1 /* 2 * BSD LICENSE 3 * 4 * Copyright (C) Cavium, Inc 2017. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 10 * * Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * * Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in 14 * the documentation and/or other materials provided with the 15 * distribution. 16 * * Neither the name of Cavium, Inc nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 23 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 24 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 25 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 26 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 27 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 28 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 29 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 30 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 #include "test_perf_common.h" 34 35 /* See http://dpdk.org/doc/guides/tools/testeventdev.html for test details */ 36 37 static inline int 38 perf_queue_nb_event_queues(struct evt_options *opt) 39 { 40 /* nb_queues = number of producers * number of stages */ 41 return evt_nr_active_lcores(opt->plcores) * opt->nb_stages; 42 } 43 44 static inline __attribute__((always_inline)) void 45 mark_fwd_latency(struct rte_event *const ev, 46 const uint8_t nb_stages) 47 { 48 if (unlikely((ev->queue_id % nb_stages) == 0)) { 49 struct perf_elt *const m = ev->event_ptr; 50 51 m->timestamp = rte_get_timer_cycles(); 52 } 53 } 54 55 static inline __attribute__((always_inline)) void 56 fwd_event(struct rte_event *const ev, uint8_t *const sched_type_list, 57 const uint8_t nb_stages) 58 { 59 ev->queue_id++; 60 ev->sched_type = sched_type_list[ev->queue_id % nb_stages]; 61 ev->op = RTE_EVENT_OP_FORWARD; 62 ev->event_type = RTE_EVENT_TYPE_CPU; 63 } 64 65 static int 66 perf_queue_worker(void *arg, const int enable_fwd_latency) 67 { 68 PERF_WORKER_INIT; 69 struct rte_event ev; 70 71 while (t->done == false) { 72 uint16_t event = rte_event_dequeue_burst(dev, port, &ev, 1, 0); 73 74 if (!event) { 75 rte_pause(); 76 continue; 77 } 78 if (enable_fwd_latency) 79 /* first q in pipeline, mark timestamp to compute fwd latency */ 80 mark_fwd_latency(&ev, nb_stages); 81 82 /* last stage in pipeline */ 83 if (unlikely((ev.queue_id % nb_stages) == laststage)) { 84 if (enable_fwd_latency) 85 cnt = perf_process_last_stage_latency(pool, 86 &ev, w, bufs, sz, cnt); 87 else 88 cnt = perf_process_last_stage(pool, 89 &ev, w, bufs, sz, cnt); 90 } else { 91 fwd_event(&ev, sched_type_list, nb_stages); 92 while (rte_event_enqueue_burst(dev, port, &ev, 1) != 1) 93 rte_pause(); 94 } 95 } 96 return 0; 97 } 98 99 static int 100 perf_queue_worker_burst(void *arg, const int enable_fwd_latency) 101 { 102 PERF_WORKER_INIT; 103 uint16_t i; 104 /* +1 to avoid prefetch out of array check */ 105 struct rte_event ev[BURST_SIZE + 1]; 106 107 while (t->done == false) { 108 uint16_t const nb_rx = rte_event_dequeue_burst(dev, port, ev, 109 BURST_SIZE, 0); 110 111 if (!nb_rx) { 112 rte_pause(); 113 continue; 114 } 115 116 for (i = 0; i < nb_rx; i++) { 117 if (enable_fwd_latency) { 118 rte_prefetch0(ev[i+1].event_ptr); 119 /* first queue in pipeline. 120 * mark time stamp to compute fwd latency 121 */ 122 mark_fwd_latency(&ev[i], nb_stages); 123 } 124 /* last stage in pipeline */ 125 if (unlikely((ev[i].queue_id % nb_stages) == 126 laststage)) { 127 if (enable_fwd_latency) 128 cnt = perf_process_last_stage_latency( 129 pool, &ev[i], w, bufs, sz, cnt); 130 else 131 cnt = perf_process_last_stage(pool, 132 &ev[i], w, bufs, sz, cnt); 133 134 ev[i].op = RTE_EVENT_OP_RELEASE; 135 } else { 136 fwd_event(&ev[i], sched_type_list, nb_stages); 137 } 138 } 139 140 uint16_t enq; 141 142 enq = rte_event_enqueue_burst(dev, port, ev, nb_rx); 143 while (enq < nb_rx) { 144 enq += rte_event_enqueue_burst(dev, port, 145 ev + enq, nb_rx - enq); 146 } 147 } 148 return 0; 149 } 150 151 static int 152 worker_wrapper(void *arg) 153 { 154 struct worker_data *w = arg; 155 struct evt_options *opt = w->t->opt; 156 157 const bool burst = evt_has_burst_mode(w->dev_id); 158 const int fwd_latency = opt->fwd_latency; 159 160 /* allow compiler to optimize */ 161 if (!burst && !fwd_latency) 162 return perf_queue_worker(arg, 0); 163 else if (!burst && fwd_latency) 164 return perf_queue_worker(arg, 1); 165 else if (burst && !fwd_latency) 166 return perf_queue_worker_burst(arg, 0); 167 else if (burst && fwd_latency) 168 return perf_queue_worker_burst(arg, 1); 169 170 rte_panic("invalid worker\n"); 171 } 172 173 static int 174 perf_queue_launch_lcores(struct evt_test *test, struct evt_options *opt) 175 { 176 return perf_launch_lcores(test, opt, worker_wrapper); 177 } 178 179 static int 180 perf_queue_eventdev_setup(struct evt_test *test, struct evt_options *opt) 181 { 182 uint8_t queue; 183 int nb_stages = opt->nb_stages; 184 int ret; 185 186 const struct rte_event_dev_config config = { 187 .nb_event_queues = perf_queue_nb_event_queues(opt), 188 .nb_event_ports = perf_nb_event_ports(opt), 189 .nb_events_limit = 4096, 190 .nb_event_queue_flows = opt->nb_flows, 191 .nb_event_port_dequeue_depth = 128, 192 .nb_event_port_enqueue_depth = 128, 193 }; 194 195 ret = rte_event_dev_configure(opt->dev_id, &config); 196 if (ret) { 197 evt_err("failed to configure eventdev %d", opt->dev_id); 198 return ret; 199 } 200 201 struct rte_event_queue_conf q_conf = { 202 .priority = RTE_EVENT_DEV_PRIORITY_NORMAL, 203 .nb_atomic_flows = opt->nb_flows, 204 .nb_atomic_order_sequences = opt->nb_flows, 205 }; 206 /* queue configurations */ 207 for (queue = 0; queue < perf_queue_nb_event_queues(opt); queue++) { 208 q_conf.schedule_type = 209 (opt->sched_type_list[queue % nb_stages]); 210 211 if (opt->q_priority) { 212 uint8_t stage_pos = queue % nb_stages; 213 /* Configure event queues(stage 0 to stage n) with 214 * RTE_EVENT_DEV_PRIORITY_LOWEST to 215 * RTE_EVENT_DEV_PRIORITY_HIGHEST. 216 */ 217 uint8_t step = RTE_EVENT_DEV_PRIORITY_LOWEST / 218 (nb_stages - 1); 219 /* Higher prio for the queues closer to last stage */ 220 q_conf.priority = RTE_EVENT_DEV_PRIORITY_LOWEST - 221 (step * stage_pos); 222 } 223 ret = rte_event_queue_setup(opt->dev_id, queue, &q_conf); 224 if (ret) { 225 evt_err("failed to setup queue=%d", queue); 226 return ret; 227 } 228 } 229 230 ret = perf_event_dev_port_setup(test, opt, nb_stages /* stride */, 231 perf_queue_nb_event_queues(opt)); 232 if (ret) 233 return ret; 234 235 ret = evt_service_setup(opt->dev_id); 236 if (ret) { 237 evt_err("No service lcore found to run event dev."); 238 return ret; 239 } 240 241 ret = rte_event_dev_start(opt->dev_id); 242 if (ret) { 243 evt_err("failed to start eventdev %d", opt->dev_id); 244 return ret; 245 } 246 247 return 0; 248 } 249 250 static void 251 perf_queue_opt_dump(struct evt_options *opt) 252 { 253 evt_dump_fwd_latency(opt); 254 perf_opt_dump(opt, perf_queue_nb_event_queues(opt)); 255 } 256 257 static int 258 perf_queue_opt_check(struct evt_options *opt) 259 { 260 return perf_opt_check(opt, perf_queue_nb_event_queues(opt)); 261 } 262 263 static bool 264 perf_queue_capability_check(struct evt_options *opt) 265 { 266 struct rte_event_dev_info dev_info; 267 268 rte_event_dev_info_get(opt->dev_id, &dev_info); 269 if (dev_info.max_event_queues < perf_queue_nb_event_queues(opt) || 270 dev_info.max_event_ports < perf_nb_event_ports(opt)) { 271 evt_err("not enough eventdev queues=%d/%d or ports=%d/%d", 272 perf_queue_nb_event_queues(opt), 273 dev_info.max_event_queues, 274 perf_nb_event_ports(opt), dev_info.max_event_ports); 275 } 276 277 return true; 278 } 279 280 static const struct evt_test_ops perf_queue = { 281 .cap_check = perf_queue_capability_check, 282 .opt_check = perf_queue_opt_check, 283 .opt_dump = perf_queue_opt_dump, 284 .test_setup = perf_test_setup, 285 .mempool_setup = perf_mempool_setup, 286 .eventdev_setup = perf_queue_eventdev_setup, 287 .launch_lcores = perf_queue_launch_lcores, 288 .eventdev_destroy = perf_eventdev_destroy, 289 .mempool_destroy = perf_mempool_destroy, 290 .test_result = perf_test_result, 291 .test_destroy = perf_test_destroy, 292 }; 293 294 EVT_TEST_REGISTER(perf_queue); 295