xref: /openbmc/linux/tools/perf/bench/futex-wake.c (revision 6aa7de05)
1 /*
2  * Copyright (C) 2013  Davidlohr Bueso <davidlohr@hp.com>
3  *
4  * futex-wake: Block a bunch of threads on a futex and wake'em up, N at a time.
5  *
6  * This program is particularly useful to measure the latency of nthread wakeups
7  * in non-error situations:  all waiters are queued and all wake calls wakeup
8  * one or more tasks, and thus the waitqueue is never empty.
9  */
10 
11 /* For the CLR_() macros */
12 #include <string.h>
13 #include <pthread.h>
14 
15 #include <signal.h>
16 #include "../util/stat.h"
17 #include <subcmd/parse-options.h>
18 #include <linux/compiler.h>
19 #include <linux/kernel.h>
20 #include <linux/time64.h>
21 #include <errno.h>
22 #include "bench.h"
23 #include "futex.h"
24 
25 #include <err.h>
26 #include <stdlib.h>
27 #include <sys/time.h>
28 
29 /* all threads will block on the same futex */
30 static u_int32_t futex1 = 0;
31 
32 /*
33  * How many wakeups to do at a time.
34  * Default to 1 in order to make the kernel work more.
35  */
36 static unsigned int nwakes = 1;
37 
38 pthread_t *worker;
39 static bool done = false, silent = false, fshared = false;
40 static pthread_mutex_t thread_lock;
41 static pthread_cond_t thread_parent, thread_worker;
42 static struct stats waketime_stats, wakeup_stats;
43 static unsigned int ncpus, threads_starting, nthreads = 0;
44 static int futex_flag = 0;
45 
46 static const struct option options[] = {
47 	OPT_UINTEGER('t', "threads", &nthreads, "Specify amount of threads"),
48 	OPT_UINTEGER('w', "nwakes",  &nwakes,   "Specify amount of threads to wake at once"),
49 	OPT_BOOLEAN( 's', "silent",  &silent,   "Silent mode: do not display data/details"),
50 	OPT_BOOLEAN( 'S', "shared",  &fshared,  "Use shared futexes instead of private ones"),
51 	OPT_END()
52 };
53 
54 static const char * const bench_futex_wake_usage[] = {
55 	"perf bench futex wake <options>",
56 	NULL
57 };
58 
59 static void *workerfn(void *arg __maybe_unused)
60 {
61 	pthread_mutex_lock(&thread_lock);
62 	threads_starting--;
63 	if (!threads_starting)
64 		pthread_cond_signal(&thread_parent);
65 	pthread_cond_wait(&thread_worker, &thread_lock);
66 	pthread_mutex_unlock(&thread_lock);
67 
68 	while (1) {
69 		if (futex_wait(&futex1, 0, NULL, futex_flag) != EINTR)
70 			break;
71 	}
72 
73 	pthread_exit(NULL);
74 	return NULL;
75 }
76 
77 static void print_summary(void)
78 {
79 	double waketime_avg = avg_stats(&waketime_stats);
80 	double waketime_stddev = stddev_stats(&waketime_stats);
81 	unsigned int wakeup_avg = avg_stats(&wakeup_stats);
82 
83 	printf("Wokeup %d of %d threads in %.4f ms (+-%.2f%%)\n",
84 	       wakeup_avg,
85 	       nthreads,
86 	       waketime_avg / USEC_PER_MSEC,
87 	       rel_stddev_stats(waketime_stddev, waketime_avg));
88 }
89 
90 static void block_threads(pthread_t *w,
91 			  pthread_attr_t thread_attr)
92 {
93 	cpu_set_t cpu;
94 	unsigned int i;
95 
96 	threads_starting = nthreads;
97 
98 	/* create and block all threads */
99 	for (i = 0; i < nthreads; i++) {
100 		CPU_ZERO(&cpu);
101 		CPU_SET(i % ncpus, &cpu);
102 
103 		if (pthread_attr_setaffinity_np(&thread_attr, sizeof(cpu_set_t), &cpu))
104 			err(EXIT_FAILURE, "pthread_attr_setaffinity_np");
105 
106 		if (pthread_create(&w[i], &thread_attr, workerfn, NULL))
107 			err(EXIT_FAILURE, "pthread_create");
108 	}
109 }
110 
111 static void toggle_done(int sig __maybe_unused,
112 			siginfo_t *info __maybe_unused,
113 			void *uc __maybe_unused)
114 {
115 	done = true;
116 }
117 
118 int bench_futex_wake(int argc, const char **argv)
119 {
120 	int ret = 0;
121 	unsigned int i, j;
122 	struct sigaction act;
123 	pthread_attr_t thread_attr;
124 
125 	argc = parse_options(argc, argv, options, bench_futex_wake_usage, 0);
126 	if (argc) {
127 		usage_with_options(bench_futex_wake_usage, options);
128 		exit(EXIT_FAILURE);
129 	}
130 
131 	ncpus = sysconf(_SC_NPROCESSORS_ONLN);
132 
133 	sigfillset(&act.sa_mask);
134 	act.sa_sigaction = toggle_done;
135 	sigaction(SIGINT, &act, NULL);
136 
137 	if (!nthreads)
138 		nthreads = ncpus;
139 
140 	worker = calloc(nthreads, sizeof(*worker));
141 	if (!worker)
142 		err(EXIT_FAILURE, "calloc");
143 
144 	if (!fshared)
145 		futex_flag = FUTEX_PRIVATE_FLAG;
146 
147 	printf("Run summary [PID %d]: blocking on %d threads (at [%s] futex %p), "
148 	       "waking up %d at a time.\n\n",
149 	       getpid(), nthreads, fshared ? "shared":"private",  &futex1, nwakes);
150 
151 	init_stats(&wakeup_stats);
152 	init_stats(&waketime_stats);
153 	pthread_attr_init(&thread_attr);
154 	pthread_mutex_init(&thread_lock, NULL);
155 	pthread_cond_init(&thread_parent, NULL);
156 	pthread_cond_init(&thread_worker, NULL);
157 
158 	for (j = 0; j < bench_repeat && !done; j++) {
159 		unsigned int nwoken = 0;
160 		struct timeval start, end, runtime;
161 
162 		/* create, launch & block all threads */
163 		block_threads(worker, thread_attr);
164 
165 		/* make sure all threads are already blocked */
166 		pthread_mutex_lock(&thread_lock);
167 		while (threads_starting)
168 			pthread_cond_wait(&thread_parent, &thread_lock);
169 		pthread_cond_broadcast(&thread_worker);
170 		pthread_mutex_unlock(&thread_lock);
171 
172 		usleep(100000);
173 
174 		/* Ok, all threads are patiently blocked, start waking folks up */
175 		gettimeofday(&start, NULL);
176 		while (nwoken != nthreads)
177 			nwoken += futex_wake(&futex1, nwakes, futex_flag);
178 		gettimeofday(&end, NULL);
179 		timersub(&end, &start, &runtime);
180 
181 		update_stats(&wakeup_stats, nwoken);
182 		update_stats(&waketime_stats, runtime.tv_usec);
183 
184 		if (!silent) {
185 			printf("[Run %d]: Wokeup %d of %d threads in %.4f ms\n",
186 			       j + 1, nwoken, nthreads, runtime.tv_usec / (double)USEC_PER_MSEC);
187 		}
188 
189 		for (i = 0; i < nthreads; i++) {
190 			ret = pthread_join(worker[i], NULL);
191 			if (ret)
192 				err(EXIT_FAILURE, "pthread_join");
193 		}
194 
195 	}
196 
197 	/* cleanup & report results */
198 	pthread_cond_destroy(&thread_parent);
199 	pthread_cond_destroy(&thread_worker);
200 	pthread_mutex_destroy(&thread_lock);
201 	pthread_attr_destroy(&thread_attr);
202 
203 	print_summary();
204 
205 	free(worker);
206 	return ret;
207 }
208