4 * Copyright(c) 2017 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.
41 #include <rte_service.h>
42 #include "include/rte_service_component.h"
45 #include <rte_lcore.h>
46 #include <rte_common.h>
47 #include <rte_debug.h>
48 #include <rte_cycles.h>
49 #include <rte_atomic.h>
50 #include <rte_memory.h>
51 #include <rte_malloc.h>
53 #define RTE_SERVICE_NUM_MAX 64
55 #define SERVICE_F_REGISTERED (1 << 0)
56 #define SERVICE_F_STATS_ENABLED (1 << 1)
58 /* runstates for services and lcores, denoting if they are active or not */
59 #define RUNSTATE_STOPPED 0
60 #define RUNSTATE_RUNNING 1
62 /* internal representation of a service */
63 struct rte_service_spec_impl {
64 /* public part of the struct */
65 struct rte_service_spec spec;
67 /* atomic lock that when set indicates a service core is currently
68 * running this service callback. When not set, a core may take the
69 * lock and then run the service callback.
71 rte_atomic32_t execute_lock;
73 /* API set/get-able variables */
75 uint8_t internal_flags;
77 /* per service statistics */
78 uint32_t num_mapped_cores;
80 uint64_t cycles_spent;
81 } __rte_cache_aligned;
83 /* the internal values of a service core */
85 /* map of services IDs are run on this core */
86 uint64_t service_mask;
87 uint8_t runstate; /* running or stopped */
88 uint8_t is_service_core; /* set if core is currently a service core */
90 /* extreme statistics */
91 uint64_t calls_per_service[RTE_SERVICE_NUM_MAX];
92 } __rte_cache_aligned;
94 static uint32_t rte_service_count;
95 static struct rte_service_spec_impl *rte_services;
96 static struct core_state *lcore_states;
97 static uint32_t rte_service_library_initialized;
99 int32_t rte_service_init(void)
101 if (rte_service_library_initialized) {
102 printf("service library init() called, init flag %d\n",
103 rte_service_library_initialized);
107 rte_services = rte_calloc("rte_services", RTE_SERVICE_NUM_MAX,
108 sizeof(struct rte_service_spec_impl),
109 RTE_CACHE_LINE_SIZE);
111 printf("error allocating rte services array\n");
115 lcore_states = rte_calloc("rte_service_core_states", RTE_MAX_LCORE,
116 sizeof(struct core_state), RTE_CACHE_LINE_SIZE);
118 printf("error allocating core states array\n");
124 struct rte_config *cfg = rte_eal_get_configuration();
125 for (i = 0; i < RTE_MAX_LCORE; i++) {
126 if (lcore_config[i].core_role == ROLE_SERVICE) {
127 if ((unsigned int)i == cfg->master_lcore)
129 rte_service_lcore_add(i);
134 rte_service_library_initialized = 1;
138 /* returns 1 if service is registered and has not been unregistered
139 * Returns 0 if service never registered, or has been unregistered
142 service_valid(uint32_t id)
144 return !!(rte_services[id].internal_flags & SERVICE_F_REGISTERED);
147 /* returns 1 if statistics should be colleced for service
148 * Returns 0 if statistics should not be collected for service
151 service_stats_enabled(struct rte_service_spec_impl *impl)
153 return !!(impl->internal_flags & SERVICE_F_STATS_ENABLED);
157 service_mt_safe(struct rte_service_spec_impl *s)
159 return s->spec.capabilities & RTE_SERVICE_CAP_MT_SAFE;
162 int32_t rte_service_set_stats_enable(struct rte_service_spec *service,
165 struct rte_service_spec_impl *impl =
166 (struct rte_service_spec_impl *)service;
171 impl->internal_flags |= SERVICE_F_STATS_ENABLED;
173 impl->internal_flags &= ~(SERVICE_F_STATS_ENABLED);
179 rte_service_get_count(void)
181 return rte_service_count;
184 struct rte_service_spec *
185 rte_service_get_by_id(uint32_t id)
187 struct rte_service_spec *service = NULL;
188 if (id < rte_service_count)
189 service = (struct rte_service_spec *)&rte_services[id];
194 struct rte_service_spec *rte_service_get_by_name(const char *name)
196 struct rte_service_spec *service = NULL;
198 for (i = 0; i < RTE_SERVICE_NUM_MAX; i++) {
199 if (service_valid(i) &&
200 strcmp(name, rte_services[i].spec.name) == 0) {
201 service = (struct rte_service_spec *)&rte_services[i];
210 rte_service_get_name(const struct rte_service_spec *service)
212 return service->name;
216 rte_service_probe_capability(const struct rte_service_spec *service,
219 return service->capabilities & capability;
223 rte_service_is_running(const struct rte_service_spec *spec)
225 const struct rte_service_spec_impl *impl =
226 (const struct rte_service_spec_impl *)spec;
230 return (impl->runstate == RUNSTATE_RUNNING) &&
231 (impl->num_mapped_cores > 0);
235 rte_service_register(const struct rte_service_spec *spec)
238 int32_t free_slot = -1;
240 if (spec->callback == NULL || strlen(spec->name) == 0)
243 for (i = 0; i < RTE_SERVICE_NUM_MAX; i++) {
244 if (!service_valid(i)) {
250 if ((free_slot < 0) || (i == RTE_SERVICE_NUM_MAX))
253 struct rte_service_spec_impl *s = &rte_services[free_slot];
255 s->internal_flags |= SERVICE_F_REGISTERED;
264 rte_service_unregister(struct rte_service_spec *spec)
266 struct rte_service_spec_impl *s = NULL;
267 struct rte_service_spec_impl *spec_impl =
268 (struct rte_service_spec_impl *)spec;
272 for (i = 0; i < RTE_SERVICE_NUM_MAX; i++) {
273 if (&rte_services[i] == spec_impl) {
286 s->internal_flags &= ~(SERVICE_F_REGISTERED);
288 for (i = 0; i < RTE_MAX_LCORE; i++)
289 lcore_states[i].service_mask &= ~(1 << service_id);
291 memset(&rte_services[service_id], 0,
292 sizeof(struct rte_service_spec_impl));
298 rte_service_start(struct rte_service_spec *service)
300 struct rte_service_spec_impl *s =
301 (struct rte_service_spec_impl *)service;
302 s->runstate = RUNSTATE_RUNNING;
308 rte_service_stop(struct rte_service_spec *service)
310 struct rte_service_spec_impl *s =
311 (struct rte_service_spec_impl *)service;
312 s->runstate = RUNSTATE_STOPPED;
318 rte_service_runner_func(void *arg)
322 const int lcore = rte_lcore_id();
323 struct core_state *cs = &lcore_states[lcore];
325 while (lcore_states[lcore].runstate == RUNSTATE_RUNNING) {
326 const uint64_t service_mask = cs->service_mask;
327 for (i = 0; i < rte_service_count; i++) {
328 struct rte_service_spec_impl *s = &rte_services[i];
329 if (s->runstate != RUNSTATE_RUNNING ||
330 !(service_mask & (1 << i)))
333 /* check do we need cmpset, if MT safe or <= 1 core
334 * mapped, atomic ops are not required.
336 const int need_cmpset = !((service_mt_safe(s) == 0) &&
337 (s->num_mapped_cores > 1));
338 uint32_t *lock = (uint32_t *)&s->execute_lock;
340 if (need_cmpset || rte_atomic32_cmpset(lock, 0, 1)) {
341 void *userdata = s->spec.callback_userdata;
343 if (service_stats_enabled(s)) {
344 uint64_t start = rte_rdtsc();
345 s->spec.callback(userdata);
346 uint64_t end = rte_rdtsc();
347 s->cycles_spent += end - start;
348 cs->calls_per_service[i]++;
351 s->spec.callback(userdata);
354 rte_atomic32_clear(&s->execute_lock);
361 lcore_config[lcore].state = WAIT;
367 rte_service_lcore_count(void)
371 for (i = 0; i < RTE_MAX_LCORE; i++)
372 count += lcore_states[i].is_service_core;
377 rte_service_lcore_list(uint32_t array[], uint32_t n)
379 uint32_t count = rte_service_lcore_count();
388 for (i = 0; i < RTE_MAX_LCORE; i++) {
389 struct core_state *cs = &lcore_states[i];
390 if (cs->is_service_core) {
400 rte_service_start_with_defaults(void)
402 /* create a default mapping from cores to services, then start the
403 * services to make them transparent to unaware applications.
407 uint32_t count = rte_service_get_count();
409 int32_t lcore_iter = 0;
410 uint32_t ids[RTE_MAX_LCORE];
411 int32_t lcore_count = rte_service_lcore_list(ids, RTE_MAX_LCORE);
413 if (lcore_count == 0)
416 for (i = 0; (int)i < lcore_count; i++)
417 rte_service_lcore_start(ids[i]);
419 for (i = 0; i < count; i++) {
420 struct rte_service_spec *s = rte_service_get_by_id(i);
424 /* do 1:1 core mapping here, with each service getting
425 * assigned a single core by default. Adding multiple services
426 * should multiplex to a single core, or 1:1 if there are the
427 * same amount of services as service-cores
429 ret = rte_service_enable_on_lcore(s, ids[lcore_iter]);
434 if (lcore_iter >= lcore_count)
437 ret = rte_service_start(s);
446 service_update(struct rte_service_spec *service, uint32_t lcore,
447 uint32_t *set, uint32_t *enabled)
452 for (i = 0; i < RTE_SERVICE_NUM_MAX; i++) {
453 if ((struct rte_service_spec *)&rte_services[i] == service &&
460 if (sid == -1 || lcore >= RTE_MAX_LCORE)
463 if (!lcore_states[lcore].is_service_core)
468 lcore_states[lcore].service_mask |= (1 << sid);
469 rte_services[sid].num_mapped_cores++;
471 lcore_states[lcore].service_mask &= ~(1 << sid);
472 rte_services[sid].num_mapped_cores--;
477 *enabled = (lcore_states[lcore].service_mask & (1 << sid));
484 int32_t rte_service_get_enabled_on_lcore(struct rte_service_spec *service,
488 int ret = service_update(service, lcore, 0, &enabled);
495 rte_service_enable_on_lcore(struct rte_service_spec *service, uint32_t lcore)
498 return service_update(service, lcore, &on, 0);
502 rte_service_disable_on_lcore(struct rte_service_spec *service, uint32_t lcore)
505 return service_update(service, lcore, &off, 0);
508 int32_t rte_service_lcore_reset_all(void)
510 /* loop over cores, reset all to mask 0 */
512 for (i = 0; i < RTE_MAX_LCORE; i++) {
513 lcore_states[i].service_mask = 0;
514 lcore_states[i].is_service_core = 0;
515 lcore_states[i].runstate = RUNSTATE_STOPPED;
517 for (i = 0; i < RTE_SERVICE_NUM_MAX; i++)
518 rte_services[i].num_mapped_cores = 0;
526 set_lcore_state(uint32_t lcore, int32_t state)
528 /* mark core state in hugepage backed config */
529 struct rte_config *cfg = rte_eal_get_configuration();
530 cfg->lcore_role[lcore] = state;
532 /* mark state in process local lcore_config */
533 lcore_config[lcore].core_role = state;
535 /* update per-lcore optimized state tracking */
536 lcore_states[lcore].is_service_core = (state == ROLE_SERVICE);
540 rte_service_lcore_add(uint32_t lcore)
542 if (lcore >= RTE_MAX_LCORE)
544 if (lcore_states[lcore].is_service_core)
547 set_lcore_state(lcore, ROLE_SERVICE);
549 /* ensure that after adding a core the mask and state are defaults */
550 lcore_states[lcore].service_mask = 0;
551 lcore_states[lcore].runstate = RUNSTATE_STOPPED;
558 rte_service_lcore_del(uint32_t lcore)
560 if (lcore >= RTE_MAX_LCORE)
563 struct core_state *cs = &lcore_states[lcore];
564 if (!cs->is_service_core)
567 if (cs->runstate != RUNSTATE_STOPPED)
570 set_lcore_state(lcore, ROLE_RTE);
577 rte_service_lcore_start(uint32_t lcore)
579 if (lcore >= RTE_MAX_LCORE)
582 struct core_state *cs = &lcore_states[lcore];
583 if (!cs->is_service_core)
586 if (cs->runstate == RUNSTATE_RUNNING)
589 /* set core to run state first, and then launch otherwise it will
590 * return immediately as runstate keeps it in the service poll loop
592 lcore_states[lcore].runstate = RUNSTATE_RUNNING;
594 int ret = rte_eal_remote_launch(rte_service_runner_func, 0, lcore);
595 /* returns -EBUSY if the core is already launched, 0 on success */
600 rte_service_lcore_stop(uint32_t lcore)
602 if (lcore >= RTE_MAX_LCORE)
605 if (lcore_states[lcore].runstate == RUNSTATE_STOPPED)
609 for (i = 0; i < RTE_SERVICE_NUM_MAX; i++) {
610 int32_t enabled = lcore_states[i].service_mask & (1 << i);
611 int32_t service_running = rte_services[i].runstate !=
613 int32_t only_core = rte_services[i].num_mapped_cores == 1;
615 /* if the core is mapped, and the service is running, and this
616 * is the only core that is mapped, the service would cease to
617 * run if this core stopped, so fail instead.
619 if (enabled && service_running && only_core)
623 lcore_states[lcore].runstate = RUNSTATE_STOPPED;
629 rte_service_dump_one(FILE *f, struct rte_service_spec_impl *s,
630 uint64_t all_cycles, uint32_t reset)
632 /* avoid divide by zero */
640 fprintf(f, " %s: stats %d\tcalls %"PRIu64"\tcycles %"
641 PRIu64"\tavg: %"PRIu64"\n",
642 s->spec.name, service_stats_enabled(s), s->calls,
643 s->cycles_spent, s->cycles_spent / calls);
652 service_dump_calls_per_lcore(FILE *f, uint32_t lcore, uint32_t reset)
655 struct core_state *cs = &lcore_states[lcore];
657 fprintf(f, "%02d\t", lcore);
658 for (i = 0; i < RTE_SERVICE_NUM_MAX; i++) {
659 if (!service_valid(i))
661 fprintf(f, "%"PRIu64"\t", cs->calls_per_service[i]);
663 cs->calls_per_service[i] = 0;
668 int32_t rte_service_dump(FILE *f, struct rte_service_spec *service)
672 uint64_t total_cycles = 0;
673 for (i = 0; i < rte_service_count; i++) {
674 if (!service_valid(i))
676 total_cycles += rte_services[i].cycles_spent;
680 struct rte_service_spec_impl *s =
681 (struct rte_service_spec_impl *)service;
682 fprintf(f, "Service %s Summary\n", s->spec.name);
684 rte_service_dump_one(f, s, total_cycles, reset);
688 fprintf(f, "Services Summary\n");
689 for (i = 0; i < rte_service_count; i++) {
691 rte_service_dump_one(f, &rte_services[i], total_cycles, reset);
694 fprintf(f, "Service Cores Summary\n");
695 for (i = 0; i < RTE_MAX_LCORE; i++) {
696 if (lcore_config[i].core_role != ROLE_SERVICE)
700 service_dump_calls_per_lcore(f, i, reset);