linux/samples/bpf/map_perf_test_user.c
Alexei Starovoitov 95ff141e52 samples/bpf: add map_lookup microbenchmark
$ map_perf_test 128
speed of HASH bpf_map_lookup_elem() in lookups per second
	w/o JIT		w/JIT
before	46M		58M
after	42M		74M

perf report
before:
    54.23%  map_perf_test  [kernel.kallsyms]  [k] __htab_map_lookup_elem
    14.24%  map_perf_test  [kernel.kallsyms]  [k] lookup_elem_raw
     8.84%  map_perf_test  [kernel.kallsyms]  [k] htab_map_lookup_elem
     5.93%  map_perf_test  [kernel.kallsyms]  [k] bpf_map_lookup_elem
     2.30%  map_perf_test  [kernel.kallsyms]  [k] bpf_prog_da4fc6a3f41761a2
     1.49%  map_perf_test  [kernel.kallsyms]  [k] kprobe_ftrace_handler

after:
    60.03%  map_perf_test  [kernel.kallsyms]  [k] __htab_map_lookup_elem
    18.07%  map_perf_test  [kernel.kallsyms]  [k] lookup_elem_raw
     2.91%  map_perf_test  [kernel.kallsyms]  [k] bpf_prog_da4fc6a3f41761a2
     1.94%  map_perf_test  [kernel.kallsyms]  [k] _einittext
     1.90%  map_perf_test  [kernel.kallsyms]  [k] __audit_syscall_exit
     1.72%  map_perf_test  [kernel.kallsyms]  [k] kprobe_ftrace_handler

Notice that bpf_map_lookup_elem() and htab_map_lookup_elem() are trivial
functions, yet they take sizeable amount of cpu time.
htab_map_gen_lookup() removes bpf_map_lookup_elem() and converts
htab_map_lookup_elem() into three BPF insns which causing cpu time
for bpf_prog_da4fc6a3f41761a2() slightly increase.

$ map_perf_test 256
speed of ARRAY bpf_map_lookup_elem() in lookups per second
	w/o JIT		w/JIT
before	97M		174M
after	64M		280M

before:
    37.33%  map_perf_test  [kernel.kallsyms]  [k] array_map_lookup_elem
    13.95%  map_perf_test  [kernel.kallsyms]  [k] bpf_map_lookup_elem
     6.54%  map_perf_test  [kernel.kallsyms]  [k] bpf_prog_da4fc6a3f41761a2
     4.57%  map_perf_test  [kernel.kallsyms]  [k] kprobe_ftrace_handler

after:
    32.86%  map_perf_test  [kernel.kallsyms]  [k] bpf_prog_da4fc6a3f41761a2
     6.54%  map_perf_test  [kernel.kallsyms]  [k] kprobe_ftrace_handler

array_map_gen_lookup() removes calls to array_map_lookup_elem()
and bpf_map_lookup_elem() and replaces them with 7 bpf insns.

The performance without JIT is slower, since executing extra insns
in the interpreter is slower than running native C code,
but with JIT the performance gains are obvious,
since native C->x86 code is replaced with fewer bpf->x86 instructions.

Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
2017-03-16 20:44:12 -07:00

270 lines
5.7 KiB
C

/* Copyright (c) 2016 Facebook
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of version 2 of the GNU General Public
* License as published by the Free Software Foundation.
*/
#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 <linux/bpf.h>
#include <string.h>
#include <time.h>
#include <sys/resource.h>
#include "libbpf.h"
#include "bpf_load.h"
#define MAX_CNT 1000000
static __u64 time_get_ns(void)
{
struct timespec ts;
clock_gettime(CLOCK_MONOTONIC, &ts);
return ts.tv_sec * 1000000000ull + ts.tv_nsec;
}
#define HASH_PREALLOC (1 << 0)
#define PERCPU_HASH_PREALLOC (1 << 1)
#define HASH_KMALLOC (1 << 2)
#define PERCPU_HASH_KMALLOC (1 << 3)
#define LRU_HASH_PREALLOC (1 << 4)
#define PERCPU_LRU_HASH_PREALLOC (1 << 5)
#define LPM_KMALLOC (1 << 6)
#define HASH_LOOKUP (1 << 7)
#define ARRAY_LOOKUP (1 << 8)
static int test_flags = ~0;
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 void test_lru_hash_prealloc(int cpu)
{
__u64 start_time;
int i;
start_time = time_get_ns();
for (i = 0; i < MAX_CNT; i++)
syscall(__NR_getpid);
printf("%d:lru_hash_map_perf pre-alloc %lld events per sec\n",
cpu, MAX_CNT * 1000000000ll / (time_get_ns() - start_time));
}
static void test_percpu_lru_hash_prealloc(int cpu)
{
__u64 start_time;
int i;
start_time = time_get_ns();
for (i = 0; i < MAX_CNT; i++)
syscall(__NR_getppid);
printf("%d:lru_hash_map_perf pre-alloc %lld events per sec\n",
cpu, MAX_CNT * 1000000000ll / (time_get_ns() - start_time));
}
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_getpgrp, 0);
printf("%d:array_lookup %lld lookups per sec\n",
cpu, MAX_CNT * 1000000000ll * 64 / (time_get_ns() - start_time));
}
static void loop(int cpu)
{
cpu_set_t cpuset;
CPU_ZERO(&cpuset);
CPU_SET(cpu, &cpuset);
sched_setaffinity(0, sizeof(cpuset), &cpuset);
if (test_flags & HASH_PREALLOC)
test_hash_prealloc(cpu);
if (test_flags & PERCPU_HASH_PREALLOC)
test_percpu_hash_prealloc(cpu);
if (test_flags & HASH_KMALLOC)
test_hash_kmalloc(cpu);
if (test_flags & PERCPU_HASH_KMALLOC)
test_percpu_hash_kmalloc(cpu);
if (test_flags & LRU_HASH_PREALLOC)
test_lru_hash_prealloc(cpu);
if (test_flags & PERCPU_LRU_HASH_PREALLOC)
test_percpu_lru_hash_prealloc(cpu);
if (test_flags & LPM_KMALLOC)
test_lpm_kmalloc(cpu);
if (test_flags & HASH_LOOKUP)
test_hash_lookup(cpu);
if (test_flags & ARRAY_LOOKUP)
test_array_lookup(cpu);
}
static void run_perf_test(int tasks)
{
pid_t pid[tasks];
int i;
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[6], 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[6], key, &value, 0);
assert(!r);
}
int main(int argc, char **argv)
{
struct rlimit r = {RLIM_INFINITY, RLIM_INFINITY};
char filename[256];
int num_cpu = 8;
snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
setrlimit(RLIMIT_MEMLOCK, &r);
if (argc > 1)
test_flags = atoi(argv[1]) ? : test_flags;
if (argc > 2)
num_cpu = atoi(argv[2]) ? : num_cpu;
if (load_bpf_file(filename)) {
printf("%s", bpf_log_buf);
return 1;
}
fill_lpm_trie();
run_perf_test(num_cpu);
return 0;
}