mempool/octeontx2: add fast path mempool ops
[dpdk.git] / 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 /* Maximum character device basename size. */
75 #define MAX_BASENAME_SZ 20
76
77 /* Maximum long option length for option parsing. */
78 #define MAX_LONG_OPT_SZ 64
79
80 /* Used to compare MAC addresses. */
81 #define MAC_ADDR_CMP 0xFFFFFFFFFFFFULL
82
83 #define CMD_LINE_OPT_NB_DEVICES "nb-devices"
84 #define CMD_LINE_OPT_UDP_PORT "udp-port"
85 #define CMD_LINE_OPT_TX_CHECKSUM "tx-checksum"
86 #define CMD_LINE_OPT_TSO_SEGSZ "tso-segsz"
87 #define CMD_LINE_OPT_FILTER_TYPE "filter-type"
88 #define CMD_LINE_OPT_ENCAP "encap"
89 #define CMD_LINE_OPT_DECAP "decap"
90 #define CMD_LINE_OPT_RX_RETRY "rx-retry"
91 #define CMD_LINE_OPT_RX_RETRY_DELAY "rx-retry-delay"
92 #define CMD_LINE_OPT_RX_RETRY_NUM "rx-retry-num"
93 #define CMD_LINE_OPT_STATS "stats"
94 #define CMD_LINE_OPT_DEV_BASENAME "dev-basename"
95
96 /* mask of enabled ports */
97 static uint32_t enabled_port_mask;
98
99 /*Number of switching cores enabled*/
100 static uint32_t nb_switching_cores;
101
102 /* number of devices/queues to support*/
103 uint16_t nb_devices = 2;
104
105 /* max ring descriptor, ixgbe, i40e, e1000 all are 4096. */
106 #define MAX_RING_DESC 4096
107
108 struct vpool {
109         struct rte_mempool *pool;
110         struct rte_ring *ring;
111         uint32_t buf_size;
112 } vpool_array[MAX_QUEUES+MAX_QUEUES];
113
114 /* UDP tunneling port */
115 uint16_t udp_port = 4789;
116
117 /* enable/disable inner TX checksum */
118 uint8_t tx_checksum = 0;
119
120 /* TCP segment size */
121 uint16_t tso_segsz = 0;
122
123 /* enable/disable decapsulation */
124 uint8_t rx_decap = 1;
125
126 /* enable/disable encapsulation */
127 uint8_t tx_encap = 1;
128
129 /* RX filter type for tunneling packet */
130 uint8_t filter_idx = 1;
131
132 /* overlay packet operation */
133 struct ol_switch_ops overlay_options = {
134         .port_configure = vxlan_port_init,
135         .tunnel_setup = vxlan_link,
136         .tunnel_destroy = vxlan_unlink,
137         .tx_handle = vxlan_tx_pkts,
138         .rx_handle = vxlan_rx_pkts,
139         .param_handle = NULL,
140 };
141
142 /* Enable stats. */
143 uint32_t enable_stats = 0;
144 /* Enable retries on RX. */
145 static uint32_t enable_retry = 1;
146 /* Specify timeout (in useconds) between retries on RX. */
147 static uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
148 /* Specify the number of retries on RX. */
149 static uint32_t burst_rx_retry_num = BURST_RX_RETRIES;
150
151 /* Character device basename. Can be set by user. */
152 static char dev_basename[MAX_BASENAME_SZ] = "vhost-net";
153
154 static unsigned lcore_ids[RTE_MAX_LCORE];
155 uint16_t ports[RTE_MAX_ETHPORTS];
156
157 static unsigned nb_ports; /**< The number of ports specified in command line */
158
159 /* ethernet addresses of ports */
160 struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
161
162 /* heads for the main used and free linked lists for the data path. */
163 static struct virtio_net_data_ll *ll_root_used;
164 static struct virtio_net_data_ll *ll_root_free;
165
166 /**
167  * Array of data core structures containing information on
168  * individual core linked lists.
169  */
170 static struct lcore_info lcore_info[RTE_MAX_LCORE];
171
172 /* Used for queueing bursts of TX packets. */
173 struct mbuf_table {
174         unsigned len;
175         unsigned txq_id;
176         struct rte_mbuf *m_table[MAX_PKT_BURST];
177 };
178
179 /* TX queue for each data core. */
180 struct mbuf_table lcore_tx_queue[RTE_MAX_LCORE];
181
182 struct device_statistics dev_statistics[MAX_DEVICES];
183
184 /**
185  * Set character device basename.
186  */
187 static int
188 us_vhost_parse_basename(const char *q_arg)
189 {
190         /* parse number string */
191         if (strlen(q_arg) >= MAX_BASENAME_SZ)
192                 return -1;
193         else
194                 strlcpy((char *)&dev_basename, q_arg, MAX_BASENAME_SZ);
195
196         return 0;
197 }
198
199 /**
200  * Parse the portmask provided at run time.
201  */
202 static int
203 parse_portmask(const char *portmask)
204 {
205         char *end = NULL;
206         unsigned long pm;
207
208         /* parse hexadecimal string */
209         pm = strtoul(portmask, &end, 16);
210         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
211                 return -1;
212
213         if (pm == 0)
214                 return -1;
215
216         return pm;
217 }
218
219 /**
220  * Parse num options at run time.
221  */
222 static int
223 parse_num_opt(const char *q_arg, uint32_t max_valid_value)
224 {
225         char *end = NULL;
226         unsigned long num;
227
228         /* parse unsigned int string */
229         num = strtoul(q_arg, &end, 10);
230         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
231                 return -1;
232
233         if (num > max_valid_value)
234                 return -1;
235
236         return num;
237 }
238
239 /**
240  * Display usage
241  */
242 static void
243 tep_termination_usage(const char *prgname)
244 {
245         RTE_LOG(INFO, VHOST_CONFIG, "%s [EAL options] -- -p PORTMASK\n"
246         "               --udp-port: UDP destination port for VXLAN packet\n"
247         "               --nb-devices[1-64]: The number of virtIO device\n"
248         "               --tx-checksum [0|1]: inner Tx checksum offload\n"
249         "               --tso-segsz [0-N]: TCP segment size\n"
250         "               --decap [0|1]: tunneling packet decapsulation\n"
251         "               --encap [0|1]: tunneling packet encapsulation\n"
252         "               --filter-type[1-3]: filter type for tunneling packet\n"
253         "                   1: Inner MAC and tenent ID\n"
254         "                   2: Inner MAC and VLAN, and tenent ID\n"
255         "                   3: Outer MAC, Inner MAC and tenent ID\n"
256         "               -p PORTMASK: Set mask for ports to be used by application\n"
257         "               --rx-retry [0|1]: disable/enable(default) retries on rx."
258         "                Enable retry if destintation queue is full\n"
259         "               --rx-retry-delay [0-N]: timeout(in usecond) between retries on RX."
260         "                This makes effect only if retries on rx enabled\n"
261         "               --rx-retry-num [0-N]: the number of retries on rx."
262         "                This makes effect only if retries on rx enabled\n"
263         "               --stats [0-N]: 0: Disable stats, N: Time in seconds to print stats\n"
264         "               --dev-basename: The basename to be used for the character device.\n",
265                prgname);
266 }
267
268 /**
269  * Parse the arguments given in the command line of the application.
270  */
271 static int
272 tep_termination_parse_args(int argc, char **argv)
273 {
274         int opt, ret;
275         int option_index;
276         unsigned i;
277         const char *prgname = argv[0];
278         static struct option long_option[] = {
279                 {CMD_LINE_OPT_NB_DEVICES, required_argument, NULL, 0},
280                 {CMD_LINE_OPT_UDP_PORT, required_argument, NULL, 0},
281                 {CMD_LINE_OPT_TX_CHECKSUM, required_argument, NULL, 0},
282                 {CMD_LINE_OPT_TSO_SEGSZ, required_argument, NULL, 0},
283                 {CMD_LINE_OPT_DECAP, required_argument, NULL, 0},
284                 {CMD_LINE_OPT_ENCAP, required_argument, NULL, 0},
285                 {CMD_LINE_OPT_FILTER_TYPE, required_argument, NULL, 0},
286                 {CMD_LINE_OPT_RX_RETRY, required_argument, NULL, 0},
287                 {CMD_LINE_OPT_RX_RETRY_DELAY, required_argument, NULL, 0},
288                 {CMD_LINE_OPT_RX_RETRY_NUM, required_argument, NULL, 0},
289                 {CMD_LINE_OPT_STATS, required_argument, NULL, 0},
290                 {CMD_LINE_OPT_DEV_BASENAME, required_argument, NULL, 0},
291                 {NULL, 0, 0, 0},
292         };
293
294         /* Parse command line */
295         while ((opt = getopt_long(argc, argv, "p:",
296                         long_option, &option_index)) != EOF) {
297                 switch (opt) {
298                 /* Portmask */
299                 case 'p':
300                         enabled_port_mask = parse_portmask(optarg);
301                         if (enabled_port_mask == 0) {
302                                 RTE_LOG(INFO, VHOST_CONFIG,
303                                         "Invalid portmask\n");
304                                 tep_termination_usage(prgname);
305                                 return -1;
306                         }
307                         break;
308                 case 0:
309                         if (!strncmp(long_option[option_index].name,
310                                 CMD_LINE_OPT_NB_DEVICES,
311                                 sizeof(CMD_LINE_OPT_NB_DEVICES))) {
312                                 ret = parse_num_opt(optarg, MAX_DEVICES);
313                                 if (ret == -1) {
314                                         RTE_LOG(INFO, VHOST_CONFIG,
315                                         "Invalid argument for nb-devices [0-%d]\n",
316                                         MAX_DEVICES);
317                                         tep_termination_usage(prgname);
318                                         return -1;
319                                 } else
320                                         nb_devices = ret;
321                         }
322
323                         /* Enable/disable retries on RX. */
324                         if (!strncmp(long_option[option_index].name,
325                                 CMD_LINE_OPT_RX_RETRY,
326                                 sizeof(CMD_LINE_OPT_RX_RETRY))) {
327                                 ret = parse_num_opt(optarg, 1);
328                                 if (ret == -1) {
329                                         RTE_LOG(INFO, VHOST_CONFIG,
330                                                 "Invalid argument for rx-retry [0|1]\n");
331                                         tep_termination_usage(prgname);
332                                         return -1;
333                                 } else
334                                         enable_retry = ret;
335                         }
336
337                         if (!strncmp(long_option[option_index].name,
338                                 CMD_LINE_OPT_TSO_SEGSZ,
339                                 sizeof(CMD_LINE_OPT_TSO_SEGSZ))) {
340                                 ret = parse_num_opt(optarg, INT16_MAX);
341                                 if (ret == -1) {
342                                         RTE_LOG(INFO, VHOST_CONFIG,
343                                                 "Invalid argument for TCP segment size [0-N]\n");
344                                         tep_termination_usage(prgname);
345                                         return -1;
346                                 } else
347                                         tso_segsz = ret;
348                         }
349
350                         if (!strncmp(long_option[option_index].name,
351                                         CMD_LINE_OPT_UDP_PORT,
352                                         sizeof(CMD_LINE_OPT_UDP_PORT))) {
353                                 ret = parse_num_opt(optarg, INT16_MAX);
354                                 if (ret == -1) {
355                                         RTE_LOG(INFO, VHOST_CONFIG,
356                                                 "Invalid argument for UDP port [0-N]\n");
357                                         tep_termination_usage(prgname);
358                                         return -1;
359                                 } else
360                                         udp_port = ret;
361                         }
362
363                         /* Specify the retries delay time (in useconds) on RX.*/
364                         if (!strncmp(long_option[option_index].name,
365                                 CMD_LINE_OPT_RX_RETRY_DELAY,
366                                 sizeof(CMD_LINE_OPT_RX_RETRY_DELAY))) {
367                                 ret = parse_num_opt(optarg, INT32_MAX);
368                                 if (ret == -1) {
369                                         RTE_LOG(INFO, VHOST_CONFIG,
370                                                 "Invalid argument for rx-retry-delay [0-N]\n");
371                                         tep_termination_usage(prgname);
372                                         return -1;
373                                 } else
374                                         burst_rx_delay_time = ret;
375                         }
376
377                         /* Specify the retries number on RX. */
378                         if (!strncmp(long_option[option_index].name,
379                                 CMD_LINE_OPT_RX_RETRY_NUM,
380                                 sizeof(CMD_LINE_OPT_RX_RETRY_NUM))) {
381                                 ret = parse_num_opt(optarg, INT32_MAX);
382                                 if (ret == -1) {
383                                         RTE_LOG(INFO, VHOST_CONFIG,
384                                                 "Invalid argument for rx-retry-num [0-N]\n");
385                                         tep_termination_usage(prgname);
386                                         return -1;
387                                 } else
388                                         burst_rx_retry_num = ret;
389                         }
390
391                         if (!strncmp(long_option[option_index].name,
392                                 CMD_LINE_OPT_TX_CHECKSUM,
393                                 sizeof(CMD_LINE_OPT_TX_CHECKSUM))) {
394                                 ret = parse_num_opt(optarg, 1);
395                                 if (ret == -1) {
396                                         RTE_LOG(INFO, VHOST_CONFIG,
397                                                 "Invalid argument for tx-checksum [0|1]\n");
398                                         tep_termination_usage(prgname);
399                                         return -1;
400                                 } else
401                                         tx_checksum = ret;
402                         }
403
404                         if (!strncmp(long_option[option_index].name,
405                                         CMD_LINE_OPT_FILTER_TYPE,
406                                         sizeof(CMD_LINE_OPT_FILTER_TYPE))) {
407                                 ret = parse_num_opt(optarg, 3);
408                                 if ((ret == -1) || (ret == 0)) {
409                                         RTE_LOG(INFO, VHOST_CONFIG,
410                                                 "Invalid argument for filter type [1-3]\n");
411                                         tep_termination_usage(prgname);
412                                         return -1;
413                                 } else
414                                         filter_idx = ret - 1;
415                         }
416
417                         /* Enable/disable encapsulation on RX. */
418                         if (!strncmp(long_option[option_index].name,
419                                 CMD_LINE_OPT_DECAP,
420                                 sizeof(CMD_LINE_OPT_DECAP))) {
421                                 ret = parse_num_opt(optarg, 1);
422                                 if (ret == -1) {
423                                         RTE_LOG(INFO, VHOST_CONFIG,
424                                                 "Invalid argument for decap [0|1]\n");
425                                         tep_termination_usage(prgname);
426                                         return -1;
427                                 } else
428                                         rx_decap = ret;
429                         }
430
431                         /* Enable/disable encapsulation on TX. */
432                         if (!strncmp(long_option[option_index].name,
433                                 CMD_LINE_OPT_ENCAP,
434                                 sizeof(CMD_LINE_OPT_ENCAP))) {
435                                 ret = parse_num_opt(optarg, 1);
436                                 if (ret == -1) {
437                                         RTE_LOG(INFO, VHOST_CONFIG,
438                                                 "Invalid argument for encap [0|1]\n");
439                                         tep_termination_usage(prgname);
440                                         return -1;
441                                 } else
442                                         tx_encap = ret;
443                         }
444
445                         /* Enable/disable stats. */
446                         if (!strncmp(long_option[option_index].name,
447                                 CMD_LINE_OPT_STATS,
448                                 sizeof(CMD_LINE_OPT_STATS))) {
449                                 ret = parse_num_opt(optarg, INT32_MAX);
450                                 if (ret == -1) {
451                                         RTE_LOG(INFO, VHOST_CONFIG,
452                                                         "Invalid argument for stats [0..N]\n");
453                                         tep_termination_usage(prgname);
454                                         return -1;
455                                 } else
456                                         enable_stats = ret;
457                         }
458
459                         /* Set character device basename. */
460                         if (!strncmp(long_option[option_index].name,
461                                 CMD_LINE_OPT_DEV_BASENAME,
462                                 sizeof(CMD_LINE_OPT_DEV_BASENAME))) {
463                                 if (us_vhost_parse_basename(optarg) == -1) {
464                                         RTE_LOG(INFO, VHOST_CONFIG,
465                                                 "Invalid argument for character "
466                                                 "device basename (Max %d characters)\n",
467                                                 MAX_BASENAME_SZ);
468                                         tep_termination_usage(prgname);
469                                         return -1;
470                                 }
471                         }
472
473                         break;
474
475                         /* Invalid option - print options. */
476                 default:
477                         tep_termination_usage(prgname);
478                         return -1;
479                 }
480         }
481
482         for (i = 0; i < RTE_MAX_ETHPORTS; i++) {
483                 if (enabled_port_mask & (1 << i))
484                         ports[nb_ports++] = (uint8_t)i;
485         }
486
487         if ((nb_ports ==  0) || (nb_ports > MAX_SUP_PORTS)) {
488                 RTE_LOG(INFO, VHOST_PORT, "Current enabled port number is %u,"
489                         "but only %u port can be enabled\n", nb_ports,
490                         MAX_SUP_PORTS);
491                 return -1;
492         }
493
494         return 0;
495 }
496
497 /**
498  * Update the global var NB_PORTS and array PORTS
499  * according to system ports number and return valid ports number
500  */
501 static unsigned
502 check_ports_num(unsigned max_nb_ports)
503 {
504         unsigned valid_nb_ports = nb_ports;
505         unsigned portid;
506
507         if (nb_ports > max_nb_ports) {
508                 RTE_LOG(INFO, VHOST_PORT, "\nSpecified port number(%u) "
509                         " exceeds total system port number(%u)\n",
510                         nb_ports, max_nb_ports);
511                 nb_ports = max_nb_ports;
512         }
513
514         for (portid = 0; portid < nb_ports; portid++) {
515                 if (!rte_eth_dev_is_valid_port(ports[portid])) {
516                         RTE_LOG(INFO, VHOST_PORT,
517                                 "\nSpecified port ID(%u) is not valid\n",
518                                 ports[portid]);
519                         ports[portid] = INVALID_PORT_ID;
520                         valid_nb_ports--;
521                 }
522         }
523         return valid_nb_ports;
524 }
525
526 /**
527  * This function routes the TX packet to the correct interface. This may be a local device
528  * or the physical port.
529  */
530 static __rte_always_inline void
531 virtio_tx_route(struct vhost_dev *vdev, struct rte_mbuf *m)
532 {
533         struct mbuf_table *tx_q;
534         struct rte_mbuf **m_table;
535         unsigned len, ret = 0;
536         const uint16_t lcore_id = rte_lcore_id();
537
538         RTE_LOG_DP(DEBUG, VHOST_DATA, "(%d) TX: MAC address is external\n",
539                 vdev->vid);
540
541         /* Add packet to the port tx queue */
542         tx_q = &lcore_tx_queue[lcore_id];
543         len = tx_q->len;
544
545         tx_q->m_table[len] = m;
546         len++;
547         if (enable_stats) {
548                 dev_statistics[vdev->vid].tx_total++;
549                 dev_statistics[vdev->vid].tx++;
550         }
551
552         if (unlikely(len == MAX_PKT_BURST)) {
553                 m_table = (struct rte_mbuf **)tx_q->m_table;
554                 ret = overlay_options.tx_handle(ports[0],
555                         (uint16_t)tx_q->txq_id, m_table,
556                         (uint16_t)tx_q->len);
557
558                 /* Free any buffers not handled by TX and update
559                  * the port stats.
560                  */
561                 if (unlikely(ret < len)) {
562                         do {
563                                 rte_pktmbuf_free(m_table[ret]);
564                         } while (++ret < len);
565                 }
566
567                 len = 0;
568         }
569
570         tx_q->len = len;
571         return;
572 }
573
574 /**
575  * This function is called by each data core. It handles all
576  * RX/TX registered with the core. For TX the specific lcore
577  * linked list is used. For RX, MAC addresses are compared
578  * with all devices in the main linked list.
579  */
580 static int
581 switch_worker(__rte_unused void *arg)
582 {
583         struct rte_mempool *mbuf_pool = arg;
584         struct vhost_dev *vdev = NULL;
585         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
586         struct virtio_net_data_ll *dev_ll;
587         struct mbuf_table *tx_q;
588         volatile struct lcore_ll_info *lcore_ll;
589         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
590                                         / US_PER_S * BURST_TX_DRAIN_US;
591         uint64_t prev_tsc, diff_tsc, cur_tsc, ret_count = 0;
592         unsigned i, ret = 0;
593         const uint16_t lcore_id = rte_lcore_id();
594         const uint16_t num_cores = (uint16_t)rte_lcore_count();
595         uint16_t rx_count = 0;
596         uint16_t tx_count;
597         uint32_t retry = 0;
598
599         RTE_LOG(INFO, VHOST_DATA, "Procesing on Core %u started\n", lcore_id);
600         lcore_ll = lcore_info[lcore_id].lcore_ll;
601         prev_tsc = 0;
602
603         tx_q = &lcore_tx_queue[lcore_id];
604         for (i = 0; i < num_cores; i++) {
605                 if (lcore_ids[i] == lcore_id) {
606                         tx_q->txq_id = i;
607                         break;
608                 }
609         }
610
611         while (1) {
612                 cur_tsc = rte_rdtsc();
613                 /*
614                  * TX burst queue drain
615                  */
616                 diff_tsc = cur_tsc - prev_tsc;
617                 if (unlikely(diff_tsc > drain_tsc)) {
618
619                         if (tx_q->len) {
620                                 RTE_LOG_DP(DEBUG, VHOST_DATA, "TX queue drained after "
621                                         "timeout with burst size %u\n",
622                                         tx_q->len);
623                                 ret = overlay_options.tx_handle(ports[0],
624                                         (uint16_t)tx_q->txq_id,
625                                         (struct rte_mbuf **)tx_q->m_table,
626                                         (uint16_t)tx_q->len);
627                                 if (unlikely(ret < tx_q->len)) {
628                                         do {
629                                                 rte_pktmbuf_free(tx_q->m_table[ret]);
630                                         } while (++ret < tx_q->len);
631                                 }
632
633                                 tx_q->len = 0;
634                         }
635
636                         prev_tsc = cur_tsc;
637
638                 }
639
640                 rte_prefetch0(lcore_ll->ll_root_used);
641
642                 /**
643                  * Inform the configuration core that we have exited
644                  * the linked list and that no devices are
645                  * in use if requested.
646                  */
647                 if (lcore_ll->dev_removal_flag == REQUEST_DEV_REMOVAL)
648                         lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL;
649
650                 /*
651                  * Process devices
652                  */
653                 dev_ll = lcore_ll->ll_root_used;
654
655                 while (dev_ll != NULL) {
656                         vdev = dev_ll->vdev;
657
658                         if (unlikely(vdev->remove)) {
659                                 dev_ll = dev_ll->next;
660                                 overlay_options.tunnel_destroy(vdev);
661                                 vdev->ready = DEVICE_SAFE_REMOVE;
662                                 continue;
663                         }
664                         if (likely(vdev->ready == DEVICE_RX)) {
665                                 /* Handle guest RX */
666                                 rx_count = rte_eth_rx_burst(ports[0],
667                                         vdev->rx_q, pkts_burst, MAX_PKT_BURST);
668
669                                 if (rx_count) {
670                                         /*
671                                         * Retry is enabled and the queue is
672                                         * full then we wait and retry to
673                                         * avoid packet loss. Here MAX_PKT_BURST
674                                         * must be less than virtio queue size
675                                         */
676                                         if (enable_retry && unlikely(rx_count >
677                                                 rte_vhost_avail_entries(vdev->vid, VIRTIO_RXQ))) {
678                                                 for (retry = 0; retry < burst_rx_retry_num;
679                                                         retry++) {
680                                                         rte_delay_us(burst_rx_delay_time);
681                                                         if (rx_count <= rte_vhost_avail_entries(vdev->vid, VIRTIO_RXQ))
682                                                                 break;
683                                                 }
684                                         }
685
686                                         ret_count = overlay_options.rx_handle(vdev->vid, pkts_burst, rx_count);
687                                         if (enable_stats) {
688                                                 rte_atomic64_add(
689                                                 &dev_statistics[vdev->vid].rx_total_atomic,
690                                                 rx_count);
691                                                 rte_atomic64_add(
692                                                 &dev_statistics[vdev->vid].rx_atomic, ret_count);
693                                         }
694                                         while (likely(rx_count)) {
695                                                 rx_count--;
696                                                 rte_pktmbuf_free(pkts_burst[rx_count]);
697                                         }
698
699                                 }
700                         }
701
702                         if (likely(!vdev->remove)) {
703                                 /* Handle guest TX*/
704                                 tx_count = rte_vhost_dequeue_burst(vdev->vid,
705                                                 VIRTIO_TXQ, mbuf_pool,
706                                                 pkts_burst, MAX_PKT_BURST);
707                                 /* If this is the first received packet we need to learn the MAC */
708                                 if (unlikely(vdev->ready == DEVICE_MAC_LEARNING) && tx_count) {
709                                         if (vdev->remove ||
710                                                 (overlay_options.tunnel_setup(vdev, pkts_burst[0]) == -1)) {
711                                                 while (tx_count)
712                                                         rte_pktmbuf_free(pkts_burst[--tx_count]);
713                                         }
714                                 }
715                                 while (tx_count)
716                                         virtio_tx_route(vdev, pkts_burst[--tx_count]);
717                         }
718
719                         /* move to the next device in the list */
720                         dev_ll = dev_ll->next;
721                 }
722         }
723
724         return 0;
725 }
726
727 /**
728  * Add an entry to a used linked list. A free entry must first be found
729  * in the free linked list using get_data_ll_free_entry();
730  */
731 static void
732 add_data_ll_entry(struct virtio_net_data_ll **ll_root_addr,
733         struct virtio_net_data_ll *ll_dev)
734 {
735         struct virtio_net_data_ll *ll = *ll_root_addr;
736
737         /* Set next as NULL and use a compiler barrier to avoid reordering. */
738         ll_dev->next = NULL;
739         rte_compiler_barrier();
740
741         /* If ll == NULL then this is the first device. */
742         if (ll) {
743                 /* Increment to the tail of the linked list. */
744                 while (ll->next != NULL)
745                         ll = ll->next;
746
747                 ll->next = ll_dev;
748         } else {
749                 *ll_root_addr = ll_dev;
750         }
751 }
752
753 /**
754  * Remove an entry from a used linked list. The entry must then be added to
755  * the free linked list using put_data_ll_free_entry().
756  */
757 static void
758 rm_data_ll_entry(struct virtio_net_data_ll **ll_root_addr,
759         struct virtio_net_data_ll *ll_dev,
760         struct virtio_net_data_ll *ll_dev_last)
761 {
762         struct virtio_net_data_ll *ll = *ll_root_addr;
763
764         if (unlikely((ll == NULL) || (ll_dev == NULL)))
765                 return;
766
767         if (ll_dev == ll)
768                 *ll_root_addr = ll_dev->next;
769         else
770                 if (likely(ll_dev_last != NULL))
771                         ll_dev_last->next = ll_dev->next;
772                 else
773                         RTE_LOG(ERR, VHOST_CONFIG,
774                                 "Remove entry form ll failed.\n");
775 }
776
777 /**
778  * Find and return an entry from the free linked list.
779  */
780 static struct virtio_net_data_ll *
781 get_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr)
782 {
783         struct virtio_net_data_ll *ll_free = *ll_root_addr;
784         struct virtio_net_data_ll *ll_dev;
785
786         if (ll_free == NULL)
787                 return NULL;
788
789         ll_dev = ll_free;
790         *ll_root_addr = ll_free->next;
791
792         return ll_dev;
793 }
794
795 /**
796  * Place an entry back on to the free linked list.
797  */
798 static void
799 put_data_ll_free_entry(struct virtio_net_data_ll **ll_root_addr,
800         struct virtio_net_data_ll *ll_dev)
801 {
802         struct virtio_net_data_ll *ll_free = *ll_root_addr;
803
804         if (ll_dev == NULL)
805                 return;
806
807         ll_dev->next = ll_free;
808         *ll_root_addr = ll_dev;
809 }
810
811 /**
812  * Creates a linked list of a given size.
813  */
814 static struct virtio_net_data_ll *
815 alloc_data_ll(uint32_t size)
816 {
817         struct virtio_net_data_ll *ll_new;
818         uint32_t i;
819
820         /* Malloc and then chain the linked list. */
821         ll_new = malloc(size * sizeof(struct virtio_net_data_ll));
822         if (ll_new == NULL) {
823                 RTE_LOG(ERR, VHOST_CONFIG,
824                         "Failed to allocate memory for ll_new.\n");
825                 return NULL;
826         }
827
828         for (i = 0; i < size - 1; i++) {
829                 ll_new[i].vdev = NULL;
830                 ll_new[i].next = &ll_new[i+1];
831         }
832         ll_new[i].next = NULL;
833
834         return ll_new;
835 }
836
837 /**
838  * Create the main linked list along with each individual cores
839  * linked list. A used and a free list are created to manage entries.
840  */
841 static int
842 init_data_ll(void)
843 {
844         int lcore;
845
846         RTE_LCORE_FOREACH_SLAVE(lcore) {
847                 lcore_info[lcore].lcore_ll =
848                         malloc(sizeof(struct lcore_ll_info));
849                 if (lcore_info[lcore].lcore_ll == NULL) {
850                         RTE_LOG(ERR, VHOST_CONFIG,
851                                 "Failed to allocate memory for lcore_ll.\n");
852                         return -1;
853                 }
854
855                 lcore_info[lcore].lcore_ll->device_num = 0;
856                 lcore_info[lcore].lcore_ll->dev_removal_flag = ACK_DEV_REMOVAL;
857                 lcore_info[lcore].lcore_ll->ll_root_used = NULL;
858                 if (nb_devices % nb_switching_cores)
859                         lcore_info[lcore].lcore_ll->ll_root_free =
860                                 alloc_data_ll((nb_devices / nb_switching_cores)
861                                                 + 1);
862                 else
863                         lcore_info[lcore].lcore_ll->ll_root_free =
864                                 alloc_data_ll(nb_devices / nb_switching_cores);
865         }
866
867         /* Allocate devices up to a maximum of MAX_DEVICES. */
868         ll_root_free = alloc_data_ll(MIN((nb_devices), MAX_DEVICES));
869
870         return 0;
871 }
872
873 /**
874  * Remove a device from the specific data core linked list and
875  * from the main linked list. Synchonization occurs through the use
876  * of the lcore dev_removal_flag.
877  */
878 static void
879 destroy_device(int vid)
880 {
881         struct virtio_net_data_ll *ll_lcore_dev_cur;
882         struct virtio_net_data_ll *ll_main_dev_cur;
883         struct virtio_net_data_ll *ll_lcore_dev_last = NULL;
884         struct virtio_net_data_ll *ll_main_dev_last = NULL;
885         struct vhost_dev *vdev = NULL;
886         int lcore;
887
888         ll_main_dev_cur = ll_root_used;
889         while (ll_main_dev_cur != NULL) {
890                 if (ll_main_dev_cur->vdev->vid == vid) {
891                         vdev = ll_main_dev_cur->vdev;
892                         break;
893                 }
894         }
895         if (!vdev)
896                 return;
897
898         /* set the remove flag. */
899         vdev->remove = 1;
900         while (vdev->ready != DEVICE_SAFE_REMOVE)
901                 rte_pause();
902
903         /* Search for entry to be removed from lcore ll */
904         ll_lcore_dev_cur = lcore_info[vdev->coreid].lcore_ll->ll_root_used;
905         while (ll_lcore_dev_cur != NULL) {
906                 if (ll_lcore_dev_cur->vdev == vdev) {
907                         break;
908                 } else {
909                         ll_lcore_dev_last = ll_lcore_dev_cur;
910                         ll_lcore_dev_cur = ll_lcore_dev_cur->next;
911                 }
912         }
913
914         if (ll_lcore_dev_cur == NULL) {
915                 RTE_LOG(ERR, VHOST_CONFIG,
916                         "(%d) Failed to find the dev to be destroy.\n", vid);
917                 return;
918         }
919
920         /* Search for entry to be removed from main ll */
921         ll_main_dev_cur = ll_root_used;
922         ll_main_dev_last = NULL;
923         while (ll_main_dev_cur != NULL) {
924                 if (ll_main_dev_cur->vdev == vdev) {
925                         break;
926                 } else {
927                         ll_main_dev_last = ll_main_dev_cur;
928                         ll_main_dev_cur = ll_main_dev_cur->next;
929                 }
930         }
931
932         /* Remove entries from the lcore and main ll. */
933         rm_data_ll_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_used,
934                         ll_lcore_dev_cur, ll_lcore_dev_last);
935         rm_data_ll_entry(&ll_root_used, ll_main_dev_cur, ll_main_dev_last);
936
937         /* Set the dev_removal_flag on each lcore. */
938         RTE_LCORE_FOREACH_SLAVE(lcore) {
939                 lcore_info[lcore].lcore_ll->dev_removal_flag =
940                         REQUEST_DEV_REMOVAL;
941         }
942
943         /*
944          * Once each core has set the dev_removal_flag to
945          * ACK_DEV_REMOVAL we can be sure that they can no longer access
946          * the device removed from the linked lists and that the devices
947          * are no longer in use.
948          */
949         RTE_LCORE_FOREACH_SLAVE(lcore) {
950                 while (lcore_info[lcore].lcore_ll->dev_removal_flag
951                         != ACK_DEV_REMOVAL)
952                         rte_pause();
953         }
954
955         /* Add the entries back to the lcore and main free ll.*/
956         put_data_ll_free_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_free,
957                                 ll_lcore_dev_cur);
958         put_data_ll_free_entry(&ll_root_free, ll_main_dev_cur);
959
960         /* Decrement number of device on the lcore. */
961         lcore_info[vdev->coreid].lcore_ll->device_num--;
962
963         RTE_LOG(INFO, VHOST_DATA, "(%d) Device has been removed "
964                 "from data core\n", vid);
965
966         rte_free(vdev);
967
968 }
969
970 /**
971  * A new device is added to a data core. First the device is added
972  * to the main linked list and the allocated to a specific data core.
973  */
974 static int
975 new_device(int vid)
976 {
977         struct virtio_net_data_ll *ll_dev;
978         int lcore, core_add = 0;
979         uint32_t device_num_min = nb_devices;
980         struct vhost_dev *vdev;
981
982         vdev = rte_zmalloc("vhost device", sizeof(*vdev), RTE_CACHE_LINE_SIZE);
983         if (vdev == NULL) {
984                 RTE_LOG(INFO, VHOST_DATA,
985                         "(%d) Couldn't allocate memory for vhost dev\n", vid);
986                 return -1;
987         }
988         vdev->vid = vid;
989         /* Add device to main ll */
990         ll_dev = get_data_ll_free_entry(&ll_root_free);
991         if (ll_dev == NULL) {
992                 RTE_LOG(INFO, VHOST_DATA, "(%d) No free entry found in"
993                         " linked list Device limit of %d devices per core"
994                         " has been reached\n", vid, nb_devices);
995                 if (vdev->regions_hpa)
996                         rte_free(vdev->regions_hpa);
997                 rte_free(vdev);
998                 return -1;
999         }
1000         ll_dev->vdev = vdev;
1001         add_data_ll_entry(&ll_root_used, ll_dev);
1002         vdev->rx_q = vid;
1003
1004         /* reset ready flag */
1005         vdev->ready = DEVICE_MAC_LEARNING;
1006         vdev->remove = 0;
1007
1008         /* Find a suitable lcore to add the device. */
1009         RTE_LCORE_FOREACH_SLAVE(lcore) {
1010                 if (lcore_info[lcore].lcore_ll->device_num < device_num_min) {
1011                         device_num_min = lcore_info[lcore].lcore_ll->device_num;
1012                         core_add = lcore;
1013                 }
1014         }
1015         /* Add device to lcore ll */
1016         ll_dev = get_data_ll_free_entry(&lcore_info[core_add].lcore_ll->ll_root_free);
1017         if (ll_dev == NULL) {
1018                 RTE_LOG(INFO, VHOST_DATA,
1019                         "(%d) Failed to add device to data core\n",
1020                         vid);
1021                 vdev->ready = DEVICE_SAFE_REMOVE;
1022                 destroy_device(vid);
1023                 rte_free(vdev->regions_hpa);
1024                 rte_free(vdev);
1025                 return -1;
1026         }
1027         ll_dev->vdev = vdev;
1028         vdev->coreid = core_add;
1029
1030         add_data_ll_entry(&lcore_info[vdev->coreid].lcore_ll->ll_root_used,
1031                         ll_dev);
1032
1033         /* Initialize device stats */
1034         memset(&dev_statistics[vid], 0,
1035                 sizeof(struct device_statistics));
1036
1037         /* Disable notifications. */
1038         rte_vhost_enable_guest_notification(vid, VIRTIO_RXQ, 0);
1039         rte_vhost_enable_guest_notification(vid, VIRTIO_TXQ, 0);
1040         lcore_info[vdev->coreid].lcore_ll->device_num++;
1041
1042         RTE_LOG(INFO, VHOST_DATA, "(%d) Device has been added to data core %d\n",
1043                 vid, vdev->coreid);
1044
1045         return 0;
1046 }
1047
1048 /**
1049  * These callback allow devices to be added to the data core when configuration
1050  * has been fully complete.
1051  */
1052 static const struct vhost_device_ops virtio_net_device_ops = {
1053         .new_device =  new_device,
1054         .destroy_device = destroy_device,
1055 };
1056
1057 /**
1058  * This is a thread will wake up after a period to print stats if the user has
1059  * enabled them.
1060  */
1061 static void *
1062 print_stats(__rte_unused void *arg)
1063 {
1064         struct virtio_net_data_ll *dev_ll;
1065         uint64_t tx_dropped, rx_dropped;
1066         uint64_t tx, tx_total, rx, rx_total, rx_ip_csum, rx_l4_csum;
1067         int vid;
1068         const char clr[] = { 27, '[', '2', 'J', '\0' };
1069         const char top_left[] = { 27, '[', '1', ';', '1', 'H', '\0' };
1070
1071         while (1) {
1072                 sleep(enable_stats);
1073
1074                 /* Clear screen and move to top left */
1075                 printf("%s%s", clr, top_left);
1076
1077                 printf("\nDevice statistics ================================");
1078
1079                 dev_ll = ll_root_used;
1080                 while (dev_ll != NULL) {
1081                         vid = dev_ll->vdev->vid;
1082                         tx_total = dev_statistics[vid].tx_total;
1083                         tx = dev_statistics[vid].tx;
1084                         tx_dropped = tx_total - tx;
1085
1086                         rx_total = rte_atomic64_read(
1087                                 &dev_statistics[vid].rx_total_atomic);
1088                         rx = rte_atomic64_read(
1089                                 &dev_statistics[vid].rx_atomic);
1090                         rx_dropped = rx_total - rx;
1091                         rx_ip_csum = rte_atomic64_read(
1092                                 &dev_statistics[vid].rx_bad_ip_csum);
1093                         rx_l4_csum = rte_atomic64_read(
1094                                 &dev_statistics[vid].rx_bad_l4_csum);
1095
1096                         printf("\nStatistics for device %d ----------"
1097                                         "\nTX total:            %"PRIu64""
1098                                         "\nTX dropped:          %"PRIu64""
1099                                         "\nTX successful:               %"PRIu64""
1100                                         "\nRX total:            %"PRIu64""
1101                                         "\nRX bad IP csum:      %"PRIu64""
1102                                         "\nRX bad L4 csum:      %"PRIu64""
1103                                         "\nRX dropped:          %"PRIu64""
1104                                         "\nRX successful:               %"PRIu64"",
1105                                         vid,
1106                                         tx_total,
1107                                         tx_dropped,
1108                                         tx,
1109                                         rx_total,
1110                                         rx_ip_csum,
1111                                         rx_l4_csum,
1112                                         rx_dropped,
1113                                         rx);
1114
1115                         dev_ll = dev_ll->next;
1116                 }
1117                 printf("\n================================================\n");
1118         }
1119
1120         return NULL;
1121 }
1122
1123 /**
1124  * Main function, does initialisation and calls the per-lcore functions.
1125  */
1126 int
1127 main(int argc, char *argv[])
1128 {
1129         struct rte_mempool *mbuf_pool = NULL;
1130         unsigned lcore_id, core_id = 0;
1131         unsigned nb_ports, valid_nb_ports;
1132         int ret;
1133         uint16_t portid;
1134         uint16_t queue_id;
1135         static pthread_t tid;
1136
1137         /* init EAL */
1138         ret = rte_eal_init(argc, argv);
1139         if (ret < 0)
1140                 rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
1141         argc -= ret;
1142         argv += ret;
1143
1144         /* parse app arguments */
1145         ret = tep_termination_parse_args(argc, argv);
1146         if (ret < 0)
1147                 rte_exit(EXIT_FAILURE, "Invalid argument\n");
1148
1149         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++)
1150                 if (rte_lcore_is_enabled(lcore_id))
1151                         lcore_ids[core_id++] = lcore_id;
1152
1153         /* set the number of swithcing cores available */
1154         nb_switching_cores = rte_lcore_count()-1;
1155
1156         /* Get the number of physical ports. */
1157         nb_ports = rte_eth_dev_count_avail();
1158
1159         /*
1160          * Update the global var NB_PORTS and global array PORTS
1161          * and get value of var VALID_NB_PORTS according to system ports number
1162          */
1163         valid_nb_ports = check_ports_num(nb_ports);
1164
1165         if ((valid_nb_ports == 0) || (valid_nb_ports > MAX_SUP_PORTS)) {
1166                 rte_exit(EXIT_FAILURE, "Current enabled port number is %u,"
1167                         "but only %u port can be enabled\n", nb_ports,
1168                         MAX_SUP_PORTS);
1169         }
1170         /* Create the mbuf pool. */
1171         mbuf_pool = rte_pktmbuf_pool_create(
1172                         "MBUF_POOL",
1173                         NUM_MBUFS_PER_PORT * valid_nb_ports,
1174                         MBUF_CACHE_SIZE,
1175                         0,
1176                         MBUF_DATA_SIZE,
1177                         rte_socket_id());
1178         if (mbuf_pool == NULL)
1179                 rte_exit(EXIT_FAILURE, "Cannot create mbuf pool\n");
1180
1181         for (queue_id = 0; queue_id < MAX_QUEUES + 1; queue_id++)
1182                 vpool_array[queue_id].pool = mbuf_pool;
1183
1184         /* initialize all ports */
1185         RTE_ETH_FOREACH_DEV(portid) {
1186                 /* skip ports that are not enabled */
1187                 if ((enabled_port_mask & (1 << portid)) == 0) {
1188                         RTE_LOG(INFO, VHOST_PORT,
1189                                 "Skipping disabled port %d\n", portid);
1190                         continue;
1191                 }
1192                 if (overlay_options.port_configure(portid, mbuf_pool) != 0)
1193                         rte_exit(EXIT_FAILURE,
1194                                 "Cannot initialize network ports\n");
1195         }
1196
1197         /* Initialise all linked lists. */
1198         if (init_data_ll() == -1)
1199                 rte_exit(EXIT_FAILURE, "Failed to initialize linked list\n");
1200
1201         /* Initialize device stats */
1202         memset(&dev_statistics, 0, sizeof(dev_statistics));
1203
1204         /* Enable stats if the user option is set. */
1205         if (enable_stats) {
1206                 ret = rte_ctrl_thread_create(&tid, "print-stats", NULL,
1207                                         print_stats, NULL);
1208                 if (ret < 0)
1209                         rte_exit(EXIT_FAILURE, "Cannot create print-stats thread\n");
1210         }
1211
1212         /* Launch all data cores. */
1213         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1214                 rte_eal_remote_launch(switch_worker,
1215                         mbuf_pool, lcore_id);
1216         }
1217
1218         ret = rte_vhost_driver_register((char *)&dev_basename, 0);
1219         if (ret != 0)
1220                 rte_exit(EXIT_FAILURE, "failed to register vhost driver.\n");
1221
1222         rte_vhost_driver_disable_features(dev_basename,
1223                 1ULL << VIRTIO_NET_F_MRG_RXBUF);
1224
1225         ret = rte_vhost_driver_callback_register(dev_basename,
1226                 &virtio_net_device_ops);
1227         if (ret != 0) {
1228                 rte_exit(EXIT_FAILURE,
1229                         "failed to register vhost driver callbacks.\n");
1230         }
1231
1232         if (rte_vhost_driver_start(dev_basename) < 0) {
1233                 rte_exit(EXIT_FAILURE,
1234                         "failed to start vhost driver.\n");
1235         }
1236
1237         RTE_LCORE_FOREACH_SLAVE(lcore_id)
1238                 rte_eal_wait_lcore(lcore_id);
1239
1240         return 0;
1241 }