4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of Intel Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39 #include <sys/queue.h>
41 #include <rte_atomic.h>
42 #include <rte_common.h>
43 #include <rte_cycles.h>
44 #include <rte_per_lcore.h>
45 #include <rte_memory.h>
46 #include <rte_launch.h>
48 #include <rte_lcore.h>
49 #include <rte_branch_prediction.h>
50 #include <rte_spinlock.h>
51 #include <rte_random.h>
52 #include <rte_pause.h>
54 #include "rte_timer.h"
56 LIST_HEAD(rte_timer_list, rte_timer);
59 struct rte_timer pending_head; /**< dummy timer instance to head up list */
60 rte_spinlock_t list_lock; /**< lock to protect list access */
62 /** per-core variable that true if a timer was updated on this
63 * core since last reset of the variable */
66 /** track the current depth of the skiplist */
67 unsigned curr_skiplist_depth;
69 unsigned prev_lcore; /**< used for lcore round robin */
71 /** running timer on this lcore now */
72 struct rte_timer *running_tim;
74 #ifdef RTE_LIBRTE_TIMER_DEBUG
75 /** per-lcore statistics */
76 struct rte_timer_debug_stats stats;
78 } __rte_cache_aligned;
80 /** per-lcore private info for timers */
81 static struct priv_timer priv_timer[RTE_MAX_LCORE];
83 /* when debug is enabled, store some statistics */
84 #ifdef RTE_LIBRTE_TIMER_DEBUG
85 #define __TIMER_STAT_ADD(name, n) do { \
86 unsigned __lcore_id = rte_lcore_id(); \
87 if (__lcore_id < RTE_MAX_LCORE) \
88 priv_timer[__lcore_id].stats.name += (n); \
91 #define __TIMER_STAT_ADD(name, n) do {} while(0)
94 /* Init the timer library. */
96 rte_timer_subsystem_init(void)
100 /* since priv_timer is static, it's zeroed by default, so only init some
103 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id ++) {
104 rte_spinlock_init(&priv_timer[lcore_id].list_lock);
105 priv_timer[lcore_id].prev_lcore = lcore_id;
109 /* Initialize the timer handle tim for use */
111 rte_timer_init(struct rte_timer *tim)
113 union rte_timer_status status;
115 status.state = RTE_TIMER_STOP;
116 status.owner = RTE_TIMER_NO_OWNER;
117 tim->status.u32 = status.u32;
121 * if timer is pending or stopped (or running on the same core than
122 * us), mark timer as configuring, and on success return the previous
123 * status of the timer
126 timer_set_config_state(struct rte_timer *tim,
127 union rte_timer_status *ret_prev_status)
129 union rte_timer_status prev_status, status;
133 lcore_id = rte_lcore_id();
135 /* wait that the timer is in correct status before update,
136 * and mark it as being configured */
137 while (success == 0) {
138 prev_status.u32 = tim->status.u32;
140 /* timer is running on another core
141 * or ready to run on local core, exit
143 if (prev_status.state == RTE_TIMER_RUNNING &&
144 (prev_status.owner != (uint16_t)lcore_id ||
145 tim != priv_timer[lcore_id].running_tim))
148 /* timer is being configured on another core */
149 if (prev_status.state == RTE_TIMER_CONFIG)
152 /* here, we know that timer is stopped or pending,
153 * mark it atomically as being configured */
154 status.state = RTE_TIMER_CONFIG;
155 status.owner = (int16_t)lcore_id;
156 success = rte_atomic32_cmpset(&tim->status.u32,
161 ret_prev_status->u32 = prev_status.u32;
166 * if timer is pending, mark timer as running
169 timer_set_running_state(struct rte_timer *tim)
171 union rte_timer_status prev_status, status;
172 unsigned lcore_id = rte_lcore_id();
175 /* wait that the timer is in correct status before update,
176 * and mark it as running */
177 while (success == 0) {
178 prev_status.u32 = tim->status.u32;
180 /* timer is not pending anymore */
181 if (prev_status.state != RTE_TIMER_PENDING)
184 /* here, we know that timer is stopped or pending,
185 * mark it atomically as being configured */
186 status.state = RTE_TIMER_RUNNING;
187 status.owner = (int16_t)lcore_id;
188 success = rte_atomic32_cmpset(&tim->status.u32,
197 * Return a skiplist level for a new entry.
198 * This probabalistically gives a level with p=1/4 that an entry at level n
199 * will also appear at level n+1.
202 timer_get_skiplist_level(unsigned curr_depth)
204 #ifdef RTE_LIBRTE_TIMER_DEBUG
205 static uint32_t i, count = 0;
206 static uint32_t levels[MAX_SKIPLIST_DEPTH] = {0};
209 /* probability value is 1/4, i.e. all at level 0, 1 in 4 is at level 1,
210 * 1 in 16 at level 2, 1 in 64 at level 3, etc. Calculated using lowest
211 * bit position of a (pseudo)random number.
213 uint32_t rand = rte_rand() & (UINT32_MAX - 1);
214 uint32_t level = rand == 0 ? MAX_SKIPLIST_DEPTH : (rte_bsf32(rand)-1) / 2;
216 /* limit the levels used to one above our current level, so we don't,
217 * for instance, have a level 0 and a level 7 without anything between
219 if (level > curr_depth)
221 if (level >= MAX_SKIPLIST_DEPTH)
222 level = MAX_SKIPLIST_DEPTH-1;
223 #ifdef RTE_LIBRTE_TIMER_DEBUG
226 if (count % 10000 == 0)
227 for (i = 0; i < MAX_SKIPLIST_DEPTH; i++)
228 printf("Level %u: %u\n", (unsigned)i, (unsigned)levels[i]);
234 * For a given time value, get the entries at each level which
235 * are <= that time value.
238 timer_get_prev_entries(uint64_t time_val, unsigned tim_lcore,
239 struct rte_timer **prev)
241 unsigned lvl = priv_timer[tim_lcore].curr_skiplist_depth;
242 prev[lvl] = &priv_timer[tim_lcore].pending_head;
245 prev[lvl] = prev[lvl+1];
246 while (prev[lvl]->sl_next[lvl] &&
247 prev[lvl]->sl_next[lvl]->expire <= time_val)
248 prev[lvl] = prev[lvl]->sl_next[lvl];
253 * Given a timer node in the skiplist, find the previous entries for it at
254 * all skiplist levels.
257 timer_get_prev_entries_for_node(struct rte_timer *tim, unsigned tim_lcore,
258 struct rte_timer **prev)
261 /* to get a specific entry in the list, look for just lower than the time
262 * values, and then increment on each level individually if necessary
264 timer_get_prev_entries(tim->expire - 1, tim_lcore, prev);
265 for (i = priv_timer[tim_lcore].curr_skiplist_depth - 1; i >= 0; i--) {
266 while (prev[i]->sl_next[i] != NULL &&
267 prev[i]->sl_next[i] != tim &&
268 prev[i]->sl_next[i]->expire <= tim->expire)
269 prev[i] = prev[i]->sl_next[i];
274 * add in list, lock if needed
275 * timer must be in config state
276 * timer must not be in a list
279 timer_add(struct rte_timer *tim, unsigned tim_lcore, int local_is_locked)
281 unsigned lcore_id = rte_lcore_id();
283 struct rte_timer *prev[MAX_SKIPLIST_DEPTH+1];
285 /* if timer needs to be scheduled on another core, we need to
286 * lock the list; if it is on local core, we need to lock if
287 * we are not called from rte_timer_manage() */
288 if (tim_lcore != lcore_id || !local_is_locked)
289 rte_spinlock_lock(&priv_timer[tim_lcore].list_lock);
291 /* find where exactly this element goes in the list of elements
293 timer_get_prev_entries(tim->expire, tim_lcore, prev);
295 /* now assign it a new level and add at that level */
296 const unsigned tim_level = timer_get_skiplist_level(
297 priv_timer[tim_lcore].curr_skiplist_depth);
298 if (tim_level == priv_timer[tim_lcore].curr_skiplist_depth)
299 priv_timer[tim_lcore].curr_skiplist_depth++;
303 tim->sl_next[lvl] = prev[lvl]->sl_next[lvl];
304 prev[lvl]->sl_next[lvl] = tim;
307 tim->sl_next[0] = prev[0]->sl_next[0];
308 prev[0]->sl_next[0] = tim;
310 /* save the lowest list entry into the expire field of the dummy hdr
311 * NOTE: this is not atomic on 32-bit*/
312 priv_timer[tim_lcore].pending_head.expire = priv_timer[tim_lcore].\
313 pending_head.sl_next[0]->expire;
315 if (tim_lcore != lcore_id || !local_is_locked)
316 rte_spinlock_unlock(&priv_timer[tim_lcore].list_lock);
320 * del from list, lock if needed
321 * timer must be in config state
322 * timer must be in a list
325 timer_del(struct rte_timer *tim, union rte_timer_status prev_status,
328 unsigned lcore_id = rte_lcore_id();
329 unsigned prev_owner = prev_status.owner;
331 struct rte_timer *prev[MAX_SKIPLIST_DEPTH+1];
333 /* if timer needs is pending another core, we need to lock the
334 * list; if it is on local core, we need to lock if we are not
335 * called from rte_timer_manage() */
336 if (prev_owner != lcore_id || !local_is_locked)
337 rte_spinlock_lock(&priv_timer[prev_owner].list_lock);
339 /* save the lowest list entry into the expire field of the dummy hdr.
340 * NOTE: this is not atomic on 32-bit */
341 if (tim == priv_timer[prev_owner].pending_head.sl_next[0])
342 priv_timer[prev_owner].pending_head.expire =
343 ((tim->sl_next[0] == NULL) ? 0 : tim->sl_next[0]->expire);
345 /* adjust pointers from previous entries to point past this */
346 timer_get_prev_entries_for_node(tim, prev_owner, prev);
347 for (i = priv_timer[prev_owner].curr_skiplist_depth - 1; i >= 0; i--) {
348 if (prev[i]->sl_next[i] == tim)
349 prev[i]->sl_next[i] = tim->sl_next[i];
352 /* in case we deleted last entry at a level, adjust down max level */
353 for (i = priv_timer[prev_owner].curr_skiplist_depth - 1; i >= 0; i--)
354 if (priv_timer[prev_owner].pending_head.sl_next[i] == NULL)
355 priv_timer[prev_owner].curr_skiplist_depth --;
359 if (prev_owner != lcore_id || !local_is_locked)
360 rte_spinlock_unlock(&priv_timer[prev_owner].list_lock);
363 /* Reset and start the timer associated with the timer handle (private func) */
365 __rte_timer_reset(struct rte_timer *tim, uint64_t expire,
366 uint64_t period, unsigned tim_lcore,
367 rte_timer_cb_t fct, void *arg,
370 union rte_timer_status prev_status, status;
372 unsigned lcore_id = rte_lcore_id();
374 /* round robin for tim_lcore */
375 if (tim_lcore == (unsigned)LCORE_ID_ANY) {
376 if (lcore_id < RTE_MAX_LCORE) {
377 /* EAL thread with valid lcore_id */
378 tim_lcore = rte_get_next_lcore(
379 priv_timer[lcore_id].prev_lcore,
381 priv_timer[lcore_id].prev_lcore = tim_lcore;
383 /* non-EAL thread do not run rte_timer_manage(),
384 * so schedule the timer on the first enabled lcore. */
385 tim_lcore = rte_get_next_lcore(LCORE_ID_ANY, 0, 1);
388 /* wait that the timer is in correct status before update,
389 * and mark it as being configured */
390 ret = timer_set_config_state(tim, &prev_status);
394 __TIMER_STAT_ADD(reset, 1);
395 if (prev_status.state == RTE_TIMER_RUNNING &&
396 lcore_id < RTE_MAX_LCORE) {
397 priv_timer[lcore_id].updated = 1;
400 /* remove it from list */
401 if (prev_status.state == RTE_TIMER_PENDING) {
402 timer_del(tim, prev_status, local_is_locked);
403 __TIMER_STAT_ADD(pending, -1);
406 tim->period = period;
407 tim->expire = expire;
411 __TIMER_STAT_ADD(pending, 1);
412 timer_add(tim, tim_lcore, local_is_locked);
414 /* update state: as we are in CONFIG state, only us can modify
415 * the state so we don't need to use cmpset() here */
417 status.state = RTE_TIMER_PENDING;
418 status.owner = (int16_t)tim_lcore;
419 tim->status.u32 = status.u32;
424 /* Reset and start the timer associated with the timer handle tim */
426 rte_timer_reset(struct rte_timer *tim, uint64_t ticks,
427 enum rte_timer_type type, unsigned tim_lcore,
428 rte_timer_cb_t fct, void *arg)
430 uint64_t cur_time = rte_get_timer_cycles();
433 if (unlikely((tim_lcore != (unsigned)LCORE_ID_ANY) &&
434 !(rte_lcore_is_enabled(tim_lcore) ||
435 rte_lcore_has_role(tim_lcore, ROLE_SERVICE))))
438 if (type == PERIODICAL)
443 return __rte_timer_reset(tim, cur_time + ticks, period, tim_lcore,
447 /* loop until rte_timer_reset() succeed */
449 rte_timer_reset_sync(struct rte_timer *tim, uint64_t ticks,
450 enum rte_timer_type type, unsigned tim_lcore,
451 rte_timer_cb_t fct, void *arg)
453 while (rte_timer_reset(tim, ticks, type, tim_lcore,
458 /* Stop the timer associated with the timer handle tim */
460 rte_timer_stop(struct rte_timer *tim)
462 union rte_timer_status prev_status, status;
463 unsigned lcore_id = rte_lcore_id();
466 /* wait that the timer is in correct status before update,
467 * and mark it as being configured */
468 ret = timer_set_config_state(tim, &prev_status);
472 __TIMER_STAT_ADD(stop, 1);
473 if (prev_status.state == RTE_TIMER_RUNNING &&
474 lcore_id < RTE_MAX_LCORE) {
475 priv_timer[lcore_id].updated = 1;
478 /* remove it from list */
479 if (prev_status.state == RTE_TIMER_PENDING) {
480 timer_del(tim, prev_status, 0);
481 __TIMER_STAT_ADD(pending, -1);
484 /* mark timer as stopped */
486 status.state = RTE_TIMER_STOP;
487 status.owner = RTE_TIMER_NO_OWNER;
488 tim->status.u32 = status.u32;
493 /* loop until rte_timer_stop() succeed */
495 rte_timer_stop_sync(struct rte_timer *tim)
497 while (rte_timer_stop(tim) != 0)
501 /* Test the PENDING status of the timer handle tim */
503 rte_timer_pending(struct rte_timer *tim)
505 return tim->status.state == RTE_TIMER_PENDING;
508 /* must be called periodically, run all timer that expired */
509 void rte_timer_manage(void)
511 union rte_timer_status status;
512 struct rte_timer *tim, *next_tim;
513 struct rte_timer *run_first_tim, **pprev;
514 unsigned lcore_id = rte_lcore_id();
515 struct rte_timer *prev[MAX_SKIPLIST_DEPTH + 1];
519 /* timer manager only runs on EAL thread with valid lcore_id */
520 assert(lcore_id < RTE_MAX_LCORE);
522 __TIMER_STAT_ADD(manage, 1);
523 /* optimize for the case where per-cpu list is empty */
524 if (priv_timer[lcore_id].pending_head.sl_next[0] == NULL)
526 cur_time = rte_get_timer_cycles();
529 /* on 64-bit the value cached in the pending_head.expired will be
530 * updated atomically, so we can consult that for a quick check here
531 * outside the lock */
532 if (likely(priv_timer[lcore_id].pending_head.expire > cur_time))
536 /* browse ordered list, add expired timers in 'expired' list */
537 rte_spinlock_lock(&priv_timer[lcore_id].list_lock);
539 /* if nothing to do just unlock and return */
540 if (priv_timer[lcore_id].pending_head.sl_next[0] == NULL ||
541 priv_timer[lcore_id].pending_head.sl_next[0]->expire > cur_time) {
542 rte_spinlock_unlock(&priv_timer[lcore_id].list_lock);
546 /* save start of list of expired timers */
547 tim = priv_timer[lcore_id].pending_head.sl_next[0];
549 /* break the existing list at current time point */
550 timer_get_prev_entries(cur_time, lcore_id, prev);
551 for (i = priv_timer[lcore_id].curr_skiplist_depth -1; i >= 0; i--) {
552 if (prev[i] == &priv_timer[lcore_id].pending_head)
554 priv_timer[lcore_id].pending_head.sl_next[i] =
556 if (prev[i]->sl_next[i] == NULL)
557 priv_timer[lcore_id].curr_skiplist_depth--;
558 prev[i] ->sl_next[i] = NULL;
561 /* transition run-list from PENDING to RUNNING */
563 pprev = &run_first_tim;
565 for ( ; tim != NULL; tim = next_tim) {
566 next_tim = tim->sl_next[0];
568 ret = timer_set_running_state(tim);
569 if (likely(ret == 0)) {
570 pprev = &tim->sl_next[0];
572 /* another core is trying to re-config this one,
573 * remove it from local expired list
579 /* update the next to expire timer value */
580 priv_timer[lcore_id].pending_head.expire =
581 (priv_timer[lcore_id].pending_head.sl_next[0] == NULL) ? 0 :
582 priv_timer[lcore_id].pending_head.sl_next[0]->expire;
584 rte_spinlock_unlock(&priv_timer[lcore_id].list_lock);
586 /* now scan expired list and call callbacks */
587 for (tim = run_first_tim; tim != NULL; tim = next_tim) {
588 next_tim = tim->sl_next[0];
589 priv_timer[lcore_id].updated = 0;
590 priv_timer[lcore_id].running_tim = tim;
592 /* execute callback function with list unlocked */
593 tim->f(tim, tim->arg);
595 __TIMER_STAT_ADD(pending, -1);
596 /* the timer was stopped or reloaded by the callback
597 * function, we have nothing to do here */
598 if (priv_timer[lcore_id].updated == 1)
601 if (tim->period == 0) {
602 /* remove from done list and mark timer as stopped */
603 status.state = RTE_TIMER_STOP;
604 status.owner = RTE_TIMER_NO_OWNER;
606 tim->status.u32 = status.u32;
609 /* keep it in list and mark timer as pending */
610 rte_spinlock_lock(&priv_timer[lcore_id].list_lock);
611 status.state = RTE_TIMER_PENDING;
612 __TIMER_STAT_ADD(pending, 1);
613 status.owner = (int16_t)lcore_id;
615 tim->status.u32 = status.u32;
616 __rte_timer_reset(tim, tim->expire + tim->period,
617 tim->period, lcore_id, tim->f, tim->arg, 1);
618 rte_spinlock_unlock(&priv_timer[lcore_id].list_lock);
621 priv_timer[lcore_id].running_tim = NULL;
624 /* dump statistics about timers */
625 void rte_timer_dump_stats(FILE *f)
627 #ifdef RTE_LIBRTE_TIMER_DEBUG
628 struct rte_timer_debug_stats sum;
631 memset(&sum, 0, sizeof(sum));
632 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
633 sum.reset += priv_timer[lcore_id].stats.reset;
634 sum.stop += priv_timer[lcore_id].stats.stop;
635 sum.manage += priv_timer[lcore_id].stats.manage;
636 sum.pending += priv_timer[lcore_id].stats.pending;
638 fprintf(f, "Timer statistics:\n");
639 fprintf(f, " reset = %"PRIu64"\n", sum.reset);
640 fprintf(f, " stop = %"PRIu64"\n", sum.stop);
641 fprintf(f, " manage = %"PRIu64"\n", sum.manage);
642 fprintf(f, " pending = %"PRIu64"\n", sum.pending);
644 fprintf(f, "No timer statistics, RTE_LIBRTE_TIMER_DEBUG is disabled\n");