]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/examples/tep_termination/main.c
update source to Ceph Pacific 16.2.2
[ceph.git] / ceph / src / spdk / dpdk / examples / tep_termination / main.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2015 Intel Corporation
3 */
4
5 #include <arpa/inet.h>
6 #include <getopt.h>
7 #include <linux/if_ether.h>
8 #include <linux/if_vlan.h>
9 #include <linux/virtio_net.h>
10 #include <linux/virtio_ring.h>
11 #include <signal.h>
12 #include <stdint.h>
13 #include <sys/eventfd.h>
14 #include <sys/param.h>
15 #include <unistd.h>
16
17 #include <rte_atomic.h>
18 #include <rte_cycles.h>
19 #include <rte_ethdev.h>
20 #include <rte_log.h>
21 #include <rte_string_fns.h>
22 #include <rte_malloc.h>
23 #include <rte_vhost.h>
24 #include <rte_pause.h>
25
26 #include "main.h"
27 #include "vxlan.h"
28 #include "vxlan_setup.h"
29
30 /* the maximum number of external ports supported */
31 #define MAX_SUP_PORTS 1
32
33 /**
34 * Calculate the number of buffers needed per port
35 */
36 #define NUM_MBUFS_PER_PORT ((MAX_QUEUES * RTE_TEST_RX_DESC_DEFAULT) +\
37 (nb_switching_cores * MAX_PKT_BURST) +\
38 (nb_switching_cores * \
39 RTE_TEST_TX_DESC_DEFAULT) +\
40 (nb_switching_cores * MBUF_CACHE_SIZE))
41
42 #define MBUF_CACHE_SIZE 128
43 #define MBUF_DATA_SIZE RTE_MBUF_DEFAULT_BUF_SIZE
44
45 #define MAX_PKT_BURST 32 /* Max burst size for RX/TX */
46 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
47
48 /* Defines how long we wait between retries on RX */
49 #define BURST_RX_WAIT_US 15
50
51 #define BURST_RX_RETRIES 4 /* Number of retries on RX. */
52
53 #define JUMBO_FRAME_MAX_SIZE 0x2600
54
55 /* Config_core_flag status definitions. */
56 #define REQUEST_DEV_REMOVAL 1
57 #define ACK_DEV_REMOVAL 0
58
59 /* Configurable number of RX/TX ring descriptors */
60 #define RTE_TEST_RX_DESC_DEFAULT 1024
61 #define RTE_TEST_TX_DESC_DEFAULT 512
62
63 /* Get first 4 bytes in mbuf headroom. */
64 #define MBUF_HEADROOM_UINT32(mbuf) (*(uint32_t *)((uint8_t *)(mbuf) \
65 + sizeof(struct rte_mbuf)))
66
67 #define INVALID_PORT_ID 0xFFFF
68
69 /* Maximum character device basename size. */
70 #define MAX_BASENAME_SZ 20
71
72 /* Maximum long option length for option parsing. */
73 #define MAX_LONG_OPT_SZ 64
74
75 /* Used to compare MAC addresses. */
76 #define MAC_ADDR_CMP 0xFFFFFFFFFFFFULL
77
78 #define CMD_LINE_OPT_NB_DEVICES "nb-devices"
79 #define CMD_LINE_OPT_UDP_PORT "udp-port"
80 #define CMD_LINE_OPT_TX_CHECKSUM "tx-checksum"
81 #define CMD_LINE_OPT_TSO_SEGSZ "tso-segsz"
82 #define CMD_LINE_OPT_FILTER_TYPE "filter-type"
83 #define CMD_LINE_OPT_ENCAP "encap"
84 #define CMD_LINE_OPT_DECAP "decap"
85 #define CMD_LINE_OPT_RX_RETRY "rx-retry"
86 #define CMD_LINE_OPT_RX_RETRY_DELAY "rx-retry-delay"
87 #define CMD_LINE_OPT_RX_RETRY_NUM "rx-retry-num"
88 #define CMD_LINE_OPT_STATS "stats"
89 #define CMD_LINE_OPT_DEV_BASENAME "dev-basename"
90
91 /* mask of enabled ports */
92 static uint32_t enabled_port_mask;
93
94 /*Number of switching cores enabled*/
95 static uint32_t nb_switching_cores;
96
97 /* number of devices/queues to support*/
98 uint16_t nb_devices = 2;
99
100 /* max ring descriptor, ixgbe, i40e, e1000 all are 4096. */
101 #define MAX_RING_DESC 4096
102
103 struct vpool {
104 struct rte_mempool *pool;
105 struct rte_ring *ring;
106 uint32_t buf_size;
107 } vpool_array[MAX_QUEUES+MAX_QUEUES];
108
109 /* UDP tunneling port */
110 uint16_t udp_port = 4789;
111
112 /* enable/disable inner TX checksum */
113 uint8_t tx_checksum = 0;
114
115 /* TCP segment size */
116 uint16_t tso_segsz = 0;
117
118 /* enable/disable decapsulation */
119 uint8_t rx_decap = 1;
120
121 /* enable/disable encapsulation */
122 uint8_t tx_encap = 1;
123
124 /* RX filter type for tunneling packet */
125 uint8_t filter_idx = 1;
126
127 /* overlay packet operation */
128 struct ol_switch_ops overlay_options = {
129 .port_configure = vxlan_port_init,
130 .tunnel_setup = vxlan_link,
131 .tunnel_destroy = vxlan_unlink,
132 .tx_handle = vxlan_tx_pkts,
133 .rx_handle = vxlan_rx_pkts,
134 .param_handle = NULL,
135 };
136
137 /* Enable stats. */
138 uint32_t enable_stats = 0;
139 /* Enable retries on RX. */
140 static uint32_t enable_retry = 1;
141 /* Specify timeout (in useconds) between retries on RX. */
142 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
143 /* Specify the number of retries on RX. */
144 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
145
146 /* Character device basename. Can be set by user. */
147 static char dev_basename[MAX_BASENAME_SZ] = "vhost-net";
148
149 static unsigned lcore_ids[RTE_MAX_LCORE];
150 uint16_t ports[RTE_MAX_ETHPORTS];
151
152 static unsigned nb_ports; /**< The number of ports specified in command line */
153
154 /* ethernet addresses of ports */
155 struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
156
157 /* heads for the main used and free linked lists for the data path. */
158 static struct virtio_net_data_ll *ll_root_used;
159 static struct virtio_net_data_ll *ll_root_free;
160
161 /**
162 * Array of data core structures containing information on
163 * individual core linked lists.
164 */
165 static struct lcore_info lcore_info[RTE_MAX_LCORE];
166
167 /* Used for queueing bursts of TX packets. */
168 struct mbuf_table {
169 unsigned len;
170 unsigned txq_id;
171 struct rte_mbuf *m_table[MAX_PKT_BURST];
172 };
173
174 /* TX queue for each data core. */
175 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE];
176
177 struct device_statistics dev_statistics[MAX_DEVICES];
178
179 /**
180 * Set character device basename.
181 */
182 static int
183 us_vhost_parse_basename(const char *q_arg)
184 {
185 /* parse number string */
186 if (strlen(q_arg) >= MAX_BASENAME_SZ)
187 return -1;
188 else
189 strlcpy((char *)&dev_basename, q_arg, MAX_BASENAME_SZ);
190
191 return 0;
192 }
193
194 /**
195 * Parse the portmask provided at run time.
196 */
197 static int
198 parse_portmask(const char *portmask)
199 {
200 char *end = NULL;
201 unsigned long pm;
202
203 /* parse hexadecimal string */
204 pm = strtoul(portmask, &end, 16);
205 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
206 return -1;
207
208 if (pm == 0)
209 return -1;
210
211 return pm;
212 }
213
214 /**
215 * Parse num options at run time.
216 */
217 static int
218 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
219 {
220 char *end = NULL;
221 unsigned long num;
222
223 /* parse unsigned int string */
224 num = strtoul(q_arg, &end, 10);
225 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
226 return -1;
227
228 if (num > max_valid_value)
229 return -1;
230
231 return num;
232 }
233
234 /**
235 * Display usage
236 */
237 static void
238 tep_termination_usage(const char *prgname)
239 {
240 RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
241 " --udp-port: UDP destination port for VXLAN packet\n"
242 " --nb-devices[1-64]: The number of virtIO device\n"
243 " --tx-checksum [0|1]: inner Tx checksum offload\n"
244 " --tso-segsz [0-N]: TCP segment size\n"
245 " --decap [0|1]: tunneling packet decapsulation\n"
246 " --encap [0|1]: tunneling packet encapsulation\n"
247 " --filter-type[1-3]: filter type for tunneling packet\n"
248 " 1: Inner MAC and tenent ID\n"
249 " 2: Inner MAC and VLAN, and tenent ID\n"
250 " 3: Outer MAC, Inner MAC and tenent ID\n"
251 " -p PORTMASK: Set mask for ports to be used by application\n"
252 " --rx-retry [0|1]: disable/enable(default) retries on rx."
253 " Enable retry if destintation queue is full\n"
254 " --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX."
255 " This makes effect only if retries on rx enabled\n"
256 " --rx-retry-num [0-N]: the number of retries on rx."
257 " This makes effect only if retries on rx enabled\n"
258 " --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
259 " --dev-basename: The basename to be used for the character device.\n",
260 prgname);
261 }
262
263 /**
264 * Parse the arguments given in the command line of the application.
265 */
266 static int
267 tep_termination_parse_args(int argc, char **argv)
268 {
269 int opt, ret;
270 int option_index;
271 unsigned i;
272 const char *prgname = argv[0];
273 static struct option long_option[] = {
274 {CMD_LINE_OPT_NB_DEVICES, required_argument, NULL, 0},
275 {CMD_LINE_OPT_UDP_PORT, required_argument, NULL, 0},
276 {CMD_LINE_OPT_TX_CHECKSUM, required_argument, NULL, 0},
277 {CMD_LINE_OPT_TSO_SEGSZ, required_argument, NULL, 0},
278 {CMD_LINE_OPT_DECAP, required_argument, NULL, 0},
279 {CMD_LINE_OPT_ENCAP, required_argument, NULL, 0},
280 {CMD_LINE_OPT_FILTER_TYPE, required_argument, NULL, 0},
281 {CMD_LINE_OPT_RX_RETRY, required_argument, NULL, 0},
282 {CMD_LINE_OPT_RX_RETRY_DELAY, required_argument, NULL, 0},
283 {CMD_LINE_OPT_RX_RETRY_NUM, required_argument, NULL, 0},
284 {CMD_LINE_OPT_STATS, required_argument, NULL, 0},
285 {CMD_LINE_OPT_DEV_BASENAME, required_argument, NULL, 0},
286 {NULL, 0, 0, 0},
287 };
288
289 /* Parse command line */
290 while ((opt = getopt_long(argc, argv, "p:",
291 long_option, &option_index)) != EOF) {
292 switch (opt) {
293 /* Portmask */
294 case 'p':
295 enabled_port_mask = parse_portmask(optarg);
296 if (enabled_port_mask == 0) {
297 RTE_LOG(INFO, VHOST_CONFIG,
298 "Invalid portmask\n");
299 tep_termination_usage(prgname);
300 return -1;
301 }
302 break;
303 case 0:
304 if (!strncmp(long_option[option_index].name,
305 CMD_LINE_OPT_NB_DEVICES,
306 sizeof(CMD_LINE_OPT_NB_DEVICES))) {
307 ret = parse_num_opt(optarg, MAX_DEVICES);
308 if (ret == -1) {
309 RTE_LOG(INFO, VHOST_CONFIG,
310 "Invalid argument for nb-devices [0-%d]\n",
311 MAX_DEVICES);
312 tep_termination_usage(prgname);
313 return -1;
314 } else
315 nb_devices = ret;
316 }
317
318 /* Enable/disable retries on RX. */
319 if (!strncmp(long_option[option_index].name,
320 CMD_LINE_OPT_RX_RETRY,
321 sizeof(CMD_LINE_OPT_RX_RETRY))) {
322 ret = parse_num_opt(optarg, 1);
323 if (ret == -1) {
324 RTE_LOG(INFO, VHOST_CONFIG,
325 "Invalid argument for rx-retry [0|1]\n");
326 tep_termination_usage(prgname);
327 return -1;
328 } else
329 enable_retry = ret;
330 }
331
332 if (!strncmp(long_option[option_index].name,
333 CMD_LINE_OPT_TSO_SEGSZ,
334 sizeof(CMD_LINE_OPT_TSO_SEGSZ))) {
335 ret = parse_num_opt(optarg, INT16_MAX);
336 if (ret == -1) {
337 RTE_LOG(INFO, VHOST_CONFIG,
338 "Invalid argument for TCP segment size [0-N]\n");
339 tep_termination_usage(prgname);
340 return -1;
341 } else
342 tso_segsz = ret;
343 }
344
345 if (!strncmp(long_option[option_index].name,
346 CMD_LINE_OPT_UDP_PORT,
347 sizeof(CMD_LINE_OPT_UDP_PORT))) {
348 ret = parse_num_opt(optarg, INT16_MAX);
349 if (ret == -1) {
350 RTE_LOG(INFO, VHOST_CONFIG,
351 "Invalid argument for UDP port [0-N]\n");
352 tep_termination_usage(prgname);
353 return -1;
354 } else
355 udp_port = ret;
356 }
357
358 /* Specify the retries delay time (in useconds) on RX.*/
359 if (!strncmp(long_option[option_index].name,
360 CMD_LINE_OPT_RX_RETRY_DELAY,
361 sizeof(CMD_LINE_OPT_RX_RETRY_DELAY))) {
362 ret = parse_num_opt(optarg, INT32_MAX);
363 if (ret == -1) {
364 RTE_LOG(INFO, VHOST_CONFIG,
365 "Invalid argument for rx-retry-delay [0-N]\n");
366 tep_termination_usage(prgname);
367 return -1;
368 } else
369 burst_rx_delay_time = ret;
370 }
371
372 /* Specify the retries number on RX. */
373 if (!strncmp(long_option[option_index].name,
374 CMD_LINE_OPT_RX_RETRY_NUM,
375 sizeof(CMD_LINE_OPT_RX_RETRY_NUM))) {
376 ret = parse_num_opt(optarg, INT32_MAX);
377 if (ret == -1) {
378 RTE_LOG(INFO, VHOST_CONFIG,
379 "Invalid argument for rx-retry-num [0-N]\n");
380 tep_termination_usage(prgname);
381 return -1;
382 } else
383 burst_rx_retry_num = ret;
384 }
385
386 if (!strncmp(long_option[option_index].name,
387 CMD_LINE_OPT_TX_CHECKSUM,
388 sizeof(CMD_LINE_OPT_TX_CHECKSUM))) {
389 ret = parse_num_opt(optarg, 1);
390 if (ret == -1) {
391 RTE_LOG(INFO, VHOST_CONFIG,
392 "Invalid argument for tx-checksum [0|1]\n");
393 tep_termination_usage(prgname);
394 return -1;
395 } else
396 tx_checksum = ret;
397 }
398
399 if (!strncmp(long_option[option_index].name,
400 CMD_LINE_OPT_FILTER_TYPE,
401 sizeof(CMD_LINE_OPT_FILTER_TYPE))) {
402 ret = parse_num_opt(optarg, 3);
403 if ((ret == -1) || (ret == 0)) {
404 RTE_LOG(INFO, VHOST_CONFIG,
405 "Invalid argument for filter type [1-3]\n");
406 tep_termination_usage(prgname);
407 return -1;
408 } else
409 filter_idx = ret - 1;
410 }
411
412 /* Enable/disable encapsulation on RX. */
413 if (!strncmp(long_option[option_index].name,
414 CMD_LINE_OPT_DECAP,
415 sizeof(CMD_LINE_OPT_DECAP))) {
416 ret = parse_num_opt(optarg, 1);
417 if (ret == -1) {
418 RTE_LOG(INFO, VHOST_CONFIG,
419 "Invalid argument for decap [0|1]\n");
420 tep_termination_usage(prgname);
421 return -1;
422 } else
423 rx_decap = ret;
424 }
425
426 /* Enable/disable encapsulation on TX. */
427 if (!strncmp(long_option[option_index].name,
428 CMD_LINE_OPT_ENCAP,
429 sizeof(CMD_LINE_OPT_ENCAP))) {
430 ret = parse_num_opt(optarg, 1);
431 if (ret == -1) {
432 RTE_LOG(INFO, VHOST_CONFIG,
433 "Invalid argument for encap [0|1]\n");
434 tep_termination_usage(prgname);
435 return -1;
436 } else
437 tx_encap = ret;
438 }
439
440 /* Enable/disable stats. */
441 if (!strncmp(long_option[option_index].name,
442 CMD_LINE_OPT_STATS,
443 sizeof(CMD_LINE_OPT_STATS))) {
444 ret = parse_num_opt(optarg, INT32_MAX);
445 if (ret == -1) {
446 RTE_LOG(INFO, VHOST_CONFIG,
447 "Invalid argument for stats [0..N]\n");
448 tep_termination_usage(prgname);
449 return -1;
450 } else
451 enable_stats = ret;
452 }
453
454 /* Set character device basename. */
455 if (!strncmp(long_option[option_index].name,
456 CMD_LINE_OPT_DEV_BASENAME,
457 sizeof(CMD_LINE_OPT_DEV_BASENAME))) {
458 if (us_vhost_parse_basename(optarg) == -1) {
459 RTE_LOG(INFO, VHOST_CONFIG,
460 "Invalid argument for character "
461 "device basename (Max %d characters)\n",
462 MAX_BASENAME_SZ);
463 tep_termination_usage(prgname);
464 return -1;
465 }
466 }
467
468 break;
469
470 /* Invalid option - print options. */
471 default:
472 tep_termination_usage(prgname);
473 return -1;
474 }
475 }
476
477 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
478 if (enabled_port_mask & (1 << i))
479 ports[nb_ports++] = (uint8_t)i;
480 }
481
482 if ((nb_ports == 0) || (nb_ports > MAX_SUP_PORTS)) {
483 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
484 "but only %u port can be enabled\n", nb_ports,
485 MAX_SUP_PORTS);
486 return -1;
487 }
488
489 return 0;
490 }
491
492 /**
493 * Update the global var NB_PORTS and array PORTS
494 * according to system ports number and return valid ports number
495 */
496 static unsigned
497 check_ports_num(unsigned max_nb_ports)
498 {
499 unsigned valid_nb_ports = nb_ports;
500 unsigned portid;
501
502 if (nb_ports > max_nb_ports) {
503 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) "
504 " exceeds total system port number(%u)\n",
505 nb_ports, max_nb_ports);
506 nb_ports = max_nb_ports;
507 }
508
509 for (portid = 0; portid < nb_ports; portid++) {
510 if (!rte_eth_dev_is_valid_port(ports[portid])) {
511 RTE_LOG(INFO, VHOST_PORT,
512 "\nSpecified port ID(%u) is not valid\n",
513 ports[portid]);
514 ports[portid] = INVALID_PORT_ID;
515 valid_nb_ports--;
516 }
517 }
518 return valid_nb_ports;
519 }
520
521 /**
522 * This function routes the TX packet to the correct interface. This may be a local device
523 * or the physical port.
524 */
525 static __rte_always_inline void
526 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m)
527 {
528 struct mbuf_table *tx_q;
529 struct rte_mbuf **m_table;
530 unsigned len, ret = 0;
531 const uint16_t lcore_id = rte_lcore_id();
532
533 RTE_LOG_DP(DEBUG, VHOST_DATA, "(%d) TX: MAC address is external\n",
534 vdev->vid);
535
536 /* Add packet to the port tx queue */
537 tx_q = &lcore_tx_queue[lcore_id];
538 len = tx_q->len;
539
540 tx_q->m_table[len] = m;
541 len++;
542 if (enable_stats) {
543 dev_statistics[vdev->vid].tx_total++;
544 dev_statistics[vdev->vid].tx++;
545 }
546
547 if (unlikely(len == MAX_PKT_BURST)) {
548 m_table = (struct rte_mbuf **)tx_q->m_table;
549 ret = overlay_options.tx_handle(ports[0],
550 (uint16_t)tx_q->txq_id, m_table,
551 (uint16_t)tx_q->len);
552
553 /* Free any buffers not handled by TX and update
554 * the port stats.
555 */
556 if (unlikely(ret < len)) {
557 do {
558 rte_pktmbuf_free(m_table[ret]);
559 } while (++ret < len);
560 }
561
562 len = 0;
563 }
564
565 tx_q->len = len;
566 return;
567 }
568
569 /**
570 * This function is called by each data core. It handles all
571 * RX/TX registered with the core. For TX the specific lcore
572 * linked list is used. For RX, MAC addresses are compared
573 * with all devices in the main linked list.
574 */
575 static int
576 switch_worker(__rte_unused void *arg)
577 {
578 struct rte_mempool *mbuf_pool = arg;
579 struct vhost_dev *vdev = NULL;
580 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
581 struct virtio_net_data_ll *dev_ll;
582 struct mbuf_table *tx_q;
583 volatile struct lcore_ll_info *lcore_ll;
584 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
585 / US_PER_S * BURST_TX_DRAIN_US;
586 uint64_t prev_tsc, diff_tsc, cur_tsc, ret_count = 0;
587 unsigned i, ret = 0;
588 const uint16_t lcore_id = rte_lcore_id();
589 const uint16_t num_cores = (uint16_t)rte_lcore_count();
590 uint16_t rx_count = 0;
591 uint16_t tx_count;
592 uint32_t retry = 0;
593
594 RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
595 lcore_ll = lcore_info[lcore_id].lcore_ll;
596 prev_tsc = 0;
597
598 tx_q = &lcore_tx_queue[lcore_id];
599 for (i = 0; i < num_cores; i++) {
600 if (lcore_ids[i] == lcore_id) {
601 tx_q->txq_id = i;
602 break;
603 }
604 }
605
606 while (1) {
607 cur_tsc = rte_rdtsc();
608 /*
609 * TX burst queue drain
610 */
611 diff_tsc = cur_tsc - prev_tsc;
612 if (unlikely(diff_tsc > drain_tsc)) {
613
614 if (tx_q->len) {
615 RTE_LOG_DP(DEBUG, VHOST_DATA, "TX queue drained after "
616 "timeout with burst size %u\n",
617 tx_q->len);
618 ret = overlay_options.tx_handle(ports[0],
619 (uint16_t)tx_q->txq_id,
620 (struct rte_mbuf **)tx_q->m_table,
621 (uint16_t)tx_q->len);
622 if (unlikely(ret < tx_q->len)) {
623 do {
624 rte_pktmbuf_free(tx_q->m_table[ret]);
625 } while (++ret < tx_q->len);
626 }
627
628 tx_q->len = 0;
629 }
630
631 prev_tsc = cur_tsc;
632
633 }
634
635 rte_prefetch0(lcore_ll->ll_root_used);
636
637 /**
638 * Inform the configuration core that we have exited
639 * the linked list and that no devices are
640 * in use if requested.
641 */
642 if (lcore_ll->dev_removal_flag == REQUEST_DEV_REMOVAL)
643 lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL;
644
645 /*
646 * Process devices
647 */
648 dev_ll = lcore_ll->ll_root_used;
649
650 while (dev_ll != NULL) {
651 vdev = dev_ll->vdev;
652
653 if (unlikely(vdev->remove)) {
654 dev_ll = dev_ll->next;
655 overlay_options.tunnel_destroy(vdev);
656 vdev->ready = DEVICE_SAFE_REMOVE;
657 continue;
658 }
659 if (likely(vdev->ready == DEVICE_RX)) {
660 /* Handle guest RX */
661 rx_count = rte_eth_rx_burst(ports[0],
662 vdev->rx_q, pkts_burst, MAX_PKT_BURST);
663
664 if (rx_count) {
665 /*
666 * Retry is enabled and the queue is
667 * full then we wait and retry to
668 * avoid packet loss. Here MAX_PKT_BURST
669 * must be less than virtio queue size
670 */
671 if (enable_retry && unlikely(rx_count >
672 rte_vhost_avail_entries(vdev->vid, VIRTIO_RXQ))) {
673 for (retry = 0; retry < burst_rx_retry_num;
674 retry++) {
675 rte_delay_us(burst_rx_delay_time);
676 if (rx_count <= rte_vhost_avail_entries(vdev->vid, VIRTIO_RXQ))
677 break;
678 }
679 }
680
681 ret_count = overlay_options.rx_handle(vdev->vid, pkts_burst, rx_count);
682 if (enable_stats) {
683 rte_atomic64_add(
684 &dev_statistics[vdev->vid].rx_total_atomic,
685 rx_count);
686 rte_atomic64_add(
687 &dev_statistics[vdev->vid].rx_atomic, ret_count);
688 }
689 while (likely(rx_count)) {
690 rx_count--;
691 rte_pktmbuf_free(pkts_burst[rx_count]);
692 }
693
694 }
695 }
696
697 if (likely(!vdev->remove)) {
698 /* Handle guest TX*/
699 tx_count = rte_vhost_dequeue_burst(vdev->vid,
700 VIRTIO_TXQ, mbuf_pool,
701 pkts_burst, MAX_PKT_BURST);
702 /* If this is the first received packet we need to learn the MAC */
703 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && tx_count) {
704 if (vdev->remove ||
705 (overlay_options.tunnel_setup(vdev, pkts_burst[0]) == -1)) {
706 while (tx_count)
707 rte_pktmbuf_free(pkts_burst[--tx_count]);
708 }
709 }
710 while (tx_count)
711 virtio_tx_route(vdev, pkts_burst[--tx_count]);
712 }
713
714 /* move to the next device in the list */
715 dev_ll = dev_ll->next;
716 }
717 }
718
719 return 0;
720 }
721
722 /**
723 * Add an entry to a used linked list. A free entry must first be found
724 * in the free linked list using get_data_ll_free_entry();
725 */
726 static void
727 add_data_ll_entry(struct virtio_net_data_ll **ll_root_addr,
728 struct virtio_net_data_ll *ll_dev)
729 {
730 struct virtio_net_data_ll *ll = *ll_root_addr;
731
732 /* Set next as NULL and use a compiler barrier to avoid reordering. */
733 ll_dev->next = NULL;
734 rte_compiler_barrier();
735
736 /* If ll == NULL then this is the first device. */
737 if (ll) {
738 /* Increment to the tail of the linked list. */
739 while (ll->next != NULL)
740 ll = ll->next;
741
742 ll->next = ll_dev;
743 } else {
744 *ll_root_addr = ll_dev;
745 }
746 }
747
748 /**
749 * Remove an entry from a used linked list. The entry must then be added to
750 * the free linked list using put_data_ll_free_entry().
751 */
752 static void
753 rm_data_ll_entry(struct virtio_net_data_ll **ll_root_addr,
754 struct virtio_net_data_ll *ll_dev,
755 struct virtio_net_data_ll *ll_dev_last)
756 {
757 struct virtio_net_data_ll *ll = *ll_root_addr;
758
759 if (unlikely((ll == NULL) || (ll_dev == NULL)))
760 return;
761
762 if (ll_dev == ll)
763 *ll_root_addr = ll_dev->next;
764 else
765 if (likely(ll_dev_last != NULL))
766 ll_dev_last->next = ll_dev->next;
767 else
768 RTE_LOG(ERR, VHOST_CONFIG,
769 "Remove entry form ll failed.\n");
770 }
771
772 /**
773 * Find and return an entry from the free linked list.
774 */
775 static struct virtio_net_data_ll *
776 get_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr)
777 {
778 struct virtio_net_data_ll *ll_free = *ll_root_addr;
779 struct virtio_net_data_ll *ll_dev;
780
781 if (ll_free == NULL)
782 return NULL;
783
784 ll_dev = ll_free;
785 *ll_root_addr = ll_free->next;
786
787 return ll_dev;
788 }
789
790 /**
791 * Place an entry back on to the free linked list.
792 */
793 static void
794 put_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr,
795 struct virtio_net_data_ll *ll_dev)
796 {
797 struct virtio_net_data_ll *ll_free = *ll_root_addr;
798
799 if (ll_dev == NULL)
800 return;
801
802 ll_dev->next = ll_free;
803 *ll_root_addr = ll_dev;
804 }
805
806 /**
807 * Creates a linked list of a given size.
808 */
809 static struct virtio_net_data_ll *
810 alloc_data_ll(uint32_t size)
811 {
812 struct virtio_net_data_ll *ll_new;
813 uint32_t i;
814
815 /* Malloc and then chain the linked list. */
816 ll_new = malloc(size * sizeof(struct virtio_net_data_ll));
817 if (ll_new == NULL) {
818 RTE_LOG(ERR, VHOST_CONFIG,
819 "Failed to allocate memory for ll_new.\n");
820 return NULL;
821 }
822
823 for (i = 0; i < size - 1; i++) {
824 ll_new[i].vdev = NULL;
825 ll_new[i].next = &ll_new[i+1];
826 }
827 ll_new[i].next = NULL;
828
829 return ll_new;
830 }
831
832 /**
833 * Create the main linked list along with each individual cores
834 * linked list. A used and a free list are created to manage entries.
835 */
836 static int
837 init_data_ll(void)
838 {
839 int lcore;
840
841 RTE_LCORE_FOREACH_SLAVE(lcore) {
842 lcore_info[lcore].lcore_ll =
843 malloc(sizeof(struct lcore_ll_info));
844 if (lcore_info[lcore].lcore_ll == NULL) {
845 RTE_LOG(ERR, VHOST_CONFIG,
846 "Failed to allocate memory for lcore_ll.\n");
847 return -1;
848 }
849
850 lcore_info[lcore].lcore_ll->device_num = 0;
851 lcore_info[lcore].lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL;
852 lcore_info[lcore].lcore_ll->ll_root_used = NULL;
853 if (nb_devices % nb_switching_cores)
854 lcore_info[lcore].lcore_ll->ll_root_free =
855 alloc_data_ll((nb_devices / nb_switching_cores)
856 + 1);
857 else
858 lcore_info[lcore].lcore_ll->ll_root_free =
859 alloc_data_ll(nb_devices / nb_switching_cores);
860 }
861
862 /* Allocate devices up to a maximum of MAX_DEVICES. */
863 ll_root_free = alloc_data_ll(MIN((nb_devices), MAX_DEVICES));
864
865 return 0;
866 }
867
868 /**
869 * Remove a device from the specific data core linked list and
870 * from the main linked list. Synchonization occurs through the use
871 * of the lcore dev_removal_flag.
872 */
873 static void
874 destroy_device(int vid)
875 {
876 struct virtio_net_data_ll *ll_lcore_dev_cur;
877 struct virtio_net_data_ll *ll_main_dev_cur;
878 struct virtio_net_data_ll *ll_lcore_dev_last = NULL;
879 struct virtio_net_data_ll *ll_main_dev_last = NULL;
880 struct vhost_dev *vdev = NULL;
881 int lcore;
882
883 ll_main_dev_cur = ll_root_used;
884 while (ll_main_dev_cur != NULL) {
885 if (ll_main_dev_cur->vdev->vid == vid) {
886 vdev = ll_main_dev_cur->vdev;
887 break;
888 }
889 }
890 if (!vdev)
891 return;
892
893 /* set the remove flag. */
894 vdev->remove = 1;
895 while (vdev->ready != DEVICE_SAFE_REMOVE)
896 rte_pause();
897
898 /* Search for entry to be removed from lcore ll */
899 ll_lcore_dev_cur = lcore_info[vdev->coreid].lcore_ll->ll_root_used;
900 while (ll_lcore_dev_cur != NULL) {
901 if (ll_lcore_dev_cur->vdev == vdev) {
902 break;
903 } else {
904 ll_lcore_dev_last = ll_lcore_dev_cur;
905 ll_lcore_dev_cur = ll_lcore_dev_cur->next;
906 }
907 }
908
909 if (ll_lcore_dev_cur == NULL) {
910 RTE_LOG(ERR, VHOST_CONFIG,
911 "(%d) Failed to find the dev to be destroy.\n", vid);
912 return;
913 }
914
915 /* Search for entry to be removed from main ll */
916 ll_main_dev_cur = ll_root_used;
917 ll_main_dev_last = NULL;
918 while (ll_main_dev_cur != NULL) {
919 if (ll_main_dev_cur->vdev == vdev) {
920 break;
921 } else {
922 ll_main_dev_last = ll_main_dev_cur;
923 ll_main_dev_cur = ll_main_dev_cur->next;
924 }
925 }
926
927 /* Remove entries from the lcore and main ll. */
928 rm_data_ll_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_used,
929 ll_lcore_dev_cur, ll_lcore_dev_last);
930 rm_data_ll_entry(&ll_root_used, ll_main_dev_cur, ll_main_dev_last);
931
932 /* Set the dev_removal_flag on each lcore. */
933 RTE_LCORE_FOREACH_SLAVE(lcore) {
934 lcore_info[lcore].lcore_ll->dev_removal_flag =
935 REQUEST_DEV_REMOVAL;
936 }
937
938 /*
939 * Once each core has set the dev_removal_flag to
940 * ACK_DEV_REMOVAL we can be sure that they can no longer access
941 * the device removed from the linked lists and that the devices
942 * are no longer in use.
943 */
944 RTE_LCORE_FOREACH_SLAVE(lcore) {
945 while (lcore_info[lcore].lcore_ll->dev_removal_flag
946 != ACK_DEV_REMOVAL)
947 rte_pause();
948 }
949
950 /* Add the entries back to the lcore and main free ll.*/
951 put_data_ll_free_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_free,
952 ll_lcore_dev_cur);
953 put_data_ll_free_entry(&ll_root_free, ll_main_dev_cur);
954
955 /* Decrement number of device on the lcore. */
956 lcore_info[vdev->coreid].lcore_ll->device_num--;
957
958 RTE_LOG(INFO, VHOST_DATA, "(%d) Device has been removed "
959 "from data core\n", vid);
960
961 rte_free(vdev);
962
963 }
964
965 /**
966 * A new device is added to a data core. First the device is added
967 * to the main linked list and the allocated to a specific data core.
968 */
969 static int
970 new_device(int vid)
971 {
972 struct virtio_net_data_ll *ll_dev;
973 int lcore, core_add = 0;
974 uint32_t device_num_min = nb_devices;
975 struct vhost_dev *vdev;
976
977 vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
978 if (vdev == NULL) {
979 RTE_LOG(INFO, VHOST_DATA,
980 "(%d) Couldn't allocate memory for vhost dev\n", vid);
981 return -1;
982 }
983 vdev->vid = vid;
984 /* Add device to main ll */
985 ll_dev = get_data_ll_free_entry(&ll_root_free);
986 if (ll_dev == NULL) {
987 RTE_LOG(INFO, VHOST_DATA, "(%d) No free entry found in"
988 " linked list Device limit of %d devices per core"
989 " has been reached\n", vid, nb_devices);
990 if (vdev->regions_hpa)
991 rte_free(vdev->regions_hpa);
992 rte_free(vdev);
993 return -1;
994 }
995 ll_dev->vdev = vdev;
996 add_data_ll_entry(&ll_root_used, ll_dev);
997 vdev->rx_q = vid;
998
999 /* reset ready flag */
1000 vdev->ready = DEVICE_MAC_LEARNING;
1001 vdev->remove = 0;
1002
1003 /* Find a suitable lcore to add the device. */
1004 RTE_LCORE_FOREACH_SLAVE(lcore) {
1005 if (lcore_info[lcore].lcore_ll->device_num < device_num_min) {
1006 device_num_min = lcore_info[lcore].lcore_ll->device_num;
1007 core_add = lcore;
1008 }
1009 }
1010 /* Add device to lcore ll */
1011 ll_dev = get_data_ll_free_entry(&lcore_info[core_add].lcore_ll->ll_root_free);
1012 if (ll_dev == NULL) {
1013 RTE_LOG(INFO, VHOST_DATA,
1014 "(%d) Failed to add device to data core\n",
1015 vid);
1016 vdev->ready = DEVICE_SAFE_REMOVE;
1017 destroy_device(vid);
1018 rte_free(vdev->regions_hpa);
1019 rte_free(vdev);
1020 return -1;
1021 }
1022 ll_dev->vdev = vdev;
1023 vdev->coreid = core_add;
1024
1025 add_data_ll_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_used,
1026 ll_dev);
1027
1028 /* Initialize device stats */
1029 memset(&dev_statistics[vid], 0,
1030 sizeof(struct device_statistics));
1031
1032 /* Disable notifications. */
1033 rte_vhost_enable_guest_notification(vid, VIRTIO_RXQ, 0);
1034 rte_vhost_enable_guest_notification(vid, VIRTIO_TXQ, 0);
1035 lcore_info[vdev->coreid].lcore_ll->device_num++;
1036
1037 RTE_LOG(INFO, VHOST_DATA, "(%d) Device has been added to data core %d\n",
1038 vid, vdev->coreid);
1039
1040 return 0;
1041 }
1042
1043 /**
1044 * These callback allow devices to be added to the data core when configuration
1045 * has been fully complete.
1046 */
1047 static const struct vhost_device_ops virtio_net_device_ops = {
1048 .new_device = new_device,
1049 .destroy_device = destroy_device,
1050 };
1051
1052 /**
1053 * This is a thread will wake up after a period to print stats if the user has
1054 * enabled them.
1055 */
1056 static void *
1057 print_stats(__rte_unused void *arg)
1058 {
1059 struct virtio_net_data_ll *dev_ll;
1060 uint64_t tx_dropped, rx_dropped;
1061 uint64_t tx, tx_total, rx, rx_total, rx_ip_csum, rx_l4_csum;
1062 int vid;
1063 const char clr[] = { 27, '[', '2', 'J', '\0' };
1064 const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
1065
1066 while (1) {
1067 sleep(enable_stats);
1068
1069 /* Clear screen and move to top left */
1070 printf("%s%s", clr, top_left);
1071
1072 printf("\nDevice statistics ================================");
1073
1074 dev_ll = ll_root_used;
1075 while (dev_ll != NULL) {
1076 vid = dev_ll->vdev->vid;
1077 tx_total = dev_statistics[vid].tx_total;
1078 tx = dev_statistics[vid].tx;
1079 tx_dropped = tx_total - tx;
1080
1081 rx_total = rte_atomic64_read(
1082 &dev_statistics[vid].rx_total_atomic);
1083 rx = rte_atomic64_read(
1084 &dev_statistics[vid].rx_atomic);
1085 rx_dropped = rx_total - rx;
1086 rx_ip_csum = rte_atomic64_read(
1087 &dev_statistics[vid].rx_bad_ip_csum);
1088 rx_l4_csum = rte_atomic64_read(
1089 &dev_statistics[vid].rx_bad_l4_csum);
1090
1091 printf("\nStatistics for device %d ----------"
1092 "\nTX total: %"PRIu64""
1093 "\nTX dropped: %"PRIu64""
1094 "\nTX successful: %"PRIu64""
1095 "\nRX total: %"PRIu64""
1096 "\nRX bad IP csum: %"PRIu64""
1097 "\nRX bad L4 csum: %"PRIu64""
1098 "\nRX dropped: %"PRIu64""
1099 "\nRX successful: %"PRIu64"",
1100 vid,
1101 tx_total,
1102 tx_dropped,
1103 tx,
1104 rx_total,
1105 rx_ip_csum,
1106 rx_l4_csum,
1107 rx_dropped,
1108 rx);
1109
1110 dev_ll = dev_ll->next;
1111 }
1112 printf("\n================================================\n");
1113 }
1114
1115 return NULL;
1116 }
1117
1118 /**
1119 * Main function, does initialisation and calls the per-lcore functions.
1120 */
1121 int
1122 main(int argc, char *argv[])
1123 {
1124 struct rte_mempool *mbuf_pool = NULL;
1125 unsigned lcore_id, core_id = 0;
1126 unsigned nb_ports, valid_nb_ports;
1127 int ret;
1128 uint16_t portid;
1129 uint16_t queue_id;
1130 static pthread_t tid;
1131
1132 /* init EAL */
1133 ret = rte_eal_init(argc, argv);
1134 if (ret < 0)
1135 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1136 argc -= ret;
1137 argv += ret;
1138
1139 /* parse app arguments */
1140 ret = tep_termination_parse_args(argc, argv);
1141 if (ret < 0)
1142 rte_exit(EXIT_FAILURE, "Invalid argument\n");
1143
1144 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
1145 if (rte_lcore_is_enabled(lcore_id))
1146 lcore_ids[core_id++] = lcore_id;
1147
1148 /* set the number of swithcing cores available */
1149 nb_switching_cores = rte_lcore_count()-1;
1150
1151 /* Get the number of physical ports. */
1152 nb_ports = rte_eth_dev_count_avail();
1153
1154 /*
1155 * Update the global var NB_PORTS and global array PORTS
1156 * and get value of var VALID_NB_PORTS according to system ports number
1157 */
1158 valid_nb_ports = check_ports_num(nb_ports);
1159
1160 if ((valid_nb_ports == 0) || (valid_nb_ports > MAX_SUP_PORTS)) {
1161 rte_exit(EXIT_FAILURE, "Current enabled port number is %u,"
1162 "but only %u port can be enabled\n", nb_ports,
1163 MAX_SUP_PORTS);
1164 }
1165 /* Create the mbuf pool. */
1166 mbuf_pool = rte_pktmbuf_pool_create(
1167 "MBUF_POOL",
1168 NUM_MBUFS_PER_PORT * valid_nb_ports,
1169 MBUF_CACHE_SIZE,
1170 0,
1171 MBUF_DATA_SIZE,
1172 rte_socket_id());
1173 if (mbuf_pool == NULL)
1174 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1175
1176 for (queue_id = 0; queue_id < MAX_QUEUES + 1; queue_id++)
1177 vpool_array[queue_id].pool = mbuf_pool;
1178
1179 /* initialize all ports */
1180 RTE_ETH_FOREACH_DEV(portid) {
1181 /* skip ports that are not enabled */
1182 if ((enabled_port_mask & (1 << portid)) == 0) {
1183 RTE_LOG(INFO, VHOST_PORT,
1184 "Skipping disabled port %d\n", portid);
1185 continue;
1186 }
1187 if (overlay_options.port_configure(portid, mbuf_pool) != 0)
1188 rte_exit(EXIT_FAILURE,
1189 "Cannot initialize network ports\n");
1190 }
1191
1192 /* Initialise all linked lists. */
1193 if (init_data_ll() == -1)
1194 rte_exit(EXIT_FAILURE, "Failed to initialize linked list\n");
1195
1196 /* Initialize device stats */
1197 memset(&dev_statistics, 0, sizeof(dev_statistics));
1198
1199 /* Enable stats if the user option is set. */
1200 if (enable_stats) {
1201 ret = rte_ctrl_thread_create(&tid, "print-stats", NULL,
1202 print_stats, NULL);
1203 if (ret < 0)
1204 rte_exit(EXIT_FAILURE, "Cannot create print-stats thread\n");
1205 }
1206
1207 /* Launch all data cores. */
1208 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1209 rte_eal_remote_launch(switch_worker,
1210 mbuf_pool, lcore_id);
1211 }
1212
1213 ret = rte_vhost_driver_register((char *)&dev_basename, 0);
1214 if (ret != 0)
1215 rte_exit(EXIT_FAILURE, "failed to register vhost driver.\n");
1216
1217 rte_vhost_driver_disable_features(dev_basename,
1218 1ULL << VIRTIO_NET_F_MRG_RXBUF);
1219
1220 ret = rte_vhost_driver_callback_register(dev_basename,
1221 &virtio_net_device_ops);
1222 if (ret != 0) {
1223 rte_exit(EXIT_FAILURE,
1224 "failed to register vhost driver callbacks.\n");
1225 }
1226
1227 if (rte_vhost_driver_start(dev_basename) < 0) {
1228 rte_exit(EXIT_FAILURE,
1229 "failed to start vhost driver.\n");
1230 }
1231
1232 RTE_LCORE_FOREACH_SLAVE(lcore_id)
1233 rte_eal_wait_lcore(lcore_id);
1234
1235 return 0;
1236 }