]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/examples/tep_termination/main.c
update sources to ceph Nautilus 14.2.1
[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 /* State of virtio device. */
56 #define DEVICE_MAC_LEARNING 0
57 #define DEVICE_RX 1
58 #define DEVICE_SAFE_REMOVE 2
59
60 /* Config_core_flag status definitions. */
61 #define REQUEST_DEV_REMOVAL 1
62 #define ACK_DEV_REMOVAL 0
63
64 /* Configurable number of RX/TX ring descriptors */
65 #define RTE_TEST_RX_DESC_DEFAULT 1024
66 #define RTE_TEST_TX_DESC_DEFAULT 512
67
68 /* Get first 4 bytes in mbuf headroom. */
69 #define MBUF_HEADROOM_UINT32(mbuf) (*(uint32_t *)((uint8_t *)(mbuf) \
70 + sizeof(struct rte_mbuf)))
71
72 #define INVALID_PORT_ID 0xFFFF
73
74 /* Size of buffers used for snprintfs. */
75 #define MAX_PRINT_BUFF 6072
76
77 /* Maximum character device basename size. */
78 #define MAX_BASENAME_SZ 20
79
80 /* Maximum long option length for option parsing. */
81 #define MAX_LONG_OPT_SZ 64
82
83 /* Used to compare MAC addresses. */
84 #define MAC_ADDR_CMP 0xFFFFFFFFFFFFULL
85
86 #define CMD_LINE_OPT_NB_DEVICES "nb-devices"
87 #define CMD_LINE_OPT_UDP_PORT "udp-port"
88 #define CMD_LINE_OPT_TX_CHECKSUM "tx-checksum"
89 #define CMD_LINE_OPT_TSO_SEGSZ "tso-segsz"
90 #define CMD_LINE_OPT_FILTER_TYPE "filter-type"
91 #define CMD_LINE_OPT_ENCAP "encap"
92 #define CMD_LINE_OPT_DECAP "decap"
93 #define CMD_LINE_OPT_RX_RETRY "rx-retry"
94 #define CMD_LINE_OPT_RX_RETRY_DELAY "rx-retry-delay"
95 #define CMD_LINE_OPT_RX_RETRY_NUM "rx-retry-num"
96 #define CMD_LINE_OPT_STATS "stats"
97 #define CMD_LINE_OPT_DEV_BASENAME "dev-basename"
98
99 /* mask of enabled ports */
100 static uint32_t enabled_port_mask;
101
102 /*Number of switching cores enabled*/
103 static uint32_t nb_switching_cores;
104
105 /* number of devices/queues to support*/
106 uint16_t nb_devices = 2;
107
108 /* max ring descriptor, ixgbe, i40e, e1000 all are 4096. */
109 #define MAX_RING_DESC 4096
110
111 struct vpool {
112 struct rte_mempool *pool;
113 struct rte_ring *ring;
114 uint32_t buf_size;
115 } vpool_array[MAX_QUEUES+MAX_QUEUES];
116
117 /* UDP tunneling port */
118 uint16_t udp_port = 4789;
119
120 /* enable/disable inner TX checksum */
121 uint8_t tx_checksum = 0;
122
123 /* TCP segment size */
124 uint16_t tso_segsz = 0;
125
126 /* enable/disable decapsulation */
127 uint8_t rx_decap = 1;
128
129 /* enable/disable encapsulation */
130 uint8_t tx_encap = 1;
131
132 /* RX filter type for tunneling packet */
133 uint8_t filter_idx = 1;
134
135 /* overlay packet operation */
136 struct ol_switch_ops overlay_options = {
137 .port_configure = vxlan_port_init,
138 .tunnel_setup = vxlan_link,
139 .tunnel_destroy = vxlan_unlink,
140 .tx_handle = vxlan_tx_pkts,
141 .rx_handle = vxlan_rx_pkts,
142 .param_handle = NULL,
143 };
144
145 /* Enable stats. */
146 uint32_t enable_stats = 0;
147 /* Enable retries on RX. */
148 static uint32_t enable_retry = 1;
149 /* Specify timeout (in useconds) between retries on RX. */
150 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
151 /* Specify the number of retries on RX. */
152 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
153
154 /* Character device basename. Can be set by user. */
155 static char dev_basename[MAX_BASENAME_SZ] = "vhost-net";
156
157 static unsigned lcore_ids[RTE_MAX_LCORE];
158 uint16_t ports[RTE_MAX_ETHPORTS];
159
160 static unsigned nb_ports; /**< The number of ports specified in command line */
161
162 /* ethernet addresses of ports */
163 struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
164
165 /* heads for the main used and free linked lists for the data path. */
166 static struct virtio_net_data_ll *ll_root_used;
167 static struct virtio_net_data_ll *ll_root_free;
168
169 /**
170 * Array of data core structures containing information on
171 * individual core linked lists.
172 */
173 static struct lcore_info lcore_info[RTE_MAX_LCORE];
174
175 /* Used for queueing bursts of TX packets. */
176 struct mbuf_table {
177 unsigned len;
178 unsigned txq_id;
179 struct rte_mbuf *m_table[MAX_PKT_BURST];
180 };
181
182 /* TX queue for each data core. */
183 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE];
184
185 struct device_statistics dev_statistics[MAX_DEVICES];
186
187 /**
188 * Set character device basename.
189 */
190 static int
191 us_vhost_parse_basename(const char *q_arg)
192 {
193 /* parse number string */
194 if (strlen(q_arg) >= MAX_BASENAME_SZ)
195 return -1;
196 else
197 snprintf((char *)&dev_basename, MAX_BASENAME_SZ, "%s", q_arg);
198
199 return 0;
200 }
201
202 /**
203 * Parse the portmask provided at run time.
204 */
205 static int
206 parse_portmask(const char *portmask)
207 {
208 char *end = NULL;
209 unsigned long pm;
210
211 /* parse hexadecimal string */
212 pm = strtoul(portmask, &end, 16);
213 if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
214 return -1;
215
216 if (pm == 0)
217 return -1;
218
219 return pm;
220 }
221
222 /**
223 * Parse num options at run time.
224 */
225 static int
226 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
227 {
228 char *end = NULL;
229 unsigned long num;
230
231 /* parse unsigned int string */
232 num = strtoul(q_arg, &end, 10);
233 if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
234 return -1;
235
236 if (num > max_valid_value)
237 return -1;
238
239 return num;
240 }
241
242 /**
243 * Display usage
244 */
245 static void
246 tep_termination_usage(const char *prgname)
247 {
248 RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
249 " --udp-port: UDP destination port for VXLAN packet\n"
250 " --nb-devices[1-64]: The number of virtIO device\n"
251 " --tx-checksum [0|1]: inner Tx checksum offload\n"
252 " --tso-segsz [0-N]: TCP segment size\n"
253 " --decap [0|1]: tunneling packet decapsulation\n"
254 " --encap [0|1]: tunneling packet encapsulation\n"
255 " --filter-type[1-3]: filter type for tunneling packet\n"
256 " 1: Inner MAC and tenent ID\n"
257 " 2: Inner MAC and VLAN, and tenent ID\n"
258 " 3: Outer MAC, Inner MAC and tenent ID\n"
259 " -p PORTMASK: Set mask for ports to be used by application\n"
260 " --rx-retry [0|1]: disable/enable(default) retries on rx."
261 " Enable retry if destintation queue is full\n"
262 " --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX."
263 " This makes effect only if retries on rx enabled\n"
264 " --rx-retry-num [0-N]: the number of retries on rx."
265 " This makes effect only if retries on rx enabled\n"
266 " --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
267 " --dev-basename: The basename to be used for the character device.\n",
268 prgname);
269 }
270
271 /**
272 * Parse the arguments given in the command line of the application.
273 */
274 static int
275 tep_termination_parse_args(int argc, char **argv)
276 {
277 int opt, ret;
278 int option_index;
279 unsigned i;
280 const char *prgname = argv[0];
281 static struct option long_option[] = {
282 {CMD_LINE_OPT_NB_DEVICES, required_argument, NULL, 0},
283 {CMD_LINE_OPT_UDP_PORT, required_argument, NULL, 0},
284 {CMD_LINE_OPT_TX_CHECKSUM, required_argument, NULL, 0},
285 {CMD_LINE_OPT_TSO_SEGSZ, required_argument, NULL, 0},
286 {CMD_LINE_OPT_DECAP, required_argument, NULL, 0},
287 {CMD_LINE_OPT_ENCAP, required_argument, NULL, 0},
288 {CMD_LINE_OPT_FILTER_TYPE, required_argument, NULL, 0},
289 {CMD_LINE_OPT_RX_RETRY, required_argument, NULL, 0},
290 {CMD_LINE_OPT_RX_RETRY_DELAY, required_argument, NULL, 0},
291 {CMD_LINE_OPT_RX_RETRY_NUM, required_argument, NULL, 0},
292 {CMD_LINE_OPT_STATS, required_argument, NULL, 0},
293 {CMD_LINE_OPT_DEV_BASENAME, required_argument, NULL, 0},
294 {NULL, 0, 0, 0},
295 };
296
297 /* Parse command line */
298 while ((opt = getopt_long(argc, argv, "p:",
299 long_option, &option_index)) != EOF) {
300 switch (opt) {
301 /* Portmask */
302 case 'p':
303 enabled_port_mask = parse_portmask(optarg);
304 if (enabled_port_mask == 0) {
305 RTE_LOG(INFO, VHOST_CONFIG,
306 "Invalid portmask\n");
307 tep_termination_usage(prgname);
308 return -1;
309 }
310 break;
311 case 0:
312 if (!strncmp(long_option[option_index].name,
313 CMD_LINE_OPT_NB_DEVICES,
314 sizeof(CMD_LINE_OPT_NB_DEVICES))) {
315 ret = parse_num_opt(optarg, MAX_DEVICES);
316 if (ret == -1) {
317 RTE_LOG(INFO, VHOST_CONFIG,
318 "Invalid argument for nb-devices [0-%d]\n",
319 MAX_DEVICES);
320 tep_termination_usage(prgname);
321 return -1;
322 } else
323 nb_devices = ret;
324 }
325
326 /* Enable/disable retries on RX. */
327 if (!strncmp(long_option[option_index].name,
328 CMD_LINE_OPT_RX_RETRY,
329 sizeof(CMD_LINE_OPT_RX_RETRY))) {
330 ret = parse_num_opt(optarg, 1);
331 if (ret == -1) {
332 RTE_LOG(INFO, VHOST_CONFIG,
333 "Invalid argument for rx-retry [0|1]\n");
334 tep_termination_usage(prgname);
335 return -1;
336 } else
337 enable_retry = ret;
338 }
339
340 if (!strncmp(long_option[option_index].name,
341 CMD_LINE_OPT_TSO_SEGSZ,
342 sizeof(CMD_LINE_OPT_TSO_SEGSZ))) {
343 ret = parse_num_opt(optarg, INT16_MAX);
344 if (ret == -1) {
345 RTE_LOG(INFO, VHOST_CONFIG,
346 "Invalid argument for TCP segment size [0-N]\n");
347 tep_termination_usage(prgname);
348 return -1;
349 } else
350 tso_segsz = ret;
351 }
352
353 if (!strncmp(long_option[option_index].name,
354 CMD_LINE_OPT_UDP_PORT,
355 sizeof(CMD_LINE_OPT_UDP_PORT))) {
356 ret = parse_num_opt(optarg, INT16_MAX);
357 if (ret == -1) {
358 RTE_LOG(INFO, VHOST_CONFIG,
359 "Invalid argument for UDP port [0-N]\n");
360 tep_termination_usage(prgname);
361 return -1;
362 } else
363 udp_port = ret;
364 }
365
366 /* Specify the retries delay time (in useconds) on RX.*/
367 if (!strncmp(long_option[option_index].name,
368 CMD_LINE_OPT_RX_RETRY_DELAY,
369 sizeof(CMD_LINE_OPT_RX_RETRY_DELAY))) {
370 ret = parse_num_opt(optarg, INT32_MAX);
371 if (ret == -1) {
372 RTE_LOG(INFO, VHOST_CONFIG,
373 "Invalid argument for rx-retry-delay [0-N]\n");
374 tep_termination_usage(prgname);
375 return -1;
376 } else
377 burst_rx_delay_time = ret;
378 }
379
380 /* Specify the retries number on RX. */
381 if (!strncmp(long_option[option_index].name,
382 CMD_LINE_OPT_RX_RETRY_NUM,
383 sizeof(CMD_LINE_OPT_RX_RETRY_NUM))) {
384 ret = parse_num_opt(optarg, INT32_MAX);
385 if (ret == -1) {
386 RTE_LOG(INFO, VHOST_CONFIG,
387 "Invalid argument for rx-retry-num [0-N]\n");
388 tep_termination_usage(prgname);
389 return -1;
390 } else
391 burst_rx_retry_num = ret;
392 }
393
394 if (!strncmp(long_option[option_index].name,
395 CMD_LINE_OPT_TX_CHECKSUM,
396 sizeof(CMD_LINE_OPT_TX_CHECKSUM))) {
397 ret = parse_num_opt(optarg, 1);
398 if (ret == -1) {
399 RTE_LOG(INFO, VHOST_CONFIG,
400 "Invalid argument for tx-checksum [0|1]\n");
401 tep_termination_usage(prgname);
402 return -1;
403 } else
404 tx_checksum = ret;
405 }
406
407 if (!strncmp(long_option[option_index].name,
408 CMD_LINE_OPT_FILTER_TYPE,
409 sizeof(CMD_LINE_OPT_FILTER_TYPE))) {
410 ret = parse_num_opt(optarg, 3);
411 if ((ret == -1) || (ret == 0)) {
412 RTE_LOG(INFO, VHOST_CONFIG,
413 "Invalid argument for filter type [1-3]\n");
414 tep_termination_usage(prgname);
415 return -1;
416 } else
417 filter_idx = ret - 1;
418 }
419
420 /* Enable/disable encapsulation on RX. */
421 if (!strncmp(long_option[option_index].name,
422 CMD_LINE_OPT_DECAP,
423 sizeof(CMD_LINE_OPT_DECAP))) {
424 ret = parse_num_opt(optarg, 1);
425 if (ret == -1) {
426 RTE_LOG(INFO, VHOST_CONFIG,
427 "Invalid argument for decap [0|1]\n");
428 tep_termination_usage(prgname);
429 return -1;
430 } else
431 rx_decap = ret;
432 }
433
434 /* Enable/disable encapsulation on TX. */
435 if (!strncmp(long_option[option_index].name,
436 CMD_LINE_OPT_ENCAP,
437 sizeof(CMD_LINE_OPT_ENCAP))) {
438 ret = parse_num_opt(optarg, 1);
439 if (ret == -1) {
440 RTE_LOG(INFO, VHOST_CONFIG,
441 "Invalid argument for encap [0|1]\n");
442 tep_termination_usage(prgname);
443 return -1;
444 } else
445 tx_encap = ret;
446 }
447
448 /* Enable/disable stats. */
449 if (!strncmp(long_option[option_index].name,
450 CMD_LINE_OPT_STATS,
451 sizeof(CMD_LINE_OPT_STATS))) {
452 ret = parse_num_opt(optarg, INT32_MAX);
453 if (ret == -1) {
454 RTE_LOG(INFO, VHOST_CONFIG,
455 "Invalid argument for stats [0..N]\n");
456 tep_termination_usage(prgname);
457 return -1;
458 } else
459 enable_stats = ret;
460 }
461
462 /* Set character device basename. */
463 if (!strncmp(long_option[option_index].name,
464 CMD_LINE_OPT_DEV_BASENAME,
465 sizeof(CMD_LINE_OPT_DEV_BASENAME))) {
466 if (us_vhost_parse_basename(optarg) == -1) {
467 RTE_LOG(INFO, VHOST_CONFIG,
468 "Invalid argument for character "
469 "device basename (Max %d characters)\n",
470 MAX_BASENAME_SZ);
471 tep_termination_usage(prgname);
472 return -1;
473 }
474 }
475
476 break;
477
478 /* Invalid option - print options. */
479 default:
480 tep_termination_usage(prgname);
481 return -1;
482 }
483 }
484
485 for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
486 if (enabled_port_mask & (1 << i))
487 ports[nb_ports++] = (uint8_t)i;
488 }
489
490 if ((nb_ports == 0) || (nb_ports > MAX_SUP_PORTS)) {
491 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
492 "but only %u port can be enabled\n", nb_ports,
493 MAX_SUP_PORTS);
494 return -1;
495 }
496
497 return 0;
498 }
499
500 /**
501 * Update the global var NB_PORTS and array PORTS
502 * according to system ports number and return valid ports number
503 */
504 static unsigned
505 check_ports_num(unsigned max_nb_ports)
506 {
507 unsigned valid_nb_ports = nb_ports;
508 unsigned portid;
509
510 if (nb_ports > max_nb_ports) {
511 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) "
512 " exceeds total system port number(%u)\n",
513 nb_ports, max_nb_ports);
514 nb_ports = max_nb_ports;
515 }
516
517 for (portid = 0; portid < nb_ports; portid++) {
518 if (!rte_eth_dev_is_valid_port(ports[portid])) {
519 RTE_LOG(INFO, VHOST_PORT,
520 "\nSpecified port ID(%u) is not valid\n",
521 ports[portid]);
522 ports[portid] = INVALID_PORT_ID;
523 valid_nb_ports--;
524 }
525 }
526 return valid_nb_ports;
527 }
528
529 /**
530 * This function routes the TX packet to the correct interface. This may be a local device
531 * or the physical port.
532 */
533 static __rte_always_inline void
534 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m)
535 {
536 struct mbuf_table *tx_q;
537 struct rte_mbuf **m_table;
538 unsigned len, ret = 0;
539 const uint16_t lcore_id = rte_lcore_id();
540
541 RTE_LOG_DP(DEBUG, VHOST_DATA, "(%d) TX: MAC address is external\n",
542 vdev->vid);
543
544 /* Add packet to the port tx queue */
545 tx_q = &lcore_tx_queue[lcore_id];
546 len = tx_q->len;
547
548 tx_q->m_table[len] = m;
549 len++;
550 if (enable_stats) {
551 dev_statistics[vdev->vid].tx_total++;
552 dev_statistics[vdev->vid].tx++;
553 }
554
555 if (unlikely(len == MAX_PKT_BURST)) {
556 m_table = (struct rte_mbuf **)tx_q->m_table;
557 ret = overlay_options.tx_handle(ports[0],
558 (uint16_t)tx_q->txq_id, m_table,
559 (uint16_t)tx_q->len);
560
561 /* Free any buffers not handled by TX and update
562 * the port stats.
563 */
564 if (unlikely(ret < len)) {
565 do {
566 rte_pktmbuf_free(m_table[ret]);
567 } while (++ret < len);
568 }
569
570 len = 0;
571 }
572
573 tx_q->len = len;
574 return;
575 }
576
577 /**
578 * This function is called by each data core. It handles all
579 * RX/TX registered with the core. For TX the specific lcore
580 * linked list is used. For RX, MAC addresses are compared
581 * with all devices in the main linked list.
582 */
583 static int
584 switch_worker(__rte_unused void *arg)
585 {
586 struct rte_mempool *mbuf_pool = arg;
587 struct vhost_dev *vdev = NULL;
588 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
589 struct virtio_net_data_ll *dev_ll;
590 struct mbuf_table *tx_q;
591 volatile struct lcore_ll_info *lcore_ll;
592 const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
593 / US_PER_S * BURST_TX_DRAIN_US;
594 uint64_t prev_tsc, diff_tsc, cur_tsc, ret_count = 0;
595 unsigned i, ret = 0;
596 const uint16_t lcore_id = rte_lcore_id();
597 const uint16_t num_cores = (uint16_t)rte_lcore_count();
598 uint16_t rx_count = 0;
599 uint16_t tx_count;
600 uint32_t retry = 0;
601
602 RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
603 lcore_ll = lcore_info[lcore_id].lcore_ll;
604 prev_tsc = 0;
605
606 tx_q = &lcore_tx_queue[lcore_id];
607 for (i = 0; i < num_cores; i++) {
608 if (lcore_ids[i] == lcore_id) {
609 tx_q->txq_id = i;
610 break;
611 }
612 }
613
614 while (1) {
615 cur_tsc = rte_rdtsc();
616 /*
617 * TX burst queue drain
618 */
619 diff_tsc = cur_tsc - prev_tsc;
620 if (unlikely(diff_tsc > drain_tsc)) {
621
622 if (tx_q->len) {
623 RTE_LOG_DP(DEBUG, VHOST_DATA, "TX queue drained after "
624 "timeout with burst size %u\n",
625 tx_q->len);
626 ret = overlay_options.tx_handle(ports[0],
627 (uint16_t)tx_q->txq_id,
628 (struct rte_mbuf **)tx_q->m_table,
629 (uint16_t)tx_q->len);
630 if (unlikely(ret < tx_q->len)) {
631 do {
632 rte_pktmbuf_free(tx_q->m_table[ret]);
633 } while (++ret < tx_q->len);
634 }
635
636 tx_q->len = 0;
637 }
638
639 prev_tsc = cur_tsc;
640
641 }
642
643 rte_prefetch0(lcore_ll->ll_root_used);
644
645 /**
646 * Inform the configuration core that we have exited
647 * the linked list and that no devices are
648 * in use if requested.
649 */
650 if (lcore_ll->dev_removal_flag == REQUEST_DEV_REMOVAL)
651 lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL;
652
653 /*
654 * Process devices
655 */
656 dev_ll = lcore_ll->ll_root_used;
657
658 while (dev_ll != NULL) {
659 vdev = dev_ll->vdev;
660
661 if (unlikely(vdev->remove)) {
662 dev_ll = dev_ll->next;
663 overlay_options.tunnel_destroy(vdev);
664 vdev->ready = DEVICE_SAFE_REMOVE;
665 continue;
666 }
667 if (likely(vdev->ready == DEVICE_RX)) {
668 /* Handle guest RX */
669 rx_count = rte_eth_rx_burst(ports[0],
670 vdev->rx_q, pkts_burst, MAX_PKT_BURST);
671
672 if (rx_count) {
673 /*
674 * Retry is enabled and the queue is
675 * full then we wait and retry to
676 * avoid packet loss. Here MAX_PKT_BURST
677 * must be less than virtio queue size
678 */
679 if (enable_retry && unlikely(rx_count >
680 rte_vhost_avail_entries(vdev->vid, VIRTIO_RXQ))) {
681 for (retry = 0; retry < burst_rx_retry_num;
682 retry++) {
683 rte_delay_us(burst_rx_delay_time);
684 if (rx_count <= rte_vhost_avail_entries(vdev->vid, VIRTIO_RXQ))
685 break;
686 }
687 }
688
689 ret_count = overlay_options.rx_handle(vdev->vid, pkts_burst, rx_count);
690 if (enable_stats) {
691 rte_atomic64_add(
692 &dev_statistics[vdev->vid].rx_total_atomic,
693 rx_count);
694 rte_atomic64_add(
695 &dev_statistics[vdev->vid].rx_atomic, ret_count);
696 }
697 while (likely(rx_count)) {
698 rx_count--;
699 rte_pktmbuf_free(pkts_burst[rx_count]);
700 }
701
702 }
703 }
704
705 if (likely(!vdev->remove)) {
706 /* Handle guest TX*/
707 tx_count = rte_vhost_dequeue_burst(vdev->vid,
708 VIRTIO_TXQ, mbuf_pool,
709 pkts_burst, MAX_PKT_BURST);
710 /* If this is the first received packet we need to learn the MAC */
711 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && tx_count) {
712 if (vdev->remove ||
713 (overlay_options.tunnel_setup(vdev, pkts_burst[0]) == -1)) {
714 while (tx_count)
715 rte_pktmbuf_free(pkts_burst[--tx_count]);
716 }
717 }
718 while (tx_count)
719 virtio_tx_route(vdev, pkts_burst[--tx_count]);
720 }
721
722 /* move to the next device in the list */
723 dev_ll = dev_ll->next;
724 }
725 }
726
727 return 0;
728 }
729
730 /**
731 * Add an entry to a used linked list. A free entry must first be found
732 * in the free linked list using get_data_ll_free_entry();
733 */
734 static void
735 add_data_ll_entry(struct virtio_net_data_ll **ll_root_addr,
736 struct virtio_net_data_ll *ll_dev)
737 {
738 struct virtio_net_data_ll *ll = *ll_root_addr;
739
740 /* Set next as NULL and use a compiler barrier to avoid reordering. */
741 ll_dev->next = NULL;
742 rte_compiler_barrier();
743
744 /* If ll == NULL then this is the first device. */
745 if (ll) {
746 /* Increment to the tail of the linked list. */
747 while (ll->next != NULL)
748 ll = ll->next;
749
750 ll->next = ll_dev;
751 } else {
752 *ll_root_addr = ll_dev;
753 }
754 }
755
756 /**
757 * Remove an entry from a used linked list. The entry must then be added to
758 * the free linked list using put_data_ll_free_entry().
759 */
760 static void
761 rm_data_ll_entry(struct virtio_net_data_ll **ll_root_addr,
762 struct virtio_net_data_ll *ll_dev,
763 struct virtio_net_data_ll *ll_dev_last)
764 {
765 struct virtio_net_data_ll *ll = *ll_root_addr;
766
767 if (unlikely((ll == NULL) || (ll_dev == NULL)))
768 return;
769
770 if (ll_dev == ll)
771 *ll_root_addr = ll_dev->next;
772 else
773 if (likely(ll_dev_last != NULL))
774 ll_dev_last->next = ll_dev->next;
775 else
776 RTE_LOG(ERR, VHOST_CONFIG,
777 "Remove entry form ll failed.\n");
778 }
779
780 /**
781 * Find and return an entry from the free linked list.
782 */
783 static struct virtio_net_data_ll *
784 get_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr)
785 {
786 struct virtio_net_data_ll *ll_free = *ll_root_addr;
787 struct virtio_net_data_ll *ll_dev;
788
789 if (ll_free == NULL)
790 return NULL;
791
792 ll_dev = ll_free;
793 *ll_root_addr = ll_free->next;
794
795 return ll_dev;
796 }
797
798 /**
799 * Place an entry back on to the free linked list.
800 */
801 static void
802 put_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr,
803 struct virtio_net_data_ll *ll_dev)
804 {
805 struct virtio_net_data_ll *ll_free = *ll_root_addr;
806
807 if (ll_dev == NULL)
808 return;
809
810 ll_dev->next = ll_free;
811 *ll_root_addr = ll_dev;
812 }
813
814 /**
815 * Creates a linked list of a given size.
816 */
817 static struct virtio_net_data_ll *
818 alloc_data_ll(uint32_t size)
819 {
820 struct virtio_net_data_ll *ll_new;
821 uint32_t i;
822
823 /* Malloc and then chain the linked list. */
824 ll_new = malloc(size * sizeof(struct virtio_net_data_ll));
825 if (ll_new == NULL) {
826 RTE_LOG(ERR, VHOST_CONFIG,
827 "Failed to allocate memory for ll_new.\n");
828 return NULL;
829 }
830
831 for (i = 0; i < size - 1; i++) {
832 ll_new[i].vdev = NULL;
833 ll_new[i].next = &ll_new[i+1];
834 }
835 ll_new[i].next = NULL;
836
837 return ll_new;
838 }
839
840 /**
841 * Create the main linked list along with each individual cores
842 * linked list. A used and a free list are created to manage entries.
843 */
844 static int
845 init_data_ll(void)
846 {
847 int lcore;
848
849 RTE_LCORE_FOREACH_SLAVE(lcore) {
850 lcore_info[lcore].lcore_ll =
851 malloc(sizeof(struct lcore_ll_info));
852 if (lcore_info[lcore].lcore_ll == NULL) {
853 RTE_LOG(ERR, VHOST_CONFIG,
854 "Failed to allocate memory for lcore_ll.\n");
855 return -1;
856 }
857
858 lcore_info[lcore].lcore_ll->device_num = 0;
859 lcore_info[lcore].lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL;
860 lcore_info[lcore].lcore_ll->ll_root_used = NULL;
861 if (nb_devices % nb_switching_cores)
862 lcore_info[lcore].lcore_ll->ll_root_free =
863 alloc_data_ll((nb_devices / nb_switching_cores)
864 + 1);
865 else
866 lcore_info[lcore].lcore_ll->ll_root_free =
867 alloc_data_ll(nb_devices / nb_switching_cores);
868 }
869
870 /* Allocate devices up to a maximum of MAX_DEVICES. */
871 ll_root_free = alloc_data_ll(MIN((nb_devices), MAX_DEVICES));
872
873 return 0;
874 }
875
876 /**
877 * Remove a device from the specific data core linked list and
878 * from the main linked list. Synchonization occurs through the use
879 * of the lcore dev_removal_flag.
880 */
881 static void
882 destroy_device(int vid)
883 {
884 struct virtio_net_data_ll *ll_lcore_dev_cur;
885 struct virtio_net_data_ll *ll_main_dev_cur;
886 struct virtio_net_data_ll *ll_lcore_dev_last = NULL;
887 struct virtio_net_data_ll *ll_main_dev_last = NULL;
888 struct vhost_dev *vdev = NULL;
889 int lcore;
890
891 ll_main_dev_cur = ll_root_used;
892 while (ll_main_dev_cur != NULL) {
893 if (ll_main_dev_cur->vdev->vid == vid) {
894 vdev = ll_main_dev_cur->vdev;
895 break;
896 }
897 }
898 if (!vdev)
899 return;
900
901 /* set the remove flag. */
902 vdev->remove = 1;
903 while (vdev->ready != DEVICE_SAFE_REMOVE)
904 rte_pause();
905
906 /* Search for entry to be removed from lcore ll */
907 ll_lcore_dev_cur = lcore_info[vdev->coreid].lcore_ll->ll_root_used;
908 while (ll_lcore_dev_cur != NULL) {
909 if (ll_lcore_dev_cur->vdev == vdev) {
910 break;
911 } else {
912 ll_lcore_dev_last = ll_lcore_dev_cur;
913 ll_lcore_dev_cur = ll_lcore_dev_cur->next;
914 }
915 }
916
917 if (ll_lcore_dev_cur == NULL) {
918 RTE_LOG(ERR, VHOST_CONFIG,
919 "(%d) Failed to find the dev to be destroy.\n", vid);
920 return;
921 }
922
923 /* Search for entry to be removed from main ll */
924 ll_main_dev_cur = ll_root_used;
925 ll_main_dev_last = NULL;
926 while (ll_main_dev_cur != NULL) {
927 if (ll_main_dev_cur->vdev == vdev) {
928 break;
929 } else {
930 ll_main_dev_last = ll_main_dev_cur;
931 ll_main_dev_cur = ll_main_dev_cur->next;
932 }
933 }
934
935 /* Remove entries from the lcore and main ll. */
936 rm_data_ll_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_used,
937 ll_lcore_dev_cur, ll_lcore_dev_last);
938 rm_data_ll_entry(&ll_root_used, ll_main_dev_cur, ll_main_dev_last);
939
940 /* Set the dev_removal_flag on each lcore. */
941 RTE_LCORE_FOREACH_SLAVE(lcore) {
942 lcore_info[lcore].lcore_ll->dev_removal_flag =
943 REQUEST_DEV_REMOVAL;
944 }
945
946 /*
947 * Once each core has set the dev_removal_flag to
948 * ACK_DEV_REMOVAL we can be sure that they can no longer access
949 * the device removed from the linked lists and that the devices
950 * are no longer in use.
951 */
952 RTE_LCORE_FOREACH_SLAVE(lcore) {
953 while (lcore_info[lcore].lcore_ll->dev_removal_flag
954 != ACK_DEV_REMOVAL)
955 rte_pause();
956 }
957
958 /* Add the entries back to the lcore and main free ll.*/
959 put_data_ll_free_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_free,
960 ll_lcore_dev_cur);
961 put_data_ll_free_entry(&ll_root_free, ll_main_dev_cur);
962
963 /* Decrement number of device on the lcore. */
964 lcore_info[vdev->coreid].lcore_ll->device_num--;
965
966 RTE_LOG(INFO, VHOST_DATA, "(%d) Device has been removed "
967 "from data core\n", vid);
968
969 rte_free(vdev);
970
971 }
972
973 /**
974 * A new device is added to a data core. First the device is added
975 * to the main linked list and the allocated to a specific data core.
976 */
977 static int
978 new_device(int vid)
979 {
980 struct virtio_net_data_ll *ll_dev;
981 int lcore, core_add = 0;
982 uint32_t device_num_min = nb_devices;
983 struct vhost_dev *vdev;
984
985 vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
986 if (vdev == NULL) {
987 RTE_LOG(INFO, VHOST_DATA,
988 "(%d) Couldn't allocate memory for vhost dev\n", vid);
989 return -1;
990 }
991 vdev->vid = vid;
992 /* Add device to main ll */
993 ll_dev = get_data_ll_free_entry(&ll_root_free);
994 if (ll_dev == NULL) {
995 RTE_LOG(INFO, VHOST_DATA, "(%d) No free entry found in"
996 " linked list Device limit of %d devices per core"
997 " has been reached\n", vid, nb_devices);
998 if (vdev->regions_hpa)
999 rte_free(vdev->regions_hpa);
1000 rte_free(vdev);
1001 return -1;
1002 }
1003 ll_dev->vdev = vdev;
1004 add_data_ll_entry(&ll_root_used, ll_dev);
1005 vdev->rx_q = vid;
1006
1007 /* reset ready flag */
1008 vdev->ready = DEVICE_MAC_LEARNING;
1009 vdev->remove = 0;
1010
1011 /* Find a suitable lcore to add the device. */
1012 RTE_LCORE_FOREACH_SLAVE(lcore) {
1013 if (lcore_info[lcore].lcore_ll->device_num < device_num_min) {
1014 device_num_min = lcore_info[lcore].lcore_ll->device_num;
1015 core_add = lcore;
1016 }
1017 }
1018 /* Add device to lcore ll */
1019 ll_dev = get_data_ll_free_entry(&lcore_info[core_add].lcore_ll->ll_root_free);
1020 if (ll_dev == NULL) {
1021 RTE_LOG(INFO, VHOST_DATA,
1022 "(%d) Failed to add device to data core\n",
1023 vid);
1024 vdev->ready = DEVICE_SAFE_REMOVE;
1025 destroy_device(vid);
1026 rte_free(vdev->regions_hpa);
1027 rte_free(vdev);
1028 return -1;
1029 }
1030 ll_dev->vdev = vdev;
1031 vdev->coreid = core_add;
1032
1033 add_data_ll_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_used,
1034 ll_dev);
1035
1036 /* Initialize device stats */
1037 memset(&dev_statistics[vid], 0,
1038 sizeof(struct device_statistics));
1039
1040 /* Disable notifications. */
1041 rte_vhost_enable_guest_notification(vid, VIRTIO_RXQ, 0);
1042 rte_vhost_enable_guest_notification(vid, VIRTIO_TXQ, 0);
1043 lcore_info[vdev->coreid].lcore_ll->device_num++;
1044
1045 RTE_LOG(INFO, VHOST_DATA, "(%d) Device has been added to data core %d\n",
1046 vid, vdev->coreid);
1047
1048 return 0;
1049 }
1050
1051 /**
1052 * These callback allow devices to be added to the data core when configuration
1053 * has been fully complete.
1054 */
1055 static const struct vhost_device_ops virtio_net_device_ops = {
1056 .new_device = new_device,
1057 .destroy_device = destroy_device,
1058 };
1059
1060 /**
1061 * This is a thread will wake up after a period to print stats if the user has
1062 * enabled them.
1063 */
1064 static void *
1065 print_stats(__rte_unused void *arg)
1066 {
1067 struct virtio_net_data_ll *dev_ll;
1068 uint64_t tx_dropped, rx_dropped;
1069 uint64_t tx, tx_total, rx, rx_total, rx_ip_csum, rx_l4_csum;
1070 int vid;
1071 const char clr[] = { 27, '[', '2', 'J', '\0' };
1072 const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
1073
1074 while (1) {
1075 sleep(enable_stats);
1076
1077 /* Clear screen and move to top left */
1078 printf("%s%s", clr, top_left);
1079
1080 printf("\nDevice statistics ================================");
1081
1082 dev_ll = ll_root_used;
1083 while (dev_ll != NULL) {
1084 vid = dev_ll->vdev->vid;
1085 tx_total = dev_statistics[vid].tx_total;
1086 tx = dev_statistics[vid].tx;
1087 tx_dropped = tx_total - tx;
1088
1089 rx_total = rte_atomic64_read(
1090 &dev_statistics[vid].rx_total_atomic);
1091 rx = rte_atomic64_read(
1092 &dev_statistics[vid].rx_atomic);
1093 rx_dropped = rx_total - rx;
1094 rx_ip_csum = rte_atomic64_read(
1095 &dev_statistics[vid].rx_bad_ip_csum);
1096 rx_l4_csum = rte_atomic64_read(
1097 &dev_statistics[vid].rx_bad_l4_csum);
1098
1099 printf("\nStatistics for device %d ----------"
1100 "\nTX total: %"PRIu64""
1101 "\nTX dropped: %"PRIu64""
1102 "\nTX successful: %"PRIu64""
1103 "\nRX total: %"PRIu64""
1104 "\nRX bad IP csum: %"PRIu64""
1105 "\nRX bad L4 csum: %"PRIu64""
1106 "\nRX dropped: %"PRIu64""
1107 "\nRX successful: %"PRIu64"",
1108 vid,
1109 tx_total,
1110 tx_dropped,
1111 tx,
1112 rx_total,
1113 rx_ip_csum,
1114 rx_l4_csum,
1115 rx_dropped,
1116 rx);
1117
1118 dev_ll = dev_ll->next;
1119 }
1120 printf("\n================================================\n");
1121 }
1122
1123 return NULL;
1124 }
1125
1126 /**
1127 * Main function, does initialisation and calls the per-lcore functions.
1128 */
1129 int
1130 main(int argc, char *argv[])
1131 {
1132 struct rte_mempool *mbuf_pool = NULL;
1133 unsigned lcore_id, core_id = 0;
1134 unsigned nb_ports, valid_nb_ports;
1135 int ret;
1136 uint16_t portid;
1137 uint16_t queue_id;
1138 static pthread_t tid;
1139
1140 /* init EAL */
1141 ret = rte_eal_init(argc, argv);
1142 if (ret < 0)
1143 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1144 argc -= ret;
1145 argv += ret;
1146
1147 /* parse app arguments */
1148 ret = tep_termination_parse_args(argc, argv);
1149 if (ret < 0)
1150 rte_exit(EXIT_FAILURE, "Invalid argument\n");
1151
1152 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
1153 if (rte_lcore_is_enabled(lcore_id))
1154 lcore_ids[core_id++] = lcore_id;
1155
1156 /* set the number of swithcing cores available */
1157 nb_switching_cores = rte_lcore_count()-1;
1158
1159 /* Get the number of physical ports. */
1160 nb_ports = rte_eth_dev_count_avail();
1161
1162 /*
1163 * Update the global var NB_PORTS and global array PORTS
1164 * and get value of var VALID_NB_PORTS according to system ports number
1165 */
1166 valid_nb_ports = check_ports_num(nb_ports);
1167
1168 if ((valid_nb_ports == 0) || (valid_nb_ports > MAX_SUP_PORTS)) {
1169 rte_exit(EXIT_FAILURE, "Current enabled port number is %u,"
1170 "but only %u port can be enabled\n", nb_ports,
1171 MAX_SUP_PORTS);
1172 }
1173 /* Create the mbuf pool. */
1174 mbuf_pool = rte_pktmbuf_pool_create(
1175 "MBUF_POOL",
1176 NUM_MBUFS_PER_PORT * valid_nb_ports,
1177 MBUF_CACHE_SIZE,
1178 0,
1179 MBUF_DATA_SIZE,
1180 rte_socket_id());
1181 if (mbuf_pool == NULL)
1182 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1183
1184 for (queue_id = 0; queue_id < MAX_QUEUES + 1; queue_id++)
1185 vpool_array[queue_id].pool = mbuf_pool;
1186
1187 /* initialize all ports */
1188 RTE_ETH_FOREACH_DEV(portid) {
1189 /* skip ports that are not enabled */
1190 if ((enabled_port_mask & (1 << portid)) == 0) {
1191 RTE_LOG(INFO, VHOST_PORT,
1192 "Skipping disabled port %d\n", portid);
1193 continue;
1194 }
1195 if (overlay_options.port_configure(portid, mbuf_pool) != 0)
1196 rte_exit(EXIT_FAILURE,
1197 "Cannot initialize network ports\n");
1198 }
1199
1200 /* Initialise all linked lists. */
1201 if (init_data_ll() == -1)
1202 rte_exit(EXIT_FAILURE, "Failed to initialize linked list\n");
1203
1204 /* Initialize device stats */
1205 memset(&dev_statistics, 0, sizeof(dev_statistics));
1206
1207 /* Enable stats if the user option is set. */
1208 if (enable_stats) {
1209 ret = rte_ctrl_thread_create(&tid, "print-stats", NULL,
1210 print_stats, NULL);
1211 if (ret < 0)
1212 rte_exit(EXIT_FAILURE, "Cannot create print-stats thread\n");
1213 }
1214
1215 /* Launch all data cores. */
1216 RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1217 rte_eal_remote_launch(switch_worker,
1218 mbuf_pool, lcore_id);
1219 }
1220
1221 ret = rte_vhost_driver_register((char *)&dev_basename, 0);
1222 if (ret != 0)
1223 rte_exit(EXIT_FAILURE, "failed to register vhost driver.\n");
1224
1225 rte_vhost_driver_disable_features(dev_basename,
1226 1ULL << VIRTIO_NET_F_MRG_RXBUF);
1227
1228 ret = rte_vhost_driver_callback_register(dev_basename,
1229 &virtio_net_device_ops);
1230 if (ret != 0) {
1231 rte_exit(EXIT_FAILURE,
1232 "failed to register vhost driver callbacks.\n");
1233 }
1234
1235 if (rte_vhost_driver_start(dev_basename) < 0) {
1236 rte_exit(EXIT_FAILURE,
1237 "failed to start vhost driver.\n");
1238 }
1239
1240 RTE_LCORE_FOREACH_SLAVE(lcore_id)
1241 rte_eal_wait_lcore(lcore_id);
1242
1243 return 0;
1244 }