]>
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 | ||
e14deea0 | 31 | #include "dp-packet.h" |
8a9562d2 PS |
32 | #include "dpif-netdev.h" |
33 | #include "list.h" | |
34 | #include "netdev-dpdk.h" | |
35 | #include "netdev-provider.h" | |
36 | #include "netdev-vport.h" | |
37 | #include "odp-util.h" | |
38 | #include "ofp-print.h" | |
94143fc4 | 39 | #include "ovs-numa.h" |
8a9562d2 PS |
40 | #include "ovs-thread.h" |
41 | #include "ovs-rcu.h" | |
42 | #include "packets.h" | |
43 | #include "shash.h" | |
8a9562d2 PS |
44 | #include "sset.h" |
45 | #include "unaligned.h" | |
46 | #include "timeval.h" | |
47 | #include "unixctl.h" | |
e6211adc | 48 | #include "openvswitch/vlog.h" |
8a9562d2 PS |
49 | |
50 | VLOG_DEFINE_THIS_MODULE(dpdk); | |
51 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
52 | ||
53 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
54 | ||
55 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
56 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
57 | ||
58 | /* | |
59 | * need to reserve tons of extra space in the mbufs so we can align the | |
60 | * DMA addresses to 4KB. | |
61 | */ | |
62 | ||
63 | #define MTU_TO_MAX_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) | |
64 | #define MBUF_SIZE(mtu) (MTU_TO_MAX_LEN(mtu) + (512) + \ | |
65 | sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM) | |
66 | ||
645b8934 | 67 | /* XXX: mempool size should be based on system resources. */ |
8a9562d2 PS |
68 | #define NB_MBUF (4096 * 64) |
69 | #define MP_CACHE_SZ (256 * 2) | |
70 | #define SOCKET0 0 | |
71 | ||
79f5354c PM |
72 | #define NIC_PORT_RX_Q_SIZE 2048 /* Size of Physical NIC RX Queue, Max (n+32<=4096)*/ |
73 | #define NIC_PORT_TX_Q_SIZE 2048 /* Size of Physical NIC TX Queue, Max (n+32<=4096)*/ | |
74 | ||
645b8934 | 75 | /* XXX: Needs per NIC value for these constants. */ |
8a9562d2 PS |
76 | #define RX_PTHRESH 32 /* Default values of RX prefetch threshold reg. */ |
77 | #define RX_HTHRESH 32 /* Default values of RX host threshold reg. */ | |
78 | #define RX_WTHRESH 16 /* Default values of RX write-back threshold reg. */ | |
79 | ||
80 | #define TX_PTHRESH 36 /* Default values of TX prefetch threshold reg. */ | |
81 | #define TX_HTHRESH 0 /* Default values of TX host threshold reg. */ | |
82 | #define TX_WTHRESH 0 /* Default values of TX write-back threshold reg. */ | |
83 | ||
84 | static const struct rte_eth_conf port_conf = { | |
a28ddd11 DDP |
85 | .rxmode = { |
86 | .mq_mode = ETH_MQ_RX_RSS, | |
87 | .split_hdr_size = 0, | |
88 | .header_split = 0, /* Header Split disabled */ | |
89 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
90 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
91 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
92 | .hw_strip_crc = 0, | |
93 | }, | |
94 | .rx_adv_conf = { | |
95 | .rss_conf = { | |
96 | .rss_key = NULL, | |
61a2647e DDP |
97 | .rss_hf = ETH_RSS_IPV4_TCP | ETH_RSS_IPV4 | ETH_RSS_IPV6 |
98 | | ETH_RSS_IPV4_UDP | ETH_RSS_IPV6_TCP | ETH_RSS_IPV6_UDP, | |
8a9562d2 | 99 | }, |
a28ddd11 DDP |
100 | }, |
101 | .txmode = { | |
102 | .mq_mode = ETH_MQ_TX_NONE, | |
103 | }, | |
8a9562d2 PS |
104 | }; |
105 | ||
106 | static const struct rte_eth_rxconf rx_conf = { | |
a28ddd11 DDP |
107 | .rx_thresh = { |
108 | .pthresh = RX_PTHRESH, | |
109 | .hthresh = RX_HTHRESH, | |
110 | .wthresh = RX_WTHRESH, | |
111 | }, | |
8a9562d2 PS |
112 | }; |
113 | ||
114 | static const struct rte_eth_txconf tx_conf = { | |
a28ddd11 DDP |
115 | .tx_thresh = { |
116 | .pthresh = TX_PTHRESH, | |
117 | .hthresh = TX_HTHRESH, | |
118 | .wthresh = TX_WTHRESH, | |
119 | }, | |
120 | .tx_free_thresh = 0, | |
121 | .tx_rs_thresh = 0, | |
94777510 | 122 | .txq_flags = ETH_TXQ_FLAGS_NOMULTSEGS|ETH_TXQ_FLAGS_NOOFFLOADS, |
8a9562d2 PS |
123 | }; |
124 | ||
3a100265 DDP |
125 | enum { MAX_RX_QUEUE_LEN = 192 }; |
126 | enum { MAX_TX_QUEUE_LEN = 384 }; | |
58f7c37b DDP |
127 | enum { DPDK_RING_SIZE = 256 }; |
128 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
129 | enum { DRAIN_TSC = 200000ULL }; |
130 | ||
131 | static int rte_eal_init_ret = ENODEV; | |
132 | ||
133 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; | |
134 | ||
135 | /* Contains all 'struct dpdk_dev's. */ | |
ca6ba700 | 136 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 137 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 138 | |
ca6ba700 | 139 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 140 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 141 | |
db73f716 DDP |
142 | /* This mutex must be used by non pmd threads when allocating or freeing |
143 | * mbufs through mempools. Since dpdk_queue_pkts() and dpdk_queue_flush() may | |
144 | * use mempools, a non pmd thread should hold this mutex while calling them */ | |
145 | struct ovs_mutex nonpmd_mempool_mutex = OVS_MUTEX_INITIALIZER; | |
146 | ||
8a9562d2 PS |
147 | struct dpdk_mp { |
148 | struct rte_mempool *mp; | |
149 | int mtu; | |
150 | int socket_id; | |
151 | int refcount; | |
ca6ba700 | 152 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
153 | }; |
154 | ||
5a034064 AW |
155 | /* There should be one 'struct dpdk_tx_queue' created for |
156 | * each cpu core. */ | |
8a9562d2 | 157 | struct dpdk_tx_queue { |
94143fc4 AW |
158 | bool flush_tx; /* Set to true to flush queue everytime */ |
159 | /* pkts are queued. */ | |
8a9562d2 PS |
160 | int count; |
161 | uint64_t tsc; | |
162 | struct rte_mbuf *burst_pkts[MAX_TX_QUEUE_LEN]; | |
163 | }; | |
164 | ||
95fb793a | 165 | /* dpdk has no way to remove dpdk ring ethernet devices |
166 | so we have to keep them around once they've been created | |
167 | */ | |
168 | ||
ca6ba700 | 169 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 170 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 171 | |
172 | struct dpdk_ring { | |
173 | /* For the client rings */ | |
174 | struct rte_ring *cring_tx; | |
175 | struct rte_ring *cring_rx; | |
176 | int user_port_id; /* User given port no, parsed from port name */ | |
177 | int eth_port_id; /* ethernet device port id */ | |
ca6ba700 | 178 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 179 | }; |
180 | ||
8a9562d2 PS |
181 | struct netdev_dpdk { |
182 | struct netdev up; | |
183 | int port_id; | |
184 | int max_packet_len; | |
185 | ||
5a034064 | 186 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
187 | |
188 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
189 | ||
190 | struct dpdk_mp *dpdk_mp; | |
191 | int mtu; | |
192 | int socket_id; | |
193 | int buf_size; | |
8a9562d2 PS |
194 | struct netdev_stats stats; |
195 | ||
196 | uint8_t hwaddr[ETH_ADDR_LEN]; | |
197 | enum netdev_flags flags; | |
198 | ||
199 | struct rte_eth_link link; | |
200 | int link_reset_cnt; | |
201 | ||
202 | /* In dpdk_list. */ | |
ca6ba700 | 203 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
7251515e | 204 | rte_spinlock_t dpdkr_tx_lock; |
8a9562d2 PS |
205 | }; |
206 | ||
207 | struct netdev_rxq_dpdk { | |
208 | struct netdev_rxq up; | |
209 | int port_id; | |
210 | }; | |
211 | ||
db73f716 DDP |
212 | static bool thread_is_pmd(void); |
213 | ||
8a9562d2 PS |
214 | static int netdev_dpdk_construct(struct netdev *); |
215 | ||
216 | static bool | |
217 | is_dpdk_class(const struct netdev_class *class) | |
218 | { | |
219 | return class->construct == netdev_dpdk_construct; | |
220 | } | |
221 | ||
645b8934 | 222 | /* XXX: use dpdk malloc for entire OVS. infact huge page shld be used |
8a9562d2 PS |
223 | * for all other sengments data, bss and text. */ |
224 | ||
225 | static void * | |
226 | dpdk_rte_mzalloc(size_t sz) | |
227 | { | |
228 | void *ptr; | |
229 | ||
230 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
231 | if (ptr == NULL) { | |
232 | out_of_memory(); | |
233 | } | |
234 | return ptr; | |
235 | } | |
236 | ||
db73f716 DDP |
237 | /* XXX this function should be called only by pmd threads (or by non pmd |
238 | * threads holding the nonpmd_mempool_mutex) */ | |
8a9562d2 | 239 | void |
e14deea0 | 240 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 241 | { |
db73f716 | 242 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 243 | |
db73f716 | 244 | rte_pktmbuf_free_seg(pkt); |
8a9562d2 PS |
245 | } |
246 | ||
b3cd9f9d PS |
247 | static void |
248 | __rte_pktmbuf_init(struct rte_mempool *mp, | |
249 | void *opaque_arg OVS_UNUSED, | |
250 | void *_m, | |
251 | unsigned i OVS_UNUSED) | |
252 | { | |
253 | struct rte_mbuf *m = _m; | |
e14deea0 | 254 | uint32_t buf_len = mp->elt_size - sizeof(struct dp_packet); |
b3cd9f9d | 255 | |
e14deea0 | 256 | RTE_MBUF_ASSERT(mp->elt_size >= sizeof(struct dp_packet)); |
b3cd9f9d PS |
257 | |
258 | memset(m, 0, mp->elt_size); | |
259 | ||
260 | /* start of buffer is just after mbuf structure */ | |
e14deea0 | 261 | m->buf_addr = (char *)m + sizeof(struct dp_packet); |
b3cd9f9d | 262 | m->buf_physaddr = rte_mempool_virt2phy(mp, m) + |
e14deea0 | 263 | sizeof(struct dp_packet); |
b3cd9f9d PS |
264 | m->buf_len = (uint16_t)buf_len; |
265 | ||
266 | /* keep some headroom between start of buffer and data */ | |
267 | m->pkt.data = (char*) m->buf_addr + RTE_MIN(RTE_PKTMBUF_HEADROOM, m->buf_len); | |
268 | ||
269 | /* init some constant fields */ | |
270 | m->type = RTE_MBUF_PKT; | |
271 | m->pool = mp; | |
272 | m->pkt.nb_segs = 1; | |
273 | m->pkt.in_port = 0xff; | |
274 | } | |
275 | ||
276 | static void | |
277 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
278 | void *opaque_arg OVS_UNUSED, | |
279 | void *_m, | |
280 | unsigned i OVS_UNUSED) | |
281 | { | |
282 | struct rte_mbuf *m = _m; | |
283 | ||
284 | __rte_pktmbuf_init(mp, opaque_arg, _m, i); | |
285 | ||
cf62fa4c | 286 | dp_packet_init_dpdk((struct dp_packet *) m, m->buf_len); |
b3cd9f9d PS |
287 | } |
288 | ||
8a9562d2 PS |
289 | static struct dpdk_mp * |
290 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
291 | { | |
292 | struct dpdk_mp *dmp = NULL; | |
293 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
294 | ||
295 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
296 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
297 | dmp->refcount++; | |
298 | return dmp; | |
299 | } | |
300 | } | |
301 | ||
302 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
303 | dmp->socket_id = socket_id; | |
304 | dmp->mtu = mtu; | |
305 | dmp->refcount = 1; | |
306 | ||
34631d72 AW |
307 | if (snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d_%d", dmp->mtu, |
308 | dmp->socket_id) < 0) { | |
95fb793a | 309 | return NULL; |
310 | } | |
311 | ||
8a9562d2 PS |
312 | dmp->mp = rte_mempool_create(mp_name, NB_MBUF, MBUF_SIZE(mtu), |
313 | MP_CACHE_SZ, | |
314 | sizeof(struct rte_pktmbuf_pool_private), | |
315 | rte_pktmbuf_pool_init, NULL, | |
b3cd9f9d | 316 | ovs_rte_pktmbuf_init, NULL, |
8a9562d2 PS |
317 | socket_id, 0); |
318 | ||
319 | if (dmp->mp == NULL) { | |
320 | return NULL; | |
321 | } | |
322 | ||
323 | list_push_back(&dpdk_mp_list, &dmp->list_node); | |
324 | return dmp; | |
325 | } | |
326 | ||
327 | static void | |
328 | dpdk_mp_put(struct dpdk_mp *dmp) | |
329 | { | |
330 | ||
331 | if (!dmp) { | |
332 | return; | |
333 | } | |
334 | ||
335 | dmp->refcount--; | |
336 | ovs_assert(dmp->refcount >= 0); | |
337 | ||
338 | #if 0 | |
339 | /* I could not find any API to destroy mp. */ | |
340 | if (dmp->refcount == 0) { | |
341 | list_delete(dmp->list_node); | |
342 | /* destroy mp-pool. */ | |
343 | } | |
344 | #endif | |
345 | } | |
346 | ||
347 | static void | |
348 | check_link_status(struct netdev_dpdk *dev) | |
349 | { | |
350 | struct rte_eth_link link; | |
351 | ||
352 | rte_eth_link_get_nowait(dev->port_id, &link); | |
353 | ||
354 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 355 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
356 | |
357 | dev->link_reset_cnt++; | |
358 | dev->link = link; | |
359 | if (dev->link.link_status) { | |
360 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
361 | dev->port_id, (unsigned)dev->link.link_speed, | |
362 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
363 | ("full-duplex") : ("half-duplex")); | |
364 | } else { | |
365 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
366 | } | |
367 | } | |
368 | } | |
369 | ||
370 | static void * | |
371 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
372 | { | |
373 | struct netdev_dpdk *dev; | |
374 | ||
375 | pthread_detach(pthread_self()); | |
376 | ||
377 | for (;;) { | |
378 | ovs_mutex_lock(&dpdk_mutex); | |
379 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
380 | ovs_mutex_lock(&dev->mutex); | |
381 | check_link_status(dev); | |
382 | ovs_mutex_unlock(&dev->mutex); | |
383 | } | |
384 | ovs_mutex_unlock(&dpdk_mutex); | |
385 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
386 | } | |
387 | ||
388 | return NULL; | |
389 | } | |
390 | ||
391 | static int | |
392 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
393 | { | |
394 | struct rte_pktmbuf_pool_private *mbp_priv; | |
395 | struct ether_addr eth_addr; | |
396 | int diag; | |
397 | int i; | |
398 | ||
399 | if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) { | |
95fb793a | 400 | return ENODEV; |
8a9562d2 PS |
401 | } |
402 | ||
5496878c AW |
403 | diag = rte_eth_dev_configure(dev->port_id, dev->up.n_rxq, dev->up.n_txq, |
404 | &port_conf); | |
8a9562d2 PS |
405 | if (diag) { |
406 | VLOG_ERR("eth dev config error %d",diag); | |
95fb793a | 407 | return -diag; |
8a9562d2 PS |
408 | } |
409 | ||
5496878c | 410 | for (i = 0; i < dev->up.n_txq; i++) { |
79f5354c | 411 | diag = rte_eth_tx_queue_setup(dev->port_id, i, NIC_PORT_TX_Q_SIZE, |
d221ffa1 | 412 | dev->socket_id, &tx_conf); |
8a9562d2 PS |
413 | if (diag) { |
414 | VLOG_ERR("eth dev tx queue setup error %d",diag); | |
95fb793a | 415 | return -diag; |
8a9562d2 PS |
416 | } |
417 | } | |
418 | ||
5496878c | 419 | for (i = 0; i < dev->up.n_rxq; i++) { |
79f5354c | 420 | diag = rte_eth_rx_queue_setup(dev->port_id, i, NIC_PORT_RX_Q_SIZE, |
d221ffa1 | 421 | dev->socket_id, |
a715f600 | 422 | &rx_conf, dev->dpdk_mp->mp); |
8a9562d2 PS |
423 | if (diag) { |
424 | VLOG_ERR("eth dev rx queue setup error %d",diag); | |
95fb793a | 425 | return -diag; |
8a9562d2 PS |
426 | } |
427 | } | |
428 | ||
429 | diag = rte_eth_dev_start(dev->port_id); | |
430 | if (diag) { | |
431 | VLOG_ERR("eth dev start error %d",diag); | |
95fb793a | 432 | return -diag; |
8a9562d2 PS |
433 | } |
434 | ||
435 | rte_eth_promiscuous_enable(dev->port_id); | |
436 | rte_eth_allmulticast_enable(dev->port_id); | |
437 | ||
438 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
439 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
440 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
441 | dev->port_id, ETH_ADDR_ARGS(eth_addr.addr_bytes)); | |
442 | ||
443 | memcpy(dev->hwaddr, eth_addr.addr_bytes, ETH_ADDR_LEN); | |
444 | rte_eth_link_get_nowait(dev->port_id, &dev->link); | |
445 | ||
446 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
447 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
448 | ||
449 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
450 | return 0; | |
451 | } | |
452 | ||
453 | static struct netdev_dpdk * | |
454 | netdev_dpdk_cast(const struct netdev *netdev) | |
455 | { | |
456 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
457 | } | |
458 | ||
459 | static struct netdev * | |
460 | netdev_dpdk_alloc(void) | |
461 | { | |
462 | struct netdev_dpdk *netdev = dpdk_rte_mzalloc(sizeof *netdev); | |
463 | return &netdev->up; | |
464 | } | |
465 | ||
5a034064 | 466 | static void |
91968eb0 | 467 | netdev_dpdk_alloc_txq(struct netdev_dpdk *netdev, unsigned int n_txqs) |
5a034064 AW |
468 | { |
469 | int i; | |
470 | ||
471 | netdev->tx_q = dpdk_rte_mzalloc(n_txqs * sizeof *netdev->tx_q); | |
94143fc4 AW |
472 | /* Each index is considered as a cpu core id, since there should |
473 | * be one tx queue for each cpu core. */ | |
5a034064 | 474 | for (i = 0; i < n_txqs; i++) { |
ba0358a1 | 475 | int numa_id = ovs_numa_get_numa_id(i); |
94143fc4 | 476 | |
94143fc4 AW |
477 | /* If the corresponding core is not on the same numa node |
478 | * as 'netdev', flags the 'flush_tx'. */ | |
ba0358a1 | 479 | netdev->tx_q[i].flush_tx = netdev->socket_id == numa_id; |
5a034064 AW |
480 | } |
481 | } | |
482 | ||
8a9562d2 | 483 | static int |
5a034064 AW |
484 | netdev_dpdk_init(struct netdev *netdev_, unsigned int port_no) |
485 | OVS_REQUIRES(dpdk_mutex) | |
8a9562d2 PS |
486 | { |
487 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1b7a04e0 | 488 | int sid; |
95fb793a | 489 | int err = 0; |
8a9562d2 | 490 | |
95fb793a | 491 | ovs_mutex_init(&netdev->mutex); |
8a9562d2 | 492 | |
95fb793a | 493 | ovs_mutex_lock(&netdev->mutex); |
8a9562d2 | 494 | |
1b7a04e0 AW |
495 | /* If the 'sid' is negative, it means that the kernel fails |
496 | * to obtain the pci numa info. In that situation, always | |
497 | * use 'SOCKET0'. */ | |
498 | sid = rte_eth_dev_socket_id(port_no); | |
499 | netdev->socket_id = sid < 0 ? SOCKET0 : sid; | |
91968eb0 | 500 | netdev_dpdk_alloc_txq(netdev, NR_QUEUE); |
95fb793a | 501 | netdev->port_id = port_no; |
8a9562d2 | 502 | netdev->flags = 0; |
8a9562d2 PS |
503 | netdev->mtu = ETHER_MTU; |
504 | netdev->max_packet_len = MTU_TO_MAX_LEN(netdev->mtu); | |
7251515e | 505 | rte_spinlock_init(&netdev->dpdkr_tx_lock); |
8a9562d2 | 506 | |
8a9562d2 PS |
507 | netdev->dpdk_mp = dpdk_mp_get(netdev->socket_id, netdev->mtu); |
508 | if (!netdev->dpdk_mp) { | |
509 | err = ENOMEM; | |
95fb793a | 510 | goto unlock; |
8a9562d2 PS |
511 | } |
512 | ||
5496878c AW |
513 | netdev_->n_txq = NR_QUEUE; |
514 | netdev_->n_rxq = NR_QUEUE; | |
8a9562d2 PS |
515 | err = dpdk_eth_dev_init(netdev); |
516 | if (err) { | |
95fb793a | 517 | goto unlock; |
8a9562d2 | 518 | } |
8a9562d2 PS |
519 | |
520 | list_push_back(&dpdk_list, &netdev->list_node); | |
521 | ||
95fb793a | 522 | unlock: |
5a034064 AW |
523 | if (err) { |
524 | rte_free(netdev->tx_q); | |
525 | } | |
8a9562d2 | 526 | ovs_mutex_unlock(&netdev->mutex); |
95fb793a | 527 | return err; |
528 | } | |
529 | ||
530 | static int | |
531 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
532 | unsigned int *port_no) | |
533 | { | |
534 | const char *cport; | |
535 | ||
536 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
537 | return ENODEV; | |
538 | } | |
539 | ||
540 | cport = dev_name + strlen(prefix); | |
541 | *port_no = strtol(cport, 0, 0); /* string must be null terminated */ | |
542 | return 0; | |
543 | } | |
544 | ||
545 | static int | |
546 | netdev_dpdk_construct(struct netdev *netdev) | |
547 | { | |
548 | unsigned int port_no; | |
549 | int err; | |
550 | ||
551 | if (rte_eal_init_ret) { | |
552 | return rte_eal_init_ret; | |
553 | } | |
554 | ||
555 | /* Names always start with "dpdk" */ | |
556 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
557 | if (err) { | |
558 | return err; | |
559 | } | |
560 | ||
561 | ovs_mutex_lock(&dpdk_mutex); | |
562 | err = netdev_dpdk_init(netdev, port_no); | |
8a9562d2 PS |
563 | ovs_mutex_unlock(&dpdk_mutex); |
564 | return err; | |
565 | } | |
566 | ||
567 | static void | |
568 | netdev_dpdk_destruct(struct netdev *netdev_) | |
569 | { | |
570 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
571 | ||
572 | ovs_mutex_lock(&dev->mutex); | |
573 | rte_eth_dev_stop(dev->port_id); | |
574 | ovs_mutex_unlock(&dev->mutex); | |
575 | ||
576 | ovs_mutex_lock(&dpdk_mutex); | |
5a034064 | 577 | rte_free(dev->tx_q); |
8a9562d2 PS |
578 | list_remove(&dev->list_node); |
579 | dpdk_mp_put(dev->dpdk_mp); | |
580 | ovs_mutex_unlock(&dpdk_mutex); | |
581 | ||
582 | ovs_mutex_destroy(&dev->mutex); | |
583 | } | |
584 | ||
585 | static void | |
586 | netdev_dpdk_dealloc(struct netdev *netdev_) | |
587 | { | |
588 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
589 | ||
590 | rte_free(netdev); | |
591 | } | |
592 | ||
593 | static int | |
594 | netdev_dpdk_get_config(const struct netdev *netdev_, struct smap *args) | |
595 | { | |
596 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
597 | ||
598 | ovs_mutex_lock(&dev->mutex); | |
599 | ||
47659062 AW |
600 | smap_add_format(args, "configured_rx_queues", "%d", netdev_->n_rxq); |
601 | smap_add_format(args, "configured_tx_queues", "%d", netdev_->n_txq); | |
8a9562d2 PS |
602 | ovs_mutex_unlock(&dev->mutex); |
603 | ||
604 | return 0; | |
605 | } | |
606 | ||
7dec44fe AW |
607 | static int |
608 | netdev_dpdk_get_numa_id(const struct netdev *netdev_) | |
609 | { | |
610 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
611 | ||
612 | return netdev->socket_id; | |
613 | } | |
614 | ||
5496878c AW |
615 | /* Sets the number of tx queues and rx queues for the dpdk interface. |
616 | * If the configuration fails, do not try restoring its old configuration | |
617 | * and just returns the error. */ | |
618 | static int | |
619 | netdev_dpdk_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
620 | unsigned int n_rxq) | |
621 | { | |
622 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
623 | int err = 0; | |
624 | ||
625 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
626 | return err; | |
627 | } | |
628 | ||
b7ccaf67 | 629 | ovs_mutex_lock(&dpdk_mutex); |
5496878c | 630 | ovs_mutex_lock(&netdev->mutex); |
91968eb0 | 631 | |
5496878c | 632 | rte_eth_dev_stop(netdev->port_id); |
91968eb0 | 633 | |
5496878c AW |
634 | netdev->up.n_txq = n_txq; |
635 | netdev->up.n_rxq = n_rxq; | |
91968eb0 AW |
636 | rte_free(netdev->tx_q); |
637 | netdev_dpdk_alloc_txq(netdev, n_txq); | |
5496878c | 638 | err = dpdk_eth_dev_init(netdev); |
91968eb0 | 639 | |
5496878c | 640 | ovs_mutex_unlock(&netdev->mutex); |
b7ccaf67 | 641 | ovs_mutex_unlock(&dpdk_mutex); |
5496878c AW |
642 | |
643 | return err; | |
644 | } | |
645 | ||
8a9562d2 PS |
646 | static struct netdev_rxq * |
647 | netdev_dpdk_rxq_alloc(void) | |
648 | { | |
649 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
650 | ||
651 | return &rx->up; | |
652 | } | |
653 | ||
654 | static struct netdev_rxq_dpdk * | |
655 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rx) | |
656 | { | |
657 | return CONTAINER_OF(rx, struct netdev_rxq_dpdk, up); | |
658 | } | |
659 | ||
660 | static int | |
661 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq_) | |
662 | { | |
663 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
664 | struct netdev_dpdk *netdev = netdev_dpdk_cast(rx->up.netdev); | |
665 | ||
666 | ovs_mutex_lock(&netdev->mutex); | |
667 | rx->port_id = netdev->port_id; | |
668 | ovs_mutex_unlock(&netdev->mutex); | |
669 | ||
670 | return 0; | |
671 | } | |
672 | ||
673 | static void | |
674 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq_ OVS_UNUSED) | |
675 | { | |
676 | } | |
677 | ||
678 | static void | |
679 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq_) | |
680 | { | |
681 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
682 | ||
683 | rte_free(rx); | |
684 | } | |
685 | ||
b170db2a RW |
686 | static inline void |
687 | dpdk_queue_flush__(struct netdev_dpdk *dev, int qid) | |
8a9562d2 PS |
688 | { |
689 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1304f1f8 DDP |
690 | uint32_t nb_tx = 0; |
691 | ||
692 | while (nb_tx != txq->count) { | |
693 | uint32_t ret; | |
694 | ||
695 | ret = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts + nb_tx, | |
696 | txq->count - nb_tx); | |
697 | if (!ret) { | |
698 | break; | |
699 | } | |
700 | ||
701 | nb_tx += ret; | |
702 | } | |
8a9562d2 | 703 | |
b170db2a | 704 | if (OVS_UNLIKELY(nb_tx != txq->count)) { |
db73f716 DDP |
705 | /* free buffers, which we couldn't transmit, one at a time (each |
706 | * packet could come from a different mempool) */ | |
707 | int i; | |
708 | ||
709 | for (i = nb_tx; i < txq->count; i++) { | |
710 | rte_pktmbuf_free_seg(txq->burst_pkts[i]); | |
711 | } | |
1304f1f8 DDP |
712 | ovs_mutex_lock(&dev->mutex); |
713 | dev->stats.tx_dropped += txq->count-nb_tx; | |
714 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 | 715 | } |
1304f1f8 | 716 | |
8a9562d2 | 717 | txq->count = 0; |
844f2d74 | 718 | txq->tsc = rte_get_timer_cycles(); |
b170db2a RW |
719 | } |
720 | ||
721 | static inline void | |
722 | dpdk_queue_flush(struct netdev_dpdk *dev, int qid) | |
723 | { | |
724 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
725 | ||
726 | if (txq->count == 0) { | |
727 | return; | |
728 | } | |
b170db2a | 729 | dpdk_queue_flush__(dev, qid); |
8a9562d2 PS |
730 | } |
731 | ||
8a9562d2 | 732 | static int |
e14deea0 | 733 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet **packets, |
91088554 | 734 | int *c) |
8a9562d2 PS |
735 | { |
736 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
737 | struct netdev *netdev = rx->up.netdev; | |
738 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 739 | int nb_rx; |
8a9562d2 | 740 | |
5496878c AW |
741 | /* There is only one tx queue for this core. Do not flush other |
742 | * queueus. */ | |
743 | if (rxq_->queue_id == rte_lcore_id()) { | |
744 | dpdk_queue_flush(dev, rxq_->queue_id); | |
745 | } | |
8a9562d2 PS |
746 | |
747 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq_->queue_id, | |
7d08d53e DDP |
748 | (struct rte_mbuf **) packets, |
749 | MIN((int)NETDEV_MAX_RX_BATCH, | |
750 | (int)MAX_RX_QUEUE_LEN)); | |
8a9562d2 PS |
751 | if (!nb_rx) { |
752 | return EAGAIN; | |
753 | } | |
754 | ||
8a9562d2 PS |
755 | *c = nb_rx; |
756 | ||
757 | return 0; | |
758 | } | |
759 | ||
760 | inline static void | |
f4fd623c DDP |
761 | dpdk_queue_pkts(struct netdev_dpdk *dev, int qid, |
762 | struct rte_mbuf **pkts, int cnt) | |
8a9562d2 PS |
763 | { |
764 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
765 | uint64_t diff_tsc; | |
8a9562d2 | 766 | |
f4fd623c DDP |
767 | int i = 0; |
768 | ||
f4fd623c DDP |
769 | while (i < cnt) { |
770 | int freeslots = MAX_TX_QUEUE_LEN - txq->count; | |
771 | int tocopy = MIN(freeslots, cnt-i); | |
8a9562d2 | 772 | |
f4fd623c DDP |
773 | memcpy(&txq->burst_pkts[txq->count], &pkts[i], |
774 | tocopy * sizeof (struct rte_mbuf *)); | |
775 | ||
776 | txq->count += tocopy; | |
777 | i += tocopy; | |
778 | ||
94143fc4 | 779 | if (txq->count == MAX_TX_QUEUE_LEN || txq->flush_tx) { |
b170db2a | 780 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 781 | } |
844f2d74 | 782 | diff_tsc = rte_get_timer_cycles() - txq->tsc; |
f4fd623c | 783 | if (diff_tsc >= DRAIN_TSC) { |
b170db2a | 784 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 785 | } |
8a9562d2 | 786 | } |
8a9562d2 PS |
787 | } |
788 | ||
789 | /* Tx function. Transmit packets indefinitely */ | |
790 | static void | |
e14deea0 | 791 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet ** pkts, |
2654cc33 | 792 | int cnt) |
db73f716 | 793 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 PS |
794 | { |
795 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
f4fd623c | 796 | struct rte_mbuf *mbufs[cnt]; |
175cf4de RW |
797 | int dropped = 0; |
798 | int newcnt = 0; | |
799 | int i; | |
8a9562d2 | 800 | |
db73f716 DDP |
801 | /* If we are on a non pmd thread we have to use the mempool mutex, because |
802 | * every non pmd thread shares the same mempool cache */ | |
803 | ||
804 | if (!thread_is_pmd()) { | |
805 | ovs_mutex_lock(&nonpmd_mempool_mutex); | |
806 | } | |
807 | ||
f4fd623c | 808 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 809 | int size = dp_packet_size(pkts[i]); |
95fb793a | 810 | |
f98d7864 | 811 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c DDP |
812 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
813 | (int)size , dev->max_packet_len); | |
814 | ||
175cf4de | 815 | dropped++; |
f4fd623c DDP |
816 | continue; |
817 | } | |
8a9562d2 | 818 | |
f4fd623c | 819 | mbufs[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 820 | |
f4fd623c | 821 | if (!mbufs[newcnt]) { |
175cf4de RW |
822 | dropped += cnt - i; |
823 | break; | |
f4fd623c DDP |
824 | } |
825 | ||
826 | /* We have to do a copy for now */ | |
cf62fa4c | 827 | memcpy(mbufs[newcnt]->pkt.data, dp_packet_data(pkts[i]), size); |
f4fd623c DDP |
828 | |
829 | rte_pktmbuf_data_len(mbufs[newcnt]) = size; | |
830 | rte_pktmbuf_pkt_len(mbufs[newcnt]) = size; | |
831 | ||
832 | newcnt++; | |
833 | } | |
8a9562d2 | 834 | |
f98d7864 | 835 | if (OVS_UNLIKELY(dropped)) { |
175cf4de RW |
836 | ovs_mutex_lock(&dev->mutex); |
837 | dev->stats.tx_dropped += dropped; | |
838 | ovs_mutex_unlock(&dev->mutex); | |
839 | } | |
840 | ||
2654cc33 AW |
841 | dpdk_queue_pkts(dev, qid, mbufs, newcnt); |
842 | dpdk_queue_flush(dev, qid); | |
db73f716 DDP |
843 | |
844 | if (!thread_is_pmd()) { | |
845 | ovs_mutex_unlock(&nonpmd_mempool_mutex); | |
846 | } | |
8a9562d2 PS |
847 | } |
848 | ||
7251515e DV |
849 | static inline void |
850 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
e14deea0 | 851 | struct dp_packet **pkts, int cnt, bool may_steal) |
8a9562d2 | 852 | { |
f4fd623c | 853 | int i; |
8a9562d2 | 854 | |
7251515e | 855 | if (OVS_UNLIKELY(!may_steal || |
cf62fa4c | 856 | pkts[0]->source != DPBUF_DPDK)) { |
7251515e DV |
857 | struct netdev *netdev = &dev->up; |
858 | ||
2654cc33 | 859 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); |
b3cd9f9d PS |
860 | |
861 | if (may_steal) { | |
f4fd623c | 862 | for (i = 0; i < cnt; i++) { |
e14deea0 | 863 | dp_packet_delete(pkts[i]); |
f4fd623c | 864 | } |
b3cd9f9d | 865 | } |
8a9562d2 | 866 | } else { |
f4fd623c DDP |
867 | int next_tx_idx = 0; |
868 | int dropped = 0; | |
8a9562d2 | 869 | |
f4fd623c | 870 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 871 | int size = dp_packet_size(pkts[i]); |
f4fd623c DDP |
872 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
873 | if (next_tx_idx != i) { | |
874 | dpdk_queue_pkts(dev, qid, | |
875 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
876 | i-next_tx_idx); | |
1ebfe1ac | 877 | } |
f4fd623c | 878 | |
1ebfe1ac DDP |
879 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
880 | (int)size , dev->max_packet_len); | |
f4fd623c | 881 | |
e14deea0 | 882 | dp_packet_delete(pkts[i]); |
1ebfe1ac | 883 | dropped++; |
f4fd623c DDP |
884 | next_tx_idx = i + 1; |
885 | } | |
886 | } | |
887 | if (next_tx_idx != cnt) { | |
888 | dpdk_queue_pkts(dev, qid, | |
889 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
890 | cnt-next_tx_idx); | |
891 | } | |
8a9562d2 | 892 | |
f4fd623c DDP |
893 | if (OVS_UNLIKELY(dropped)) { |
894 | ovs_mutex_lock(&dev->mutex); | |
895 | dev->stats.tx_dropped += dropped; | |
896 | ovs_mutex_unlock(&dev->mutex); | |
897 | } | |
8a9562d2 | 898 | } |
7251515e DV |
899 | } |
900 | ||
901 | static int | |
902 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
e14deea0 | 903 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
904 | { |
905 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 906 | |
7251515e DV |
907 | netdev_dpdk_send__(dev, qid, pkts, cnt, may_steal); |
908 | return 0; | |
8a9562d2 PS |
909 | } |
910 | ||
911 | static int | |
912 | netdev_dpdk_set_etheraddr(struct netdev *netdev, | |
913 | const uint8_t mac[ETH_ADDR_LEN]) | |
914 | { | |
915 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
916 | ||
917 | ovs_mutex_lock(&dev->mutex); | |
918 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
919 | memcpy(dev->hwaddr, mac, ETH_ADDR_LEN); | |
045c0d1a | 920 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
921 | } |
922 | ovs_mutex_unlock(&dev->mutex); | |
923 | ||
924 | return 0; | |
925 | } | |
926 | ||
927 | static int | |
928 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, | |
929 | uint8_t mac[ETH_ADDR_LEN]) | |
930 | { | |
931 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
932 | ||
933 | ovs_mutex_lock(&dev->mutex); | |
934 | memcpy(mac, dev->hwaddr, ETH_ADDR_LEN); | |
935 | ovs_mutex_unlock(&dev->mutex); | |
936 | ||
937 | return 0; | |
938 | } | |
939 | ||
940 | static int | |
941 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
942 | { | |
943 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
944 | ||
945 | ovs_mutex_lock(&dev->mutex); | |
946 | *mtup = dev->mtu; | |
947 | ovs_mutex_unlock(&dev->mutex); | |
948 | ||
949 | return 0; | |
950 | } | |
951 | ||
952 | static int | |
953 | netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu) | |
954 | { | |
955 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
956 | int old_mtu, err; | |
957 | struct dpdk_mp *old_mp; | |
958 | struct dpdk_mp *mp; | |
959 | ||
960 | ovs_mutex_lock(&dpdk_mutex); | |
961 | ovs_mutex_lock(&dev->mutex); | |
962 | if (dev->mtu == mtu) { | |
963 | err = 0; | |
964 | goto out; | |
965 | } | |
966 | ||
967 | mp = dpdk_mp_get(dev->socket_id, dev->mtu); | |
968 | if (!mp) { | |
969 | err = ENOMEM; | |
970 | goto out; | |
971 | } | |
972 | ||
973 | rte_eth_dev_stop(dev->port_id); | |
974 | ||
975 | old_mtu = dev->mtu; | |
976 | old_mp = dev->dpdk_mp; | |
977 | dev->dpdk_mp = mp; | |
978 | dev->mtu = mtu; | |
979 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
980 | ||
981 | err = dpdk_eth_dev_init(dev); | |
982 | if (err) { | |
8a9562d2 PS |
983 | dpdk_mp_put(mp); |
984 | dev->mtu = old_mtu; | |
985 | dev->dpdk_mp = old_mp; | |
986 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
987 | dpdk_eth_dev_init(dev); | |
988 | goto out; | |
989 | } | |
990 | ||
991 | dpdk_mp_put(old_mp); | |
045c0d1a | 992 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
993 | out: |
994 | ovs_mutex_unlock(&dev->mutex); | |
995 | ovs_mutex_unlock(&dpdk_mutex); | |
996 | return err; | |
997 | } | |
998 | ||
999 | static int | |
1000 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier); | |
1001 | ||
1002 | static int | |
1003 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1004 | { | |
1005 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1006 | struct rte_eth_stats rte_stats; | |
1007 | bool gg; | |
1008 | ||
1009 | netdev_dpdk_get_carrier(netdev, &gg); | |
1010 | ovs_mutex_lock(&dev->mutex); | |
1011 | rte_eth_stats_get(dev->port_id, &rte_stats); | |
1012 | ||
2f9dd77f | 1013 | memset(stats, 0, sizeof(*stats)); |
8a9562d2 | 1014 | |
2f9dd77f PS |
1015 | stats->rx_packets = rte_stats.ipackets; |
1016 | stats->tx_packets = rte_stats.opackets; | |
1017 | stats->rx_bytes = rte_stats.ibytes; | |
1018 | stats->tx_bytes = rte_stats.obytes; | |
1019 | stats->rx_errors = rte_stats.ierrors; | |
1020 | stats->tx_errors = rte_stats.oerrors; | |
1021 | stats->multicast = rte_stats.imcasts; | |
8a9562d2 | 1022 | |
2f9dd77f | 1023 | stats->tx_dropped = dev->stats.tx_dropped; |
8a9562d2 PS |
1024 | ovs_mutex_unlock(&dev->mutex); |
1025 | ||
1026 | return 0; | |
1027 | } | |
1028 | ||
1029 | static int | |
1030 | netdev_dpdk_get_features(const struct netdev *netdev_, | |
1031 | enum netdev_features *current, | |
1032 | enum netdev_features *advertised OVS_UNUSED, | |
1033 | enum netdev_features *supported OVS_UNUSED, | |
1034 | enum netdev_features *peer OVS_UNUSED) | |
1035 | { | |
1036 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1037 | struct rte_eth_link link; | |
1038 | ||
1039 | ovs_mutex_lock(&dev->mutex); | |
1040 | link = dev->link; | |
1041 | ovs_mutex_unlock(&dev->mutex); | |
1042 | ||
1043 | if (link.link_duplex == ETH_LINK_AUTONEG_DUPLEX) { | |
1044 | if (link.link_speed == ETH_LINK_SPEED_AUTONEG) { | |
1045 | *current = NETDEV_F_AUTONEG; | |
1046 | } | |
1047 | } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { | |
1048 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1049 | *current = NETDEV_F_10MB_HD; | |
1050 | } | |
1051 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1052 | *current = NETDEV_F_100MB_HD; | |
1053 | } | |
1054 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1055 | *current = NETDEV_F_1GB_HD; | |
1056 | } | |
1057 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
1058 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1059 | *current = NETDEV_F_10MB_FD; | |
1060 | } | |
1061 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1062 | *current = NETDEV_F_100MB_FD; | |
1063 | } | |
1064 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1065 | *current = NETDEV_F_1GB_FD; | |
1066 | } | |
1067 | if (link.link_speed == ETH_LINK_SPEED_10000) { | |
1068 | *current = NETDEV_F_10GB_FD; | |
1069 | } | |
1070 | } | |
1071 | ||
1072 | return 0; | |
1073 | } | |
1074 | ||
1075 | static int | |
1076 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
1077 | { | |
1078 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1079 | int ifindex; | |
1080 | ||
1081 | ovs_mutex_lock(&dev->mutex); | |
1082 | ifindex = dev->port_id; | |
1083 | ovs_mutex_unlock(&dev->mutex); | |
1084 | ||
1085 | return ifindex; | |
1086 | } | |
1087 | ||
1088 | static int | |
1089 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1090 | { | |
1091 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1092 | ||
1093 | ovs_mutex_lock(&dev->mutex); | |
1094 | check_link_status(dev); | |
1095 | *carrier = dev->link.link_status; | |
1096 | ovs_mutex_unlock(&dev->mutex); | |
1097 | ||
1098 | return 0; | |
1099 | } | |
1100 | ||
1101 | static long long int | |
1102 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev_) | |
1103 | { | |
1104 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1105 | long long int carrier_resets; | |
1106 | ||
1107 | ovs_mutex_lock(&dev->mutex); | |
1108 | carrier_resets = dev->link_reset_cnt; | |
1109 | ovs_mutex_unlock(&dev->mutex); | |
1110 | ||
1111 | return carrier_resets; | |
1112 | } | |
1113 | ||
1114 | static int | |
1115 | netdev_dpdk_set_miimon(struct netdev *netdev_ OVS_UNUSED, | |
1116 | long long int interval OVS_UNUSED) | |
1117 | { | |
ee32150e | 1118 | return EOPNOTSUPP; |
8a9562d2 PS |
1119 | } |
1120 | ||
1121 | static int | |
1122 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
1123 | enum netdev_flags off, enum netdev_flags on, | |
95fb793a | 1124 | enum netdev_flags *old_flagsp) OVS_REQUIRES(dev->mutex) |
8a9562d2 PS |
1125 | { |
1126 | int err; | |
1127 | ||
1128 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
1129 | return EINVAL; | |
1130 | } | |
1131 | ||
1132 | *old_flagsp = dev->flags; | |
1133 | dev->flags |= on; | |
1134 | dev->flags &= ~off; | |
1135 | ||
1136 | if (dev->flags == *old_flagsp) { | |
1137 | return 0; | |
1138 | } | |
1139 | ||
1140 | if (dev->flags & NETDEV_UP) { | |
1141 | err = rte_eth_dev_start(dev->port_id); | |
1142 | if (err) | |
95fb793a | 1143 | return -err; |
8a9562d2 PS |
1144 | } |
1145 | ||
1146 | if (dev->flags & NETDEV_PROMISC) { | |
1147 | rte_eth_promiscuous_enable(dev->port_id); | |
1148 | } | |
1149 | ||
1150 | if (!(dev->flags & NETDEV_UP)) { | |
1151 | rte_eth_dev_stop(dev->port_id); | |
1152 | } | |
1153 | ||
1154 | return 0; | |
1155 | } | |
1156 | ||
1157 | static int | |
1158 | netdev_dpdk_update_flags(struct netdev *netdev_, | |
1159 | enum netdev_flags off, enum netdev_flags on, | |
1160 | enum netdev_flags *old_flagsp) | |
1161 | { | |
1162 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1163 | int error; | |
1164 | ||
1165 | ovs_mutex_lock(&netdev->mutex); | |
1166 | error = netdev_dpdk_update_flags__(netdev, off, on, old_flagsp); | |
1167 | ovs_mutex_unlock(&netdev->mutex); | |
1168 | ||
1169 | return error; | |
1170 | } | |
1171 | ||
1172 | static int | |
1173 | netdev_dpdk_get_status(const struct netdev *netdev_, struct smap *args) | |
1174 | { | |
1175 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1176 | struct rte_eth_dev_info dev_info; | |
1177 | ||
e0a801c7 | 1178 | if (dev->port_id < 0) |
8a9562d2 PS |
1179 | return ENODEV; |
1180 | ||
1181 | ovs_mutex_lock(&dev->mutex); | |
1182 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
1183 | ovs_mutex_unlock(&dev->mutex); | |
1184 | ||
1185 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1186 | ||
95fb793a | 1187 | smap_add_format(args, "port_no", "%d", dev->port_id); |
8a9562d2 PS |
1188 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); |
1189 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1190 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
1191 | smap_add_format(args, "max_rx_pktlen", "%u", dev_info.max_rx_pktlen); | |
1192 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); | |
1193 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
1194 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
1195 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
1196 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
1197 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
1198 | ||
1199 | smap_add_format(args, "pci-vendor_id", "0x%u", dev_info.pci_dev->id.vendor_id); | |
1200 | smap_add_format(args, "pci-device_id", "0x%x", dev_info.pci_dev->id.device_id); | |
1201 | ||
1202 | return 0; | |
1203 | } | |
1204 | ||
1205 | static void | |
1206 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
1207 | OVS_REQUIRES(dev->mutex) | |
1208 | { | |
1209 | enum netdev_flags old_flags; | |
1210 | ||
1211 | if (admin_state) { | |
1212 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
1213 | } else { | |
1214 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
1215 | } | |
1216 | } | |
1217 | ||
1218 | static void | |
1219 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
1220 | const char *argv[], void *aux OVS_UNUSED) | |
1221 | { | |
1222 | bool up; | |
1223 | ||
1224 | if (!strcasecmp(argv[argc - 1], "up")) { | |
1225 | up = true; | |
1226 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
1227 | up = false; | |
1228 | } else { | |
1229 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
1230 | return; | |
1231 | } | |
1232 | ||
1233 | if (argc > 2) { | |
1234 | struct netdev *netdev = netdev_from_name(argv[1]); | |
1235 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
1236 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
1237 | ||
1238 | ovs_mutex_lock(&dpdk_dev->mutex); | |
1239 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
1240 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
1241 | ||
1242 | netdev_close(netdev); | |
1243 | } else { | |
1244 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
1245 | netdev_close(netdev); | |
1246 | return; | |
1247 | } | |
1248 | } else { | |
1249 | struct netdev_dpdk *netdev; | |
1250 | ||
1251 | ovs_mutex_lock(&dpdk_mutex); | |
1252 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
1253 | ovs_mutex_lock(&netdev->mutex); | |
1254 | netdev_dpdk_set_admin_state__(netdev, up); | |
1255 | ovs_mutex_unlock(&netdev->mutex); | |
1256 | } | |
1257 | ovs_mutex_unlock(&dpdk_mutex); | |
1258 | } | |
1259 | unixctl_command_reply(conn, "OK"); | |
1260 | } | |
1261 | ||
033e9df2 DDP |
1262 | static void |
1263 | dpdk_common_init(void) | |
1264 | { | |
1265 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
1266 | "[netdev] up|down", 1, 2, | |
1267 | netdev_dpdk_set_admin_state, NULL); | |
1268 | ||
1269 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
1270 | } | |
1271 | ||
8a9562d2 PS |
1272 | static int |
1273 | dpdk_class_init(void) | |
1274 | { | |
1275 | int result; | |
1276 | ||
8a9562d2 PS |
1277 | result = rte_eal_pci_probe(); |
1278 | if (result) { | |
1279 | VLOG_ERR("Cannot probe PCI"); | |
95fb793a | 1280 | return -result; |
8a9562d2 PS |
1281 | } |
1282 | ||
8a9562d2 PS |
1283 | VLOG_INFO("Ethernet Device Count: %d", (int)rte_eth_dev_count()); |
1284 | ||
8a9562d2 PS |
1285 | return 0; |
1286 | } | |
1287 | ||
95fb793a | 1288 | /* Client Rings */ |
1289 | ||
95fb793a | 1290 | static int |
1291 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
1292 | unsigned int *eth_port_id) | |
1293 | { | |
1294 | struct dpdk_ring *ivshmem; | |
1295 | char ring_name[10]; | |
1296 | int err; | |
1297 | ||
1298 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
1299 | if (ivshmem == NULL) { | |
1300 | return ENOMEM; | |
1301 | } | |
1302 | ||
7251515e | 1303 | /* XXX: Add support for multiquque ring. */ |
95fb793a | 1304 | err = snprintf(ring_name, 10, "%s_tx", dev_name); |
1305 | if (err < 0) { | |
1306 | return -err; | |
1307 | } | |
1308 | ||
7251515e DV |
1309 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1310 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1311 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1312 | if (ivshmem->cring_tx == NULL) { |
1313 | rte_free(ivshmem); | |
1314 | return ENOMEM; | |
1315 | } | |
1316 | ||
1317 | err = snprintf(ring_name, 10, "%s_rx", dev_name); | |
1318 | if (err < 0) { | |
1319 | return -err; | |
1320 | } | |
1321 | ||
7251515e DV |
1322 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1323 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1324 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1325 | if (ivshmem->cring_rx == NULL) { |
1326 | rte_free(ivshmem); | |
1327 | return ENOMEM; | |
1328 | } | |
1329 | ||
d7310583 DDP |
1330 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
1331 | &ivshmem->cring_tx, 1, SOCKET0); | |
1332 | ||
95fb793a | 1333 | if (err < 0) { |
1334 | rte_free(ivshmem); | |
1335 | return ENODEV; | |
1336 | } | |
1337 | ||
1338 | ivshmem->user_port_id = port_no; | |
1339 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
1340 | list_push_back(&dpdk_ring_list, &ivshmem->list_node); | |
1341 | ||
1342 | *eth_port_id = ivshmem->eth_port_id; | |
1343 | return 0; | |
1344 | } | |
1345 | ||
1346 | static int | |
1347 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) OVS_REQUIRES(dpdk_mutex) | |
1348 | { | |
1349 | struct dpdk_ring *ivshmem; | |
1350 | unsigned int port_no; | |
1351 | int err = 0; | |
1352 | ||
1353 | /* Names always start with "dpdkr" */ | |
1354 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
1355 | if (err) { | |
1356 | return err; | |
1357 | } | |
1358 | ||
1359 | /* look through our list to find the device */ | |
1360 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { | |
1361 | if (ivshmem->user_port_id == port_no) { | |
1362 | VLOG_INFO("Found dpdk ring device %s:\n", dev_name); | |
1363 | *eth_port_id = ivshmem->eth_port_id; /* really all that is needed */ | |
1364 | return 0; | |
1365 | } | |
1366 | } | |
1367 | /* Need to create the device rings */ | |
1368 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
1369 | } | |
1370 | ||
7251515e DV |
1371 | static int |
1372 | netdev_dpdk_ring_send(struct netdev *netdev, int qid OVS_UNUSED, | |
e14deea0 | 1373 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1374 | { |
1375 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1376 | ||
1377 | /* DPDK Rings have a single TX queue, Therefore needs locking. */ | |
1378 | rte_spinlock_lock(&dev->dpdkr_tx_lock); | |
1379 | netdev_dpdk_send__(dev, 0, pkts, cnt, may_steal); | |
1380 | rte_spinlock_unlock(&dev->dpdkr_tx_lock); | |
1381 | return 0; | |
1382 | } | |
1383 | ||
95fb793a | 1384 | static int |
1385 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
1386 | { | |
1387 | unsigned int port_no = 0; | |
1388 | int err = 0; | |
1389 | ||
1390 | if (rte_eal_init_ret) { | |
1391 | return rte_eal_init_ret; | |
1392 | } | |
1393 | ||
1394 | ovs_mutex_lock(&dpdk_mutex); | |
1395 | ||
1396 | err = dpdk_ring_open(netdev->name, &port_no); | |
1397 | if (err) { | |
1398 | goto unlock_dpdk; | |
1399 | } | |
1400 | ||
1401 | err = netdev_dpdk_init(netdev, port_no); | |
1402 | ||
1403 | unlock_dpdk: | |
1404 | ovs_mutex_unlock(&dpdk_mutex); | |
1405 | return err; | |
1406 | } | |
1407 | ||
7251515e | 1408 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, MULTIQ, SEND) \ |
95fb793a | 1409 | { \ |
1410 | NAME, \ | |
1411 | INIT, /* init */ \ | |
1412 | NULL, /* netdev_dpdk_run */ \ | |
1413 | NULL, /* netdev_dpdk_wait */ \ | |
1414 | \ | |
1415 | netdev_dpdk_alloc, \ | |
1416 | CONSTRUCT, \ | |
1417 | netdev_dpdk_destruct, \ | |
1418 | netdev_dpdk_dealloc, \ | |
1419 | netdev_dpdk_get_config, \ | |
1420 | NULL, /* netdev_dpdk_set_config */ \ | |
1421 | NULL, /* get_tunnel_config */ \ | |
a36de779 PS |
1422 | NULL, /* build header */ \ |
1423 | NULL, /* push header */ \ | |
1424 | NULL, /* pop header */ \ | |
7dec44fe | 1425 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
5496878c | 1426 | MULTIQ, /* set_multiq */ \ |
95fb793a | 1427 | \ |
7251515e | 1428 | SEND, /* send */ \ |
95fb793a | 1429 | NULL, /* send_wait */ \ |
1430 | \ | |
1431 | netdev_dpdk_set_etheraddr, \ | |
1432 | netdev_dpdk_get_etheraddr, \ | |
1433 | netdev_dpdk_get_mtu, \ | |
1434 | netdev_dpdk_set_mtu, \ | |
1435 | netdev_dpdk_get_ifindex, \ | |
1436 | netdev_dpdk_get_carrier, \ | |
1437 | netdev_dpdk_get_carrier_resets, \ | |
1438 | netdev_dpdk_set_miimon, \ | |
1439 | netdev_dpdk_get_stats, \ | |
95fb793a | 1440 | netdev_dpdk_get_features, \ |
1441 | NULL, /* set_advertisements */ \ | |
1442 | \ | |
1443 | NULL, /* set_policing */ \ | |
1444 | NULL, /* get_qos_types */ \ | |
1445 | NULL, /* get_qos_capabilities */ \ | |
1446 | NULL, /* get_qos */ \ | |
1447 | NULL, /* set_qos */ \ | |
1448 | NULL, /* get_queue */ \ | |
1449 | NULL, /* set_queue */ \ | |
1450 | NULL, /* delete_queue */ \ | |
1451 | NULL, /* get_queue_stats */ \ | |
1452 | NULL, /* queue_dump_start */ \ | |
1453 | NULL, /* queue_dump_next */ \ | |
1454 | NULL, /* queue_dump_done */ \ | |
1455 | NULL, /* dump_queue_stats */ \ | |
1456 | \ | |
1457 | NULL, /* get_in4 */ \ | |
1458 | NULL, /* set_in4 */ \ | |
1459 | NULL, /* get_in6 */ \ | |
1460 | NULL, /* add_router */ \ | |
1461 | NULL, /* get_next_hop */ \ | |
1462 | netdev_dpdk_get_status, \ | |
1463 | NULL, /* arp_lookup */ \ | |
1464 | \ | |
1465 | netdev_dpdk_update_flags, \ | |
1466 | \ | |
1467 | netdev_dpdk_rxq_alloc, \ | |
1468 | netdev_dpdk_rxq_construct, \ | |
1469 | netdev_dpdk_rxq_destruct, \ | |
1470 | netdev_dpdk_rxq_dealloc, \ | |
1471 | netdev_dpdk_rxq_recv, \ | |
1472 | NULL, /* rx_wait */ \ | |
1473 | NULL, /* rxq_drain */ \ | |
1474 | } | |
8a9562d2 PS |
1475 | |
1476 | int | |
1477 | dpdk_init(int argc, char **argv) | |
1478 | { | |
1479 | int result; | |
1480 | ||
9441caf3 | 1481 | if (argc < 2 || strcmp(argv[1], "--dpdk")) |
8a9562d2 PS |
1482 | return 0; |
1483 | ||
9441caf3 DDP |
1484 | /* Make sure program name passed to rte_eal_init() is vswitchd. */ |
1485 | argv[1] = argv[0]; | |
1486 | ||
8a9562d2 PS |
1487 | argc--; |
1488 | argv++; | |
1489 | ||
1490 | /* Make sure things are initialized ... */ | |
1491 | result = rte_eal_init(argc, argv); | |
451450fa | 1492 | if (result < 0) { |
8a9562d2 | 1493 | ovs_abort(result, "Cannot init EAL\n"); |
451450fa | 1494 | } |
8a9562d2 | 1495 | |
d7310583 | 1496 | rte_memzone_dump(stdout); |
8a9562d2 PS |
1497 | rte_eal_init_ret = 0; |
1498 | ||
451450fa | 1499 | if (argc > result) { |
9441caf3 | 1500 | argv[result] = argv[0]; |
451450fa | 1501 | } |
9441caf3 | 1502 | |
db73f716 DDP |
1503 | /* We are called from the main thread here */ |
1504 | thread_set_nonpmd(); | |
1505 | ||
9441caf3 | 1506 | return result + 1; |
8a9562d2 PS |
1507 | } |
1508 | ||
95fb793a | 1509 | const struct netdev_class dpdk_class = |
1510 | NETDEV_DPDK_CLASS( | |
1511 | "dpdk", | |
1512 | dpdk_class_init, | |
5496878c | 1513 | netdev_dpdk_construct, |
7251515e DV |
1514 | netdev_dpdk_set_multiq, |
1515 | netdev_dpdk_eth_send); | |
95fb793a | 1516 | |
1517 | const struct netdev_class dpdk_ring_class = | |
1518 | NETDEV_DPDK_CLASS( | |
1519 | "dpdkr", | |
033e9df2 | 1520 | NULL, |
5496878c | 1521 | netdev_dpdk_ring_construct, |
7251515e DV |
1522 | NULL, |
1523 | netdev_dpdk_ring_send); | |
95fb793a | 1524 | |
8a9562d2 PS |
1525 | void |
1526 | netdev_dpdk_register(void) | |
1527 | { | |
95fb793a | 1528 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
1529 | ||
033e9df2 DDP |
1530 | if (rte_eal_init_ret) { |
1531 | return; | |
1532 | } | |
1533 | ||
95fb793a | 1534 | if (ovsthread_once_start(&once)) { |
033e9df2 | 1535 | dpdk_common_init(); |
95fb793a | 1536 | netdev_register_provider(&dpdk_class); |
1537 | netdev_register_provider(&dpdk_ring_class); | |
1538 | ovsthread_once_done(&once); | |
1539 | } | |
8a9562d2 | 1540 | } |
8617afff PS |
1541 | |
1542 | int | |
1543 | pmd_thread_setaffinity_cpu(int cpu) | |
1544 | { | |
1545 | cpu_set_t cpuset; | |
1546 | int err; | |
1547 | ||
1548 | CPU_ZERO(&cpuset); | |
1549 | CPU_SET(cpu, &cpuset); | |
1550 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset); | |
1551 | if (err) { | |
1552 | VLOG_ERR("Thread affinity error %d",err); | |
1553 | return err; | |
1554 | } | |
abb5943d AW |
1555 | /* NON_PMD_CORE_ID is reserved for use by non pmd threads. */ |
1556 | ovs_assert(cpu != NON_PMD_CORE_ID); | |
65f13b50 | 1557 | RTE_PER_LCORE(_lcore_id) = cpu; |
8617afff PS |
1558 | |
1559 | return 0; | |
1560 | } | |
db73f716 DDP |
1561 | |
1562 | void | |
1563 | thread_set_nonpmd(void) | |
1564 | { | |
abb5943d AW |
1565 | /* We have to use NON_PMD_CORE_ID to allow non-pmd threads to perform |
1566 | * certain DPDK operations, like rte_eth_dev_configure(). */ | |
1567 | RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID; | |
db73f716 DDP |
1568 | } |
1569 | ||
1570 | static bool | |
1571 | thread_is_pmd(void) | |
1572 | { | |
abb5943d | 1573 | return rte_lcore_id() != NON_PMD_CORE_ID; |
db73f716 | 1574 | } |