do_test_lru(INNER_LRU_HASH_PREALLOC, cpu);
}

static void test_lru_hash_lookup(int cpu)
{
	do_test_lru(LRU_HASH_LOOKUP, cpu);
}

static void test_percpu_hash_prealloc(int cpu)
{
	__u64 start_time;
	int i;

	start_time = time_get_ns();
	for (i = 0; i < max_cnt; i++)
		syscall(__NR_geteuid);
	printf("%d:percpu_hash_map_perf pre-alloc %lld events per sec\n",
	       cpu, max_cnt * 1000000000ll / (time_get_ns() - start_time));
}

static void test_hash_kmalloc(int cpu)
{
	__u64 start_time;
	int i;

	start_time = time_get_ns();
	for (i = 0; i < max_cnt; i++)
		syscall(__NR_getgid);
	printf("%d:hash_map_perf kmalloc %lld events per sec\n",
	       cpu, max_cnt * 1000000000ll / (time_get_ns() - start_time));
}

static void test_percpu_hash_kmalloc(int cpu)
{
	__u64 start_time;
	int i;

	start_time = time_get_ns();
	for (i = 0; i < max_cnt; i++)
		syscall(__NR_getegid);
	printf("%d:percpu_hash_map_perf kmalloc %lld events per sec\n",
	       cpu, max_cnt * 1000000000ll / (time_get_ns() - start_time));
}

static void test_lpm_kmalloc(int cpu)
{
	__u64 start_time;
	int i;

	start_time = time_get_ns();
	for (i = 0; i < max_cnt; i++)
		syscall(__NR_gettid);
	printf("%d:lpm_perf kmalloc %lld events per sec\n",
	       cpu, max_cnt * 1000000000ll / (time_get_ns() - start_time));
}

static void test_hash_lookup(int cpu)
{
	__u64 start_time;
	int i;

	start_time = time_get_ns();
	for (i = 0; i < max_cnt; i++)
		syscall(__NR_getpgid, 0);
	printf("%d:hash_lookup %lld lookups per sec\n",
	       cpu, max_cnt * 1000000000ll * 64 / (time_get_ns() - start_time));
}

static void test_array_lookup(int cpu)
{
	__u64 start_time;
	int i;

	start_time = time_get_ns();
	for (i = 0; i < max_cnt; i++)
		syscall(__NR_getppid, 0);
	printf("%d:array_lookup %lld lookups per sec\n",
	       cpu, max_cnt * 1000000000ll * 64 / (time_get_ns() - start_time));
}

typedef int (*pre_test_func)(int tasks);
const pre_test_func pre_test_funcs[] = {
	[LRU_HASH_LOOKUP] = pre_test_lru_hash_lookup,
};

typedef void (*test_func)(int cpu);
const test_func test_funcs[] = {
	[HASH_PREALLOC] = test_hash_prealloc,
	[PERCPU_HASH_PREALLOC] = test_percpu_hash_prealloc,
	[HASH_KMALLOC] = test_hash_kmalloc,
	[PERCPU_HASH_KMALLOC] = test_percpu_hash_kmalloc,
	[LRU_HASH_PREALLOC] = test_lru_hash_prealloc,
	[NOCOMMON_LRU_HASH_PREALLOC] = test_nocommon_lru_hash_prealloc,
	[LPM_KMALLOC] = test_lpm_kmalloc,
	[HASH_LOOKUP] = test_hash_lookup,
	[ARRAY_LOOKUP] = test_array_lookup,
	[INNER_LRU_HASH_PREALLOC] = test_inner_lru_hash_prealloc,
	[LRU_HASH_LOOKUP] = test_lru_hash_lookup,
};

static int pre_test(int tasks)
{
	int i;

	for (i = 0; i < NR_TESTS; i++) {
		if (pre_test_funcs[i] && check_test_flags(i)) {
			int ret = pre_test_funcs[i](tasks);

			if (ret)
				return ret;
		}
	}

	return 0;
}

static void loop(int cpu)
{
	cpu_set_t cpuset;
	int i;

	CPU_ZERO(&cpuset);
	CPU_SET(cpu, &cpuset);
	sched_setaffinity(0, sizeof(cpuset), &cpuset);

	for (i = 0; i < NR_TESTS; i++) {
		if (check_test_flags(i))
			test_funcs[i](cpu);
	}
}

static void run_perf_test(int tasks)
{
	pid_t pid[tasks];
	int i;

	assert(!pre_test(tasks));

	for (i = 0; i < tasks; i++) {
		pid[i] = fork();
		if (pid[i] == 0) {
			loop(i);
			exit(0);
		} else if (pid[i] == -1) {
			printf("couldn't spawn #%d process\n", i);
			exit(1);
		}
	}
	for (i = 0; i < tasks; i++) {
		int status;

		assert(waitpid(pid[i], &status, 0) == pid[i]);
		assert(status == 0);
	}
}

static void fill_lpm_trie(void)
{
	struct bpf_lpm_trie_key *key;
	unsigned long value = 0;
	unsigned int i;
	int r;

	key = alloca(sizeof(*key) + 4);
	key->prefixlen = 32;

	for (i = 0; i < 512; ++i) {
		key->prefixlen = rand() % 33;
		key->data[0] = rand() & 0xff;
		key->data[1] = rand() & 0xff;
		key->data[2] = rand() & 0xff;
		key->data[3] = rand() & 0xff;
		r = bpf_map_update_elem(map_fd[hash_map_alloc_idx],
					key, &value, 0);
		assert(!r);
	}

	key->prefixlen = 32;
	key->data[0] = 192;
	key->data[1] = 168;
	key->data[2] = 0;
	key->data[3] = 1;
	value = 128;

	r = bpf_map_update_elem(map_fd[hash_map_alloc_idx], key, &value, 0);
	assert(!r);
}

static void fixup_map(struct bpf_object *obj)
{
	struct bpf_map *map;
	int i;

	bpf_object__for_each_map(map, obj) {
		const char *name = bpf_map__name(map);

		/* Only change the max_entries for the enabled test(s) */
		for (i = 0; i < NR_TESTS; i++) {
			if (!strcmp(test_map_names[i], name) &&
			    (check_test_flags(i))) {
				bpf_map__set_max_entries(map, num_map_entries);
				continue;
			}
		}
	}

	inner_lru_hash_size = num_map_entries;
}

int main(int argc, char **argv)
{
	int nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
	struct bpf_link *links[8];
	struct bpf_program *prog;
	struct bpf_object *obj;
	struct bpf_map *map;
	char filename[256];
	int i = 0;

	if (argc > 1)
		test_flags = atoi(argv[1]) ? : test_flags;

	if (argc > 2)
		nr_cpus = atoi(argv[2]) ? : nr_cpus;

	if (argc > 3)
		num_map_entries = atoi(argv[3]);

	if (argc > 4)
		max_cnt = atoi(argv[4]);

	snprintf(filename, sizeof(filename), "%s.bpf.o", argv[0]);
	obj = bpf_object__open_file(filename, NULL);
	if (libbpf_get_error(obj)) {
		fprintf(stderr, "ERROR: opening BPF object file failed\n");
		return 0;
	}

	map = bpf_object__find_map_by_name(obj, "inner_lru_hash_map");
	if (libbpf_get_error(map)) {
		fprintf(stderr, "ERROR: finding a map in obj file failed\n");
		goto cleanup;
	}

	inner_lru_hash_size = bpf_map__max_entries(map);
	if (!inner_lru_hash_size) {
		fprintf(stderr, "ERROR: failed to get map attribute\n");
		goto cleanup;
	}

	/* resize BPF map prior to loading */
	if (num_map_entries > 0)
		fixup_map(obj);

	/* load BPF program */
	if (bpf_object__load(obj)) {
		fprintf(stderr, "ERROR: loading BPF object file failed\n");
		goto cleanup;
	}

	map_fd[0] = bpf_object__find_map_fd_by_name(obj, "array_of_lru_hashs");
	map_fd[1] = bpf_object__find_map_fd_by_name(obj, "hash_map_alloc");
	map_fd[2] = bpf_object__find_map_fd_by_name(obj, "lru_hash_lookup_map");
	if (map_fd[0] < 0 || map_fd[1] < 0 || map_fd[2] < 0) {
		fprintf(stderr, "ERROR: finding a map in obj file failed\n");
		goto cleanup;
	}

	bpf_object__for_each_program(prog, obj) {
		links[i] = bpf_program__attach(prog);
		if (libbpf_get_error(links[i])) {
			fprintf(stderr, "ERROR: bpf_program__attach failed\n");
			links[i] = NULL;
			goto cleanup;
		}
		i++;
	}

	fill_lpm_trie();

	run_perf_test(nr_cpus);

cleanup:
	for (i--; i >= 0; i--)
		bpf_link__destroy(links[i]);

	bpf_object__close(obj);
	return 0;
}
Commit	Line	Data
25763b3c	1	// SPDX-License-Identifier: GPL-2.0-only
26e90931	2	/* Copyright (c) 2016 Facebook
26e90931 AS	3	*/
	4	#define _GNU_SOURCE
	5	#include <sched.h>
	6	#include <stdio.h>
	7	#include <sys/types.h>
	8	#include <asm/unistd.h>
	9	#include <unistd.h>
	10	#include <assert.h>
	11	#include <sys/wait.h>
	12	#include <stdlib.h>
	13	#include <signal.h>
26e90931 AS	14	#include <string.h>
26e90931 AS	15	#include <time.h>
bf8db5d2 MKL	16	#include <arpa/inet.h>
	17	#include <errno.h>
	18
2bf3e2ef	19	#include <bpf/bpf.h>
cc7f641d	20	#include <bpf/libbpf.h>
26e90931	21
9fd63d05	22	#define TEST_BIT(t) (1U << (t))
3a5795b8	23	#define MAX_NR_CPUS 1024
26e90931 AS	24
	25	static __u64 time_get_ns(void)
	26	{
	27	struct timespec ts;
	28
	29	clock_gettime(CLOCK_MONOTONIC, &ts);
	30	return ts.tv_sec * 1000000000ull + ts.tv_nsec;
	31	}
	32
9fd63d05 MKL	33	enum test_type {
	34	HASH_PREALLOC,
	35	PERCPU_HASH_PREALLOC,
	36	HASH_KMALLOC,
	37	PERCPU_HASH_KMALLOC,
	38	LRU_HASH_PREALLOC,
	39	NOCOMMON_LRU_HASH_PREALLOC,
	40	LPM_KMALLOC,
	41	HASH_LOOKUP,
	42	ARRAY_LOOKUP,
3a5795b8	43	INNER_LRU_HASH_PREALLOC,
637cd8c3	44	LRU_HASH_LOOKUP,
9fd63d05 MKL	45	NR_TESTS,
	46	};
	47
	48	const char *test_map_names[NR_TESTS] = {
	49	[HASH_PREALLOC] = "hash_map",
	50	[PERCPU_HASH_PREALLOC] = "percpu_hash_map",
	51	[HASH_KMALLOC] = "hash_map_alloc",
	52	[PERCPU_HASH_KMALLOC] = "percpu_hash_map_alloc",
	53	[LRU_HASH_PREALLOC] = "lru_hash_map",
	54	[NOCOMMON_LRU_HASH_PREALLOC] = "nocommon_lru_hash_map",
	55	[LPM_KMALLOC] = "lpm_trie_map_alloc",
	56	[HASH_LOOKUP] = "hash_map",
	57	[ARRAY_LOOKUP] = "array_map",
3a5795b8	58	[INNER_LRU_HASH_PREALLOC] = "inner_lru_hash_map",
637cd8c3	59	[LRU_HASH_LOOKUP] = "lru_hash_lookup_map",
9fd63d05	60	};
26e90931	61
cc7f641d DL	62	enum map_idx {
	63	array_of_lru_hashs_idx,
	64	hash_map_alloc_idx,
	65	lru_hash_lookup_idx,
	66	NR_IDXES,
	67	};
	68
	69	static int map_fd[NR_IDXES];
	70
26e90931	71	static int test_flags = ~0;
9fd63d05	72	static uint32_t num_map_entries;
3a5795b8	73	static uint32_t inner_lru_hash_size;
637cd8c3	74	static int lru_hash_lookup_test_entries = 32;
89dc8d0c	75	static uint32_t max_cnt = 10000;
9fd63d05 MKL	76
	77	static int check_test_flags(enum test_type t)
	78	{
	79	return test_flags & TEST_BIT(t);
	80	}
26e90931 AS	81
	82	static void test_hash_prealloc(int cpu)
	83	{
	84	__u64 start_time;
	85	int i;
	86
	87	start_time = time_get_ns();
9fd63d05	88	for (i = 0; i < max_cnt; i++)
26e90931 AS	89	syscall(__NR_getuid);
26e90931 AS	90	printf("%d:hash_map_perf pre-alloc %lld events per sec\n",
9fd63d05	91	cpu, max_cnt * 1000000000ll / (time_get_ns() - start_time));
26e90931 AS	92	}
26e90931 AS	93
637cd8c3 MKL	94	static int pre_test_lru_hash_lookup(int tasks)
	95	{
	96	int fd = map_fd[lru_hash_lookup_idx];
	97	uint32_t key;
	98	long val = 1;
	99	int ret;
	100
	101	if (num_map_entries > lru_hash_lookup_test_entries)
	102	lru_hash_lookup_test_entries = num_map_entries;
	103
	104	/* Populate the lru_hash_map for LRU_HASH_LOOKUP perf test.
	105	*
	106	* It is fine that the user requests for a map with
	107	* num_map_entries < 32 and some of the later lru hash lookup
	108	* may return not found. For LRU map, we are not interested
	109	* in such small map performance.
	110	*/
	111	for (key = 0; key < lru_hash_lookup_test_entries; key++) {
	112	ret = bpf_map_update_elem(fd, &key, &val, BPF_NOEXIST);
	113	if (ret)
	114	return ret;
	115	}
	116
	117	return 0;
	118	}
	119
9fd63d05	120	static void do_test_lru(enum test_type test, int cpu)
5db58faf	121	{
3a5795b8 MKL	122	static int inner_lru_map_fds[MAX_NR_CPUS];
3a5795b8 MKL	123
bf8db5d2 MKL	124	struct sockaddr_in6 in6 = { .sin6_family = AF_INET6 };
bf8db5d2 MKL	125	const char *test_name;
5db58faf	126	__u64 start_time;
bf8db5d2 MKL	127	int i, ret;
bf8db5d2 MKL	128
cc7f641d DL	129	if (test == INNER_LRU_HASH_PREALLOC && cpu) {
	130	/* If CPU is not 0, create inner_lru hash map and insert the fd
	131	* value into the array_of_lru_hash map. In case of CPU 0,
	132	* 'inner_lru_hash_map' was statically inserted on the map init
	133	*/
3a5795b8	134	int outer_fd = map_fd[array_of_lru_hashs_idx];
ad17d0e6	135	unsigned int mycpu, mynode;
c58f9815 AN	136	LIBBPF_OPTS(bpf_map_create_opts, opts,
	137	.map_flags = BPF_F_NUMA_NODE,
	138	);
3a5795b8 MKL	139
	140	assert(cpu < MAX_NR_CPUS);
	141
cc7f641d DL	142	ret = syscall(__NR_getcpu, &mycpu, &mynode, NULL);
	143	assert(!ret);
	144
c58f9815	145	opts.numa_node = mynode;
cc7f641d	146	inner_lru_map_fds[cpu] =
c58f9815 AN	147	bpf_map_create(BPF_MAP_TYPE_LRU_HASH,
	148	test_map_names[INNER_LRU_HASH_PREALLOC],
	149	sizeof(uint32_t),
	150	sizeof(long),
	151	inner_lru_hash_size, &opts);
cc7f641d DL	152	if (inner_lru_map_fds[cpu] == -1) {
	153	printf("cannot create BPF_MAP_TYPE_LRU_HASH %s(%d)\n",
	154	strerror(errno), errno);
	155	exit(1);
3a5795b8 MKL	156	}
	157
	158	ret = bpf_map_update_elem(outer_fd, &cpu,
	159	&inner_lru_map_fds[cpu],
	160	BPF_ANY);
	161	if (ret) {
	162	printf("cannot update ARRAY_OF_LRU_HASHS with key:%u. %s(%d)\n",
	163	cpu, strerror(errno), errno);
	164	exit(1);
	165	}
	166	}
	167
bf8db5d2 MKL	168	in6.sin6_addr.s6_addr16[0] = 0xdead;
	169	in6.sin6_addr.s6_addr16[1] = 0xbeef;
	170
9fd63d05	171	if (test == LRU_HASH_PREALLOC) {
bf8db5d2	172	test_name = "lru_hash_map_perf";
637cd8c3	173	in6.sin6_addr.s6_addr16[2] = 0;
9fd63d05	174	} else if (test == NOCOMMON_LRU_HASH_PREALLOC) {
bf8db5d2	175	test_name = "nocommon_lru_hash_map_perf";
637cd8c3	176	in6.sin6_addr.s6_addr16[2] = 1;
3a5795b8 MKL	177	} else if (test == INNER_LRU_HASH_PREALLOC) {
3a5795b8 MKL	178	test_name = "inner_lru_hash_map_perf";
637cd8c3 MKL	179	in6.sin6_addr.s6_addr16[2] = 2;
	180	} else if (test == LRU_HASH_LOOKUP) {
	181	test_name = "lru_hash_lookup_perf";
	182	in6.sin6_addr.s6_addr16[2] = 3;
	183	in6.sin6_addr.s6_addr32[3] = 0;
bf8db5d2 MKL	184	} else {
	185	assert(0);
	186	}
5db58faf MKL	187
5db58faf MKL	188	start_time = time_get_ns();
9fd63d05	189	for (i = 0; i < max_cnt; i++) {
bf8db5d2 MKL	190	ret = connect(-1, (const struct sockaddr *)&in6, sizeof(in6));
bf8db5d2 MKL	191	assert(ret == -1 && errno == EBADF);
637cd8c3 MKL	192	if (in6.sin6_addr.s6_addr32[3] <
	193	lru_hash_lookup_test_entries - 32)
	194	in6.sin6_addr.s6_addr32[3] += 32;
	195	else
	196	in6.sin6_addr.s6_addr32[3] = 0;
bf8db5d2 MKL	197	}
	198	printf("%d:%s pre-alloc %lld events per sec\n",
	199	cpu, test_name,
9fd63d05	200	max_cnt * 1000000000ll / (time_get_ns() - start_time));
5db58faf MKL	201	}
5db58faf MKL	202
bf8db5d2	203	static void test_lru_hash_prealloc(int cpu)
5db58faf	204	{
bf8db5d2 MKL	205	do_test_lru(LRU_HASH_PREALLOC, cpu);
bf8db5d2 MKL	206	}
5db58faf	207
bf8db5d2 MKL	208	static void test_nocommon_lru_hash_prealloc(int cpu)
	209	{
	210	do_test_lru(NOCOMMON_LRU_HASH_PREALLOC, cpu);
5db58faf MKL	211	}
5db58faf MKL	212
3a5795b8 MKL	213	static void test_inner_lru_hash_prealloc(int cpu)
	214	{
	215	do_test_lru(INNER_LRU_HASH_PREALLOC, cpu);
	216	}
	217
637cd8c3 MKL	218	static void test_lru_hash_lookup(int cpu)
	219	{
	220	do_test_lru(LRU_HASH_LOOKUP, cpu);
	221	}
	222
26e90931 AS	223	static void test_percpu_hash_prealloc(int cpu)
	224	{
	225	__u64 start_time;
	226	int i;
	227
	228	start_time = time_get_ns();
9fd63d05	229	for (i = 0; i < max_cnt; i++)
26e90931 AS	230	syscall(__NR_geteuid);
26e90931 AS	231	printf("%d:percpu_hash_map_perf pre-alloc %lld events per sec\n",
9fd63d05	232	cpu, max_cnt * 1000000000ll / (time_get_ns() - start_time));
26e90931 AS	233	}
	234
	235	static void test_hash_kmalloc(int cpu)
	236	{
	237	__u64 start_time;
	238	int i;
	239
	240	start_time = time_get_ns();
9fd63d05	241	for (i = 0; i < max_cnt; i++)
26e90931 AS	242	syscall(__NR_getgid);
26e90931 AS	243	printf("%d:hash_map_perf kmalloc %lld events per sec\n",
9fd63d05	244	cpu, max_cnt * 1000000000ll / (time_get_ns() - start_time));
26e90931 AS	245	}
	246
	247	static void test_percpu_hash_kmalloc(int cpu)
	248	{
	249	__u64 start_time;
	250	int i;
	251
	252	start_time = time_get_ns();
9fd63d05	253	for (i = 0; i < max_cnt; i++)
26e90931 AS	254	syscall(__NR_getegid);
26e90931 AS	255	printf("%d:percpu_hash_map_perf kmalloc %lld events per sec\n",
9fd63d05	256	cpu, max_cnt * 1000000000ll / (time_get_ns() - start_time));
26e90931 AS	257	}
26e90931 AS	258
b8a943e2 DH	259	static void test_lpm_kmalloc(int cpu)
	260	{
	261	__u64 start_time;
	262	int i;
	263
	264	start_time = time_get_ns();
9fd63d05	265	for (i = 0; i < max_cnt; i++)
b8a943e2 DH	266	syscall(__NR_gettid);
b8a943e2 DH	267	printf("%d:lpm_perf kmalloc %lld events per sec\n",
9fd63d05	268	cpu, max_cnt * 1000000000ll / (time_get_ns() - start_time));
b8a943e2 DH	269	}
b8a943e2 DH	270
95ff141e AS	271	static void test_hash_lookup(int cpu)
	272	{
	273	__u64 start_time;
	274	int i;
	275
	276	start_time = time_get_ns();
9fd63d05	277	for (i = 0; i < max_cnt; i++)
95ff141e AS	278	syscall(__NR_getpgid, 0);
95ff141e AS	279	printf("%d:hash_lookup %lld lookups per sec\n",
9fd63d05	280	cpu, max_cnt * 1000000000ll * 64 / (time_get_ns() - start_time));
95ff141e AS	281	}
	282
	283	static void test_array_lookup(int cpu)
	284	{
	285	__u64 start_time;
	286	int i;
	287
	288	start_time = time_get_ns();
9fd63d05	289	for (i = 0; i < max_cnt; i++)
95ec6696	290	syscall(__NR_getppid, 0);
95ff141e	291	printf("%d:array_lookup %lld lookups per sec\n",
9fd63d05	292	cpu, max_cnt * 1000000000ll * 64 / (time_get_ns() - start_time));
95ff141e AS	293	}
95ff141e AS	294
637cd8c3 MKL	295	typedef int (*pre_test_func)(int tasks);
	296	const pre_test_func pre_test_funcs[] = {
	297	[LRU_HASH_LOOKUP] = pre_test_lru_hash_lookup,
	298	};
	299
9fd63d05 MKL	300	typedef void (*test_func)(int cpu);
	301	const test_func test_funcs[] = {
	302	[HASH_PREALLOC] = test_hash_prealloc,
	303	[PERCPU_HASH_PREALLOC] = test_percpu_hash_prealloc,
	304	[HASH_KMALLOC] = test_hash_kmalloc,
	305	[PERCPU_HASH_KMALLOC] = test_percpu_hash_kmalloc,
	306	[LRU_HASH_PREALLOC] = test_lru_hash_prealloc,
	307	[NOCOMMON_LRU_HASH_PREALLOC] = test_nocommon_lru_hash_prealloc,
	308	[LPM_KMALLOC] = test_lpm_kmalloc,
	309	[HASH_LOOKUP] = test_hash_lookup,
	310	[ARRAY_LOOKUP] = test_array_lookup,
3a5795b8	311	[INNER_LRU_HASH_PREALLOC] = test_inner_lru_hash_prealloc,
637cd8c3	312	[LRU_HASH_LOOKUP] = test_lru_hash_lookup,
9fd63d05 MKL	313	};
9fd63d05 MKL	314
637cd8c3 MKL	315	static int pre_test(int tasks)
	316	{
	317	int i;
	318
	319	for (i = 0; i < NR_TESTS; i++) {
	320	if (pre_test_funcs[i] && check_test_flags(i)) {
	321	int ret = pre_test_funcs[i](tasks);
	322
	323	if (ret)
	324	return ret;
	325	}
	326	}
	327
	328	return 0;
	329	}
	330
26e90931 AS	331	static void loop(int cpu)
	332	{
	333	cpu_set_t cpuset;
9fd63d05	334	int i;
26e90931 AS	335
	336	CPU_ZERO(&cpuset);
	337	CPU_SET(cpu, &cpuset);
	338	sched_setaffinity(0, sizeof(cpuset), &cpuset);
	339
9fd63d05 MKL	340	for (i = 0; i < NR_TESTS; i++) {
	341	if (check_test_flags(i))
	342	test_funcs[i](cpu);
	343	}
26e90931 AS	344	}
	345
	346	static void run_perf_test(int tasks)
	347	{
	348	pid_t pid[tasks];
	349	int i;
	350
637cd8c3 MKL	351	assert(!pre_test(tasks));
637cd8c3 MKL	352
26e90931 AS	353	for (i = 0; i < tasks; i++) {
	354	pid[i] = fork();
	355	if (pid[i] == 0) {
	356	loop(i);
	357	exit(0);
	358	} else if (pid[i] == -1) {
	359	printf("couldn't spawn #%d process\n", i);
	360	exit(1);
	361	}
	362	}
	363	for (i = 0; i < tasks; i++) {
	364	int status;
	365
	366	assert(waitpid(pid[i], &status, 0) == pid[i]);
	367	assert(status == 0);
	368	}
	369	}
	370
b8a943e2 DH	371	static void fill_lpm_trie(void)
	372	{
	373	struct bpf_lpm_trie_key *key;
	374	unsigned long value = 0;
	375	unsigned int i;
	376	int r;
	377
	378	key = alloca(sizeof(*key) + 4);
	379	key->prefixlen = 32;
	380
	381	for (i = 0; i < 512; ++i) {
	382	key->prefixlen = rand() % 33;
	383	key->data[0] = rand() & 0xff;
	384	key->data[1] = rand() & 0xff;
	385	key->data[2] = rand() & 0xff;
	386	key->data[3] = rand() & 0xff;
cc7f641d DL	387	r = bpf_map_update_elem(map_fd[hash_map_alloc_idx],
cc7f641d DL	388	key, &value, 0);
b8a943e2 DH	389	assert(!r);
	390	}
	391
	392	key->prefixlen = 32;
	393	key->data[0] = 192;
	394	key->data[1] = 168;
	395	key->data[2] = 0;
	396	key->data[3] = 1;
	397	value = 128;
	398
cc7f641d	399	r = bpf_map_update_elem(map_fd[hash_map_alloc_idx], key, &value, 0);
b8a943e2 DH	400	assert(!r);
	401	}
	402
cc7f641d	403	static void fixup_map(struct bpf_object *obj)
9fd63d05	404	{
cc7f641d	405	struct bpf_map *map;
9fd63d05 MKL	406	int i;
9fd63d05 MKL	407
cc7f641d DL	408	bpf_object__for_each_map(map, obj) {
cc7f641d DL	409	const char *name = bpf_map__name(map);
3a5795b8	410
cc7f641d DL	411	/* Only change the max_entries for the enabled test(s) */
	412	for (i = 0; i < NR_TESTS; i++) {
	413	if (!strcmp(test_map_names[i], name) &&
	414	(check_test_flags(i))) {
61afd3da	415	bpf_map__set_max_entries(map, num_map_entries);
cc7f641d DL	416	continue;
cc7f641d DL	417	}
3a5795b8	418	}
3a5795b8 MKL	419	}
3a5795b8 MKL	420
3a5795b8	421	inner_lru_hash_size = num_map_entries;
9fd63d05 MKL	422	}
9fd63d05 MKL	423
26e90931 AS	424	int main(int argc, char **argv)
26e90931 AS	425	{
cc7f641d DL	426	int nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
	427	struct bpf_link *links[8];
	428	struct bpf_program *prog;
	429	struct bpf_object *obj;
	430	struct bpf_map *map;
26e90931	431	char filename[256];
cc7f641d	432	int i = 0;
26e90931	433
26e90931 AS	434	if (argc > 1)
	435	test_flags = atoi(argv[1]) ? : test_flags;
	436
	437	if (argc > 2)
cc7f641d	438	nr_cpus = atoi(argv[2]) ? : nr_cpus;
26e90931	439
9fd63d05 MKL	440	if (argc > 3)
	441	num_map_entries = atoi(argv[3]);
	442
	443	if (argc > 4)
	444	max_cnt = atoi(argv[4]);
	445
d4fffba4	446	snprintf(filename, sizeof(filename), "%s.bpf.o", argv[0]);
cc7f641d DL	447	obj = bpf_object__open_file(filename, NULL);
	448	if (libbpf_get_error(obj)) {
	449	fprintf(stderr, "ERROR: opening BPF object file failed\n");
	450	return 0;
	451	}
	452
	453	map = bpf_object__find_map_by_name(obj, "inner_lru_hash_map");
	454	if (libbpf_get_error(map)) {
	455	fprintf(stderr, "ERROR: finding a map in obj file failed\n");
	456	goto cleanup;
	457	}
	458
	459	inner_lru_hash_size = bpf_map__max_entries(map);
	460	if (!inner_lru_hash_size) {
	461	fprintf(stderr, "ERROR: failed to get map attribute\n");
	462	goto cleanup;
	463	}
	464
	465	/* resize BPF map prior to loading */
	466	if (num_map_entries > 0)
	467	fixup_map(obj);
	468
	469	/* load BPF program */
	470	if (bpf_object__load(obj)) {
	471	fprintf(stderr, "ERROR: loading BPF object file failed\n");
	472	goto cleanup;
	473	}
	474
	475	map_fd[0] = bpf_object__find_map_fd_by_name(obj, "array_of_lru_hashs");
	476	map_fd[1] = bpf_object__find_map_fd_by_name(obj, "hash_map_alloc");
	477	map_fd[2] = bpf_object__find_map_fd_by_name(obj, "lru_hash_lookup_map");
	478	if (map_fd[0] < 0 \|\| map_fd[1] < 0 \|\| map_fd[2] < 0) {
	479	fprintf(stderr, "ERROR: finding a map in obj file failed\n");
	480	goto cleanup;
	481	}
	482
	483	bpf_object__for_each_program(prog, obj) {
	484	links[i] = bpf_program__attach(prog);
	485	if (libbpf_get_error(links[i])) {
	486	fprintf(stderr, "ERROR: bpf_program__attach failed\n");
	487	links[i] = NULL;
	488	goto cleanup;
	489	}
	490	i++;
26e90931 AS	491	}
26e90931 AS	492
b8a943e2 DH	493	fill_lpm_trie();
b8a943e2 DH	494
cc7f641d DL	495	run_perf_test(nr_cpus);
	496
	497	cleanup:
	498	for (i--; i >= 0; i--)
	499	bpf_link__destroy(links[i]);
26e90931	500
cc7f641d	501	bpf_object__close(obj);
26e90931 AS	502	return 0;
26e90931 AS	503	}