1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* Copyright (c) 2020 Facebook */ |
3 | #define _GNU_SOURCE |
4 | #include <argp.h> |
5 | #include <linux/compiler.h> |
6 | #include <sys/time.h> |
7 | #include <sched.h> |
8 | #include <fcntl.h> |
9 | #include <pthread.h> |
10 | #include <sys/sysinfo.h> |
11 | #include <signal.h> |
12 | #include "bench.h" |
13 | #include "testing_helpers.h" |
14 | |
15 | struct env env = { |
16 | .warmup_sec = 1, |
17 | .duration_sec = 5, |
18 | .affinity = false, |
19 | .quiet = false, |
20 | .consumer_cnt = 0, |
21 | .producer_cnt = 1, |
22 | }; |
23 | |
24 | static int libbpf_print_fn(enum libbpf_print_level level, |
25 | const char *format, va_list args) |
26 | { |
27 | if (level == LIBBPF_DEBUG && !env.verbose) |
28 | return 0; |
29 | return vfprintf(stderr, format, args); |
30 | } |
31 | |
32 | void setup_libbpf(void) |
33 | { |
34 | libbpf_set_strict_mode(LIBBPF_STRICT_ALL); |
35 | libbpf_set_print(libbpf_print_fn); |
36 | } |
37 | |
38 | void false_hits_report_progress(int iter, struct bench_res *res, long delta_ns) |
39 | { |
40 | long total = res->false_hits + res->hits + res->drops; |
41 | |
42 | printf("Iter %3d (%7.3lfus): " , |
43 | iter, (delta_ns - 1000000000) / 1000.0); |
44 | |
45 | printf("%ld false hits of %ld total operations. Percentage = %2.2f %%\n" , |
46 | res->false_hits, total, ((float)res->false_hits / total) * 100); |
47 | } |
48 | |
49 | void false_hits_report_final(struct bench_res res[], int res_cnt) |
50 | { |
51 | long total_hits = 0, total_drops = 0, total_false_hits = 0, total_ops = 0; |
52 | int i; |
53 | |
54 | for (i = 0; i < res_cnt; i++) { |
55 | total_hits += res[i].hits; |
56 | total_false_hits += res[i].false_hits; |
57 | total_drops += res[i].drops; |
58 | } |
59 | total_ops = total_hits + total_false_hits + total_drops; |
60 | |
61 | printf("Summary: %ld false hits of %ld total operations. " , |
62 | total_false_hits, total_ops); |
63 | printf("Percentage = %2.2f %%\n" , |
64 | ((float)total_false_hits / total_ops) * 100); |
65 | } |
66 | |
67 | void hits_drops_report_progress(int iter, struct bench_res *res, long delta_ns) |
68 | { |
69 | double hits_per_sec, drops_per_sec; |
70 | double hits_per_prod; |
71 | |
72 | hits_per_sec = res->hits / 1000000.0 / (delta_ns / 1000000000.0); |
73 | hits_per_prod = hits_per_sec / env.producer_cnt; |
74 | drops_per_sec = res->drops / 1000000.0 / (delta_ns / 1000000000.0); |
75 | |
76 | printf("Iter %3d (%7.3lfus): " , |
77 | iter, (delta_ns - 1000000000) / 1000.0); |
78 | |
79 | printf("hits %8.3lfM/s (%7.3lfM/prod), drops %8.3lfM/s, total operations %8.3lfM/s\n" , |
80 | hits_per_sec, hits_per_prod, drops_per_sec, hits_per_sec + drops_per_sec); |
81 | } |
82 | |
83 | void |
84 | grace_period_latency_basic_stats(struct bench_res res[], int res_cnt, struct basic_stats *gp_stat) |
85 | { |
86 | int i; |
87 | |
88 | memset(gp_stat, 0, sizeof(struct basic_stats)); |
89 | |
90 | for (i = 0; i < res_cnt; i++) |
91 | gp_stat->mean += res[i].gp_ns / 1000.0 / (double)res[i].gp_ct / (0.0 + res_cnt); |
92 | |
93 | #define IT_MEAN_DIFF (res[i].gp_ns / 1000.0 / (double)res[i].gp_ct - gp_stat->mean) |
94 | if (res_cnt > 1) { |
95 | for (i = 0; i < res_cnt; i++) |
96 | gp_stat->stddev += (IT_MEAN_DIFF * IT_MEAN_DIFF) / (res_cnt - 1.0); |
97 | } |
98 | gp_stat->stddev = sqrt(gp_stat->stddev); |
99 | #undef IT_MEAN_DIFF |
100 | } |
101 | |
102 | void |
103 | grace_period_ticks_basic_stats(struct bench_res res[], int res_cnt, struct basic_stats *gp_stat) |
104 | { |
105 | int i; |
106 | |
107 | memset(gp_stat, 0, sizeof(struct basic_stats)); |
108 | for (i = 0; i < res_cnt; i++) |
109 | gp_stat->mean += res[i].stime / (double)res[i].gp_ct / (0.0 + res_cnt); |
110 | |
111 | #define IT_MEAN_DIFF (res[i].stime / (double)res[i].gp_ct - gp_stat->mean) |
112 | if (res_cnt > 1) { |
113 | for (i = 0; i < res_cnt; i++) |
114 | gp_stat->stddev += (IT_MEAN_DIFF * IT_MEAN_DIFF) / (res_cnt - 1.0); |
115 | } |
116 | gp_stat->stddev = sqrt(gp_stat->stddev); |
117 | #undef IT_MEAN_DIFF |
118 | } |
119 | |
120 | void hits_drops_report_final(struct bench_res res[], int res_cnt) |
121 | { |
122 | int i; |
123 | double hits_mean = 0.0, drops_mean = 0.0, total_ops_mean = 0.0; |
124 | double hits_stddev = 0.0, drops_stddev = 0.0, total_ops_stddev = 0.0; |
125 | double total_ops; |
126 | |
127 | for (i = 0; i < res_cnt; i++) { |
128 | hits_mean += res[i].hits / 1000000.0 / (0.0 + res_cnt); |
129 | drops_mean += res[i].drops / 1000000.0 / (0.0 + res_cnt); |
130 | } |
131 | total_ops_mean = hits_mean + drops_mean; |
132 | |
133 | if (res_cnt > 1) { |
134 | for (i = 0; i < res_cnt; i++) { |
135 | hits_stddev += (hits_mean - res[i].hits / 1000000.0) * |
136 | (hits_mean - res[i].hits / 1000000.0) / |
137 | (res_cnt - 1.0); |
138 | drops_stddev += (drops_mean - res[i].drops / 1000000.0) * |
139 | (drops_mean - res[i].drops / 1000000.0) / |
140 | (res_cnt - 1.0); |
141 | total_ops = res[i].hits + res[i].drops; |
142 | total_ops_stddev += (total_ops_mean - total_ops / 1000000.0) * |
143 | (total_ops_mean - total_ops / 1000000.0) / |
144 | (res_cnt - 1.0); |
145 | } |
146 | hits_stddev = sqrt(hits_stddev); |
147 | drops_stddev = sqrt(drops_stddev); |
148 | total_ops_stddev = sqrt(total_ops_stddev); |
149 | } |
150 | printf("Summary: hits %8.3lf \u00B1 %5.3lfM/s (%7.3lfM/prod), " , |
151 | hits_mean, hits_stddev, hits_mean / env.producer_cnt); |
152 | printf("drops %8.3lf \u00B1 %5.3lfM/s, " , |
153 | drops_mean, drops_stddev); |
154 | printf("total operations %8.3lf \u00B1 %5.3lfM/s\n" , |
155 | total_ops_mean, total_ops_stddev); |
156 | } |
157 | |
158 | void ops_report_progress(int iter, struct bench_res *res, long delta_ns) |
159 | { |
160 | double hits_per_sec, hits_per_prod; |
161 | |
162 | hits_per_sec = res->hits / 1000000.0 / (delta_ns / 1000000000.0); |
163 | hits_per_prod = hits_per_sec / env.producer_cnt; |
164 | |
165 | printf("Iter %3d (%7.3lfus): " , iter, (delta_ns - 1000000000) / 1000.0); |
166 | |
167 | printf("hits %8.3lfM/s (%7.3lfM/prod)\n" , hits_per_sec, hits_per_prod); |
168 | } |
169 | |
170 | void ops_report_final(struct bench_res res[], int res_cnt) |
171 | { |
172 | double hits_mean = 0.0, hits_stddev = 0.0; |
173 | int i; |
174 | |
175 | for (i = 0; i < res_cnt; i++) |
176 | hits_mean += res[i].hits / 1000000.0 / (0.0 + res_cnt); |
177 | |
178 | if (res_cnt > 1) { |
179 | for (i = 0; i < res_cnt; i++) |
180 | hits_stddev += (hits_mean - res[i].hits / 1000000.0) * |
181 | (hits_mean - res[i].hits / 1000000.0) / |
182 | (res_cnt - 1.0); |
183 | |
184 | hits_stddev = sqrt(hits_stddev); |
185 | } |
186 | printf("Summary: throughput %8.3lf \u00B1 %5.3lf M ops/s (%7.3lfM ops/prod), " , |
187 | hits_mean, hits_stddev, hits_mean / env.producer_cnt); |
188 | printf("latency %8.3lf ns/op\n" , 1000.0 / hits_mean * env.producer_cnt); |
189 | } |
190 | |
191 | void local_storage_report_progress(int iter, struct bench_res *res, |
192 | long delta_ns) |
193 | { |
194 | double important_hits_per_sec, hits_per_sec; |
195 | double delta_sec = delta_ns / 1000000000.0; |
196 | |
197 | hits_per_sec = res->hits / 1000000.0 / delta_sec; |
198 | important_hits_per_sec = res->important_hits / 1000000.0 / delta_sec; |
199 | |
200 | printf("Iter %3d (%7.3lfus): " , iter, (delta_ns - 1000000000) / 1000.0); |
201 | |
202 | printf("hits %8.3lfM/s " , hits_per_sec); |
203 | printf("important_hits %8.3lfM/s\n" , important_hits_per_sec); |
204 | } |
205 | |
206 | void local_storage_report_final(struct bench_res res[], int res_cnt) |
207 | { |
208 | double important_hits_mean = 0.0, important_hits_stddev = 0.0; |
209 | double hits_mean = 0.0, hits_stddev = 0.0; |
210 | int i; |
211 | |
212 | for (i = 0; i < res_cnt; i++) { |
213 | hits_mean += res[i].hits / 1000000.0 / (0.0 + res_cnt); |
214 | important_hits_mean += res[i].important_hits / 1000000.0 / (0.0 + res_cnt); |
215 | } |
216 | |
217 | if (res_cnt > 1) { |
218 | for (i = 0; i < res_cnt; i++) { |
219 | hits_stddev += (hits_mean - res[i].hits / 1000000.0) * |
220 | (hits_mean - res[i].hits / 1000000.0) / |
221 | (res_cnt - 1.0); |
222 | important_hits_stddev += |
223 | (important_hits_mean - res[i].important_hits / 1000000.0) * |
224 | (important_hits_mean - res[i].important_hits / 1000000.0) / |
225 | (res_cnt - 1.0); |
226 | } |
227 | |
228 | hits_stddev = sqrt(hits_stddev); |
229 | important_hits_stddev = sqrt(important_hits_stddev); |
230 | } |
231 | printf("Summary: hits throughput %8.3lf \u00B1 %5.3lf M ops/s, " , |
232 | hits_mean, hits_stddev); |
233 | printf("hits latency %8.3lf ns/op, " , 1000.0 / hits_mean); |
234 | printf("important_hits throughput %8.3lf \u00B1 %5.3lf M ops/s\n" , |
235 | important_hits_mean, important_hits_stddev); |
236 | } |
237 | |
238 | const char *argp_program_version = "benchmark" ; |
239 | const char *argp_program_bug_address = "<bpf@vger.kernel.org>" ; |
240 | const char argp_program_doc[] = |
241 | "benchmark Generic benchmarking framework.\n" |
242 | "\n" |
243 | "This tool runs benchmarks.\n" |
244 | "\n" |
245 | "USAGE: benchmark <bench-name>\n" |
246 | "\n" |
247 | "EXAMPLES:\n" |
248 | " # run 'count-local' benchmark with 1 producer and 1 consumer\n" |
249 | " benchmark count-local\n" |
250 | " # run 'count-local' with 16 producer and 8 consumer thread, pinned to CPUs\n" |
251 | " benchmark -p16 -c8 -a count-local\n" ; |
252 | |
253 | enum { |
254 | ARG_PROD_AFFINITY_SET = 1000, |
255 | ARG_CONS_AFFINITY_SET = 1001, |
256 | }; |
257 | |
258 | static const struct argp_option opts[] = { |
259 | { "list" , 'l', NULL, 0, "List available benchmarks" }, |
260 | { "duration" , 'd', "SEC" , 0, "Duration of benchmark, seconds" }, |
261 | { "warmup" , 'w', "SEC" , 0, "Warm-up period, seconds" }, |
262 | { "producers" , 'p', "NUM" , 0, "Number of producer threads" }, |
263 | { "consumers" , 'c', "NUM" , 0, "Number of consumer threads" }, |
264 | { "verbose" , 'v', NULL, 0, "Verbose debug output" }, |
265 | { "affinity" , 'a', NULL, 0, "Set consumer/producer thread affinity" }, |
266 | { "quiet" , 'q', NULL, 0, "Be more quiet" }, |
267 | { "prod-affinity" , ARG_PROD_AFFINITY_SET, "CPUSET" , 0, |
268 | "Set of CPUs for producer threads; implies --affinity" }, |
269 | { "cons-affinity" , ARG_CONS_AFFINITY_SET, "CPUSET" , 0, |
270 | "Set of CPUs for consumer threads; implies --affinity" }, |
271 | {}, |
272 | }; |
273 | |
274 | extern struct argp bench_ringbufs_argp; |
275 | extern struct argp bench_bloom_map_argp; |
276 | extern struct argp bench_bpf_loop_argp; |
277 | extern struct argp bench_local_storage_argp; |
278 | extern struct argp bench_local_storage_rcu_tasks_trace_argp; |
279 | extern struct argp bench_strncmp_argp; |
280 | extern struct argp bench_hashmap_lookup_argp; |
281 | extern struct argp bench_local_storage_create_argp; |
282 | extern struct argp bench_htab_mem_argp; |
283 | |
284 | static const struct argp_child bench_parsers[] = { |
285 | { &bench_ringbufs_argp, 0, "Ring buffers benchmark" , 0 }, |
286 | { &bench_bloom_map_argp, 0, "Bloom filter map benchmark" , 0 }, |
287 | { &bench_bpf_loop_argp, 0, "bpf_loop helper benchmark" , 0 }, |
288 | { &bench_local_storage_argp, 0, "local_storage benchmark" , 0 }, |
289 | { &bench_strncmp_argp, 0, "bpf_strncmp helper benchmark" , 0 }, |
290 | { &bench_local_storage_rcu_tasks_trace_argp, 0, |
291 | "local_storage RCU Tasks Trace slowdown benchmark" , 0 }, |
292 | { &bench_hashmap_lookup_argp, 0, "Hashmap lookup benchmark" , 0 }, |
293 | { &bench_local_storage_create_argp, 0, "local-storage-create benchmark" , 0 }, |
294 | { &bench_htab_mem_argp, 0, "hash map memory benchmark" , 0 }, |
295 | {}, |
296 | }; |
297 | |
298 | /* Make pos_args global, so that we can run argp_parse twice, if necessary */ |
299 | static int pos_args; |
300 | |
301 | static error_t parse_arg(int key, char *arg, struct argp_state *state) |
302 | { |
303 | switch (key) { |
304 | case 'v': |
305 | env.verbose = true; |
306 | break; |
307 | case 'l': |
308 | env.list = true; |
309 | break; |
310 | case 'd': |
311 | env.duration_sec = strtol(arg, NULL, 10); |
312 | if (env.duration_sec <= 0) { |
313 | fprintf(stderr, "Invalid duration: %s\n" , arg); |
314 | argp_usage(state); |
315 | } |
316 | break; |
317 | case 'w': |
318 | env.warmup_sec = strtol(arg, NULL, 10); |
319 | if (env.warmup_sec <= 0) { |
320 | fprintf(stderr, "Invalid warm-up duration: %s\n" , arg); |
321 | argp_usage(state); |
322 | } |
323 | break; |
324 | case 'p': |
325 | env.producer_cnt = strtol(arg, NULL, 10); |
326 | if (env.producer_cnt < 0) { |
327 | fprintf(stderr, "Invalid producer count: %s\n" , arg); |
328 | argp_usage(state); |
329 | } |
330 | break; |
331 | case 'c': |
332 | env.consumer_cnt = strtol(arg, NULL, 10); |
333 | if (env.consumer_cnt < 0) { |
334 | fprintf(stderr, "Invalid consumer count: %s\n" , arg); |
335 | argp_usage(state); |
336 | } |
337 | break; |
338 | case 'a': |
339 | env.affinity = true; |
340 | break; |
341 | case 'q': |
342 | env.quiet = true; |
343 | break; |
344 | case ARG_PROD_AFFINITY_SET: |
345 | env.affinity = true; |
346 | if (parse_num_list(s: arg, set: &env.prod_cpus.cpus, |
347 | set_len: &env.prod_cpus.cpus_len)) { |
348 | fprintf(stderr, "Invalid format of CPU set for producers." ); |
349 | argp_usage(state); |
350 | } |
351 | break; |
352 | case ARG_CONS_AFFINITY_SET: |
353 | env.affinity = true; |
354 | if (parse_num_list(s: arg, set: &env.cons_cpus.cpus, |
355 | set_len: &env.cons_cpus.cpus_len)) { |
356 | fprintf(stderr, "Invalid format of CPU set for consumers." ); |
357 | argp_usage(state); |
358 | } |
359 | break; |
360 | case ARGP_KEY_ARG: |
361 | if (pos_args++) { |
362 | fprintf(stderr, |
363 | "Unrecognized positional argument: %s\n" , arg); |
364 | argp_usage(state); |
365 | } |
366 | env.bench_name = strdup(arg); |
367 | break; |
368 | default: |
369 | return ARGP_ERR_UNKNOWN; |
370 | } |
371 | return 0; |
372 | } |
373 | |
374 | static void parse_cmdline_args_init(int argc, char **argv) |
375 | { |
376 | static const struct argp argp = { |
377 | .options = opts, |
378 | .parser = parse_arg, |
379 | .doc = argp_program_doc, |
380 | .children = bench_parsers, |
381 | }; |
382 | if (argp_parse(&argp, argc, argv, 0, NULL, NULL)) |
383 | exit(1); |
384 | } |
385 | |
386 | static void parse_cmdline_args_final(int argc, char **argv) |
387 | { |
388 | struct argp_child bench_parsers[2] = {}; |
389 | const struct argp argp = { |
390 | .options = opts, |
391 | .parser = parse_arg, |
392 | .doc = argp_program_doc, |
393 | .children = bench_parsers, |
394 | }; |
395 | |
396 | /* Parse arguments the second time with the correct set of parsers */ |
397 | if (bench->argp) { |
398 | bench_parsers[0].argp = bench->argp; |
399 | bench_parsers[0].header = bench->name; |
400 | pos_args = 0; |
401 | if (argp_parse(&argp, argc, argv, 0, NULL, NULL)) |
402 | exit(1); |
403 | } |
404 | } |
405 | |
406 | static void collect_measurements(long delta_ns); |
407 | |
408 | static __u64 last_time_ns; |
409 | static void sigalarm_handler(int signo) |
410 | { |
411 | long new_time_ns = get_time_ns(); |
412 | long delta_ns = new_time_ns - last_time_ns; |
413 | |
414 | collect_measurements(delta_ns); |
415 | |
416 | last_time_ns = new_time_ns; |
417 | } |
418 | |
419 | /* set up periodic 1-second timer */ |
420 | static void setup_timer() |
421 | { |
422 | static struct sigaction sigalarm_action = { |
423 | .sa_handler = sigalarm_handler, |
424 | }; |
425 | struct itimerval timer_settings = {}; |
426 | int err; |
427 | |
428 | last_time_ns = get_time_ns(); |
429 | err = sigaction(SIGALRM, &sigalarm_action, NULL); |
430 | if (err < 0) { |
431 | fprintf(stderr, "failed to install SIGALRM handler: %d\n" , -errno); |
432 | exit(1); |
433 | } |
434 | timer_settings.it_interval.tv_sec = 1; |
435 | timer_settings.it_value.tv_sec = 1; |
436 | err = setitimer(ITIMER_REAL, &timer_settings, NULL); |
437 | if (err < 0) { |
438 | fprintf(stderr, "failed to arm interval timer: %d\n" , -errno); |
439 | exit(1); |
440 | } |
441 | } |
442 | |
443 | static void set_thread_affinity(pthread_t thread, int cpu) |
444 | { |
445 | cpu_set_t cpuset; |
446 | int err; |
447 | |
448 | CPU_ZERO(&cpuset); |
449 | CPU_SET(cpu, &cpuset); |
450 | err = pthread_setaffinity_np(thread, sizeof(cpuset), &cpuset); |
451 | if (err) { |
452 | fprintf(stderr, "setting affinity to CPU #%d failed: %d\n" , |
453 | cpu, -err); |
454 | exit(1); |
455 | } |
456 | } |
457 | |
458 | static int next_cpu(struct cpu_set *cpu_set) |
459 | { |
460 | if (cpu_set->cpus) { |
461 | int i; |
462 | |
463 | /* find next available CPU */ |
464 | for (i = cpu_set->next_cpu; i < cpu_set->cpus_len; i++) { |
465 | if (cpu_set->cpus[i]) { |
466 | cpu_set->next_cpu = i + 1; |
467 | return i; |
468 | } |
469 | } |
470 | fprintf(stderr, "Not enough CPUs specified, need CPU #%d or higher.\n" , i); |
471 | exit(1); |
472 | } |
473 | |
474 | return cpu_set->next_cpu++ % env.nr_cpus; |
475 | } |
476 | |
477 | static struct bench_state { |
478 | int res_cnt; |
479 | struct bench_res *results; |
480 | pthread_t *consumers; |
481 | pthread_t *producers; |
482 | } state; |
483 | |
484 | const struct bench *bench = NULL; |
485 | |
486 | extern const struct bench bench_count_global; |
487 | extern const struct bench bench_count_local; |
488 | extern const struct bench bench_rename_base; |
489 | extern const struct bench bench_rename_kprobe; |
490 | extern const struct bench bench_rename_kretprobe; |
491 | extern const struct bench bench_rename_rawtp; |
492 | extern const struct bench bench_rename_fentry; |
493 | extern const struct bench bench_rename_fexit; |
494 | extern const struct bench bench_trig_base; |
495 | extern const struct bench bench_trig_tp; |
496 | extern const struct bench bench_trig_rawtp; |
497 | extern const struct bench bench_trig_kprobe; |
498 | extern const struct bench bench_trig_kretprobe; |
499 | extern const struct bench bench_trig_kprobe_multi; |
500 | extern const struct bench bench_trig_kretprobe_multi; |
501 | extern const struct bench bench_trig_fentry; |
502 | extern const struct bench bench_trig_fexit; |
503 | extern const struct bench bench_trig_fentry_sleep; |
504 | extern const struct bench bench_trig_fmodret; |
505 | extern const struct bench bench_trig_uprobe_base; |
506 | extern const struct bench bench_trig_uprobe_nop; |
507 | extern const struct bench bench_trig_uretprobe_nop; |
508 | extern const struct bench bench_trig_uprobe_push; |
509 | extern const struct bench bench_trig_uretprobe_push; |
510 | extern const struct bench bench_trig_uprobe_ret; |
511 | extern const struct bench bench_trig_uretprobe_ret; |
512 | extern const struct bench bench_rb_libbpf; |
513 | extern const struct bench bench_rb_custom; |
514 | extern const struct bench bench_pb_libbpf; |
515 | extern const struct bench bench_pb_custom; |
516 | extern const struct bench bench_bloom_lookup; |
517 | extern const struct bench bench_bloom_update; |
518 | extern const struct bench bench_bloom_false_positive; |
519 | extern const struct bench bench_hashmap_without_bloom; |
520 | extern const struct bench bench_hashmap_with_bloom; |
521 | extern const struct bench bench_bpf_loop; |
522 | extern const struct bench bench_strncmp_no_helper; |
523 | extern const struct bench bench_strncmp_helper; |
524 | extern const struct bench bench_bpf_hashmap_full_update; |
525 | extern const struct bench bench_local_storage_cache_seq_get; |
526 | extern const struct bench bench_local_storage_cache_interleaved_get; |
527 | extern const struct bench bench_local_storage_cache_hashmap_control; |
528 | extern const struct bench bench_local_storage_tasks_trace; |
529 | extern const struct bench bench_bpf_hashmap_lookup; |
530 | extern const struct bench bench_local_storage_create; |
531 | extern const struct bench bench_htab_mem; |
532 | |
533 | static const struct bench *benchs[] = { |
534 | &bench_count_global, |
535 | &bench_count_local, |
536 | &bench_rename_base, |
537 | &bench_rename_kprobe, |
538 | &bench_rename_kretprobe, |
539 | &bench_rename_rawtp, |
540 | &bench_rename_fentry, |
541 | &bench_rename_fexit, |
542 | &bench_trig_base, |
543 | &bench_trig_tp, |
544 | &bench_trig_rawtp, |
545 | &bench_trig_kprobe, |
546 | &bench_trig_kretprobe, |
547 | &bench_trig_kprobe_multi, |
548 | &bench_trig_kretprobe_multi, |
549 | &bench_trig_fentry, |
550 | &bench_trig_fexit, |
551 | &bench_trig_fentry_sleep, |
552 | &bench_trig_fmodret, |
553 | &bench_trig_uprobe_base, |
554 | &bench_trig_uprobe_nop, |
555 | &bench_trig_uretprobe_nop, |
556 | &bench_trig_uprobe_push, |
557 | &bench_trig_uretprobe_push, |
558 | &bench_trig_uprobe_ret, |
559 | &bench_trig_uretprobe_ret, |
560 | &bench_rb_libbpf, |
561 | &bench_rb_custom, |
562 | &bench_pb_libbpf, |
563 | &bench_pb_custom, |
564 | &bench_bloom_lookup, |
565 | &bench_bloom_update, |
566 | &bench_bloom_false_positive, |
567 | &bench_hashmap_without_bloom, |
568 | &bench_hashmap_with_bloom, |
569 | &bench_bpf_loop, |
570 | &bench_strncmp_no_helper, |
571 | &bench_strncmp_helper, |
572 | &bench_bpf_hashmap_full_update, |
573 | &bench_local_storage_cache_seq_get, |
574 | &bench_local_storage_cache_interleaved_get, |
575 | &bench_local_storage_cache_hashmap_control, |
576 | &bench_local_storage_tasks_trace, |
577 | &bench_bpf_hashmap_lookup, |
578 | &bench_local_storage_create, |
579 | &bench_htab_mem, |
580 | }; |
581 | |
582 | static void find_benchmark(void) |
583 | { |
584 | int i; |
585 | |
586 | if (!env.bench_name) { |
587 | fprintf(stderr, "benchmark name is not specified\n" ); |
588 | exit(1); |
589 | } |
590 | for (i = 0; i < ARRAY_SIZE(benchs); i++) { |
591 | if (strcmp(benchs[i]->name, env.bench_name) == 0) { |
592 | bench = benchs[i]; |
593 | break; |
594 | } |
595 | } |
596 | if (!bench) { |
597 | fprintf(stderr, "benchmark '%s' not found\n" , env.bench_name); |
598 | exit(1); |
599 | } |
600 | } |
601 | |
602 | static void setup_benchmark(void) |
603 | { |
604 | int i, err; |
605 | |
606 | if (!env.quiet) |
607 | printf("Setting up benchmark '%s'...\n" , bench->name); |
608 | |
609 | state.producers = calloc(env.producer_cnt, sizeof(*state.producers)); |
610 | state.consumers = calloc(env.consumer_cnt, sizeof(*state.consumers)); |
611 | state.results = calloc(env.duration_sec + env.warmup_sec + 2, |
612 | sizeof(*state.results)); |
613 | if (!state.producers || !state.consumers || !state.results) |
614 | exit(1); |
615 | |
616 | if (bench->validate) |
617 | bench->validate(); |
618 | if (bench->setup) |
619 | bench->setup(); |
620 | |
621 | for (i = 0; i < env.consumer_cnt; i++) { |
622 | if (!bench->consumer_thread) { |
623 | fprintf(stderr, "benchmark doesn't support consumers!\n" ); |
624 | exit(1); |
625 | } |
626 | err = pthread_create(&state.consumers[i], NULL, |
627 | bench->consumer_thread, (void *)(long)i); |
628 | if (err) { |
629 | fprintf(stderr, "failed to create consumer thread #%d: %d\n" , |
630 | i, -err); |
631 | exit(1); |
632 | } |
633 | if (env.affinity) |
634 | set_thread_affinity(state.consumers[i], |
635 | next_cpu(cpu_set: &env.cons_cpus)); |
636 | } |
637 | |
638 | /* unless explicit producer CPU list is specified, continue after |
639 | * last consumer CPU |
640 | */ |
641 | if (!env.prod_cpus.cpus) |
642 | env.prod_cpus.next_cpu = env.cons_cpus.next_cpu; |
643 | |
644 | for (i = 0; i < env.producer_cnt; i++) { |
645 | if (!bench->producer_thread) { |
646 | fprintf(stderr, "benchmark doesn't support producers!\n" ); |
647 | exit(1); |
648 | } |
649 | err = pthread_create(&state.producers[i], NULL, |
650 | bench->producer_thread, (void *)(long)i); |
651 | if (err) { |
652 | fprintf(stderr, "failed to create producer thread #%d: %d\n" , |
653 | i, -err); |
654 | exit(1); |
655 | } |
656 | if (env.affinity) |
657 | set_thread_affinity(state.producers[i], |
658 | next_cpu(cpu_set: &env.prod_cpus)); |
659 | } |
660 | |
661 | if (!env.quiet) |
662 | printf("Benchmark '%s' started.\n" , bench->name); |
663 | } |
664 | |
665 | static pthread_mutex_t bench_done_mtx = PTHREAD_MUTEX_INITIALIZER; |
666 | static pthread_cond_t bench_done = PTHREAD_COND_INITIALIZER; |
667 | |
668 | static void collect_measurements(long delta_ns) { |
669 | int iter = state.res_cnt++; |
670 | struct bench_res *res = &state.results[iter]; |
671 | |
672 | bench->measure(res); |
673 | |
674 | if (bench->report_progress) |
675 | bench->report_progress(iter, res, delta_ns); |
676 | |
677 | if (iter == env.duration_sec + env.warmup_sec) { |
678 | pthread_mutex_lock(&bench_done_mtx); |
679 | pthread_cond_signal(&bench_done); |
680 | pthread_mutex_unlock(&bench_done_mtx); |
681 | } |
682 | } |
683 | |
684 | int main(int argc, char **argv) |
685 | { |
686 | env.nr_cpus = get_nprocs(); |
687 | parse_cmdline_args_init(argc, argv); |
688 | |
689 | if (env.list) { |
690 | int i; |
691 | |
692 | printf("Available benchmarks:\n" ); |
693 | for (i = 0; i < ARRAY_SIZE(benchs); i++) { |
694 | printf("- %s\n" , benchs[i]->name); |
695 | } |
696 | return 0; |
697 | } |
698 | |
699 | find_benchmark(); |
700 | parse_cmdline_args_final(argc, argv); |
701 | |
702 | setup_benchmark(); |
703 | |
704 | setup_timer(); |
705 | |
706 | pthread_mutex_lock(&bench_done_mtx); |
707 | pthread_cond_wait(&bench_done, &bench_done_mtx); |
708 | pthread_mutex_unlock(&bench_done_mtx); |
709 | |
710 | if (bench->report_final) |
711 | /* skip first sample */ |
712 | bench->report_final(state.results + env.warmup_sec, |
713 | state.res_cnt - env.warmup_sec); |
714 | |
715 | return 0; |
716 | } |
717 | |