Commit | Line | Data |
---|---|---|
27db7830 DB |
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 | #include "../perf.h" | |
12 | #include "../util/util.h" | |
13 | #include "../util/stat.h" | |
14 | #include "../util/parse-options.h" | |
15 | #include "../util/header.h" | |
16 | #include "bench.h" | |
17 | #include "futex.h" | |
18 | ||
19 | #include <err.h> | |
20 | #include <stdlib.h> | |
21 | #include <sys/time.h> | |
22 | #include <pthread.h> | |
23 | ||
24 | /* all threads will block on the same futex */ | |
25 | static u_int32_t futex1 = 0; | |
26 | ||
27 | /* | |
28 | * How many wakeups to do at a time. | |
29 | * Default to 1 in order to make the kernel work more. | |
30 | */ | |
31 | static unsigned int nwakes = 1; | |
32 | ||
33 | /* | |
34 | * There can be significant variance from run to run, | |
35 | * the more repeats, the more exact the overall avg and | |
36 | * the better idea of the futex latency. | |
37 | */ | |
38 | static unsigned int repeat = 10; | |
39 | ||
40 | pthread_t *worker; | |
41 | static bool done = 0, silent = 0; | |
42 | static pthread_mutex_t thread_lock; | |
43 | static pthread_cond_t thread_parent, thread_worker; | |
44 | static struct stats waketime_stats, wakeup_stats; | |
45 | static unsigned int ncpus, threads_starting, nthreads = 0; | |
46 | ||
47 | static const struct option options[] = { | |
48 | OPT_UINTEGER('t', "threads", &nthreads, "Specify amount of threads"), | |
49 | OPT_UINTEGER('w', "nwakes", &nwakes, "Specify amount of threads to wake at once"), | |
50 | OPT_UINTEGER('r', "repeat", &repeat, "Specify amount of times to repeat the run"), | |
51 | OPT_BOOLEAN( 's', "silent", &silent, "Silent mode: do not display data/details"), | |
52 | OPT_END() | |
53 | }; | |
54 | ||
55 | static const char * const bench_futex_wake_usage[] = { | |
56 | "perf bench futex wake <options>", | |
57 | NULL | |
58 | }; | |
59 | ||
60 | static void *workerfn(void *arg __maybe_unused) | |
61 | { | |
62 | pthread_mutex_lock(&thread_lock); | |
63 | threads_starting--; | |
64 | if (!threads_starting) | |
65 | pthread_cond_signal(&thread_parent); | |
66 | pthread_cond_wait(&thread_worker, &thread_lock); | |
67 | pthread_mutex_unlock(&thread_lock); | |
68 | ||
69 | futex_wait(&futex1, 0, NULL, FUTEX_PRIVATE_FLAG); | |
70 | return NULL; | |
71 | } | |
72 | ||
73 | static void print_summary(void) | |
74 | { | |
75 | double waketime_avg = avg_stats(&waketime_stats); | |
76 | double waketime_stddev = stddev_stats(&waketime_stats); | |
77 | unsigned int wakeup_avg = avg_stats(&wakeup_stats); | |
78 | ||
79 | printf("Wokeup %d of %d threads in %.4f ms (+-%.2f%%)\n", | |
80 | wakeup_avg, | |
81 | nthreads, | |
82 | waketime_avg/1e3, | |
83 | rel_stddev_stats(waketime_stddev, waketime_avg)); | |
84 | } | |
85 | ||
86 | static void block_threads(pthread_t *w, | |
87 | pthread_attr_t thread_attr) | |
88 | { | |
89 | cpu_set_t cpu; | |
90 | unsigned int i; | |
91 | ||
92 | threads_starting = nthreads; | |
93 | ||
94 | /* create and block all threads */ | |
95 | for (i = 0; i < nthreads; i++) { | |
96 | CPU_ZERO(&cpu); | |
97 | CPU_SET(i % ncpus, &cpu); | |
98 | ||
99 | if (pthread_attr_setaffinity_np(&thread_attr, sizeof(cpu_set_t), &cpu)) | |
100 | err(EXIT_FAILURE, "pthread_attr_setaffinity_np"); | |
101 | ||
102 | if (pthread_create(&w[i], &thread_attr, workerfn, NULL)) | |
103 | err(EXIT_FAILURE, "pthread_create"); | |
104 | } | |
105 | } | |
106 | ||
107 | static void toggle_done(int sig __maybe_unused, | |
108 | siginfo_t *info __maybe_unused, | |
109 | void *uc __maybe_unused) | |
110 | { | |
111 | done = true; | |
112 | } | |
113 | ||
114 | int bench_futex_wake(int argc, const char **argv, | |
115 | const char *prefix __maybe_unused) | |
116 | { | |
117 | int ret = 0; | |
118 | unsigned int i, j; | |
119 | struct sigaction act; | |
120 | pthread_attr_t thread_attr; | |
121 | ||
122 | argc = parse_options(argc, argv, options, bench_futex_wake_usage, 0); | |
123 | if (argc) { | |
124 | usage_with_options(bench_futex_wake_usage, options); | |
125 | exit(EXIT_FAILURE); | |
126 | } | |
127 | ||
128 | ncpus = sysconf(_SC_NPROCESSORS_ONLN); | |
129 | ||
130 | sigfillset(&act.sa_mask); | |
131 | act.sa_sigaction = toggle_done; | |
132 | sigaction(SIGINT, &act, NULL); | |
133 | ||
134 | if (!nthreads) | |
135 | nthreads = ncpus; | |
136 | ||
137 | worker = calloc(nthreads, sizeof(*worker)); | |
138 | if (!worker) | |
139 | err(EXIT_FAILURE, "calloc"); | |
140 | ||
141 | printf("Run summary [PID %d]: blocking on %d threads (at futex %p), " | |
142 | "waking up %d at a time.\n\n", | |
143 | getpid(), nthreads, &futex1, nwakes); | |
144 | ||
145 | init_stats(&wakeup_stats); | |
146 | init_stats(&waketime_stats); | |
147 | pthread_attr_init(&thread_attr); | |
148 | pthread_mutex_init(&thread_lock, NULL); | |
149 | pthread_cond_init(&thread_parent, NULL); | |
150 | pthread_cond_init(&thread_worker, NULL); | |
151 | ||
152 | for (j = 0; j < repeat && !done; j++) { | |
153 | unsigned int nwoken = 0; | |
154 | struct timeval start, end, runtime; | |
155 | ||
156 | /* create, launch & block all threads */ | |
157 | block_threads(worker, thread_attr); | |
158 | ||
159 | /* make sure all threads are already blocked */ | |
160 | pthread_mutex_lock(&thread_lock); | |
161 | while (threads_starting) | |
162 | pthread_cond_wait(&thread_parent, &thread_lock); | |
163 | pthread_cond_broadcast(&thread_worker); | |
164 | pthread_mutex_unlock(&thread_lock); | |
165 | ||
166 | usleep(100000); | |
167 | ||
168 | /* Ok, all threads are patiently blocked, start waking folks up */ | |
169 | gettimeofday(&start, NULL); | |
170 | while (nwoken != nthreads) | |
171 | nwoken += futex_wake(&futex1, nwakes, FUTEX_PRIVATE_FLAG); | |
172 | gettimeofday(&end, NULL); | |
173 | timersub(&end, &start, &runtime); | |
174 | ||
175 | update_stats(&wakeup_stats, nwoken); | |
176 | update_stats(&waketime_stats, runtime.tv_usec); | |
177 | ||
178 | if (!silent) { | |
179 | printf("[Run %d]: Wokeup %d of %d threads in %.4f ms\n", | |
180 | j + 1, nwoken, nthreads, runtime.tv_usec/1e3); | |
181 | } | |
182 | ||
183 | for (i = 0; i < nthreads; i++) { | |
184 | ret = pthread_join(worker[i], NULL); | |
185 | if (ret) | |
186 | err(EXIT_FAILURE, "pthread_join"); | |
187 | } | |
188 | ||
189 | } | |
190 | ||
191 | /* cleanup & report results */ | |
192 | pthread_cond_destroy(&thread_parent); | |
193 | pthread_cond_destroy(&thread_worker); | |
194 | pthread_mutex_destroy(&thread_lock); | |
195 | pthread_attr_destroy(&thread_attr); | |
196 | ||
197 | print_summary(); | |
198 | ||
199 | free(worker); | |
200 | return ret; | |
201 | } |