ring: make bulk and burst return values consistent
[dpdk.git] / examples / load_balancer / runtime.c
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5  *   All rights reserved.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
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
16  *       distribution.
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.
20  *
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.
32  */
33
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <stdint.h>
37 #include <inttypes.h>
38 #include <sys/types.h>
39 #include <string.h>
40 #include <sys/queue.h>
41 #include <stdarg.h>
42 #include <errno.h>
43 #include <getopt.h>
44
45 #include <rte_common.h>
46 #include <rte_byteorder.h>
47 #include <rte_log.h>
48 #include <rte_memory.h>
49 #include <rte_memcpy.h>
50 #include <rte_memzone.h>
51 #include <rte_eal.h>
52 #include <rte_per_lcore.h>
53 #include <rte_launch.h>
54 #include <rte_atomic.h>
55 #include <rte_cycles.h>
56 #include <rte_prefetch.h>
57 #include <rte_lcore.h>
58 #include <rte_per_lcore.h>
59 #include <rte_branch_prediction.h>
60 #include <rte_interrupts.h>
61 #include <rte_pci.h>
62 #include <rte_random.h>
63 #include <rte_debug.h>
64 #include <rte_ether.h>
65 #include <rte_ethdev.h>
66 #include <rte_ring.h>
67 #include <rte_mempool.h>
68 #include <rte_mbuf.h>
69 #include <rte_ip.h>
70 #include <rte_tcp.h>
71 #include <rte_lpm.h>
72
73 #include "main.h"
74
75 #ifndef APP_LCORE_IO_FLUSH
76 #define APP_LCORE_IO_FLUSH           1000000
77 #endif
78
79 #ifndef APP_LCORE_WORKER_FLUSH
80 #define APP_LCORE_WORKER_FLUSH       1000000
81 #endif
82
83 #ifndef APP_STATS
84 #define APP_STATS                    1000000
85 #endif
86
87 #define APP_IO_RX_DROP_ALL_PACKETS   0
88 #define APP_WORKER_DROP_ALL_PACKETS  0
89 #define APP_IO_TX_DROP_ALL_PACKETS   0
90
91 #ifndef APP_IO_RX_PREFETCH_ENABLE
92 #define APP_IO_RX_PREFETCH_ENABLE    1
93 #endif
94
95 #ifndef APP_WORKER_PREFETCH_ENABLE
96 #define APP_WORKER_PREFETCH_ENABLE   1
97 #endif
98
99 #ifndef APP_IO_TX_PREFETCH_ENABLE
100 #define APP_IO_TX_PREFETCH_ENABLE    1
101 #endif
102
103 #if APP_IO_RX_PREFETCH_ENABLE
104 #define APP_IO_RX_PREFETCH0(p)       rte_prefetch0(p)
105 #define APP_IO_RX_PREFETCH1(p)       rte_prefetch1(p)
106 #else
107 #define APP_IO_RX_PREFETCH0(p)
108 #define APP_IO_RX_PREFETCH1(p)
109 #endif
110
111 #if APP_WORKER_PREFETCH_ENABLE
112 #define APP_WORKER_PREFETCH0(p)      rte_prefetch0(p)
113 #define APP_WORKER_PREFETCH1(p)      rte_prefetch1(p)
114 #else
115 #define APP_WORKER_PREFETCH0(p)
116 #define APP_WORKER_PREFETCH1(p)
117 #endif
118
119 #if APP_IO_TX_PREFETCH_ENABLE
120 #define APP_IO_TX_PREFETCH0(p)       rte_prefetch0(p)
121 #define APP_IO_TX_PREFETCH1(p)       rte_prefetch1(p)
122 #else
123 #define APP_IO_TX_PREFETCH0(p)
124 #define APP_IO_TX_PREFETCH1(p)
125 #endif
126
127 static inline void
128 app_lcore_io_rx_buffer_to_send (
129         struct app_lcore_params_io *lp,
130         uint32_t worker,
131         struct rte_mbuf *mbuf,
132         uint32_t bsz)
133 {
134         uint32_t pos;
135         int ret;
136
137         pos = lp->rx.mbuf_out[worker].n_mbufs;
138         lp->rx.mbuf_out[worker].array[pos ++] = mbuf;
139         if (likely(pos < bsz)) {
140                 lp->rx.mbuf_out[worker].n_mbufs = pos;
141                 return;
142         }
143
144         ret = rte_ring_sp_enqueue_bulk(
145                 lp->rx.rings[worker],
146                 (void **) lp->rx.mbuf_out[worker].array,
147                 bsz);
148
149         if (unlikely(ret == 0)) {
150                 uint32_t k;
151                 for (k = 0; k < bsz; k ++) {
152                         struct rte_mbuf *m = lp->rx.mbuf_out[worker].array[k];
153                         rte_pktmbuf_free(m);
154                 }
155         }
156
157         lp->rx.mbuf_out[worker].n_mbufs = 0;
158         lp->rx.mbuf_out_flush[worker] = 0;
159
160 #if APP_STATS
161         lp->rx.rings_iters[worker] ++;
162         if (likely(ret == 0)) {
163                 lp->rx.rings_count[worker] ++;
164         }
165         if (unlikely(lp->rx.rings_iters[worker] == APP_STATS)) {
166                 unsigned lcore = rte_lcore_id();
167
168                 printf("\tI/O RX %u out (worker %u): enq success rate = %.2f\n",
169                         lcore,
170                         (unsigned)worker,
171                         ((double) lp->rx.rings_count[worker]) / ((double) lp->rx.rings_iters[worker]));
172                 lp->rx.rings_iters[worker] = 0;
173                 lp->rx.rings_count[worker] = 0;
174         }
175 #endif
176 }
177
178 static inline void
179 app_lcore_io_rx(
180         struct app_lcore_params_io *lp,
181         uint32_t n_workers,
182         uint32_t bsz_rd,
183         uint32_t bsz_wr,
184         uint8_t pos_lb)
185 {
186         struct rte_mbuf *mbuf_1_0, *mbuf_1_1, *mbuf_2_0, *mbuf_2_1;
187         uint8_t *data_1_0, *data_1_1 = NULL;
188         uint32_t i;
189
190         for (i = 0; i < lp->rx.n_nic_queues; i ++) {
191                 uint8_t port = lp->rx.nic_queues[i].port;
192                 uint8_t queue = lp->rx.nic_queues[i].queue;
193                 uint32_t n_mbufs, j;
194
195                 n_mbufs = rte_eth_rx_burst(
196                         port,
197                         queue,
198                         lp->rx.mbuf_in.array,
199                         (uint16_t) bsz_rd);
200
201                 if (unlikely(n_mbufs == 0)) {
202                         continue;
203                 }
204
205 #if APP_STATS
206                 lp->rx.nic_queues_iters[i] ++;
207                 lp->rx.nic_queues_count[i] += n_mbufs;
208                 if (unlikely(lp->rx.nic_queues_iters[i] == APP_STATS)) {
209                         struct rte_eth_stats stats;
210                         unsigned lcore = rte_lcore_id();
211
212                         rte_eth_stats_get(port, &stats);
213
214                         printf("I/O RX %u in (NIC port %u): NIC drop ratio = %.2f avg burst size = %.2f\n",
215                                 lcore,
216                                 (unsigned) port,
217                                 (double) stats.imissed / (double) (stats.imissed + stats.ipackets),
218                                 ((double) lp->rx.nic_queues_count[i]) / ((double) lp->rx.nic_queues_iters[i]));
219                         lp->rx.nic_queues_iters[i] = 0;
220                         lp->rx.nic_queues_count[i] = 0;
221                 }
222 #endif
223
224 #if APP_IO_RX_DROP_ALL_PACKETS
225                 for (j = 0; j < n_mbufs; j ++) {
226                         struct rte_mbuf *pkt = lp->rx.mbuf_in.array[j];
227                         rte_pktmbuf_free(pkt);
228                 }
229
230                 continue;
231 #endif
232
233                 mbuf_1_0 = lp->rx.mbuf_in.array[0];
234                 mbuf_1_1 = lp->rx.mbuf_in.array[1];
235                 data_1_0 = rte_pktmbuf_mtod(mbuf_1_0, uint8_t *);
236                 if (likely(n_mbufs > 1)) {
237                         data_1_1 = rte_pktmbuf_mtod(mbuf_1_1, uint8_t *);
238                 }
239
240                 mbuf_2_0 = lp->rx.mbuf_in.array[2];
241                 mbuf_2_1 = lp->rx.mbuf_in.array[3];
242                 APP_IO_RX_PREFETCH0(mbuf_2_0);
243                 APP_IO_RX_PREFETCH0(mbuf_2_1);
244
245                 for (j = 0; j + 3 < n_mbufs; j += 2) {
246                         struct rte_mbuf *mbuf_0_0, *mbuf_0_1;
247                         uint8_t *data_0_0, *data_0_1;
248                         uint32_t worker_0, worker_1;
249
250                         mbuf_0_0 = mbuf_1_0;
251                         mbuf_0_1 = mbuf_1_1;
252                         data_0_0 = data_1_0;
253                         data_0_1 = data_1_1;
254
255                         mbuf_1_0 = mbuf_2_0;
256                         mbuf_1_1 = mbuf_2_1;
257                         data_1_0 = rte_pktmbuf_mtod(mbuf_2_0, uint8_t *);
258                         data_1_1 = rte_pktmbuf_mtod(mbuf_2_1, uint8_t *);
259                         APP_IO_RX_PREFETCH0(data_1_0);
260                         APP_IO_RX_PREFETCH0(data_1_1);
261
262                         mbuf_2_0 = lp->rx.mbuf_in.array[j+4];
263                         mbuf_2_1 = lp->rx.mbuf_in.array[j+5];
264                         APP_IO_RX_PREFETCH0(mbuf_2_0);
265                         APP_IO_RX_PREFETCH0(mbuf_2_1);
266
267                         worker_0 = data_0_0[pos_lb] & (n_workers - 1);
268                         worker_1 = data_0_1[pos_lb] & (n_workers - 1);
269
270                         app_lcore_io_rx_buffer_to_send(lp, worker_0, mbuf_0_0, bsz_wr);
271                         app_lcore_io_rx_buffer_to_send(lp, worker_1, mbuf_0_1, bsz_wr);
272                 }
273
274                 /* Handle the last 1, 2 (when n_mbufs is even) or 3 (when n_mbufs is odd) packets  */
275                 for ( ; j < n_mbufs; j += 1) {
276                         struct rte_mbuf *mbuf;
277                         uint8_t *data;
278                         uint32_t worker;
279
280                         mbuf = mbuf_1_0;
281                         mbuf_1_0 = mbuf_1_1;
282                         mbuf_1_1 = mbuf_2_0;
283                         mbuf_2_0 = mbuf_2_1;
284
285                         data = rte_pktmbuf_mtod(mbuf, uint8_t *);
286
287                         APP_IO_RX_PREFETCH0(mbuf_1_0);
288
289                         worker = data[pos_lb] & (n_workers - 1);
290
291                         app_lcore_io_rx_buffer_to_send(lp, worker, mbuf, bsz_wr);
292                 }
293         }
294 }
295
296 static inline void
297 app_lcore_io_rx_flush(struct app_lcore_params_io *lp, uint32_t n_workers)
298 {
299         uint32_t worker;
300
301         for (worker = 0; worker < n_workers; worker ++) {
302                 int ret;
303
304                 if (likely((lp->rx.mbuf_out_flush[worker] == 0) ||
305                            (lp->rx.mbuf_out[worker].n_mbufs == 0))) {
306                         lp->rx.mbuf_out_flush[worker] = 1;
307                         continue;
308                 }
309
310                 ret = rte_ring_sp_enqueue_bulk(
311                         lp->rx.rings[worker],
312                         (void **) lp->rx.mbuf_out[worker].array,
313                         lp->rx.mbuf_out[worker].n_mbufs);
314
315                 if (unlikely(ret == 0)) {
316                         uint32_t k;
317                         for (k = 0; k < lp->rx.mbuf_out[worker].n_mbufs; k ++) {
318                                 struct rte_mbuf *pkt_to_free = lp->rx.mbuf_out[worker].array[k];
319                                 rte_pktmbuf_free(pkt_to_free);
320                         }
321                 }
322
323                 lp->rx.mbuf_out[worker].n_mbufs = 0;
324                 lp->rx.mbuf_out_flush[worker] = 1;
325         }
326 }
327
328 static inline void
329 app_lcore_io_tx(
330         struct app_lcore_params_io *lp,
331         uint32_t n_workers,
332         uint32_t bsz_rd,
333         uint32_t bsz_wr)
334 {
335         uint32_t worker;
336
337         for (worker = 0; worker < n_workers; worker ++) {
338                 uint32_t i;
339
340                 for (i = 0; i < lp->tx.n_nic_ports; i ++) {
341                         uint8_t port = lp->tx.nic_ports[i];
342                         struct rte_ring *ring = lp->tx.rings[port][worker];
343                         uint32_t n_mbufs, n_pkts;
344                         int ret;
345
346                         n_mbufs = lp->tx.mbuf_out[port].n_mbufs;
347                         ret = rte_ring_sc_dequeue_bulk(
348                                 ring,
349                                 (void **) &lp->tx.mbuf_out[port].array[n_mbufs],
350                                 bsz_rd);
351
352                         if (unlikely(ret == 0))
353                                 continue;
354
355                         n_mbufs += bsz_rd;
356
357 #if APP_IO_TX_DROP_ALL_PACKETS
358                         {
359                                 uint32_t j;
360                                 APP_IO_TX_PREFETCH0(lp->tx.mbuf_out[port].array[0]);
361                                 APP_IO_TX_PREFETCH0(lp->tx.mbuf_out[port].array[1]);
362
363                                 for (j = 0; j < n_mbufs; j ++) {
364                                         if (likely(j < n_mbufs - 2)) {
365                                                 APP_IO_TX_PREFETCH0(lp->tx.mbuf_out[port].array[j + 2]);
366                                         }
367
368                                         rte_pktmbuf_free(lp->tx.mbuf_out[port].array[j]);
369                                 }
370
371                                 lp->tx.mbuf_out[port].n_mbufs = 0;
372
373                                 continue;
374                         }
375 #endif
376
377                         if (unlikely(n_mbufs < bsz_wr)) {
378                                 lp->tx.mbuf_out[port].n_mbufs = n_mbufs;
379                                 continue;
380                         }
381
382                         n_pkts = rte_eth_tx_burst(
383                                 port,
384                                 0,
385                                 lp->tx.mbuf_out[port].array,
386                                 (uint16_t) n_mbufs);
387
388 #if APP_STATS
389                         lp->tx.nic_ports_iters[port] ++;
390                         lp->tx.nic_ports_count[port] += n_pkts;
391                         if (unlikely(lp->tx.nic_ports_iters[port] == APP_STATS)) {
392                                 unsigned lcore = rte_lcore_id();
393
394                                 printf("\t\t\tI/O TX %u out (port %u): avg burst size = %.2f\n",
395                                         lcore,
396                                         (unsigned) port,
397                                         ((double) lp->tx.nic_ports_count[port]) / ((double) lp->tx.nic_ports_iters[port]));
398                                 lp->tx.nic_ports_iters[port] = 0;
399                                 lp->tx.nic_ports_count[port] = 0;
400                         }
401 #endif
402
403                         if (unlikely(n_pkts < n_mbufs)) {
404                                 uint32_t k;
405                                 for (k = n_pkts; k < n_mbufs; k ++) {
406                                         struct rte_mbuf *pkt_to_free = lp->tx.mbuf_out[port].array[k];
407                                         rte_pktmbuf_free(pkt_to_free);
408                                 }
409                         }
410                         lp->tx.mbuf_out[port].n_mbufs = 0;
411                         lp->tx.mbuf_out_flush[port] = 0;
412                 }
413         }
414 }
415
416 static inline void
417 app_lcore_io_tx_flush(struct app_lcore_params_io *lp)
418 {
419         uint8_t port;
420
421         for (port = 0; port < lp->tx.n_nic_ports; port ++) {
422                 uint32_t n_pkts;
423
424                 if (likely((lp->tx.mbuf_out_flush[port] == 0) ||
425                            (lp->tx.mbuf_out[port].n_mbufs == 0))) {
426                         lp->tx.mbuf_out_flush[port] = 1;
427                         continue;
428                 }
429
430                 n_pkts = rte_eth_tx_burst(
431                         port,
432                         0,
433                         lp->tx.mbuf_out[port].array,
434                         (uint16_t) lp->tx.mbuf_out[port].n_mbufs);
435
436                 if (unlikely(n_pkts < lp->tx.mbuf_out[port].n_mbufs)) {
437                         uint32_t k;
438                         for (k = n_pkts; k < lp->tx.mbuf_out[port].n_mbufs; k ++) {
439                                 struct rte_mbuf *pkt_to_free = lp->tx.mbuf_out[port].array[k];
440                                 rte_pktmbuf_free(pkt_to_free);
441                         }
442                 }
443
444                 lp->tx.mbuf_out[port].n_mbufs = 0;
445                 lp->tx.mbuf_out_flush[port] = 1;
446         }
447 }
448
449 static void
450 app_lcore_main_loop_io(void)
451 {
452         uint32_t lcore = rte_lcore_id();
453         struct app_lcore_params_io *lp = &app.lcore_params[lcore].io;
454         uint32_t n_workers = app_get_lcores_worker();
455         uint64_t i = 0;
456
457         uint32_t bsz_rx_rd = app.burst_size_io_rx_read;
458         uint32_t bsz_rx_wr = app.burst_size_io_rx_write;
459         uint32_t bsz_tx_rd = app.burst_size_io_tx_read;
460         uint32_t bsz_tx_wr = app.burst_size_io_tx_write;
461
462         uint8_t pos_lb = app.pos_lb;
463
464         for ( ; ; ) {
465                 if (APP_LCORE_IO_FLUSH && (unlikely(i == APP_LCORE_IO_FLUSH))) {
466                         if (likely(lp->rx.n_nic_queues > 0)) {
467                                 app_lcore_io_rx_flush(lp, n_workers);
468                         }
469
470                         if (likely(lp->tx.n_nic_ports > 0)) {
471                                 app_lcore_io_tx_flush(lp);
472                         }
473
474                         i = 0;
475                 }
476
477                 if (likely(lp->rx.n_nic_queues > 0)) {
478                         app_lcore_io_rx(lp, n_workers, bsz_rx_rd, bsz_rx_wr, pos_lb);
479                 }
480
481                 if (likely(lp->tx.n_nic_ports > 0)) {
482                         app_lcore_io_tx(lp, n_workers, bsz_tx_rd, bsz_tx_wr);
483                 }
484
485                 i ++;
486         }
487 }
488
489 static inline void
490 app_lcore_worker(
491         struct app_lcore_params_worker *lp,
492         uint32_t bsz_rd,
493         uint32_t bsz_wr)
494 {
495         uint32_t i;
496
497         for (i = 0; i < lp->n_rings_in; i ++) {
498                 struct rte_ring *ring_in = lp->rings_in[i];
499                 uint32_t j;
500                 int ret;
501
502                 ret = rte_ring_sc_dequeue_bulk(
503                         ring_in,
504                         (void **) lp->mbuf_in.array,
505                         bsz_rd);
506
507                 if (unlikely(ret == 0))
508                         continue;
509
510 #if APP_WORKER_DROP_ALL_PACKETS
511                 for (j = 0; j < bsz_rd; j ++) {
512                         struct rte_mbuf *pkt = lp->mbuf_in.array[j];
513                         rte_pktmbuf_free(pkt);
514                 }
515
516                 continue;
517 #endif
518
519                 APP_WORKER_PREFETCH1(rte_pktmbuf_mtod(lp->mbuf_in.array[0], unsigned char *));
520                 APP_WORKER_PREFETCH0(lp->mbuf_in.array[1]);
521
522                 for (j = 0; j < bsz_rd; j ++) {
523                         struct rte_mbuf *pkt;
524                         struct ipv4_hdr *ipv4_hdr;
525                         uint32_t ipv4_dst, pos;
526                         uint32_t port;
527
528                         if (likely(j < bsz_rd - 1)) {
529                                 APP_WORKER_PREFETCH1(rte_pktmbuf_mtod(lp->mbuf_in.array[j+1], unsigned char *));
530                         }
531                         if (likely(j < bsz_rd - 2)) {
532                                 APP_WORKER_PREFETCH0(lp->mbuf_in.array[j+2]);
533                         }
534
535                         pkt = lp->mbuf_in.array[j];
536                         ipv4_hdr = rte_pktmbuf_mtod_offset(pkt,
537                                                            struct ipv4_hdr *,
538                                                            sizeof(struct ether_hdr));
539                         ipv4_dst = rte_be_to_cpu_32(ipv4_hdr->dst_addr);
540
541                         if (unlikely(rte_lpm_lookup(lp->lpm_table, ipv4_dst, &port) != 0)) {
542                                 port = pkt->port;
543                         }
544
545                         pos = lp->mbuf_out[port].n_mbufs;
546
547                         lp->mbuf_out[port].array[pos ++] = pkt;
548                         if (likely(pos < bsz_wr)) {
549                                 lp->mbuf_out[port].n_mbufs = pos;
550                                 continue;
551                         }
552
553                         ret = rte_ring_sp_enqueue_bulk(
554                                 lp->rings_out[port],
555                                 (void **) lp->mbuf_out[port].array,
556                                 bsz_wr);
557
558 #if APP_STATS
559                         lp->rings_out_iters[port] ++;
560                         if (ret > 0) {
561                                 lp->rings_out_count[port] += 1;
562                         }
563                         if (lp->rings_out_iters[port] == APP_STATS){
564                                 printf("\t\tWorker %u out (NIC port %u): enq success rate = %.2f\n",
565                                         (unsigned) lp->worker_id,
566                                         (unsigned) port,
567                                         ((double) lp->rings_out_count[port]) / ((double) lp->rings_out_iters[port]));
568                                 lp->rings_out_iters[port] = 0;
569                                 lp->rings_out_count[port] = 0;
570                         }
571 #endif
572
573                         if (unlikely(ret == 0)) {
574                                 uint32_t k;
575                                 for (k = 0; k < bsz_wr; k ++) {
576                                         struct rte_mbuf *pkt_to_free = lp->mbuf_out[port].array[k];
577                                         rte_pktmbuf_free(pkt_to_free);
578                                 }
579                         }
580
581                         lp->mbuf_out[port].n_mbufs = 0;
582                         lp->mbuf_out_flush[port] = 0;
583                 }
584         }
585 }
586
587 static inline void
588 app_lcore_worker_flush(struct app_lcore_params_worker *lp)
589 {
590         uint32_t port;
591
592         for (port = 0; port < APP_MAX_NIC_PORTS; port ++) {
593                 int ret;
594
595                 if (unlikely(lp->rings_out[port] == NULL)) {
596                         continue;
597                 }
598
599                 if (likely((lp->mbuf_out_flush[port] == 0) ||
600                            (lp->mbuf_out[port].n_mbufs == 0))) {
601                         lp->mbuf_out_flush[port] = 1;
602                         continue;
603                 }
604
605                 ret = rte_ring_sp_enqueue_bulk(
606                         lp->rings_out[port],
607                         (void **) lp->mbuf_out[port].array,
608                         lp->mbuf_out[port].n_mbufs);
609
610                 if (unlikely(ret == 0)) {
611                         uint32_t k;
612                         for (k = 0; k < lp->mbuf_out[port].n_mbufs; k ++) {
613                                 struct rte_mbuf *pkt_to_free = lp->mbuf_out[port].array[k];
614                                 rte_pktmbuf_free(pkt_to_free);
615                         }
616                 }
617
618                 lp->mbuf_out[port].n_mbufs = 0;
619                 lp->mbuf_out_flush[port] = 1;
620         }
621 }
622
623 static void
624 app_lcore_main_loop_worker(void) {
625         uint32_t lcore = rte_lcore_id();
626         struct app_lcore_params_worker *lp = &app.lcore_params[lcore].worker;
627         uint64_t i = 0;
628
629         uint32_t bsz_rd = app.burst_size_worker_read;
630         uint32_t bsz_wr = app.burst_size_worker_write;
631
632         for ( ; ; ) {
633                 if (APP_LCORE_WORKER_FLUSH && (unlikely(i == APP_LCORE_WORKER_FLUSH))) {
634                         app_lcore_worker_flush(lp);
635                         i = 0;
636                 }
637
638                 app_lcore_worker(lp, bsz_rd, bsz_wr);
639
640                 i ++;
641         }
642 }
643
644 int
645 app_lcore_main_loop(__attribute__((unused)) void *arg)
646 {
647         struct app_lcore_params *lp;
648         unsigned lcore;
649
650         lcore = rte_lcore_id();
651         lp = &app.lcore_params[lcore];
652
653         if (lp->type == e_APP_LCORE_IO) {
654                 printf("Logical core %u (I/O) main loop.\n", lcore);
655                 app_lcore_main_loop_io();
656         }
657
658         if (lp->type == e_APP_LCORE_WORKER) {
659                 printf("Logical core %u (worker %u) main loop.\n",
660                         lcore,
661                         (unsigned) lp->worker.worker_id);
662                 app_lcore_main_loop_worker();
663         }
664
665         return 0;
666 }