]>
Commit | Line | Data |
---|---|---|
8a9562d2 PS |
1 | /* |
2 | * Copyright (c) 2014 Nicira, Inc. | |
3 | * | |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
19 | #include <stdio.h> | |
20 | #include <string.h> | |
21 | #include <signal.h> | |
22 | #include <stdlib.h> | |
23 | #include <pthread.h> | |
24 | #include <config.h> | |
25 | #include <errno.h> | |
26 | #include <sched.h> | |
27 | #include <stdlib.h> | |
28 | #include <unistd.h> | |
29 | #include <stdio.h> | |
30 | ||
8a9562d2 PS |
31 | #include "dpif-netdev.h" |
32 | #include "list.h" | |
33 | #include "netdev-dpdk.h" | |
34 | #include "netdev-provider.h" | |
35 | #include "netdev-vport.h" | |
36 | #include "odp-util.h" | |
37 | #include "ofp-print.h" | |
38 | #include "ofpbuf.h" | |
39 | #include "ovs-thread.h" | |
40 | #include "ovs-rcu.h" | |
41 | #include "packets.h" | |
42 | #include "shash.h" | |
8a9562d2 PS |
43 | #include "sset.h" |
44 | #include "unaligned.h" | |
45 | #include "timeval.h" | |
46 | #include "unixctl.h" | |
47 | #include "vlog.h" | |
48 | ||
49 | VLOG_DEFINE_THIS_MODULE(dpdk); | |
50 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
51 | ||
52 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
53 | ||
54 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
55 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
56 | ||
57 | /* | |
58 | * need to reserve tons of extra space in the mbufs so we can align the | |
59 | * DMA addresses to 4KB. | |
60 | */ | |
61 | ||
62 | #define MTU_TO_MAX_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) | |
63 | #define MBUF_SIZE(mtu) (MTU_TO_MAX_LEN(mtu) + (512) + \ | |
64 | sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM) | |
65 | ||
66 | /* TODO: mempool size should be based on system resources. */ | |
67 | #define NB_MBUF (4096 * 64) | |
68 | #define MP_CACHE_SZ (256 * 2) | |
69 | #define SOCKET0 0 | |
70 | ||
71 | #define NON_PMD_THREAD_TX_QUEUE 0 | |
72 | ||
73 | /* TODO: Needs per NIC value for these constants. */ | |
74 | #define RX_PTHRESH 32 /* Default values of RX prefetch threshold reg. */ | |
75 | #define RX_HTHRESH 32 /* Default values of RX host threshold reg. */ | |
76 | #define RX_WTHRESH 16 /* Default values of RX write-back threshold reg. */ | |
77 | ||
78 | #define TX_PTHRESH 36 /* Default values of TX prefetch threshold reg. */ | |
79 | #define TX_HTHRESH 0 /* Default values of TX host threshold reg. */ | |
80 | #define TX_WTHRESH 0 /* Default values of TX write-back threshold reg. */ | |
81 | ||
82 | static const struct rte_eth_conf port_conf = { | |
83 | .rxmode = { | |
84 | .mq_mode = ETH_MQ_RX_RSS, | |
85 | .split_hdr_size = 0, | |
86 | .header_split = 0, /* Header Split disabled */ | |
87 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
88 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
89 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
90 | .hw_strip_crc = 0, | |
91 | }, | |
92 | .rx_adv_conf = { | |
93 | .rss_conf = { | |
94 | .rss_key = NULL, | |
95 | .rss_hf = ETH_RSS_IPV4_TCP | ETH_RSS_IPV4 | ETH_RSS_IPV6, | |
96 | }, | |
97 | }, | |
98 | .txmode = { | |
99 | .mq_mode = ETH_MQ_TX_NONE, | |
100 | }, | |
101 | }; | |
102 | ||
103 | static const struct rte_eth_rxconf rx_conf = { | |
104 | .rx_thresh = { | |
105 | .pthresh = RX_PTHRESH, | |
106 | .hthresh = RX_HTHRESH, | |
107 | .wthresh = RX_WTHRESH, | |
108 | }, | |
109 | }; | |
110 | ||
111 | static const struct rte_eth_txconf tx_conf = { | |
112 | .tx_thresh = { | |
113 | .pthresh = TX_PTHRESH, | |
114 | .hthresh = TX_HTHRESH, | |
115 | .wthresh = TX_WTHRESH, | |
116 | }, | |
117 | .tx_free_thresh = 0, | |
118 | .tx_rs_thresh = 0, | |
119 | }; | |
120 | ||
121 | enum { MAX_RX_QUEUE_LEN = 64 }; | |
122 | enum { MAX_TX_QUEUE_LEN = 64 }; | |
123 | enum { DRAIN_TSC = 200000ULL }; | |
124 | ||
125 | static int rte_eal_init_ret = ENODEV; | |
126 | ||
127 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; | |
128 | ||
129 | /* Contains all 'struct dpdk_dev's. */ | |
130 | static struct list dpdk_list OVS_GUARDED_BY(dpdk_mutex) | |
131 | = LIST_INITIALIZER(&dpdk_list); | |
132 | ||
133 | static struct list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex) | |
134 | = LIST_INITIALIZER(&dpdk_mp_list); | |
135 | ||
8a9562d2 PS |
136 | struct dpdk_mp { |
137 | struct rte_mempool *mp; | |
138 | int mtu; | |
139 | int socket_id; | |
140 | int refcount; | |
141 | struct list list_node OVS_GUARDED_BY(dpdk_mutex); | |
142 | }; | |
143 | ||
144 | struct dpdk_tx_queue { | |
145 | rte_spinlock_t tx_lock; | |
146 | int count; | |
147 | uint64_t tsc; | |
148 | struct rte_mbuf *burst_pkts[MAX_TX_QUEUE_LEN]; | |
149 | }; | |
150 | ||
151 | struct netdev_dpdk { | |
152 | struct netdev up; | |
153 | int port_id; | |
154 | int max_packet_len; | |
155 | ||
156 | struct dpdk_tx_queue tx_q[NR_QUEUE]; | |
157 | ||
158 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
159 | ||
160 | struct dpdk_mp *dpdk_mp; | |
161 | int mtu; | |
162 | int socket_id; | |
163 | int buf_size; | |
164 | struct netdev_stats stats_offset; | |
165 | struct netdev_stats stats; | |
166 | ||
167 | uint8_t hwaddr[ETH_ADDR_LEN]; | |
168 | enum netdev_flags flags; | |
169 | ||
170 | struct rte_eth_link link; | |
171 | int link_reset_cnt; | |
172 | ||
173 | /* In dpdk_list. */ | |
174 | struct list list_node OVS_GUARDED_BY(dpdk_mutex); | |
175 | }; | |
176 | ||
177 | struct netdev_rxq_dpdk { | |
178 | struct netdev_rxq up; | |
179 | int port_id; | |
180 | }; | |
181 | ||
182 | static int netdev_dpdk_construct(struct netdev *); | |
183 | ||
184 | static bool | |
185 | is_dpdk_class(const struct netdev_class *class) | |
186 | { | |
187 | return class->construct == netdev_dpdk_construct; | |
188 | } | |
189 | ||
190 | /* TODO: use dpdk malloc for entire OVS. infact huge page shld be used | |
191 | * for all other sengments data, bss and text. */ | |
192 | ||
193 | static void * | |
194 | dpdk_rte_mzalloc(size_t sz) | |
195 | { | |
196 | void *ptr; | |
197 | ||
198 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
199 | if (ptr == NULL) { | |
200 | out_of_memory(); | |
201 | } | |
202 | return ptr; | |
203 | } | |
204 | ||
205 | void | |
206 | free_dpdk_buf(struct ofpbuf *b) | |
207 | { | |
b3cd9f9d | 208 | struct rte_mbuf *pkt = (struct rte_mbuf *) b; |
8a9562d2 PS |
209 | |
210 | rte_mempool_put(pkt->pool, pkt); | |
211 | } | |
212 | ||
b3cd9f9d PS |
213 | static void |
214 | __rte_pktmbuf_init(struct rte_mempool *mp, | |
215 | void *opaque_arg OVS_UNUSED, | |
216 | void *_m, | |
217 | unsigned i OVS_UNUSED) | |
218 | { | |
219 | struct rte_mbuf *m = _m; | |
220 | uint32_t buf_len = mp->elt_size - sizeof(struct ofpbuf); | |
221 | ||
222 | RTE_MBUF_ASSERT(mp->elt_size >= sizeof(struct ofpbuf)); | |
223 | ||
224 | memset(m, 0, mp->elt_size); | |
225 | ||
226 | /* start of buffer is just after mbuf structure */ | |
227 | m->buf_addr = (char *)m + sizeof(struct ofpbuf); | |
228 | m->buf_physaddr = rte_mempool_virt2phy(mp, m) + | |
229 | sizeof(struct ofpbuf); | |
230 | m->buf_len = (uint16_t)buf_len; | |
231 | ||
232 | /* keep some headroom between start of buffer and data */ | |
233 | m->pkt.data = (char*) m->buf_addr + RTE_MIN(RTE_PKTMBUF_HEADROOM, m->buf_len); | |
234 | ||
235 | /* init some constant fields */ | |
236 | m->type = RTE_MBUF_PKT; | |
237 | m->pool = mp; | |
238 | m->pkt.nb_segs = 1; | |
239 | m->pkt.in_port = 0xff; | |
240 | } | |
241 | ||
242 | static void | |
243 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
244 | void *opaque_arg OVS_UNUSED, | |
245 | void *_m, | |
246 | unsigned i OVS_UNUSED) | |
247 | { | |
248 | struct rte_mbuf *m = _m; | |
249 | ||
250 | __rte_pktmbuf_init(mp, opaque_arg, _m, i); | |
251 | ||
252 | ofpbuf_init_dpdk((struct ofpbuf *) m, m->buf_len); | |
253 | } | |
254 | ||
8a9562d2 PS |
255 | static struct dpdk_mp * |
256 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
257 | { | |
258 | struct dpdk_mp *dmp = NULL; | |
259 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
260 | ||
261 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
262 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
263 | dmp->refcount++; | |
264 | return dmp; | |
265 | } | |
266 | } | |
267 | ||
268 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
269 | dmp->socket_id = socket_id; | |
270 | dmp->mtu = mtu; | |
271 | dmp->refcount = 1; | |
272 | ||
273 | snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d", dmp->mtu); | |
274 | dmp->mp = rte_mempool_create(mp_name, NB_MBUF, MBUF_SIZE(mtu), | |
275 | MP_CACHE_SZ, | |
276 | sizeof(struct rte_pktmbuf_pool_private), | |
277 | rte_pktmbuf_pool_init, NULL, | |
b3cd9f9d | 278 | ovs_rte_pktmbuf_init, NULL, |
8a9562d2 PS |
279 | socket_id, 0); |
280 | ||
281 | if (dmp->mp == NULL) { | |
282 | return NULL; | |
283 | } | |
284 | ||
285 | list_push_back(&dpdk_mp_list, &dmp->list_node); | |
286 | return dmp; | |
287 | } | |
288 | ||
289 | static void | |
290 | dpdk_mp_put(struct dpdk_mp *dmp) | |
291 | { | |
292 | ||
293 | if (!dmp) { | |
294 | return; | |
295 | } | |
296 | ||
297 | dmp->refcount--; | |
298 | ovs_assert(dmp->refcount >= 0); | |
299 | ||
300 | #if 0 | |
301 | /* I could not find any API to destroy mp. */ | |
302 | if (dmp->refcount == 0) { | |
303 | list_delete(dmp->list_node); | |
304 | /* destroy mp-pool. */ | |
305 | } | |
306 | #endif | |
307 | } | |
308 | ||
309 | static void | |
310 | check_link_status(struct netdev_dpdk *dev) | |
311 | { | |
312 | struct rte_eth_link link; | |
313 | ||
314 | rte_eth_link_get_nowait(dev->port_id, &link); | |
315 | ||
316 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 317 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
318 | |
319 | dev->link_reset_cnt++; | |
320 | dev->link = link; | |
321 | if (dev->link.link_status) { | |
322 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
323 | dev->port_id, (unsigned)dev->link.link_speed, | |
324 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
325 | ("full-duplex") : ("half-duplex")); | |
326 | } else { | |
327 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
328 | } | |
329 | } | |
330 | } | |
331 | ||
332 | static void * | |
333 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
334 | { | |
335 | struct netdev_dpdk *dev; | |
336 | ||
337 | pthread_detach(pthread_self()); | |
338 | ||
339 | for (;;) { | |
340 | ovs_mutex_lock(&dpdk_mutex); | |
341 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
342 | ovs_mutex_lock(&dev->mutex); | |
343 | check_link_status(dev); | |
344 | ovs_mutex_unlock(&dev->mutex); | |
345 | } | |
346 | ovs_mutex_unlock(&dpdk_mutex); | |
347 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
348 | } | |
349 | ||
350 | return NULL; | |
351 | } | |
352 | ||
353 | static int | |
354 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
355 | { | |
356 | struct rte_pktmbuf_pool_private *mbp_priv; | |
357 | struct ether_addr eth_addr; | |
358 | int diag; | |
359 | int i; | |
360 | ||
361 | if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) { | |
362 | return -ENODEV; | |
363 | } | |
364 | ||
365 | diag = rte_eth_dev_configure(dev->port_id, NR_QUEUE, NR_QUEUE, &port_conf); | |
366 | if (diag) { | |
367 | VLOG_ERR("eth dev config error %d",diag); | |
368 | return diag; | |
369 | } | |
370 | ||
371 | for (i = 0; i < NR_QUEUE; i++) { | |
372 | diag = rte_eth_tx_queue_setup(dev->port_id, i, 64, 0, &tx_conf); | |
373 | if (diag) { | |
374 | VLOG_ERR("eth dev tx queue setup error %d",diag); | |
375 | return diag; | |
376 | } | |
377 | } | |
378 | ||
379 | for (i = 0; i < NR_QUEUE; i++) { | |
380 | diag = rte_eth_rx_queue_setup(dev->port_id, i, 64, 0, &rx_conf, | |
381 | dev->dpdk_mp->mp); | |
382 | if (diag) { | |
383 | VLOG_ERR("eth dev rx queue setup error %d",diag); | |
384 | return diag; | |
385 | } | |
386 | } | |
387 | ||
388 | diag = rte_eth_dev_start(dev->port_id); | |
389 | if (diag) { | |
390 | VLOG_ERR("eth dev start error %d",diag); | |
391 | return diag; | |
392 | } | |
393 | ||
394 | rte_eth_promiscuous_enable(dev->port_id); | |
395 | rte_eth_allmulticast_enable(dev->port_id); | |
396 | ||
397 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
398 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
399 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
400 | dev->port_id, ETH_ADDR_ARGS(eth_addr.addr_bytes)); | |
401 | ||
402 | memcpy(dev->hwaddr, eth_addr.addr_bytes, ETH_ADDR_LEN); | |
403 | rte_eth_link_get_nowait(dev->port_id, &dev->link); | |
404 | ||
405 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
406 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
407 | ||
408 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
409 | return 0; | |
410 | } | |
411 | ||
412 | static struct netdev_dpdk * | |
413 | netdev_dpdk_cast(const struct netdev *netdev) | |
414 | { | |
415 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
416 | } | |
417 | ||
418 | static struct netdev * | |
419 | netdev_dpdk_alloc(void) | |
420 | { | |
421 | struct netdev_dpdk *netdev = dpdk_rte_mzalloc(sizeof *netdev); | |
422 | return &netdev->up; | |
423 | } | |
424 | ||
425 | static int | |
426 | netdev_dpdk_construct(struct netdev *netdev_) | |
427 | { | |
428 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
429 | unsigned int port_no; | |
430 | char *cport; | |
431 | int err; | |
432 | int i; | |
433 | ||
434 | if (rte_eal_init_ret) { | |
435 | return rte_eal_init_ret; | |
436 | } | |
437 | ||
438 | ovs_mutex_lock(&dpdk_mutex); | |
439 | cport = netdev_->name + 4; /* Names always start with "dpdk" */ | |
440 | ||
441 | if (strncmp(netdev_->name, "dpdk", 4)) { | |
442 | err = ENODEV; | |
443 | goto unlock_dpdk; | |
444 | } | |
445 | ||
446 | port_no = strtol(cport, 0, 0); /* string must be null terminated */ | |
447 | ||
448 | for (i = 0; i < NR_QUEUE; i++) { | |
449 | rte_spinlock_init(&netdev->tx_q[i].tx_lock); | |
450 | } | |
451 | ||
452 | ovs_mutex_init(&netdev->mutex); | |
453 | ||
454 | ovs_mutex_lock(&netdev->mutex); | |
455 | netdev->flags = 0; | |
456 | ||
457 | netdev->mtu = ETHER_MTU; | |
458 | netdev->max_packet_len = MTU_TO_MAX_LEN(netdev->mtu); | |
459 | ||
460 | /* TODO: need to discover device node at run time. */ | |
461 | netdev->socket_id = SOCKET0; | |
462 | netdev->port_id = port_no; | |
463 | ||
464 | netdev->dpdk_mp = dpdk_mp_get(netdev->socket_id, netdev->mtu); | |
465 | if (!netdev->dpdk_mp) { | |
466 | err = ENOMEM; | |
467 | goto unlock_dev; | |
468 | } | |
469 | ||
470 | err = dpdk_eth_dev_init(netdev); | |
471 | if (err) { | |
472 | goto unlock_dev; | |
473 | } | |
474 | netdev_->n_rxq = NR_QUEUE; | |
475 | ||
476 | list_push_back(&dpdk_list, &netdev->list_node); | |
477 | ||
478 | unlock_dev: | |
479 | ovs_mutex_unlock(&netdev->mutex); | |
480 | unlock_dpdk: | |
481 | ovs_mutex_unlock(&dpdk_mutex); | |
482 | return err; | |
483 | } | |
484 | ||
485 | static void | |
486 | netdev_dpdk_destruct(struct netdev *netdev_) | |
487 | { | |
488 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
489 | ||
490 | ovs_mutex_lock(&dev->mutex); | |
491 | rte_eth_dev_stop(dev->port_id); | |
492 | ovs_mutex_unlock(&dev->mutex); | |
493 | ||
494 | ovs_mutex_lock(&dpdk_mutex); | |
495 | list_remove(&dev->list_node); | |
496 | dpdk_mp_put(dev->dpdk_mp); | |
497 | ovs_mutex_unlock(&dpdk_mutex); | |
498 | ||
499 | ovs_mutex_destroy(&dev->mutex); | |
500 | } | |
501 | ||
502 | static void | |
503 | netdev_dpdk_dealloc(struct netdev *netdev_) | |
504 | { | |
505 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
506 | ||
507 | rte_free(netdev); | |
508 | } | |
509 | ||
510 | static int | |
511 | netdev_dpdk_get_config(const struct netdev *netdev_, struct smap *args) | |
512 | { | |
513 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
514 | ||
515 | ovs_mutex_lock(&dev->mutex); | |
516 | ||
517 | /* TODO: Allow to configure number of queues. */ | |
518 | smap_add_format(args, "configured_rx_queues", "%u", netdev_->n_rxq); | |
519 | smap_add_format(args, "configured_tx_queues", "%u", netdev_->n_rxq); | |
520 | ovs_mutex_unlock(&dev->mutex); | |
521 | ||
522 | return 0; | |
523 | } | |
524 | ||
525 | static struct netdev_rxq * | |
526 | netdev_dpdk_rxq_alloc(void) | |
527 | { | |
528 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
529 | ||
530 | return &rx->up; | |
531 | } | |
532 | ||
533 | static struct netdev_rxq_dpdk * | |
534 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rx) | |
535 | { | |
536 | return CONTAINER_OF(rx, struct netdev_rxq_dpdk, up); | |
537 | } | |
538 | ||
539 | static int | |
540 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq_) | |
541 | { | |
542 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
543 | struct netdev_dpdk *netdev = netdev_dpdk_cast(rx->up.netdev); | |
544 | ||
545 | ovs_mutex_lock(&netdev->mutex); | |
546 | rx->port_id = netdev->port_id; | |
547 | ovs_mutex_unlock(&netdev->mutex); | |
548 | ||
549 | return 0; | |
550 | } | |
551 | ||
552 | static void | |
553 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq_ OVS_UNUSED) | |
554 | { | |
555 | } | |
556 | ||
557 | static void | |
558 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq_) | |
559 | { | |
560 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
561 | ||
562 | rte_free(rx); | |
563 | } | |
564 | ||
565 | inline static void | |
566 | dpdk_queue_flush(struct netdev_dpdk *dev, int qid) | |
567 | { | |
568 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
569 | uint32_t nb_tx; | |
570 | ||
571 | if (txq->count == 0) { | |
572 | return; | |
573 | } | |
574 | rte_spinlock_lock(&txq->tx_lock); | |
575 | nb_tx = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts, txq->count); | |
576 | if (nb_tx != txq->count) { | |
577 | /* free buffers if we couldn't transmit packets */ | |
578 | rte_mempool_put_bulk(dev->dpdk_mp->mp, | |
579 | (void **) &txq->burst_pkts[nb_tx], | |
580 | (txq->count - nb_tx)); | |
581 | } | |
582 | txq->count = 0; | |
583 | rte_spinlock_unlock(&txq->tx_lock); | |
584 | } | |
585 | ||
8a9562d2 | 586 | static int |
b3cd9f9d | 587 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq_, struct ofpbuf **packets, int *c) |
8a9562d2 PS |
588 | { |
589 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
590 | struct netdev *netdev = rx->up.netdev; | |
591 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 592 | int nb_rx; |
8a9562d2 PS |
593 | |
594 | dpdk_queue_flush(dev, rxq_->queue_id); | |
595 | ||
596 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq_->queue_id, | |
b3cd9f9d | 597 | (struct rte_mbuf **) packets, MAX_RX_QUEUE_LEN); |
8a9562d2 PS |
598 | if (!nb_rx) { |
599 | return EAGAIN; | |
600 | } | |
601 | ||
8a9562d2 PS |
602 | *c = nb_rx; |
603 | ||
604 | return 0; | |
605 | } | |
606 | ||
607 | inline static void | |
608 | dpdk_queue_pkt(struct netdev_dpdk *dev, int qid, | |
609 | struct rte_mbuf *pkt) | |
610 | { | |
611 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
612 | uint64_t diff_tsc; | |
613 | uint64_t cur_tsc; | |
614 | uint32_t nb_tx; | |
615 | ||
616 | rte_spinlock_lock(&txq->tx_lock); | |
617 | txq->burst_pkts[txq->count++] = pkt; | |
618 | if (txq->count == MAX_TX_QUEUE_LEN) { | |
619 | goto flush; | |
620 | } | |
621 | cur_tsc = rte_get_timer_cycles(); | |
622 | if (txq->count == 1) { | |
623 | txq->tsc = cur_tsc; | |
624 | } | |
625 | diff_tsc = cur_tsc - txq->tsc; | |
626 | if (diff_tsc >= DRAIN_TSC) { | |
627 | goto flush; | |
628 | } | |
629 | rte_spinlock_unlock(&txq->tx_lock); | |
630 | return; | |
631 | ||
632 | flush: | |
633 | nb_tx = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts, txq->count); | |
634 | if (nb_tx != txq->count) { | |
635 | /* free buffers if we couldn't transmit packets */ | |
636 | rte_mempool_put_bulk(dev->dpdk_mp->mp, | |
637 | (void **) &txq->burst_pkts[nb_tx], | |
638 | (txq->count - nb_tx)); | |
639 | } | |
640 | txq->count = 0; | |
641 | rte_spinlock_unlock(&txq->tx_lock); | |
642 | } | |
643 | ||
644 | /* Tx function. Transmit packets indefinitely */ | |
645 | static void | |
646 | dpdk_do_tx_copy(struct netdev *netdev, char *buf, int size) | |
647 | { | |
648 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
649 | struct rte_mbuf *pkt; | |
650 | ||
651 | pkt = rte_pktmbuf_alloc(dev->dpdk_mp->mp); | |
652 | if (!pkt) { | |
653 | ovs_mutex_lock(&dev->mutex); | |
654 | dev->stats.tx_dropped++; | |
655 | ovs_mutex_unlock(&dev->mutex); | |
656 | return; | |
657 | } | |
658 | ||
659 | /* We have to do a copy for now */ | |
660 | memcpy(pkt->pkt.data, buf, size); | |
661 | ||
662 | rte_pktmbuf_data_len(pkt) = size; | |
663 | rte_pktmbuf_pkt_len(pkt) = size; | |
664 | ||
665 | dpdk_queue_pkt(dev, NON_PMD_THREAD_TX_QUEUE, pkt); | |
666 | dpdk_queue_flush(dev, NON_PMD_THREAD_TX_QUEUE); | |
667 | } | |
668 | ||
669 | static int | |
670 | netdev_dpdk_send(struct netdev *netdev, | |
671 | struct ofpbuf *ofpbuf, bool may_steal) | |
672 | { | |
673 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
674 | int ret; | |
675 | ||
1f317cb5 | 676 | if (ofpbuf_size(ofpbuf) > dev->max_packet_len) { |
8a9562d2 | 677 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1f317cb5 | 678 | (int)ofpbuf_size(ofpbuf) , dev->max_packet_len); |
8a9562d2 PS |
679 | |
680 | ovs_mutex_lock(&dev->mutex); | |
681 | dev->stats.tx_dropped++; | |
682 | ovs_mutex_unlock(&dev->mutex); | |
683 | ||
684 | ret = E2BIG; | |
685 | goto out; | |
686 | } | |
687 | ||
b3cd9f9d | 688 | if (!may_steal || ofpbuf->source != OFPBUF_DPDK) { |
1f317cb5 | 689 | dpdk_do_tx_copy(netdev, (char *) ofpbuf_data(ofpbuf), ofpbuf_size(ofpbuf)); |
b3cd9f9d PS |
690 | |
691 | if (may_steal) { | |
692 | ofpbuf_delete(ofpbuf); | |
693 | } | |
8a9562d2 | 694 | } else { |
8a9562d2 PS |
695 | int qid; |
696 | ||
8a9562d2 PS |
697 | qid = rte_lcore_id() % NR_QUEUE; |
698 | ||
b3cd9f9d | 699 | dpdk_queue_pkt(dev, qid, (struct rte_mbuf *)ofpbuf); |
8a9562d2 | 700 | |
8a9562d2 PS |
701 | } |
702 | ret = 0; | |
703 | ||
704 | out: | |
8a9562d2 PS |
705 | return ret; |
706 | } | |
707 | ||
708 | static int | |
709 | netdev_dpdk_set_etheraddr(struct netdev *netdev, | |
710 | const uint8_t mac[ETH_ADDR_LEN]) | |
711 | { | |
712 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
713 | ||
714 | ovs_mutex_lock(&dev->mutex); | |
715 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
716 | memcpy(dev->hwaddr, mac, ETH_ADDR_LEN); | |
045c0d1a | 717 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
718 | } |
719 | ovs_mutex_unlock(&dev->mutex); | |
720 | ||
721 | return 0; | |
722 | } | |
723 | ||
724 | static int | |
725 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, | |
726 | uint8_t mac[ETH_ADDR_LEN]) | |
727 | { | |
728 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
729 | ||
730 | ovs_mutex_lock(&dev->mutex); | |
731 | memcpy(mac, dev->hwaddr, ETH_ADDR_LEN); | |
732 | ovs_mutex_unlock(&dev->mutex); | |
733 | ||
734 | return 0; | |
735 | } | |
736 | ||
737 | static int | |
738 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
739 | { | |
740 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
741 | ||
742 | ovs_mutex_lock(&dev->mutex); | |
743 | *mtup = dev->mtu; | |
744 | ovs_mutex_unlock(&dev->mutex); | |
745 | ||
746 | return 0; | |
747 | } | |
748 | ||
749 | static int | |
750 | netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu) | |
751 | { | |
752 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
753 | int old_mtu, err; | |
754 | struct dpdk_mp *old_mp; | |
755 | struct dpdk_mp *mp; | |
756 | ||
757 | ovs_mutex_lock(&dpdk_mutex); | |
758 | ovs_mutex_lock(&dev->mutex); | |
759 | if (dev->mtu == mtu) { | |
760 | err = 0; | |
761 | goto out; | |
762 | } | |
763 | ||
764 | mp = dpdk_mp_get(dev->socket_id, dev->mtu); | |
765 | if (!mp) { | |
766 | err = ENOMEM; | |
767 | goto out; | |
768 | } | |
769 | ||
770 | rte_eth_dev_stop(dev->port_id); | |
771 | ||
772 | old_mtu = dev->mtu; | |
773 | old_mp = dev->dpdk_mp; | |
774 | dev->dpdk_mp = mp; | |
775 | dev->mtu = mtu; | |
776 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
777 | ||
778 | err = dpdk_eth_dev_init(dev); | |
779 | if (err) { | |
780 | ||
781 | dpdk_mp_put(mp); | |
782 | dev->mtu = old_mtu; | |
783 | dev->dpdk_mp = old_mp; | |
784 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
785 | dpdk_eth_dev_init(dev); | |
786 | goto out; | |
787 | } | |
788 | ||
789 | dpdk_mp_put(old_mp); | |
045c0d1a | 790 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
791 | out: |
792 | ovs_mutex_unlock(&dev->mutex); | |
793 | ovs_mutex_unlock(&dpdk_mutex); | |
794 | return err; | |
795 | } | |
796 | ||
797 | static int | |
798 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier); | |
799 | ||
800 | static int | |
801 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
802 | { | |
803 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
804 | struct rte_eth_stats rte_stats; | |
805 | bool gg; | |
806 | ||
807 | netdev_dpdk_get_carrier(netdev, &gg); | |
808 | ovs_mutex_lock(&dev->mutex); | |
809 | rte_eth_stats_get(dev->port_id, &rte_stats); | |
810 | ||
811 | *stats = dev->stats_offset; | |
812 | ||
813 | stats->rx_packets += rte_stats.ipackets; | |
814 | stats->tx_packets += rte_stats.opackets; | |
815 | stats->rx_bytes += rte_stats.ibytes; | |
816 | stats->tx_bytes += rte_stats.obytes; | |
817 | stats->rx_errors += rte_stats.ierrors; | |
818 | stats->tx_errors += rte_stats.oerrors; | |
819 | stats->multicast += rte_stats.imcasts; | |
820 | ||
821 | stats->tx_dropped += dev->stats.tx_dropped; | |
822 | ovs_mutex_unlock(&dev->mutex); | |
823 | ||
824 | return 0; | |
825 | } | |
826 | ||
827 | static int | |
828 | netdev_dpdk_set_stats(struct netdev *netdev, const struct netdev_stats *stats) | |
829 | { | |
830 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
831 | ||
832 | ovs_mutex_lock(&dev->mutex); | |
833 | dev->stats_offset = *stats; | |
834 | ovs_mutex_unlock(&dev->mutex); | |
835 | ||
836 | return 0; | |
837 | } | |
838 | ||
839 | static int | |
840 | netdev_dpdk_get_features(const struct netdev *netdev_, | |
841 | enum netdev_features *current, | |
842 | enum netdev_features *advertised OVS_UNUSED, | |
843 | enum netdev_features *supported OVS_UNUSED, | |
844 | enum netdev_features *peer OVS_UNUSED) | |
845 | { | |
846 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
847 | struct rte_eth_link link; | |
848 | ||
849 | ovs_mutex_lock(&dev->mutex); | |
850 | link = dev->link; | |
851 | ovs_mutex_unlock(&dev->mutex); | |
852 | ||
853 | if (link.link_duplex == ETH_LINK_AUTONEG_DUPLEX) { | |
854 | if (link.link_speed == ETH_LINK_SPEED_AUTONEG) { | |
855 | *current = NETDEV_F_AUTONEG; | |
856 | } | |
857 | } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { | |
858 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
859 | *current = NETDEV_F_10MB_HD; | |
860 | } | |
861 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
862 | *current = NETDEV_F_100MB_HD; | |
863 | } | |
864 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
865 | *current = NETDEV_F_1GB_HD; | |
866 | } | |
867 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
868 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
869 | *current = NETDEV_F_10MB_FD; | |
870 | } | |
871 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
872 | *current = NETDEV_F_100MB_FD; | |
873 | } | |
874 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
875 | *current = NETDEV_F_1GB_FD; | |
876 | } | |
877 | if (link.link_speed == ETH_LINK_SPEED_10000) { | |
878 | *current = NETDEV_F_10GB_FD; | |
879 | } | |
880 | } | |
881 | ||
882 | return 0; | |
883 | } | |
884 | ||
885 | static int | |
886 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
887 | { | |
888 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
889 | int ifindex; | |
890 | ||
891 | ovs_mutex_lock(&dev->mutex); | |
892 | ifindex = dev->port_id; | |
893 | ovs_mutex_unlock(&dev->mutex); | |
894 | ||
895 | return ifindex; | |
896 | } | |
897 | ||
898 | static int | |
899 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier) | |
900 | { | |
901 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
902 | ||
903 | ovs_mutex_lock(&dev->mutex); | |
904 | check_link_status(dev); | |
905 | *carrier = dev->link.link_status; | |
906 | ovs_mutex_unlock(&dev->mutex); | |
907 | ||
908 | return 0; | |
909 | } | |
910 | ||
911 | static long long int | |
912 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev_) | |
913 | { | |
914 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
915 | long long int carrier_resets; | |
916 | ||
917 | ovs_mutex_lock(&dev->mutex); | |
918 | carrier_resets = dev->link_reset_cnt; | |
919 | ovs_mutex_unlock(&dev->mutex); | |
920 | ||
921 | return carrier_resets; | |
922 | } | |
923 | ||
924 | static int | |
925 | netdev_dpdk_set_miimon(struct netdev *netdev_ OVS_UNUSED, | |
926 | long long int interval OVS_UNUSED) | |
927 | { | |
928 | return 0; | |
929 | } | |
930 | ||
931 | static int | |
932 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
933 | enum netdev_flags off, enum netdev_flags on, | |
934 | enum netdev_flags *old_flagsp) | |
935 | OVS_REQUIRES(dev->mutex) | |
936 | { | |
937 | int err; | |
938 | ||
939 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
940 | return EINVAL; | |
941 | } | |
942 | ||
943 | *old_flagsp = dev->flags; | |
944 | dev->flags |= on; | |
945 | dev->flags &= ~off; | |
946 | ||
947 | if (dev->flags == *old_flagsp) { | |
948 | return 0; | |
949 | } | |
950 | ||
951 | if (dev->flags & NETDEV_UP) { | |
952 | err = rte_eth_dev_start(dev->port_id); | |
953 | if (err) | |
954 | return err; | |
955 | } | |
956 | ||
957 | if (dev->flags & NETDEV_PROMISC) { | |
958 | rte_eth_promiscuous_enable(dev->port_id); | |
959 | } | |
960 | ||
961 | if (!(dev->flags & NETDEV_UP)) { | |
962 | rte_eth_dev_stop(dev->port_id); | |
963 | } | |
964 | ||
965 | return 0; | |
966 | } | |
967 | ||
968 | static int | |
969 | netdev_dpdk_update_flags(struct netdev *netdev_, | |
970 | enum netdev_flags off, enum netdev_flags on, | |
971 | enum netdev_flags *old_flagsp) | |
972 | { | |
973 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
974 | int error; | |
975 | ||
976 | ovs_mutex_lock(&netdev->mutex); | |
977 | error = netdev_dpdk_update_flags__(netdev, off, on, old_flagsp); | |
978 | ovs_mutex_unlock(&netdev->mutex); | |
979 | ||
980 | return error; | |
981 | } | |
982 | ||
983 | static int | |
984 | netdev_dpdk_get_status(const struct netdev *netdev_, struct smap *args) | |
985 | { | |
986 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
987 | struct rte_eth_dev_info dev_info; | |
988 | ||
989 | if (dev->port_id <= 0) | |
990 | return ENODEV; | |
991 | ||
992 | ovs_mutex_lock(&dev->mutex); | |
993 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
994 | ovs_mutex_unlock(&dev->mutex); | |
995 | ||
996 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
997 | ||
998 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); | |
999 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1000 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
1001 | smap_add_format(args, "max_rx_pktlen", "%u", dev_info.max_rx_pktlen); | |
1002 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); | |
1003 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
1004 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
1005 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
1006 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
1007 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
1008 | ||
1009 | smap_add_format(args, "pci-vendor_id", "0x%u", dev_info.pci_dev->id.vendor_id); | |
1010 | smap_add_format(args, "pci-device_id", "0x%x", dev_info.pci_dev->id.device_id); | |
1011 | ||
1012 | return 0; | |
1013 | } | |
1014 | ||
1015 | static void | |
1016 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
1017 | OVS_REQUIRES(dev->mutex) | |
1018 | { | |
1019 | enum netdev_flags old_flags; | |
1020 | ||
1021 | if (admin_state) { | |
1022 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
1023 | } else { | |
1024 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
1025 | } | |
1026 | } | |
1027 | ||
1028 | static void | |
1029 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
1030 | const char *argv[], void *aux OVS_UNUSED) | |
1031 | { | |
1032 | bool up; | |
1033 | ||
1034 | if (!strcasecmp(argv[argc - 1], "up")) { | |
1035 | up = true; | |
1036 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
1037 | up = false; | |
1038 | } else { | |
1039 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
1040 | return; | |
1041 | } | |
1042 | ||
1043 | if (argc > 2) { | |
1044 | struct netdev *netdev = netdev_from_name(argv[1]); | |
1045 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
1046 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
1047 | ||
1048 | ovs_mutex_lock(&dpdk_dev->mutex); | |
1049 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
1050 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
1051 | ||
1052 | netdev_close(netdev); | |
1053 | } else { | |
1054 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
1055 | netdev_close(netdev); | |
1056 | return; | |
1057 | } | |
1058 | } else { | |
1059 | struct netdev_dpdk *netdev; | |
1060 | ||
1061 | ovs_mutex_lock(&dpdk_mutex); | |
1062 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
1063 | ovs_mutex_lock(&netdev->mutex); | |
1064 | netdev_dpdk_set_admin_state__(netdev, up); | |
1065 | ovs_mutex_unlock(&netdev->mutex); | |
1066 | } | |
1067 | ovs_mutex_unlock(&dpdk_mutex); | |
1068 | } | |
1069 | unixctl_command_reply(conn, "OK"); | |
1070 | } | |
1071 | ||
1072 | static int | |
1073 | dpdk_class_init(void) | |
1074 | { | |
1075 | int result; | |
1076 | ||
1077 | if (rte_eal_init_ret) { | |
1078 | return 0; | |
1079 | } | |
1080 | ||
1081 | result = rte_pmd_init_all(); | |
1082 | if (result) { | |
1083 | VLOG_ERR("Cannot init PMD"); | |
1084 | return result; | |
1085 | } | |
1086 | ||
1087 | result = rte_eal_pci_probe(); | |
1088 | if (result) { | |
1089 | VLOG_ERR("Cannot probe PCI"); | |
1090 | return result; | |
1091 | } | |
1092 | ||
1093 | if (rte_eth_dev_count() < 1) { | |
1094 | VLOG_ERR("No Ethernet devices found. Try assigning ports to UIO."); | |
1095 | } | |
1096 | ||
1097 | VLOG_INFO("Ethernet Device Count: %d", (int)rte_eth_dev_count()); | |
1098 | ||
1099 | list_init(&dpdk_list); | |
1100 | list_init(&dpdk_mp_list); | |
1101 | ||
1102 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
1103 | "[netdev] up|down", 1, 2, | |
1104 | netdev_dpdk_set_admin_state, NULL); | |
1105 | ||
8ba0a522 | 1106 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); |
8a9562d2 PS |
1107 | return 0; |
1108 | } | |
1109 | ||
1110 | static struct netdev_class netdev_dpdk_class = { | |
1111 | "dpdk", | |
1112 | dpdk_class_init, /* init */ | |
1113 | NULL, /* netdev_dpdk_run */ | |
1114 | NULL, /* netdev_dpdk_wait */ | |
1115 | ||
1116 | netdev_dpdk_alloc, | |
1117 | netdev_dpdk_construct, | |
1118 | netdev_dpdk_destruct, | |
1119 | netdev_dpdk_dealloc, | |
1120 | netdev_dpdk_get_config, | |
1121 | NULL, /* netdev_dpdk_set_config */ | |
1122 | NULL, /* get_tunnel_config */ | |
1123 | ||
1124 | netdev_dpdk_send, /* send */ | |
1125 | NULL, /* send_wait */ | |
1126 | ||
1127 | netdev_dpdk_set_etheraddr, | |
1128 | netdev_dpdk_get_etheraddr, | |
1129 | netdev_dpdk_get_mtu, | |
1130 | netdev_dpdk_set_mtu, | |
1131 | netdev_dpdk_get_ifindex, | |
1132 | netdev_dpdk_get_carrier, | |
1133 | netdev_dpdk_get_carrier_resets, | |
1134 | netdev_dpdk_set_miimon, | |
1135 | netdev_dpdk_get_stats, | |
1136 | netdev_dpdk_set_stats, | |
1137 | netdev_dpdk_get_features, | |
1138 | NULL, /* set_advertisements */ | |
1139 | ||
1140 | NULL, /* set_policing */ | |
1141 | NULL, /* get_qos_types */ | |
1142 | NULL, /* get_qos_capabilities */ | |
1143 | NULL, /* get_qos */ | |
1144 | NULL, /* set_qos */ | |
1145 | NULL, /* get_queue */ | |
1146 | NULL, /* set_queue */ | |
1147 | NULL, /* delete_queue */ | |
1148 | NULL, /* get_queue_stats */ | |
1149 | NULL, /* queue_dump_start */ | |
1150 | NULL, /* queue_dump_next */ | |
1151 | NULL, /* queue_dump_done */ | |
1152 | NULL, /* dump_queue_stats */ | |
1153 | ||
1154 | NULL, /* get_in4 */ | |
1155 | NULL, /* set_in4 */ | |
1156 | NULL, /* get_in6 */ | |
1157 | NULL, /* add_router */ | |
1158 | NULL, /* get_next_hop */ | |
1159 | netdev_dpdk_get_status, | |
1160 | NULL, /* arp_lookup */ | |
1161 | ||
1162 | netdev_dpdk_update_flags, | |
1163 | ||
1164 | netdev_dpdk_rxq_alloc, | |
1165 | netdev_dpdk_rxq_construct, | |
1166 | netdev_dpdk_rxq_destruct, | |
1167 | netdev_dpdk_rxq_dealloc, | |
1168 | netdev_dpdk_rxq_recv, | |
1169 | NULL, /* rxq_wait */ | |
1170 | NULL, /* rxq_drain */ | |
1171 | }; | |
1172 | ||
1173 | int | |
1174 | dpdk_init(int argc, char **argv) | |
1175 | { | |
1176 | int result; | |
1177 | ||
1178 | if (strcmp(argv[1], "--dpdk")) | |
1179 | return 0; | |
1180 | ||
1181 | argc--; | |
1182 | argv++; | |
1183 | ||
1184 | /* Make sure things are initialized ... */ | |
1185 | result = rte_eal_init(argc, argv); | |
1186 | if (result < 0) | |
1187 | ovs_abort(result, "Cannot init EAL\n"); | |
1188 | ||
1189 | rte_memzone_dump(); | |
1190 | rte_eal_init_ret = 0; | |
1191 | ||
1192 | return result; | |
1193 | } | |
1194 | ||
1195 | void | |
1196 | netdev_dpdk_register(void) | |
1197 | { | |
1198 | netdev_register_provider(&netdev_dpdk_class); | |
1199 | } | |
8617afff PS |
1200 | |
1201 | int | |
1202 | pmd_thread_setaffinity_cpu(int cpu) | |
1203 | { | |
1204 | cpu_set_t cpuset; | |
1205 | int err; | |
1206 | ||
1207 | CPU_ZERO(&cpuset); | |
1208 | CPU_SET(cpu, &cpuset); | |
1209 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset); | |
1210 | if (err) { | |
1211 | VLOG_ERR("Thread affinity error %d",err); | |
1212 | return err; | |
1213 | } | |
1214 | RTE_PER_LCORE(_lcore_id) = cpu; | |
1215 | ||
1216 | return 0; | |
1217 | } |