| 1 | |
| 2 | * ring buffer tester and benchmark
|
| 3 | *
|
| 4 | * Copyright (C) 2009 Steven Rostedt <[email protected]>
|
| 5 |
|
| 6 | #include <linux/ring_buffer.h>
|
| 7 | #include <linux/completion.h>
|
| 8 | #include <linux/kthread.h>
|
| 9 | #include <linux/module.h>
|
| 10 | #include <linux/time.h>
|
| 11 |
|
| 12 | struct rb_page {
|
| 13 | u64 ts;
|
| 14 | local_t commit;
|
| 15 | char data[4080];
|
| 16 | };
|
| 17 |
|
| 18 |
|
| 19 | #define RUN_TIME 10
|
| 20 | #define SLEEP_TIME 10
|
| 21 |
|
| 22 |
|
| 23 | static int wakeup_interval = 100;
|
| 24 |
|
| 25 | static int reader_finish;
|
| 26 | static struct completion read_start;
|
| 27 | static struct completion read_done;
|
| 28 |
|
| 29 | static struct ring_buffer *buffer;
|
| 30 | static struct task_struct *producer;
|
| 31 | static struct task_struct *consumer;
|
| 32 | static unsigned long read;
|
| 33 |
|
| 34 | static int disable_reader;
|
| 35 | module_param(disable_reader, uint, 0644);
|
| 36 | MODULE_PARM_DESC(disable_reader, "only run producer");
|
| 37 |
|
| 38 | static int read_events;
|
| 39 |
|
| 40 | static int kill_test;
|
| 41 |
|
| 42 | #define KILL_TEST() \
|
| 43 | do { \
|
| 44 | if (!kill_test) { \
|
| 45 | kill_test = 1; \
|
| 46 | WARN_ON(1); \
|
| 47 | } \
|
| 48 | } while (0)
|
| 49 |
|
| 50 | enum event_status {
|
| 51 | EVENT_FOUND,
|
| 52 | EVENT_DROPPED,
|
| 53 | };
|
| 54 |
|
| 55 | static enum event_status read_event(int cpu)
|
| 56 | {
|
| 57 | struct ring_buffer_event *event;
|
| 58 | int *entry;
|
| 59 | u64 ts;
|
| 60 |
|
| 61 | event = ring_buffer_consume(buffer, cpu, &ts);
|
| 62 | if (!event)
|
| 63 | return EVENT_DROPPED;
|
| 64 |
|
| 65 | entry = ring_buffer_event_data(event);
|
| 66 | if (*entry != cpu) {
|
| 67 | KILL_TEST();
|
| 68 | return EVENT_DROPPED;
|
| 69 | }
|
| 70 |
|
| 71 | read++;
|
| 72 | return EVENT_FOUND;
|
| 73 | }
|
| 74 |
|
| 75 | static enum event_status read_page(int cpu)
|
| 76 | {
|
| 77 | struct ring_buffer_event *event;
|
| 78 | struct rb_page *rpage;
|
| 79 | unsigned long commit;
|
| 80 | void *bpage;
|
| 81 | int *entry;
|
| 82 | int ret;
|
| 83 | int inc;
|
| 84 | int i;
|
| 85 |
|
| 86 | bpage = ring_buffer_alloc_read_page(buffer);
|
| 87 | if (!bpage)
|
| 88 | return EVENT_DROPPED;
|
| 89 |
|
| 90 | ret = ring_buffer_read_page(buffer, &bpage, PAGE_SIZE, cpu, 1);
|
| 91 | if (ret >= 0) {
|
| 92 | rpage = bpage;
|
| 93 | commit = local_read(&rpage->commit);
|
| 94 | for (i = 0; i < commit && !kill_test; i += inc) {
|
| 95 |
|
| 96 | if (i >= (PAGE_SIZE - offsetof(struct rb_page, data))) {
|
| 97 | KILL_TEST();
|
| 98 | break;
|
| 99 | }
|
| 100 |
|
| 101 | inc = -1;
|
| 102 | event = (void *)&rpage->data[i];
|
| 103 | switch (event->type_len) {
|
| 104 | case RINGBUF_TYPE_PADDING:
|
| 105 |
|
| 106 | if (!event->time_delta)
|
| 107 | KILL_TEST();
|
| 108 | inc = event->array[0] + 4;
|
| 109 | break;
|
| 110 | case RINGBUF_TYPE_TIME_EXTEND:
|
| 111 | inc = 8;
|
| 112 | break;
|
| 113 | case 0:
|
| 114 | entry = ring_buffer_event_data(event);
|
| 115 | if (*entry != cpu) {
|
| 116 | KILL_TEST();
|
| 117 | break;
|
| 118 | }
|
| 119 | read++;
|
| 120 | if (!event->array[0]) {
|
| 121 | KILL_TEST();
|
| 122 | break;
|
| 123 | }
|
| 124 | inc = event->array[0] + 4;
|
| 125 | break;
|
| 126 | default:
|
| 127 | entry = ring_buffer_event_data(event);
|
| 128 | if (*entry != cpu) {
|
| 129 | KILL_TEST();
|
| 130 | break;
|
| 131 | }
|
| 132 | read++;
|
| 133 | inc = ((event->type_len + 1) * 4);
|
| 134 | }
|
| 135 | if (kill_test)
|
| 136 | break;
|
| 137 |
|
| 138 | if (inc <= 0) {
|
| 139 | KILL_TEST();
|
| 140 | break;
|
| 141 | }
|
| 142 | }
|
| 143 | }
|
| 144 | ring_buffer_free_read_page(buffer, bpage);
|
| 145 |
|
| 146 | if (ret < 0)
|
| 147 | return EVENT_DROPPED;
|
| 148 | return EVENT_FOUND;
|
| 149 | }
|
| 150 |
|
| 151 | static void ring_buffer_consumer(void)
|
| 152 | {
|
| 153 |
|
| 154 | read_events ^= 1;
|
| 155 |
|
| 156 | read = 0;
|
| 157 | while (!reader_finish && !kill_test) {
|
| 158 | int found;
|
| 159 |
|
| 160 | do {
|
| 161 | int cpu;
|
| 162 |
|
| 163 | found = 0;
|
| 164 | for_each_online_cpu(cpu) {
|
| 165 | enum event_status stat;
|
| 166 |
|
| 167 | if (read_events)
|
| 168 | stat = read_event(cpu);
|
| 169 | else
|
| 170 | stat = read_page(cpu);
|
| 171 |
|
| 172 | if (kill_test)
|
| 173 | break;
|
| 174 | if (stat == EVENT_FOUND)
|
| 175 | found = 1;
|
| 176 | }
|
| 177 | } while (found && !kill_test);
|
| 178 |
|
| 179 | set_current_state(TASK_INTERRUPTIBLE);
|
| 180 | if (reader_finish)
|
| 181 | break;
|
| 182 |
|
| 183 | schedule();
|
| 184 | __set_current_state(TASK_RUNNING);
|
| 185 | }
|
| 186 | reader_finish = 0;
|
| 187 | complete(&read_done);
|
| 188 | }
|
| 189 |
|
| 190 | static void ring_buffer_producer(void)
|
| 191 | {
|
| 192 | struct timeval start_tv;
|
| 193 | struct timeval end_tv;
|
| 194 | unsigned long long time;
|
| 195 | unsigned long long entries;
|
| 196 | unsigned long long overruns;
|
| 197 | unsigned long missed = 0;
|
| 198 | unsigned long hit = 0;
|
| 199 | unsigned long avg;
|
| 200 | int cnt = 0;
|
| 201 |
|
| 202 | |
| 203 | * Hammer the buffer for 10 secs (this may
|
| 204 | * make the system stall)
|
| 205 |
|
| 206 | trace_printk("Starting ring buffer hammer\n");
|
| 207 | do_gettimeofday(&start_tv);
|
| 208 | do {
|
| 209 | struct ring_buffer_event *event;
|
| 210 | int *entry;
|
| 211 |
|
| 212 | event = ring_buffer_lock_reserve(buffer, 10);
|
| 213 | if (!event) {
|
| 214 | missed++;
|
| 215 | } else {
|
| 216 | hit++;
|
| 217 | entry = ring_buffer_event_data(event);
|
| 218 | *entry = smp_processor_id();
|
| 219 | ring_buffer_unlock_commit(buffer, event);
|
| 220 | }
|
| 221 | do_gettimeofday(&end_tv);
|
| 222 |
|
| 223 | cnt++;
|
| 224 | if (consumer && !(cnt % wakeup_interval))
|
| 225 | wake_up_process(consumer);
|
| 226 |
|
| 227 | #ifndef CONFIG_PREEMPT
|
| 228 | |
| 229 | * If we are a non preempt kernel, the 10 second run will
|
| 230 | * stop everything while it runs. Instead, we will call
|
| 231 | * cond_resched and also add any time that was lost by a
|
| 232 | * rescedule.
|
| 233 | *
|
| 234 | * Do a cond resched at the same frequency we would wake up
|
| 235 | * the reader.
|
| 236 |
|
| 237 | if (cnt % wakeup_interval)
|
| 238 | cond_resched();
|
| 239 | #endif
|
| 240 |
|
| 241 | } while (end_tv.tv_sec < (start_tv.tv_sec + RUN_TIME) && !kill_test);
|
| 242 | trace_printk("End ring buffer hammer\n");
|
| 243 |
|
| 244 | if (consumer) {
|
| 245 |
|
| 246 | init_completion(&read_start);
|
| 247 | init_completion(&read_done);
|
| 248 |
|
| 249 | smp_wmb();
|
| 250 | reader_finish = 1;
|
| 251 |
|
| 252 | smp_wmb();
|
| 253 | wake_up_process(consumer);
|
| 254 | wait_for_completion(&read_done);
|
| 255 | }
|
| 256 |
|
| 257 | time = end_tv.tv_sec - start_tv.tv_sec;
|
| 258 | time *= USEC_PER_SEC;
|
| 259 | time += (long long)((long)end_tv.tv_usec - (long)start_tv.tv_usec);
|
| 260 |
|
| 261 | entries = ring_buffer_entries(buffer);
|
| 262 | overruns = ring_buffer_overruns(buffer);
|
| 263 |
|
| 264 | if (kill_test)
|
| 265 | trace_printk("ERROR!\n");
|
| 266 | trace_printk("Time: %lld (usecs)\n", time);
|
| 267 | trace_printk("Overruns: %lld\n", overruns);
|
| 268 | if (disable_reader)
|
| 269 | trace_printk("Read: (reader disabled)\n");
|
| 270 | else
|
| 271 | trace_printk("Read: %ld (by %s)\n", read,
|
| 272 | read_events ? "events" : "pages");
|
| 273 | trace_printk("Entries: %lld\n", entries);
|
| 274 | trace_printk("Total: %lld\n", entries + overruns + read);
|
| 275 | trace_printk("Missed: %ld\n", missed);
|
| 276 | trace_printk("Hit: %ld\n", hit);
|
| 277 |
|
| 278 |
|
| 279 | do_div(time, USEC_PER_MSEC);
|
| 280 | if (time)
|
| 281 | hit /= (long)time;
|
| 282 | else
|
| 283 | trace_printk("TIME IS ZERO??\n");
|
| 284 |
|
| 285 | trace_printk("Entries per millisec: %ld\n", hit);
|
| 286 |
|
| 287 | if (hit) {
|
| 288 |
|
| 289 | avg = NSEC_PER_MSEC / hit;
|
| 290 | trace_printk("%ld ns per entry\n", avg);
|
| 291 | }
|
| 292 |
|
| 293 | if (missed) {
|
| 294 | if (time)
|
| 295 | missed /= (long)time;
|
| 296 |
|
| 297 | trace_printk("Total iterations per millisec: %ld\n",
|
| 298 | hit + missed);
|
| 299 |
|
| 300 |
|
| 301 | if (!(hit + missed)) {
|
| 302 | trace_printk("hit + missed overflowed and totalled zero!\n");
|
| 303 | hit--;
|
| 304 | }
|
| 305 |
|
| 306 |
|
| 307 | avg = NSEC_PER_MSEC / (hit + missed);
|
| 308 | trace_printk("%ld ns per entry\n", avg);
|
| 309 | }
|
| 310 | }
|
| 311 |
|
| 312 | static void wait_to_die(void)
|
| 313 | {
|
| 314 | set_current_state(TASK_INTERRUPTIBLE);
|
| 315 | while (!kthread_should_stop()) {
|
| 316 | schedule();
|
| 317 | set_current_state(TASK_INTERRUPTIBLE);
|
| 318 | }
|
| 319 | __set_current_state(TASK_RUNNING);
|
| 320 | }
|
| 321 |
|
| 322 | static int ring_buffer_consumer_thread(void *arg)
|
| 323 | {
|
| 324 | while (!kthread_should_stop() && !kill_test) {
|
| 325 | complete(&read_start);
|
| 326 |
|
| 327 | ring_buffer_consumer();
|
| 328 |
|
| 329 | set_current_state(TASK_INTERRUPTIBLE);
|
| 330 | if (kthread_should_stop() || kill_test)
|
| 331 | break;
|
| 332 |
|
| 333 | schedule();
|
| 334 | __set_current_state(TASK_RUNNING);
|
| 335 | }
|
| 336 | __set_current_state(TASK_RUNNING);
|
| 337 |
|
| 338 | if (kill_test)
|
| 339 | wait_to_die();
|
| 340 |
|
| 341 | return 0;
|
| 342 | }
|
| 343 |
|
| 344 | static int ring_buffer_producer_thread(void *arg)
|
| 345 | {
|
| 346 | init_completion(&read_start);
|
| 347 |
|
| 348 | while (!kthread_should_stop() && !kill_test) {
|
| 349 | ring_buffer_reset(buffer);
|
| 350 |
|
| 351 | if (consumer) {
|
| 352 | smp_wmb();
|
| 353 | wake_up_process(consumer);
|
| 354 | wait_for_completion(&read_start);
|
| 355 | }
|
| 356 |
|
| 357 | ring_buffer_producer();
|
| 358 |
|
| 359 | trace_printk("Sleeping for 10 secs\n");
|
| 360 | set_current_state(TASK_INTERRUPTIBLE);
|
| 361 | schedule_timeout(HZ * SLEEP_TIME);
|
| 362 | __set_current_state(TASK_RUNNING);
|
| 363 | }
|
| 364 |
|
| 365 | if (kill_test)
|
| 366 | wait_to_die();
|
| 367 |
|
| 368 | return 0;
|
| 369 | }
|
| 370 |
|
| 371 | static int __init ring_buffer_benchmark_init(void)
|
| 372 | {
|
| 373 | int ret;
|
| 374 |
|
| 375 |
|
| 376 | buffer = ring_buffer_alloc(1000000, RB_FL_OVERWRITE);
|
| 377 | if (!buffer)
|
| 378 | return -ENOMEM;
|
| 379 |
|
| 380 | if (!disable_reader) {
|
| 381 | consumer = kthread_create(ring_buffer_consumer_thread,
|
| 382 | NULL, "rb_consumer");
|
| 383 | ret = PTR_ERR(consumer);
|
| 384 | if (IS_ERR(consumer))
|
| 385 | goto out_fail;
|
| 386 | }
|
| 387 |
|
| 388 | producer = kthread_run(ring_buffer_producer_thread,
|
| 389 | NULL, "rb_producer");
|
| 390 | ret = PTR_ERR(producer);
|
| 391 |
|
| 392 | if (IS_ERR(producer))
|
| 393 | goto out_kill;
|
| 394 |
|
| 395 | return 0;
|
| 396 |
|
| 397 | out_kill:
|
| 398 | if (consumer)
|
| 399 | kthread_stop(consumer);
|
| 400 |
|
| 401 | out_fail:
|
| 402 | ring_buffer_free(buffer);
|
| 403 | return ret;
|
| 404 | }
|
| 405 |
|
| 406 | static void __exit ring_buffer_benchmark_exit(void)
|
| 407 | {
|
| 408 | kthread_stop(producer);
|
| 409 | if (consumer)
|
| 410 | kthread_stop(consumer);
|
| 411 | ring_buffer_free(buffer);
|
| 412 | }
|
| 413 |
|
| 414 | module_init(ring_buffer_benchmark_init);
|
| 415 | module_exit(ring_buffer_benchmark_exit);
|
| 416 |
|
| 417 | MODULE_AUTHOR("Steven Rostedt");
|
| 418 | MODULE_DESCRIPTION("ring_buffer_benchmark");
|
| 419 | MODULE_LICENSE("GPL");
|