]>
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 | 49 | |
b8e57534 MK |
50 | #include "rte_config.h" |
51 | #include "rte_mbuf.h" | |
58397e6c | 52 | #include "rte_virtio_net.h" |
b8e57534 | 53 | |
8a9562d2 PS |
54 | VLOG_DEFINE_THIS_MODULE(dpdk); |
55 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
56 | ||
57 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
58 | ||
59 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
60 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
61 | ||
62 | /* | |
63 | * need to reserve tons of extra space in the mbufs so we can align the | |
64 | * DMA addresses to 4KB. | |
65 | */ | |
66 | ||
67 | #define MTU_TO_MAX_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) | |
68 | #define MBUF_SIZE(mtu) (MTU_TO_MAX_LEN(mtu) + (512) + \ | |
69 | sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM) | |
70 | ||
da79ce2b DDP |
71 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
72 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
73 | * enough hugepages) we keep halving the number until the allocation succeeds | |
74 | * or we reach MIN_NB_MBUF */ | |
75 | ||
76 | #define MAX_NB_MBUF (4096 * 64) | |
77 | #define MIN_NB_MBUF (4096 * 4) | |
78 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
79 | ||
80 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
81 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
82 | ||
83 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
84 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
85 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
86 | % MP_CACHE_SZ == 0); | |
87 | ||
8a9562d2 PS |
88 | #define SOCKET0 0 |
89 | ||
79f5354c PM |
90 | #define NIC_PORT_RX_Q_SIZE 2048 /* Size of Physical NIC RX Queue, Max (n+32<=4096)*/ |
91 | #define NIC_PORT_TX_Q_SIZE 2048 /* Size of Physical NIC TX Queue, Max (n+32<=4096)*/ | |
92 | ||
58397e6c | 93 | /* Character device cuse_dev_name. */ |
bce01e3a | 94 | static char *cuse_dev_name = NULL; |
58397e6c | 95 | |
95e9881f KT |
96 | /* |
97 | * Maximum amount of time in micro seconds to try and enqueue to vhost. | |
98 | */ | |
99 | #define VHOST_ENQ_RETRY_USECS 100 | |
100 | ||
8a9562d2 | 101 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
102 | .rxmode = { |
103 | .mq_mode = ETH_MQ_RX_RSS, | |
104 | .split_hdr_size = 0, | |
105 | .header_split = 0, /* Header Split disabled */ | |
106 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
107 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
108 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
109 | .hw_strip_crc = 0, | |
110 | }, | |
111 | .rx_adv_conf = { | |
112 | .rss_conf = { | |
113 | .rss_key = NULL, | |
543342a4 | 114 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 115 | }, |
a28ddd11 DDP |
116 | }, |
117 | .txmode = { | |
118 | .mq_mode = ETH_MQ_TX_NONE, | |
119 | }, | |
8a9562d2 PS |
120 | }; |
121 | ||
3a100265 | 122 | enum { MAX_TX_QUEUE_LEN = 384 }; |
58f7c37b DDP |
123 | enum { DPDK_RING_SIZE = 256 }; |
124 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
125 | enum { DRAIN_TSC = 200000ULL }; |
126 | ||
58397e6c KT |
127 | enum dpdk_dev_type { |
128 | DPDK_DEV_ETH = 0, | |
129 | DPDK_DEV_VHOST = 1 | |
130 | }; | |
131 | ||
8a9562d2 PS |
132 | static int rte_eal_init_ret = ENODEV; |
133 | ||
134 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; | |
135 | ||
136 | /* Contains all 'struct dpdk_dev's. */ | |
ca6ba700 | 137 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 138 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 139 | |
ca6ba700 | 140 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 141 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 142 | |
db73f716 DDP |
143 | /* This mutex must be used by non pmd threads when allocating or freeing |
144 | * mbufs through mempools. Since dpdk_queue_pkts() and dpdk_queue_flush() may | |
145 | * use mempools, a non pmd thread should hold this mutex while calling them */ | |
bce01e3a | 146 | static struct ovs_mutex nonpmd_mempool_mutex = OVS_MUTEX_INITIALIZER; |
db73f716 | 147 | |
8a9562d2 PS |
148 | struct dpdk_mp { |
149 | struct rte_mempool *mp; | |
150 | int mtu; | |
151 | int socket_id; | |
152 | int refcount; | |
ca6ba700 | 153 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
154 | }; |
155 | ||
5a034064 AW |
156 | /* There should be one 'struct dpdk_tx_queue' created for |
157 | * each cpu core. */ | |
8a9562d2 | 158 | struct dpdk_tx_queue { |
94143fc4 AW |
159 | bool flush_tx; /* Set to true to flush queue everytime */ |
160 | /* pkts are queued. */ | |
8a9562d2 PS |
161 | int count; |
162 | uint64_t tsc; | |
163 | struct rte_mbuf *burst_pkts[MAX_TX_QUEUE_LEN]; | |
164 | }; | |
165 | ||
95fb793a | 166 | /* dpdk has no way to remove dpdk ring ethernet devices |
167 | so we have to keep them around once they've been created | |
168 | */ | |
169 | ||
ca6ba700 | 170 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 171 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 172 | |
173 | struct dpdk_ring { | |
174 | /* For the client rings */ | |
175 | struct rte_ring *cring_tx; | |
176 | struct rte_ring *cring_rx; | |
177 | int user_port_id; /* User given port no, parsed from port name */ | |
178 | int eth_port_id; /* ethernet device port id */ | |
ca6ba700 | 179 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 180 | }; |
181 | ||
8a9562d2 PS |
182 | struct netdev_dpdk { |
183 | struct netdev up; | |
184 | int port_id; | |
185 | int max_packet_len; | |
58397e6c | 186 | enum dpdk_dev_type type; |
8a9562d2 | 187 | |
5a034064 | 188 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
189 | |
190 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
191 | ||
192 | struct dpdk_mp *dpdk_mp; | |
193 | int mtu; | |
194 | int socket_id; | |
195 | int buf_size; | |
8a9562d2 | 196 | struct netdev_stats stats; |
45d947c4 DDP |
197 | /* Protects stats */ |
198 | rte_spinlock_t stats_lock; | |
8a9562d2 PS |
199 | |
200 | uint8_t hwaddr[ETH_ADDR_LEN]; | |
201 | enum netdev_flags flags; | |
202 | ||
203 | struct rte_eth_link link; | |
204 | int link_reset_cnt; | |
205 | ||
58397e6c KT |
206 | /* virtio-net structure for vhost device */ |
207 | OVSRCU_TYPE(struct virtio_net *) virtio_dev; | |
208 | ||
8a9562d2 | 209 | /* In dpdk_list. */ |
ca6ba700 | 210 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
58397e6c | 211 | rte_spinlock_t txq_lock; |
8a9562d2 PS |
212 | }; |
213 | ||
214 | struct netdev_rxq_dpdk { | |
215 | struct netdev_rxq up; | |
216 | int port_id; | |
217 | }; | |
218 | ||
db73f716 DDP |
219 | static bool thread_is_pmd(void); |
220 | ||
8a9562d2 PS |
221 | static int netdev_dpdk_construct(struct netdev *); |
222 | ||
58397e6c KT |
223 | struct virtio_net * netdev_dpdk_get_virtio(const struct netdev_dpdk *dev); |
224 | ||
8a9562d2 PS |
225 | static bool |
226 | is_dpdk_class(const struct netdev_class *class) | |
227 | { | |
228 | return class->construct == netdev_dpdk_construct; | |
229 | } | |
230 | ||
58397e6c KT |
231 | /* XXX: use dpdk malloc for entire OVS. in fact huge page should be used |
232 | * for all other segments data, bss and text. */ | |
8a9562d2 PS |
233 | |
234 | static void * | |
235 | dpdk_rte_mzalloc(size_t sz) | |
236 | { | |
237 | void *ptr; | |
238 | ||
239 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
240 | if (ptr == NULL) { | |
241 | out_of_memory(); | |
242 | } | |
243 | return ptr; | |
244 | } | |
245 | ||
db73f716 DDP |
246 | /* XXX this function should be called only by pmd threads (or by non pmd |
247 | * threads holding the nonpmd_mempool_mutex) */ | |
8a9562d2 | 248 | void |
e14deea0 | 249 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 250 | { |
db73f716 | 251 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 252 | |
db73f716 | 253 | rte_pktmbuf_free_seg(pkt); |
8a9562d2 PS |
254 | } |
255 | ||
b3cd9f9d PS |
256 | static void |
257 | __rte_pktmbuf_init(struct rte_mempool *mp, | |
258 | void *opaque_arg OVS_UNUSED, | |
259 | void *_m, | |
260 | unsigned i OVS_UNUSED) | |
261 | { | |
262 | struct rte_mbuf *m = _m; | |
e14deea0 | 263 | uint32_t buf_len = mp->elt_size - sizeof(struct dp_packet); |
b3cd9f9d | 264 | |
e14deea0 | 265 | RTE_MBUF_ASSERT(mp->elt_size >= sizeof(struct dp_packet)); |
b3cd9f9d PS |
266 | |
267 | memset(m, 0, mp->elt_size); | |
268 | ||
269 | /* start of buffer is just after mbuf structure */ | |
e14deea0 | 270 | m->buf_addr = (char *)m + sizeof(struct dp_packet); |
b3cd9f9d | 271 | m->buf_physaddr = rte_mempool_virt2phy(mp, m) + |
e14deea0 | 272 | sizeof(struct dp_packet); |
b3cd9f9d PS |
273 | m->buf_len = (uint16_t)buf_len; |
274 | ||
275 | /* keep some headroom between start of buffer and data */ | |
b8e57534 | 276 | m->data_off = RTE_MIN(RTE_PKTMBUF_HEADROOM, m->buf_len); |
b3cd9f9d PS |
277 | |
278 | /* init some constant fields */ | |
b3cd9f9d | 279 | m->pool = mp; |
b8e57534 MK |
280 | m->nb_segs = 1; |
281 | m->port = 0xff; | |
b3cd9f9d PS |
282 | } |
283 | ||
284 | static void | |
285 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
286 | void *opaque_arg OVS_UNUSED, | |
287 | void *_m, | |
288 | unsigned i OVS_UNUSED) | |
289 | { | |
290 | struct rte_mbuf *m = _m; | |
291 | ||
292 | __rte_pktmbuf_init(mp, opaque_arg, _m, i); | |
293 | ||
cf62fa4c | 294 | dp_packet_init_dpdk((struct dp_packet *) m, m->buf_len); |
b3cd9f9d PS |
295 | } |
296 | ||
8a9562d2 PS |
297 | static struct dpdk_mp * |
298 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
299 | { | |
300 | struct dpdk_mp *dmp = NULL; | |
301 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
da79ce2b | 302 | unsigned mp_size; |
8a9562d2 PS |
303 | |
304 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
305 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
306 | dmp->refcount++; | |
307 | return dmp; | |
308 | } | |
309 | } | |
310 | ||
311 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
312 | dmp->socket_id = socket_id; | |
313 | dmp->mtu = mtu; | |
314 | dmp->refcount = 1; | |
315 | ||
da79ce2b DDP |
316 | mp_size = MAX_NB_MBUF; |
317 | do { | |
318 | if (snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d_%d_%u", | |
319 | dmp->mtu, dmp->socket_id, mp_size) < 0) { | |
320 | return NULL; | |
321 | } | |
95fb793a | 322 | |
da79ce2b DDP |
323 | dmp->mp = rte_mempool_create(mp_name, mp_size, MBUF_SIZE(mtu), |
324 | MP_CACHE_SZ, | |
325 | sizeof(struct rte_pktmbuf_pool_private), | |
326 | rte_pktmbuf_pool_init, NULL, | |
327 | ovs_rte_pktmbuf_init, NULL, | |
328 | socket_id, 0); | |
329 | } while (!dmp->mp && rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
8a9562d2 PS |
330 | |
331 | if (dmp->mp == NULL) { | |
332 | return NULL; | |
da79ce2b DDP |
333 | } else { |
334 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", mp_name, mp_size ); | |
8a9562d2 PS |
335 | } |
336 | ||
337 | list_push_back(&dpdk_mp_list, &dmp->list_node); | |
338 | return dmp; | |
339 | } | |
340 | ||
341 | static void | |
342 | dpdk_mp_put(struct dpdk_mp *dmp) | |
343 | { | |
344 | ||
345 | if (!dmp) { | |
346 | return; | |
347 | } | |
348 | ||
349 | dmp->refcount--; | |
350 | ovs_assert(dmp->refcount >= 0); | |
351 | ||
352 | #if 0 | |
353 | /* I could not find any API to destroy mp. */ | |
354 | if (dmp->refcount == 0) { | |
355 | list_delete(dmp->list_node); | |
356 | /* destroy mp-pool. */ | |
357 | } | |
358 | #endif | |
359 | } | |
360 | ||
361 | static void | |
362 | check_link_status(struct netdev_dpdk *dev) | |
363 | { | |
364 | struct rte_eth_link link; | |
365 | ||
366 | rte_eth_link_get_nowait(dev->port_id, &link); | |
367 | ||
368 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 369 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
370 | |
371 | dev->link_reset_cnt++; | |
372 | dev->link = link; | |
373 | if (dev->link.link_status) { | |
374 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
375 | dev->port_id, (unsigned)dev->link.link_speed, | |
376 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
377 | ("full-duplex") : ("half-duplex")); | |
378 | } else { | |
379 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
380 | } | |
381 | } | |
382 | } | |
383 | ||
384 | static void * | |
385 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
386 | { | |
387 | struct netdev_dpdk *dev; | |
388 | ||
389 | pthread_detach(pthread_self()); | |
390 | ||
391 | for (;;) { | |
392 | ovs_mutex_lock(&dpdk_mutex); | |
393 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
394 | ovs_mutex_lock(&dev->mutex); | |
395 | check_link_status(dev); | |
396 | ovs_mutex_unlock(&dev->mutex); | |
397 | } | |
398 | ovs_mutex_unlock(&dpdk_mutex); | |
399 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
400 | } | |
401 | ||
402 | return NULL; | |
403 | } | |
404 | ||
405 | static int | |
406 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
407 | { | |
408 | struct rte_pktmbuf_pool_private *mbp_priv; | |
409 | struct ether_addr eth_addr; | |
410 | int diag; | |
411 | int i; | |
412 | ||
413 | if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) { | |
95fb793a | 414 | return ENODEV; |
8a9562d2 PS |
415 | } |
416 | ||
5496878c AW |
417 | diag = rte_eth_dev_configure(dev->port_id, dev->up.n_rxq, dev->up.n_txq, |
418 | &port_conf); | |
8a9562d2 PS |
419 | if (diag) { |
420 | VLOG_ERR("eth dev config error %d",diag); | |
95fb793a | 421 | return -diag; |
8a9562d2 PS |
422 | } |
423 | ||
5496878c | 424 | for (i = 0; i < dev->up.n_txq; i++) { |
79f5354c | 425 | diag = rte_eth_tx_queue_setup(dev->port_id, i, NIC_PORT_TX_Q_SIZE, |
9154f798 | 426 | dev->socket_id, NULL); |
8a9562d2 PS |
427 | if (diag) { |
428 | VLOG_ERR("eth dev tx queue setup error %d",diag); | |
95fb793a | 429 | return -diag; |
8a9562d2 PS |
430 | } |
431 | } | |
432 | ||
5496878c | 433 | for (i = 0; i < dev->up.n_rxq; i++) { |
79f5354c | 434 | diag = rte_eth_rx_queue_setup(dev->port_id, i, NIC_PORT_RX_Q_SIZE, |
d221ffa1 | 435 | dev->socket_id, |
9154f798 | 436 | NULL, dev->dpdk_mp->mp); |
8a9562d2 PS |
437 | if (diag) { |
438 | VLOG_ERR("eth dev rx queue setup error %d",diag); | |
95fb793a | 439 | return -diag; |
8a9562d2 PS |
440 | } |
441 | } | |
442 | ||
443 | diag = rte_eth_dev_start(dev->port_id); | |
444 | if (diag) { | |
445 | VLOG_ERR("eth dev start error %d",diag); | |
95fb793a | 446 | return -diag; |
8a9562d2 PS |
447 | } |
448 | ||
449 | rte_eth_promiscuous_enable(dev->port_id); | |
450 | rte_eth_allmulticast_enable(dev->port_id); | |
451 | ||
452 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
453 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
454 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
455 | dev->port_id, ETH_ADDR_ARGS(eth_addr.addr_bytes)); | |
456 | ||
457 | memcpy(dev->hwaddr, eth_addr.addr_bytes, ETH_ADDR_LEN); | |
458 | rte_eth_link_get_nowait(dev->port_id, &dev->link); | |
459 | ||
460 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
461 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
462 | ||
463 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
464 | return 0; | |
465 | } | |
466 | ||
467 | static struct netdev_dpdk * | |
468 | netdev_dpdk_cast(const struct netdev *netdev) | |
469 | { | |
470 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
471 | } | |
472 | ||
473 | static struct netdev * | |
474 | netdev_dpdk_alloc(void) | |
475 | { | |
476 | struct netdev_dpdk *netdev = dpdk_rte_mzalloc(sizeof *netdev); | |
477 | return &netdev->up; | |
478 | } | |
479 | ||
5a034064 | 480 | static void |
91968eb0 | 481 | netdev_dpdk_alloc_txq(struct netdev_dpdk *netdev, unsigned int n_txqs) |
5a034064 | 482 | { |
bd5131ba | 483 | unsigned i; |
5a034064 AW |
484 | |
485 | netdev->tx_q = dpdk_rte_mzalloc(n_txqs * sizeof *netdev->tx_q); | |
94143fc4 AW |
486 | /* Each index is considered as a cpu core id, since there should |
487 | * be one tx queue for each cpu core. */ | |
5a034064 | 488 | for (i = 0; i < n_txqs; i++) { |
ba0358a1 | 489 | int numa_id = ovs_numa_get_numa_id(i); |
94143fc4 | 490 | |
94143fc4 AW |
491 | /* If the corresponding core is not on the same numa node |
492 | * as 'netdev', flags the 'flush_tx'. */ | |
ba0358a1 | 493 | netdev->tx_q[i].flush_tx = netdev->socket_id == numa_id; |
5a034064 AW |
494 | } |
495 | } | |
496 | ||
8a9562d2 | 497 | static int |
58397e6c KT |
498 | netdev_dpdk_init(struct netdev *netdev_, unsigned int port_no, |
499 | enum dpdk_dev_type type) | |
5a034064 | 500 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 PS |
501 | { |
502 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1b7a04e0 | 503 | int sid; |
95fb793a | 504 | int err = 0; |
8a9562d2 | 505 | |
95fb793a | 506 | ovs_mutex_init(&netdev->mutex); |
95fb793a | 507 | ovs_mutex_lock(&netdev->mutex); |
8a9562d2 | 508 | |
45d947c4 DDP |
509 | rte_spinlock_init(&netdev->stats_lock); |
510 | ||
1b7a04e0 AW |
511 | /* If the 'sid' is negative, it means that the kernel fails |
512 | * to obtain the pci numa info. In that situation, always | |
513 | * use 'SOCKET0'. */ | |
58397e6c KT |
514 | if (type == DPDK_DEV_ETH) { |
515 | sid = rte_eth_dev_socket_id(port_no); | |
516 | } else { | |
517 | sid = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
518 | } | |
519 | ||
1b7a04e0 | 520 | netdev->socket_id = sid < 0 ? SOCKET0 : sid; |
95fb793a | 521 | netdev->port_id = port_no; |
58397e6c | 522 | netdev->type = type; |
8a9562d2 | 523 | netdev->flags = 0; |
8a9562d2 PS |
524 | netdev->mtu = ETHER_MTU; |
525 | netdev->max_packet_len = MTU_TO_MAX_LEN(netdev->mtu); | |
58397e6c | 526 | rte_spinlock_init(&netdev->txq_lock); |
8a9562d2 | 527 | |
8a9562d2 PS |
528 | netdev->dpdk_mp = dpdk_mp_get(netdev->socket_id, netdev->mtu); |
529 | if (!netdev->dpdk_mp) { | |
530 | err = ENOMEM; | |
95fb793a | 531 | goto unlock; |
8a9562d2 PS |
532 | } |
533 | ||
5496878c AW |
534 | netdev_->n_txq = NR_QUEUE; |
535 | netdev_->n_rxq = NR_QUEUE; | |
58397e6c KT |
536 | |
537 | if (type == DPDK_DEV_ETH) { | |
1b99bb05 MG |
538 | netdev_dpdk_alloc_txq(netdev, NR_QUEUE); |
539 | err = dpdk_eth_dev_init(netdev); | |
540 | if (err) { | |
541 | goto unlock; | |
542 | } | |
8a9562d2 | 543 | } |
8a9562d2 PS |
544 | |
545 | list_push_back(&dpdk_list, &netdev->list_node); | |
546 | ||
95fb793a | 547 | unlock: |
5a034064 AW |
548 | if (err) { |
549 | rte_free(netdev->tx_q); | |
550 | } | |
8a9562d2 | 551 | ovs_mutex_unlock(&netdev->mutex); |
95fb793a | 552 | return err; |
553 | } | |
554 | ||
555 | static int | |
556 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
557 | unsigned int *port_no) | |
558 | { | |
559 | const char *cport; | |
560 | ||
561 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
562 | return ENODEV; | |
563 | } | |
564 | ||
565 | cport = dev_name + strlen(prefix); | |
bce01e3a | 566 | *port_no = strtol(cport, NULL, 0); /* string must be null terminated */ |
95fb793a | 567 | return 0; |
568 | } | |
569 | ||
58397e6c KT |
570 | static int |
571 | netdev_dpdk_vhost_construct(struct netdev *netdev_) | |
572 | { | |
573 | int err; | |
574 | ||
575 | if (rte_eal_init_ret) { | |
576 | return rte_eal_init_ret; | |
577 | } | |
578 | ||
579 | ovs_mutex_lock(&dpdk_mutex); | |
580 | err = netdev_dpdk_init(netdev_, -1, DPDK_DEV_VHOST); | |
581 | ovs_mutex_unlock(&dpdk_mutex); | |
582 | ||
583 | return err; | |
584 | } | |
585 | ||
95fb793a | 586 | static int |
587 | netdev_dpdk_construct(struct netdev *netdev) | |
588 | { | |
589 | unsigned int port_no; | |
590 | int err; | |
591 | ||
592 | if (rte_eal_init_ret) { | |
593 | return rte_eal_init_ret; | |
594 | } | |
595 | ||
596 | /* Names always start with "dpdk" */ | |
597 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
598 | if (err) { | |
599 | return err; | |
600 | } | |
601 | ||
602 | ovs_mutex_lock(&dpdk_mutex); | |
58397e6c | 603 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
8a9562d2 PS |
604 | ovs_mutex_unlock(&dpdk_mutex); |
605 | return err; | |
606 | } | |
607 | ||
608 | static void | |
609 | netdev_dpdk_destruct(struct netdev *netdev_) | |
610 | { | |
611 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
612 | ||
613 | ovs_mutex_lock(&dev->mutex); | |
614 | rte_eth_dev_stop(dev->port_id); | |
615 | ovs_mutex_unlock(&dev->mutex); | |
616 | ||
617 | ovs_mutex_lock(&dpdk_mutex); | |
5a034064 | 618 | rte_free(dev->tx_q); |
8a9562d2 PS |
619 | list_remove(&dev->list_node); |
620 | dpdk_mp_put(dev->dpdk_mp); | |
621 | ovs_mutex_unlock(&dpdk_mutex); | |
58397e6c | 622 | } |
8a9562d2 | 623 | |
58397e6c KT |
624 | static void |
625 | netdev_dpdk_vhost_destruct(struct netdev *netdev_) | |
626 | { | |
627 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
628 | ||
629 | /* Can't remove a port while a guest is attached to it. */ | |
630 | if (netdev_dpdk_get_virtio(dev) != NULL) { | |
631 | VLOG_ERR("Can not remove port, vhost device still attached"); | |
632 | return; | |
633 | } | |
634 | ||
635 | ovs_mutex_lock(&dpdk_mutex); | |
636 | list_remove(&dev->list_node); | |
637 | dpdk_mp_put(dev->dpdk_mp); | |
638 | ovs_mutex_unlock(&dpdk_mutex); | |
8a9562d2 PS |
639 | } |
640 | ||
641 | static void | |
642 | netdev_dpdk_dealloc(struct netdev *netdev_) | |
643 | { | |
644 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
645 | ||
646 | rte_free(netdev); | |
647 | } | |
648 | ||
649 | static int | |
650 | netdev_dpdk_get_config(const struct netdev *netdev_, struct smap *args) | |
651 | { | |
652 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
653 | ||
654 | ovs_mutex_lock(&dev->mutex); | |
655 | ||
47659062 AW |
656 | smap_add_format(args, "configured_rx_queues", "%d", netdev_->n_rxq); |
657 | smap_add_format(args, "configured_tx_queues", "%d", netdev_->n_txq); | |
8a9562d2 PS |
658 | ovs_mutex_unlock(&dev->mutex); |
659 | ||
660 | return 0; | |
661 | } | |
662 | ||
7dec44fe AW |
663 | static int |
664 | netdev_dpdk_get_numa_id(const struct netdev *netdev_) | |
665 | { | |
666 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
667 | ||
668 | return netdev->socket_id; | |
669 | } | |
670 | ||
5496878c AW |
671 | /* Sets the number of tx queues and rx queues for the dpdk interface. |
672 | * If the configuration fails, do not try restoring its old configuration | |
673 | * and just returns the error. */ | |
674 | static int | |
675 | netdev_dpdk_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
676 | unsigned int n_rxq) | |
677 | { | |
678 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
679 | int err = 0; | |
680 | ||
681 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
682 | return err; | |
683 | } | |
684 | ||
b7ccaf67 | 685 | ovs_mutex_lock(&dpdk_mutex); |
5496878c | 686 | ovs_mutex_lock(&netdev->mutex); |
91968eb0 | 687 | |
5496878c | 688 | rte_eth_dev_stop(netdev->port_id); |
91968eb0 | 689 | |
5496878c AW |
690 | netdev->up.n_txq = n_txq; |
691 | netdev->up.n_rxq = n_rxq; | |
58397e6c | 692 | |
91968eb0 AW |
693 | rte_free(netdev->tx_q); |
694 | netdev_dpdk_alloc_txq(netdev, n_txq); | |
5496878c | 695 | err = dpdk_eth_dev_init(netdev); |
91968eb0 | 696 | |
5496878c | 697 | ovs_mutex_unlock(&netdev->mutex); |
b7ccaf67 | 698 | ovs_mutex_unlock(&dpdk_mutex); |
5496878c AW |
699 | |
700 | return err; | |
701 | } | |
702 | ||
58397e6c KT |
703 | static int |
704 | netdev_dpdk_vhost_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
705 | unsigned int n_rxq) | |
706 | { | |
707 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
708 | int err = 0; | |
709 | ||
710 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
711 | return err; | |
712 | } | |
713 | ||
714 | ovs_mutex_lock(&dpdk_mutex); | |
715 | ovs_mutex_lock(&netdev->mutex); | |
716 | ||
717 | netdev->up.n_txq = n_txq; | |
718 | netdev->up.n_rxq = n_rxq; | |
719 | ||
720 | ovs_mutex_unlock(&netdev->mutex); | |
721 | ovs_mutex_unlock(&dpdk_mutex); | |
722 | ||
723 | return err; | |
724 | } | |
725 | ||
8a9562d2 PS |
726 | static struct netdev_rxq * |
727 | netdev_dpdk_rxq_alloc(void) | |
728 | { | |
729 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
730 | ||
731 | return &rx->up; | |
732 | } | |
733 | ||
734 | static struct netdev_rxq_dpdk * | |
735 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rx) | |
736 | { | |
737 | return CONTAINER_OF(rx, struct netdev_rxq_dpdk, up); | |
738 | } | |
739 | ||
740 | static int | |
741 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq_) | |
742 | { | |
743 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
744 | struct netdev_dpdk *netdev = netdev_dpdk_cast(rx->up.netdev); | |
745 | ||
746 | ovs_mutex_lock(&netdev->mutex); | |
747 | rx->port_id = netdev->port_id; | |
748 | ovs_mutex_unlock(&netdev->mutex); | |
749 | ||
750 | return 0; | |
751 | } | |
752 | ||
753 | static void | |
754 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq_ OVS_UNUSED) | |
755 | { | |
756 | } | |
757 | ||
758 | static void | |
759 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq_) | |
760 | { | |
761 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
762 | ||
763 | rte_free(rx); | |
764 | } | |
765 | ||
b170db2a RW |
766 | static inline void |
767 | dpdk_queue_flush__(struct netdev_dpdk *dev, int qid) | |
8a9562d2 PS |
768 | { |
769 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1304f1f8 DDP |
770 | uint32_t nb_tx = 0; |
771 | ||
772 | while (nb_tx != txq->count) { | |
773 | uint32_t ret; | |
774 | ||
775 | ret = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts + nb_tx, | |
776 | txq->count - nb_tx); | |
777 | if (!ret) { | |
778 | break; | |
779 | } | |
780 | ||
781 | nb_tx += ret; | |
782 | } | |
8a9562d2 | 783 | |
b170db2a | 784 | if (OVS_UNLIKELY(nb_tx != txq->count)) { |
db73f716 DDP |
785 | /* free buffers, which we couldn't transmit, one at a time (each |
786 | * packet could come from a different mempool) */ | |
787 | int i; | |
788 | ||
789 | for (i = nb_tx; i < txq->count; i++) { | |
790 | rte_pktmbuf_free_seg(txq->burst_pkts[i]); | |
791 | } | |
45d947c4 | 792 | rte_spinlock_lock(&dev->stats_lock); |
1304f1f8 | 793 | dev->stats.tx_dropped += txq->count-nb_tx; |
45d947c4 | 794 | rte_spinlock_unlock(&dev->stats_lock); |
8a9562d2 | 795 | } |
1304f1f8 | 796 | |
8a9562d2 | 797 | txq->count = 0; |
844f2d74 | 798 | txq->tsc = rte_get_timer_cycles(); |
b170db2a RW |
799 | } |
800 | ||
801 | static inline void | |
802 | dpdk_queue_flush(struct netdev_dpdk *dev, int qid) | |
803 | { | |
804 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
805 | ||
806 | if (txq->count == 0) { | |
807 | return; | |
808 | } | |
b170db2a | 809 | dpdk_queue_flush__(dev, qid); |
8a9562d2 PS |
810 | } |
811 | ||
58397e6c KT |
812 | static bool |
813 | is_vhost_running(struct virtio_net *dev) | |
814 | { | |
815 | return (dev != NULL && (dev->flags & VIRTIO_DEV_RUNNING)); | |
816 | } | |
817 | ||
818 | /* | |
819 | * The receive path for the vhost port is the TX path out from guest. | |
820 | */ | |
821 | static int | |
822 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq_, | |
823 | struct dp_packet **packets, int *c) | |
824 | { | |
825 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
826 | struct netdev *netdev = rx->up.netdev; | |
827 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
828 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
829 | int qid = 1; | |
830 | uint16_t nb_rx = 0; | |
831 | ||
832 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
833 | return EAGAIN; | |
834 | } | |
835 | ||
836 | nb_rx = rte_vhost_dequeue_burst(virtio_dev, qid, | |
837 | vhost_dev->dpdk_mp->mp, | |
838 | (struct rte_mbuf **)packets, | |
cd159f1a | 839 | NETDEV_MAX_BURST); |
58397e6c KT |
840 | if (!nb_rx) { |
841 | return EAGAIN; | |
842 | } | |
843 | ||
45d947c4 | 844 | rte_spinlock_lock(&vhost_dev->stats_lock); |
58397e6c | 845 | vhost_dev->stats.rx_packets += (uint64_t)nb_rx; |
45d947c4 DDP |
846 | rte_spinlock_unlock(&vhost_dev->stats_lock); |
847 | ||
58397e6c KT |
848 | *c = (int) nb_rx; |
849 | return 0; | |
850 | } | |
851 | ||
8a9562d2 | 852 | static int |
e14deea0 | 853 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet **packets, |
91088554 | 854 | int *c) |
8a9562d2 PS |
855 | { |
856 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
857 | struct netdev *netdev = rx->up.netdev; | |
858 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 859 | int nb_rx; |
8a9562d2 | 860 | |
5496878c AW |
861 | /* There is only one tx queue for this core. Do not flush other |
862 | * queueus. */ | |
863 | if (rxq_->queue_id == rte_lcore_id()) { | |
864 | dpdk_queue_flush(dev, rxq_->queue_id); | |
865 | } | |
8a9562d2 PS |
866 | |
867 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq_->queue_id, | |
7d08d53e | 868 | (struct rte_mbuf **) packets, |
cd159f1a | 869 | NETDEV_MAX_BURST); |
8a9562d2 PS |
870 | if (!nb_rx) { |
871 | return EAGAIN; | |
872 | } | |
873 | ||
8a9562d2 PS |
874 | *c = nb_rx; |
875 | ||
876 | return 0; | |
877 | } | |
878 | ||
58397e6c KT |
879 | static void |
880 | __netdev_dpdk_vhost_send(struct netdev *netdev, struct dp_packet **pkts, | |
881 | int cnt, bool may_steal) | |
882 | { | |
883 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
884 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
95e9881f KT |
885 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
886 | unsigned int total_pkts = cnt; | |
887 | uint64_t start = 0; | |
58397e6c KT |
888 | |
889 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
45d947c4 | 890 | rte_spinlock_lock(&vhost_dev->stats_lock); |
1b99bb05 | 891 | vhost_dev->stats.tx_dropped+= cnt; |
45d947c4 | 892 | rte_spinlock_unlock(&vhost_dev->stats_lock); |
1b99bb05 | 893 | goto out; |
58397e6c KT |
894 | } |
895 | ||
896 | /* There is vHost TX single queue, So we need to lock it for TX. */ | |
897 | rte_spinlock_lock(&vhost_dev->txq_lock); | |
58397e6c | 898 | |
95e9881f KT |
899 | do { |
900 | unsigned int tx_pkts; | |
901 | ||
902 | tx_pkts = rte_vhost_enqueue_burst(virtio_dev, VIRTIO_RXQ, | |
903 | cur_pkts, cnt); | |
904 | if (OVS_LIKELY(tx_pkts)) { | |
905 | /* Packets have been sent.*/ | |
906 | cnt -= tx_pkts; | |
907 | /* Prepare for possible next iteration.*/ | |
908 | cur_pkts = &cur_pkts[tx_pkts]; | |
909 | } else { | |
910 | uint64_t timeout = VHOST_ENQ_RETRY_USECS * rte_get_timer_hz() / 1E6; | |
911 | unsigned int expired = 0; | |
912 | ||
913 | if (!start) { | |
914 | start = rte_get_timer_cycles(); | |
915 | } | |
916 | ||
917 | /* | |
918 | * Unable to enqueue packets to vhost interface. | |
919 | * Check available entries before retrying. | |
920 | */ | |
921 | while (!rte_vring_available_entries(virtio_dev, VIRTIO_RXQ)) { | |
922 | if (OVS_UNLIKELY((rte_get_timer_cycles() - start) > timeout)) { | |
923 | expired = 1; | |
924 | break; | |
925 | } | |
926 | } | |
927 | if (expired) { | |
928 | /* break out of main loop. */ | |
929 | break; | |
930 | } | |
931 | } | |
932 | } while (cnt); | |
933 | ||
45d947c4 | 934 | rte_spinlock_lock(&vhost_dev->stats_lock); |
95e9881f KT |
935 | vhost_dev->stats.tx_packets += (total_pkts - cnt); |
936 | vhost_dev->stats.tx_dropped += cnt; | |
45d947c4 | 937 | rte_spinlock_unlock(&vhost_dev->stats_lock); |
58397e6c KT |
938 | rte_spinlock_unlock(&vhost_dev->txq_lock); |
939 | ||
940 | out: | |
941 | if (may_steal) { | |
95e9881f KT |
942 | int i; |
943 | ||
944 | for (i = 0; i < total_pkts; i++) { | |
1b99bb05 MG |
945 | dp_packet_delete(pkts[i]); |
946 | } | |
58397e6c KT |
947 | } |
948 | } | |
949 | ||
8a9562d2 | 950 | inline static void |
f4fd623c DDP |
951 | dpdk_queue_pkts(struct netdev_dpdk *dev, int qid, |
952 | struct rte_mbuf **pkts, int cnt) | |
8a9562d2 PS |
953 | { |
954 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
955 | uint64_t diff_tsc; | |
8a9562d2 | 956 | |
f4fd623c DDP |
957 | int i = 0; |
958 | ||
f4fd623c DDP |
959 | while (i < cnt) { |
960 | int freeslots = MAX_TX_QUEUE_LEN - txq->count; | |
961 | int tocopy = MIN(freeslots, cnt-i); | |
8a9562d2 | 962 | |
f4fd623c DDP |
963 | memcpy(&txq->burst_pkts[txq->count], &pkts[i], |
964 | tocopy * sizeof (struct rte_mbuf *)); | |
965 | ||
966 | txq->count += tocopy; | |
967 | i += tocopy; | |
968 | ||
94143fc4 | 969 | if (txq->count == MAX_TX_QUEUE_LEN || txq->flush_tx) { |
b170db2a | 970 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 971 | } |
844f2d74 | 972 | diff_tsc = rte_get_timer_cycles() - txq->tsc; |
f4fd623c | 973 | if (diff_tsc >= DRAIN_TSC) { |
b170db2a | 974 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 975 | } |
8a9562d2 | 976 | } |
8a9562d2 PS |
977 | } |
978 | ||
979 | /* Tx function. Transmit packets indefinitely */ | |
980 | static void | |
58397e6c | 981 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet **pkts, |
2654cc33 | 982 | int cnt) |
db73f716 | 983 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 984 | { |
bce01e3a EJ |
985 | #if !defined(__CHECKER__) && !defined(_WIN32) |
986 | const size_t PKT_ARRAY_SIZE = cnt; | |
987 | #else | |
988 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 989 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 990 | #endif |
8a9562d2 | 991 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
bce01e3a | 992 | struct rte_mbuf *mbufs[PKT_ARRAY_SIZE]; |
175cf4de RW |
993 | int dropped = 0; |
994 | int newcnt = 0; | |
995 | int i; | |
8a9562d2 | 996 | |
db73f716 DDP |
997 | /* If we are on a non pmd thread we have to use the mempool mutex, because |
998 | * every non pmd thread shares the same mempool cache */ | |
999 | ||
1000 | if (!thread_is_pmd()) { | |
1001 | ovs_mutex_lock(&nonpmd_mempool_mutex); | |
1002 | } | |
1003 | ||
f4fd623c | 1004 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1005 | int size = dp_packet_size(pkts[i]); |
95fb793a | 1006 | |
f98d7864 | 1007 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c DDP |
1008 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1009 | (int)size , dev->max_packet_len); | |
1010 | ||
175cf4de | 1011 | dropped++; |
f4fd623c DDP |
1012 | continue; |
1013 | } | |
8a9562d2 | 1014 | |
f4fd623c | 1015 | mbufs[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1016 | |
f4fd623c | 1017 | if (!mbufs[newcnt]) { |
175cf4de RW |
1018 | dropped += cnt - i; |
1019 | break; | |
f4fd623c DDP |
1020 | } |
1021 | ||
1022 | /* We have to do a copy for now */ | |
b8e57534 | 1023 | memcpy(rte_pktmbuf_mtod(mbufs[newcnt], void *), dp_packet_data(pkts[i]), size); |
f4fd623c DDP |
1024 | |
1025 | rte_pktmbuf_data_len(mbufs[newcnt]) = size; | |
1026 | rte_pktmbuf_pkt_len(mbufs[newcnt]) = size; | |
1027 | ||
1028 | newcnt++; | |
1029 | } | |
8a9562d2 | 1030 | |
f98d7864 | 1031 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1032 | rte_spinlock_lock(&dev->stats_lock); |
175cf4de | 1033 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1034 | rte_spinlock_unlock(&dev->stats_lock); |
175cf4de RW |
1035 | } |
1036 | ||
58397e6c KT |
1037 | if (dev->type == DPDK_DEV_VHOST) { |
1038 | __netdev_dpdk_vhost_send(netdev, (struct dp_packet **) mbufs, newcnt, true); | |
1039 | } else { | |
1040 | dpdk_queue_pkts(dev, qid, mbufs, newcnt); | |
1041 | dpdk_queue_flush(dev, qid); | |
1042 | } | |
db73f716 DDP |
1043 | |
1044 | if (!thread_is_pmd()) { | |
1045 | ovs_mutex_unlock(&nonpmd_mempool_mutex); | |
1046 | } | |
8a9562d2 PS |
1047 | } |
1048 | ||
58397e6c KT |
1049 | static int |
1050 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid OVS_UNUSED, struct dp_packet **pkts, | |
1051 | int cnt, bool may_steal) | |
1052 | { | |
1053 | if (OVS_UNLIKELY(pkts[0]->source != DPBUF_DPDK)) { | |
1054 | int i; | |
1055 | ||
1056 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); | |
1057 | if (may_steal) { | |
1058 | for (i = 0; i < cnt; i++) { | |
1059 | dp_packet_delete(pkts[i]); | |
1060 | } | |
1061 | } | |
1062 | } else { | |
1063 | __netdev_dpdk_vhost_send(netdev, pkts, cnt, may_steal); | |
1064 | } | |
1065 | return 0; | |
1066 | } | |
1067 | ||
7251515e DV |
1068 | static inline void |
1069 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
e14deea0 | 1070 | struct dp_packet **pkts, int cnt, bool may_steal) |
8a9562d2 | 1071 | { |
f4fd623c | 1072 | int i; |
8a9562d2 | 1073 | |
7251515e | 1074 | if (OVS_UNLIKELY(!may_steal || |
cf62fa4c | 1075 | pkts[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1076 | struct netdev *netdev = &dev->up; |
1077 | ||
2654cc33 | 1078 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); |
b3cd9f9d PS |
1079 | |
1080 | if (may_steal) { | |
f4fd623c | 1081 | for (i = 0; i < cnt; i++) { |
e14deea0 | 1082 | dp_packet_delete(pkts[i]); |
f4fd623c | 1083 | } |
b3cd9f9d | 1084 | } |
8a9562d2 | 1085 | } else { |
f4fd623c DDP |
1086 | int next_tx_idx = 0; |
1087 | int dropped = 0; | |
8a9562d2 | 1088 | |
f4fd623c | 1089 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1090 | int size = dp_packet_size(pkts[i]); |
1b99bb05 | 1091 | |
f4fd623c DDP |
1092 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
1093 | if (next_tx_idx != i) { | |
1094 | dpdk_queue_pkts(dev, qid, | |
1095 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1096 | i-next_tx_idx); | |
1ebfe1ac | 1097 | } |
f4fd623c | 1098 | |
1ebfe1ac DDP |
1099 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1100 | (int)size , dev->max_packet_len); | |
f4fd623c | 1101 | |
e14deea0 | 1102 | dp_packet_delete(pkts[i]); |
1ebfe1ac | 1103 | dropped++; |
f4fd623c DDP |
1104 | next_tx_idx = i + 1; |
1105 | } | |
1106 | } | |
1107 | if (next_tx_idx != cnt) { | |
1108 | dpdk_queue_pkts(dev, qid, | |
1109 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1110 | cnt-next_tx_idx); | |
1111 | } | |
8a9562d2 | 1112 | |
f4fd623c | 1113 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1114 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1115 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1116 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1117 | } |
8a9562d2 | 1118 | } |
7251515e DV |
1119 | } |
1120 | ||
1121 | static int | |
1122 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
e14deea0 | 1123 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1124 | { |
1125 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1126 | |
7251515e DV |
1127 | netdev_dpdk_send__(dev, qid, pkts, cnt, may_steal); |
1128 | return 0; | |
8a9562d2 PS |
1129 | } |
1130 | ||
1131 | static int | |
1132 | netdev_dpdk_set_etheraddr(struct netdev *netdev, | |
1133 | const uint8_t mac[ETH_ADDR_LEN]) | |
1134 | { | |
1135 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1136 | ||
1137 | ovs_mutex_lock(&dev->mutex); | |
1138 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
1139 | memcpy(dev->hwaddr, mac, ETH_ADDR_LEN); | |
045c0d1a | 1140 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1141 | } |
1142 | ovs_mutex_unlock(&dev->mutex); | |
1143 | ||
1144 | return 0; | |
1145 | } | |
1146 | ||
1147 | static int | |
1148 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, | |
1149 | uint8_t mac[ETH_ADDR_LEN]) | |
1150 | { | |
1151 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1152 | ||
1153 | ovs_mutex_lock(&dev->mutex); | |
1154 | memcpy(mac, dev->hwaddr, ETH_ADDR_LEN); | |
1155 | ovs_mutex_unlock(&dev->mutex); | |
1156 | ||
1157 | return 0; | |
1158 | } | |
1159 | ||
1160 | static int | |
1161 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1162 | { | |
1163 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1164 | ||
1165 | ovs_mutex_lock(&dev->mutex); | |
1166 | *mtup = dev->mtu; | |
1167 | ovs_mutex_unlock(&dev->mutex); | |
1168 | ||
1169 | return 0; | |
1170 | } | |
1171 | ||
1172 | static int | |
1173 | netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu) | |
1174 | { | |
1175 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1176 | int old_mtu, err; | |
1177 | struct dpdk_mp *old_mp; | |
1178 | struct dpdk_mp *mp; | |
1179 | ||
1180 | ovs_mutex_lock(&dpdk_mutex); | |
1181 | ovs_mutex_lock(&dev->mutex); | |
1182 | if (dev->mtu == mtu) { | |
1183 | err = 0; | |
1184 | goto out; | |
1185 | } | |
1186 | ||
1187 | mp = dpdk_mp_get(dev->socket_id, dev->mtu); | |
1188 | if (!mp) { | |
1189 | err = ENOMEM; | |
1190 | goto out; | |
1191 | } | |
1192 | ||
1193 | rte_eth_dev_stop(dev->port_id); | |
1194 | ||
1195 | old_mtu = dev->mtu; | |
1196 | old_mp = dev->dpdk_mp; | |
1197 | dev->dpdk_mp = mp; | |
1198 | dev->mtu = mtu; | |
1199 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
1200 | ||
1201 | err = dpdk_eth_dev_init(dev); | |
1202 | if (err) { | |
8a9562d2 PS |
1203 | dpdk_mp_put(mp); |
1204 | dev->mtu = old_mtu; | |
1205 | dev->dpdk_mp = old_mp; | |
1206 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
1207 | dpdk_eth_dev_init(dev); | |
1208 | goto out; | |
1209 | } | |
1210 | ||
1211 | dpdk_mp_put(old_mp); | |
045c0d1a | 1212 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1213 | out: |
1214 | ovs_mutex_unlock(&dev->mutex); | |
1215 | ovs_mutex_unlock(&dpdk_mutex); | |
1216 | return err; | |
1217 | } | |
1218 | ||
1219 | static int | |
1220 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier); | |
1221 | ||
58397e6c KT |
1222 | static int |
1223 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1224 | struct netdev_stats *stats) | |
1225 | { | |
1226 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1227 | ||
1228 | ovs_mutex_lock(&dev->mutex); | |
1229 | memset(stats, 0, sizeof(*stats)); | |
1230 | /* Unsupported Stats */ | |
1231 | stats->rx_errors = UINT64_MAX; | |
1232 | stats->tx_errors = UINT64_MAX; | |
1233 | stats->multicast = UINT64_MAX; | |
1234 | stats->collisions = UINT64_MAX; | |
1235 | stats->rx_crc_errors = UINT64_MAX; | |
1236 | stats->rx_fifo_errors = UINT64_MAX; | |
1237 | stats->rx_frame_errors = UINT64_MAX; | |
1238 | stats->rx_length_errors = UINT64_MAX; | |
1239 | stats->rx_missed_errors = UINT64_MAX; | |
1240 | stats->rx_over_errors = UINT64_MAX; | |
1241 | stats->tx_aborted_errors = UINT64_MAX; | |
1242 | stats->tx_carrier_errors = UINT64_MAX; | |
1243 | stats->tx_errors = UINT64_MAX; | |
1244 | stats->tx_fifo_errors = UINT64_MAX; | |
1245 | stats->tx_heartbeat_errors = UINT64_MAX; | |
1246 | stats->tx_window_errors = UINT64_MAX; | |
1247 | stats->rx_bytes += UINT64_MAX; | |
1248 | stats->rx_dropped += UINT64_MAX; | |
1249 | stats->tx_bytes += UINT64_MAX; | |
1250 | ||
45d947c4 | 1251 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
1252 | /* Supported Stats */ |
1253 | stats->rx_packets += dev->stats.rx_packets; | |
1254 | stats->tx_packets += dev->stats.tx_packets; | |
1255 | stats->tx_dropped += dev->stats.tx_dropped; | |
45d947c4 | 1256 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1257 | ovs_mutex_unlock(&dev->mutex); |
1258 | ||
1259 | return 0; | |
1260 | } | |
1261 | ||
8a9562d2 PS |
1262 | static int |
1263 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1264 | { | |
1265 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1266 | struct rte_eth_stats rte_stats; | |
1267 | bool gg; | |
1268 | ||
1269 | netdev_dpdk_get_carrier(netdev, &gg); | |
1270 | ovs_mutex_lock(&dev->mutex); | |
1271 | rte_eth_stats_get(dev->port_id, &rte_stats); | |
1272 | ||
2f9dd77f | 1273 | memset(stats, 0, sizeof(*stats)); |
8a9562d2 | 1274 | |
2f9dd77f PS |
1275 | stats->rx_packets = rte_stats.ipackets; |
1276 | stats->tx_packets = rte_stats.opackets; | |
1277 | stats->rx_bytes = rte_stats.ibytes; | |
1278 | stats->tx_bytes = rte_stats.obytes; | |
1279 | stats->rx_errors = rte_stats.ierrors; | |
1280 | stats->tx_errors = rte_stats.oerrors; | |
1281 | stats->multicast = rte_stats.imcasts; | |
8a9562d2 | 1282 | |
45d947c4 | 1283 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 1284 | stats->tx_dropped = dev->stats.tx_dropped; |
45d947c4 | 1285 | rte_spinlock_unlock(&dev->stats_lock); |
8a9562d2 PS |
1286 | ovs_mutex_unlock(&dev->mutex); |
1287 | ||
1288 | return 0; | |
1289 | } | |
1290 | ||
1291 | static int | |
1292 | netdev_dpdk_get_features(const struct netdev *netdev_, | |
1293 | enum netdev_features *current, | |
1294 | enum netdev_features *advertised OVS_UNUSED, | |
1295 | enum netdev_features *supported OVS_UNUSED, | |
1296 | enum netdev_features *peer OVS_UNUSED) | |
1297 | { | |
1298 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1299 | struct rte_eth_link link; | |
1300 | ||
1301 | ovs_mutex_lock(&dev->mutex); | |
1302 | link = dev->link; | |
1303 | ovs_mutex_unlock(&dev->mutex); | |
1304 | ||
1305 | if (link.link_duplex == ETH_LINK_AUTONEG_DUPLEX) { | |
1306 | if (link.link_speed == ETH_LINK_SPEED_AUTONEG) { | |
1307 | *current = NETDEV_F_AUTONEG; | |
1308 | } | |
1309 | } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { | |
1310 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1311 | *current = NETDEV_F_10MB_HD; | |
1312 | } | |
1313 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1314 | *current = NETDEV_F_100MB_HD; | |
1315 | } | |
1316 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1317 | *current = NETDEV_F_1GB_HD; | |
1318 | } | |
1319 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
1320 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1321 | *current = NETDEV_F_10MB_FD; | |
1322 | } | |
1323 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1324 | *current = NETDEV_F_100MB_FD; | |
1325 | } | |
1326 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1327 | *current = NETDEV_F_1GB_FD; | |
1328 | } | |
1329 | if (link.link_speed == ETH_LINK_SPEED_10000) { | |
1330 | *current = NETDEV_F_10GB_FD; | |
1331 | } | |
1332 | } | |
1333 | ||
1334 | return 0; | |
1335 | } | |
1336 | ||
1337 | static int | |
1338 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
1339 | { | |
1340 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1341 | int ifindex; | |
1342 | ||
1343 | ovs_mutex_lock(&dev->mutex); | |
1344 | ifindex = dev->port_id; | |
1345 | ovs_mutex_unlock(&dev->mutex); | |
1346 | ||
1347 | return ifindex; | |
1348 | } | |
1349 | ||
1350 | static int | |
1351 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1352 | { | |
1353 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1354 | ||
1355 | ovs_mutex_lock(&dev->mutex); | |
1356 | check_link_status(dev); | |
1357 | *carrier = dev->link.link_status; | |
58397e6c KT |
1358 | |
1359 | ovs_mutex_unlock(&dev->mutex); | |
1360 | ||
1361 | return 0; | |
1362 | } | |
1363 | ||
1364 | static int | |
1365 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1366 | { | |
1367 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1368 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(dev); | |
1369 | ||
1370 | ovs_mutex_lock(&dev->mutex); | |
1371 | ||
1372 | if (is_vhost_running(virtio_dev)) { | |
1373 | *carrier = 1; | |
1374 | } else { | |
1375 | *carrier = 0; | |
1376 | } | |
1377 | ||
8a9562d2 PS |
1378 | ovs_mutex_unlock(&dev->mutex); |
1379 | ||
1380 | return 0; | |
1381 | } | |
1382 | ||
1383 | static long long int | |
1384 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev_) | |
1385 | { | |
1386 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1387 | long long int carrier_resets; | |
1388 | ||
1389 | ovs_mutex_lock(&dev->mutex); | |
1390 | carrier_resets = dev->link_reset_cnt; | |
1391 | ovs_mutex_unlock(&dev->mutex); | |
1392 | ||
1393 | return carrier_resets; | |
1394 | } | |
1395 | ||
1396 | static int | |
1397 | netdev_dpdk_set_miimon(struct netdev *netdev_ OVS_UNUSED, | |
1398 | long long int interval OVS_UNUSED) | |
1399 | { | |
ee32150e | 1400 | return EOPNOTSUPP; |
8a9562d2 PS |
1401 | } |
1402 | ||
1403 | static int | |
1404 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
1405 | enum netdev_flags off, enum netdev_flags on, | |
95fb793a | 1406 | enum netdev_flags *old_flagsp) OVS_REQUIRES(dev->mutex) |
8a9562d2 PS |
1407 | { |
1408 | int err; | |
1409 | ||
1410 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
1411 | return EINVAL; | |
1412 | } | |
1413 | ||
1414 | *old_flagsp = dev->flags; | |
1415 | dev->flags |= on; | |
1416 | dev->flags &= ~off; | |
1417 | ||
1418 | if (dev->flags == *old_flagsp) { | |
1419 | return 0; | |
1420 | } | |
1421 | ||
58397e6c KT |
1422 | if (dev->type == DPDK_DEV_ETH) { |
1423 | if (dev->flags & NETDEV_UP) { | |
1424 | err = rte_eth_dev_start(dev->port_id); | |
1425 | if (err) | |
1426 | return -err; | |
1427 | } | |
8a9562d2 | 1428 | |
58397e6c KT |
1429 | if (dev->flags & NETDEV_PROMISC) { |
1430 | rte_eth_promiscuous_enable(dev->port_id); | |
1431 | } | |
8a9562d2 | 1432 | |
58397e6c KT |
1433 | if (!(dev->flags & NETDEV_UP)) { |
1434 | rte_eth_dev_stop(dev->port_id); | |
1435 | } | |
8a9562d2 PS |
1436 | } |
1437 | ||
1438 | return 0; | |
1439 | } | |
1440 | ||
1441 | static int | |
1442 | netdev_dpdk_update_flags(struct netdev *netdev_, | |
1443 | enum netdev_flags off, enum netdev_flags on, | |
1444 | enum netdev_flags *old_flagsp) | |
1445 | { | |
1446 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1447 | int error; | |
1448 | ||
1449 | ovs_mutex_lock(&netdev->mutex); | |
1450 | error = netdev_dpdk_update_flags__(netdev, off, on, old_flagsp); | |
1451 | ovs_mutex_unlock(&netdev->mutex); | |
1452 | ||
1453 | return error; | |
1454 | } | |
1455 | ||
1456 | static int | |
1457 | netdev_dpdk_get_status(const struct netdev *netdev_, struct smap *args) | |
1458 | { | |
1459 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1460 | struct rte_eth_dev_info dev_info; | |
1461 | ||
e0a801c7 | 1462 | if (dev->port_id < 0) |
8a9562d2 PS |
1463 | return ENODEV; |
1464 | ||
1465 | ovs_mutex_lock(&dev->mutex); | |
1466 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
1467 | ovs_mutex_unlock(&dev->mutex); | |
1468 | ||
1469 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1470 | ||
95fb793a | 1471 | smap_add_format(args, "port_no", "%d", dev->port_id); |
8a9562d2 PS |
1472 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); |
1473 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1474 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
1475 | smap_add_format(args, "max_rx_pktlen", "%u", dev_info.max_rx_pktlen); | |
1476 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); | |
1477 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
1478 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
1479 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
1480 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
1481 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
1482 | ||
1483 | smap_add_format(args, "pci-vendor_id", "0x%u", dev_info.pci_dev->id.vendor_id); | |
1484 | smap_add_format(args, "pci-device_id", "0x%x", dev_info.pci_dev->id.device_id); | |
1485 | ||
1486 | return 0; | |
1487 | } | |
1488 | ||
1489 | static void | |
1490 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
1491 | OVS_REQUIRES(dev->mutex) | |
1492 | { | |
1493 | enum netdev_flags old_flags; | |
1494 | ||
1495 | if (admin_state) { | |
1496 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
1497 | } else { | |
1498 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
1499 | } | |
1500 | } | |
1501 | ||
1502 | static void | |
1503 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
1504 | const char *argv[], void *aux OVS_UNUSED) | |
1505 | { | |
1506 | bool up; | |
1507 | ||
1508 | if (!strcasecmp(argv[argc - 1], "up")) { | |
1509 | up = true; | |
1510 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
1511 | up = false; | |
1512 | } else { | |
1513 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
1514 | return; | |
1515 | } | |
1516 | ||
1517 | if (argc > 2) { | |
1518 | struct netdev *netdev = netdev_from_name(argv[1]); | |
1519 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
1520 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
1521 | ||
1522 | ovs_mutex_lock(&dpdk_dev->mutex); | |
1523 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
1524 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
1525 | ||
1526 | netdev_close(netdev); | |
1527 | } else { | |
1528 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
1529 | netdev_close(netdev); | |
1530 | return; | |
1531 | } | |
1532 | } else { | |
1533 | struct netdev_dpdk *netdev; | |
1534 | ||
1535 | ovs_mutex_lock(&dpdk_mutex); | |
1536 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
1537 | ovs_mutex_lock(&netdev->mutex); | |
1538 | netdev_dpdk_set_admin_state__(netdev, up); | |
1539 | ovs_mutex_unlock(&netdev->mutex); | |
1540 | } | |
1541 | ovs_mutex_unlock(&dpdk_mutex); | |
1542 | } | |
1543 | unixctl_command_reply(conn, "OK"); | |
1544 | } | |
1545 | ||
58397e6c KT |
1546 | /* |
1547 | * Set virtqueue flags so that we do not receive interrupts. | |
1548 | */ | |
1549 | static void | |
1550 | set_irq_status(struct virtio_net *dev) | |
1551 | { | |
1552 | dev->virtqueue[VIRTIO_RXQ]->used->flags = VRING_USED_F_NO_NOTIFY; | |
1553 | dev->virtqueue[VIRTIO_TXQ]->used->flags = VRING_USED_F_NO_NOTIFY; | |
1554 | } | |
1555 | ||
1556 | /* | |
1557 | * A new virtio-net device is added to a vhost port. | |
1558 | */ | |
1559 | static int | |
1560 | new_device(struct virtio_net *dev) | |
1561 | { | |
1562 | struct netdev_dpdk *netdev; | |
1563 | bool exists = false; | |
1564 | ||
1565 | ovs_mutex_lock(&dpdk_mutex); | |
1566 | /* Add device to the vhost port with the same name as that passed down. */ | |
1567 | LIST_FOR_EACH(netdev, list_node, &dpdk_list) { | |
1568 | if (strncmp(dev->ifname, netdev->up.name, IFNAMSIZ) == 0) { | |
1569 | ovs_mutex_lock(&netdev->mutex); | |
1570 | ovsrcu_set(&netdev->virtio_dev, dev); | |
1571 | ovs_mutex_unlock(&netdev->mutex); | |
1572 | exists = true; | |
1573 | dev->flags |= VIRTIO_DEV_RUNNING; | |
1574 | /* Disable notifications. */ | |
1575 | set_irq_status(dev); | |
1576 | break; | |
1577 | } | |
1578 | } | |
1579 | ovs_mutex_unlock(&dpdk_mutex); | |
1580 | ||
1581 | if (!exists) { | |
1582 | VLOG_INFO("vHost Device '%s' (%ld) can't be added - name not found", | |
1583 | dev->ifname, dev->device_fh); | |
1584 | ||
1585 | return -1; | |
1586 | } | |
1587 | ||
1588 | VLOG_INFO("vHost Device '%s' (%ld) has been added", | |
1589 | dev->ifname, dev->device_fh); | |
1590 | return 0; | |
1591 | } | |
1592 | ||
1593 | /* | |
1594 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
1595 | * flag to stop any more packets from being sent or received to/from a VM and | |
1596 | * ensure all currently queued packets have been sent/received before removing | |
1597 | * the device. | |
1598 | */ | |
1599 | static void | |
1600 | destroy_device(volatile struct virtio_net *dev) | |
1601 | { | |
1602 | struct netdev_dpdk *vhost_dev; | |
1603 | ||
1604 | ovs_mutex_lock(&dpdk_mutex); | |
1605 | LIST_FOR_EACH (vhost_dev, list_node, &dpdk_list) { | |
1606 | if (netdev_dpdk_get_virtio(vhost_dev) == dev) { | |
1607 | ||
1608 | ovs_mutex_lock(&vhost_dev->mutex); | |
1609 | dev->flags &= ~VIRTIO_DEV_RUNNING; | |
1610 | ovsrcu_set(&vhost_dev->virtio_dev, NULL); | |
1611 | ovs_mutex_unlock(&vhost_dev->mutex); | |
1612 | ||
1613 | /* | |
1614 | * Wait for other threads to quiesce before | |
1615 | * setting the virtio_dev to NULL. | |
1616 | */ | |
1617 | ovsrcu_synchronize(); | |
618f44f7 KT |
1618 | /* |
1619 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
1620 | * put thread back into quiescent state before returning. | |
1621 | */ | |
1622 | ovsrcu_quiesce_start(); | |
58397e6c KT |
1623 | } |
1624 | } | |
1625 | ovs_mutex_unlock(&dpdk_mutex); | |
1626 | ||
1627 | VLOG_INFO("vHost Device '%s' (%ld) has been removed", | |
1628 | dev->ifname, dev->device_fh); | |
1629 | } | |
1630 | ||
1631 | struct virtio_net * | |
1632 | netdev_dpdk_get_virtio(const struct netdev_dpdk *dev) | |
1633 | { | |
1634 | return ovsrcu_get(struct virtio_net *, &dev->virtio_dev); | |
1635 | } | |
1636 | ||
1637 | /* | |
1638 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
1639 | * configuration has been fully complete. | |
1640 | */ | |
bce01e3a | 1641 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
1642 | { |
1643 | .new_device = new_device, | |
1644 | .destroy_device = destroy_device, | |
1645 | }; | |
1646 | ||
1647 | static void * | |
1648 | start_cuse_session_loop(void *dummy OVS_UNUSED) | |
1649 | { | |
1650 | pthread_detach(pthread_self()); | |
618f44f7 KT |
1651 | /* Put the cuse thread into quiescent state. */ |
1652 | ovsrcu_quiesce_start(); | |
58397e6c KT |
1653 | rte_vhost_driver_session_start(); |
1654 | return NULL; | |
1655 | } | |
1656 | ||
1657 | static int | |
1658 | dpdk_vhost_class_init(void) | |
1659 | { | |
58397e6c KT |
1660 | int err = -1; |
1661 | ||
1662 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
1663 | ||
1664 | /* Register CUSE device to handle IOCTLs. | |
1665 | * Unless otherwise specified on the vswitchd command line, cuse_dev_name | |
1666 | * is set to vhost-net. | |
1667 | */ | |
1668 | err = rte_vhost_driver_register(cuse_dev_name); | |
1669 | ||
1670 | if (err != 0) { | |
1671 | VLOG_ERR("CUSE device setup failure."); | |
1672 | return -1; | |
1673 | } | |
1674 | ||
618f44f7 KT |
1675 | ovs_thread_create("cuse_thread", start_cuse_session_loop, NULL); |
1676 | return 0; | |
58397e6c KT |
1677 | } |
1678 | ||
033e9df2 DDP |
1679 | static void |
1680 | dpdk_common_init(void) | |
1681 | { | |
1682 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
1683 | "[netdev] up|down", 1, 2, | |
1684 | netdev_dpdk_set_admin_state, NULL); | |
1685 | ||
1686 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
1687 | } | |
1688 | ||
95fb793a | 1689 | /* Client Rings */ |
1690 | ||
95fb793a | 1691 | static int |
1692 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
1693 | unsigned int *eth_port_id) | |
1694 | { | |
1695 | struct dpdk_ring *ivshmem; | |
1696 | char ring_name[10]; | |
1697 | int err; | |
1698 | ||
1699 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
1700 | if (ivshmem == NULL) { | |
1701 | return ENOMEM; | |
1702 | } | |
1703 | ||
7251515e | 1704 | /* XXX: Add support for multiquque ring. */ |
95fb793a | 1705 | err = snprintf(ring_name, 10, "%s_tx", dev_name); |
1706 | if (err < 0) { | |
1707 | return -err; | |
1708 | } | |
1709 | ||
7251515e DV |
1710 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1711 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1712 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1713 | if (ivshmem->cring_tx == NULL) { |
1714 | rte_free(ivshmem); | |
1715 | return ENOMEM; | |
1716 | } | |
1717 | ||
1718 | err = snprintf(ring_name, 10, "%s_rx", dev_name); | |
1719 | if (err < 0) { | |
1720 | return -err; | |
1721 | } | |
1722 | ||
7251515e DV |
1723 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1724 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1725 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1726 | if (ivshmem->cring_rx == NULL) { |
1727 | rte_free(ivshmem); | |
1728 | return ENOMEM; | |
1729 | } | |
1730 | ||
d7310583 DDP |
1731 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
1732 | &ivshmem->cring_tx, 1, SOCKET0); | |
1733 | ||
95fb793a | 1734 | if (err < 0) { |
1735 | rte_free(ivshmem); | |
1736 | return ENODEV; | |
1737 | } | |
1738 | ||
1739 | ivshmem->user_port_id = port_no; | |
1740 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
1741 | list_push_back(&dpdk_ring_list, &ivshmem->list_node); | |
1742 | ||
1743 | *eth_port_id = ivshmem->eth_port_id; | |
1744 | return 0; | |
1745 | } | |
1746 | ||
1747 | static int | |
1748 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) OVS_REQUIRES(dpdk_mutex) | |
1749 | { | |
1750 | struct dpdk_ring *ivshmem; | |
1751 | unsigned int port_no; | |
1752 | int err = 0; | |
1753 | ||
1754 | /* Names always start with "dpdkr" */ | |
1755 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
1756 | if (err) { | |
1757 | return err; | |
1758 | } | |
1759 | ||
1760 | /* look through our list to find the device */ | |
1761 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { | |
1762 | if (ivshmem->user_port_id == port_no) { | |
58397e6c | 1763 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
95fb793a | 1764 | *eth_port_id = ivshmem->eth_port_id; /* really all that is needed */ |
1765 | return 0; | |
1766 | } | |
1767 | } | |
1768 | /* Need to create the device rings */ | |
1769 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
1770 | } | |
1771 | ||
7251515e DV |
1772 | static int |
1773 | netdev_dpdk_ring_send(struct netdev *netdev, int qid OVS_UNUSED, | |
e14deea0 | 1774 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1775 | { |
1776 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1b99bb05 MG |
1777 | unsigned i; |
1778 | ||
1779 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that the | |
1780 | * rss hash field is clear. This is because the same mbuf may be modified by | |
1781 | * the consumer of the ring and return into the datapath without recalculating | |
1782 | * the RSS hash. */ | |
1783 | for (i = 0; i < cnt; i++) { | |
1784 | dp_packet_set_rss_hash(pkts[i], 0); | |
1785 | } | |
7251515e DV |
1786 | |
1787 | /* DPDK Rings have a single TX queue, Therefore needs locking. */ | |
58397e6c | 1788 | rte_spinlock_lock(&dev->txq_lock); |
7251515e | 1789 | netdev_dpdk_send__(dev, 0, pkts, cnt, may_steal); |
58397e6c | 1790 | rte_spinlock_unlock(&dev->txq_lock); |
7251515e DV |
1791 | return 0; |
1792 | } | |
1793 | ||
95fb793a | 1794 | static int |
1795 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
1796 | { | |
1797 | unsigned int port_no = 0; | |
1798 | int err = 0; | |
1799 | ||
1800 | if (rte_eal_init_ret) { | |
1801 | return rte_eal_init_ret; | |
1802 | } | |
1803 | ||
1804 | ovs_mutex_lock(&dpdk_mutex); | |
1805 | ||
1806 | err = dpdk_ring_open(netdev->name, &port_no); | |
1807 | if (err) { | |
1808 | goto unlock_dpdk; | |
1809 | } | |
1810 | ||
58397e6c | 1811 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
95fb793a | 1812 | |
1813 | unlock_dpdk: | |
1814 | ovs_mutex_unlock(&dpdk_mutex); | |
1815 | return err; | |
1816 | } | |
1817 | ||
58397e6c KT |
1818 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, MULTIQ, SEND, \ |
1819 | GET_CARRIER, GET_STATS, GET_FEATURES, GET_STATUS, RXQ_RECV) \ | |
95fb793a | 1820 | { \ |
1821 | NAME, \ | |
1822 | INIT, /* init */ \ | |
1823 | NULL, /* netdev_dpdk_run */ \ | |
1824 | NULL, /* netdev_dpdk_wait */ \ | |
1825 | \ | |
1826 | netdev_dpdk_alloc, \ | |
1827 | CONSTRUCT, \ | |
58397e6c | 1828 | DESTRUCT, \ |
95fb793a | 1829 | netdev_dpdk_dealloc, \ |
1830 | netdev_dpdk_get_config, \ | |
1831 | NULL, /* netdev_dpdk_set_config */ \ | |
1832 | NULL, /* get_tunnel_config */ \ | |
58397e6c KT |
1833 | NULL, /* build header */ \ |
1834 | NULL, /* push header */ \ | |
1835 | NULL, /* pop header */ \ | |
7dec44fe | 1836 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
5496878c | 1837 | MULTIQ, /* set_multiq */ \ |
95fb793a | 1838 | \ |
7251515e | 1839 | SEND, /* send */ \ |
95fb793a | 1840 | NULL, /* send_wait */ \ |
1841 | \ | |
1842 | netdev_dpdk_set_etheraddr, \ | |
1843 | netdev_dpdk_get_etheraddr, \ | |
1844 | netdev_dpdk_get_mtu, \ | |
1845 | netdev_dpdk_set_mtu, \ | |
1846 | netdev_dpdk_get_ifindex, \ | |
58397e6c | 1847 | GET_CARRIER, \ |
95fb793a | 1848 | netdev_dpdk_get_carrier_resets, \ |
1849 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
1850 | GET_STATS, \ |
1851 | GET_FEATURES, \ | |
95fb793a | 1852 | NULL, /* set_advertisements */ \ |
1853 | \ | |
1854 | NULL, /* set_policing */ \ | |
1855 | NULL, /* get_qos_types */ \ | |
1856 | NULL, /* get_qos_capabilities */ \ | |
1857 | NULL, /* get_qos */ \ | |
1858 | NULL, /* set_qos */ \ | |
1859 | NULL, /* get_queue */ \ | |
1860 | NULL, /* set_queue */ \ | |
1861 | NULL, /* delete_queue */ \ | |
1862 | NULL, /* get_queue_stats */ \ | |
1863 | NULL, /* queue_dump_start */ \ | |
1864 | NULL, /* queue_dump_next */ \ | |
1865 | NULL, /* queue_dump_done */ \ | |
1866 | NULL, /* dump_queue_stats */ \ | |
1867 | \ | |
1868 | NULL, /* get_in4 */ \ | |
1869 | NULL, /* set_in4 */ \ | |
1870 | NULL, /* get_in6 */ \ | |
1871 | NULL, /* add_router */ \ | |
1872 | NULL, /* get_next_hop */ \ | |
58397e6c | 1873 | GET_STATUS, \ |
95fb793a | 1874 | NULL, /* arp_lookup */ \ |
1875 | \ | |
1876 | netdev_dpdk_update_flags, \ | |
1877 | \ | |
1878 | netdev_dpdk_rxq_alloc, \ | |
1879 | netdev_dpdk_rxq_construct, \ | |
1880 | netdev_dpdk_rxq_destruct, \ | |
1881 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 1882 | RXQ_RECV, \ |
95fb793a | 1883 | NULL, /* rx_wait */ \ |
1884 | NULL, /* rxq_drain */ \ | |
1885 | } | |
8a9562d2 PS |
1886 | |
1887 | int | |
1888 | dpdk_init(int argc, char **argv) | |
1889 | { | |
1890 | int result; | |
58397e6c KT |
1891 | int base = 0; |
1892 | char *pragram_name = argv[0]; | |
8a9562d2 | 1893 | |
9441caf3 | 1894 | if (argc < 2 || strcmp(argv[1], "--dpdk")) |
8a9562d2 PS |
1895 | return 0; |
1896 | ||
58397e6c | 1897 | /* Remove the --dpdk argument from arg list.*/ |
8a9562d2 PS |
1898 | argc--; |
1899 | argv++; | |
1900 | ||
58397e6c KT |
1901 | /* If the cuse_dev_name parameter has been provided, set 'cuse_dev_name' to |
1902 | * this string if it meets the correct criteria. Otherwise, set it to the | |
1903 | * default (vhost-net). | |
1904 | */ | |
1905 | if (!strcmp(argv[1], "--cuse_dev_name") && | |
1906 | (strlen(argv[2]) <= NAME_MAX)) { | |
1907 | ||
1908 | cuse_dev_name = strdup(argv[2]); | |
1909 | ||
1910 | /* Remove the cuse_dev_name configuration parameters from the argument | |
1911 | * list, so that the correct elements are passed to the DPDK | |
1912 | * initialization function | |
1913 | */ | |
1914 | argc -= 2; | |
1915 | argv += 2; /* Increment by two to bypass the cuse_dev_name arguments */ | |
1916 | base = 2; | |
1917 | ||
1918 | VLOG_ERR("User-provided cuse_dev_name in use: /dev/%s", cuse_dev_name); | |
1919 | } else { | |
1920 | cuse_dev_name = "vhost-net"; | |
1921 | VLOG_INFO("No cuse_dev_name provided - defaulting to /dev/vhost-net"); | |
1922 | } | |
1923 | ||
1924 | /* Keep the program name argument as this is needed for call to | |
1925 | * rte_eal_init() | |
1926 | */ | |
1927 | argv[0] = pragram_name; | |
1928 | ||
8a9562d2 PS |
1929 | /* Make sure things are initialized ... */ |
1930 | result = rte_eal_init(argc, argv); | |
451450fa | 1931 | if (result < 0) { |
58397e6c | 1932 | ovs_abort(result, "Cannot init EAL"); |
451450fa | 1933 | } |
8a9562d2 | 1934 | |
d7310583 | 1935 | rte_memzone_dump(stdout); |
8a9562d2 PS |
1936 | rte_eal_init_ret = 0; |
1937 | ||
451450fa | 1938 | if (argc > result) { |
9441caf3 | 1939 | argv[result] = argv[0]; |
451450fa | 1940 | } |
9441caf3 | 1941 | |
db73f716 | 1942 | /* We are called from the main thread here */ |
d5c199ea | 1943 | RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID; |
db73f716 | 1944 | |
58397e6c | 1945 | return result + 1 + base; |
8a9562d2 PS |
1946 | } |
1947 | ||
bce01e3a | 1948 | static const struct netdev_class dpdk_class = |
95fb793a | 1949 | NETDEV_DPDK_CLASS( |
1950 | "dpdk", | |
b8e57534 | 1951 | NULL, |
5496878c | 1952 | netdev_dpdk_construct, |
58397e6c | 1953 | netdev_dpdk_destruct, |
7251515e | 1954 | netdev_dpdk_set_multiq, |
58397e6c KT |
1955 | netdev_dpdk_eth_send, |
1956 | netdev_dpdk_get_carrier, | |
1957 | netdev_dpdk_get_stats, | |
1958 | netdev_dpdk_get_features, | |
1959 | netdev_dpdk_get_status, | |
1960 | netdev_dpdk_rxq_recv); | |
95fb793a | 1961 | |
bce01e3a | 1962 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 1963 | NETDEV_DPDK_CLASS( |
1964 | "dpdkr", | |
033e9df2 | 1965 | NULL, |
5496878c | 1966 | netdev_dpdk_ring_construct, |
58397e6c KT |
1967 | netdev_dpdk_destruct, |
1968 | NULL, | |
1969 | netdev_dpdk_ring_send, | |
1970 | netdev_dpdk_get_carrier, | |
1971 | netdev_dpdk_get_stats, | |
1972 | netdev_dpdk_get_features, | |
1973 | netdev_dpdk_get_status, | |
1974 | netdev_dpdk_rxq_recv); | |
1975 | ||
bce01e3a | 1976 | static const struct netdev_class dpdk_vhost_class = |
58397e6c KT |
1977 | NETDEV_DPDK_CLASS( |
1978 | "dpdkvhost", | |
1979 | dpdk_vhost_class_init, | |
1980 | netdev_dpdk_vhost_construct, | |
1981 | netdev_dpdk_vhost_destruct, | |
1982 | netdev_dpdk_vhost_set_multiq, | |
1983 | netdev_dpdk_vhost_send, | |
1984 | netdev_dpdk_vhost_get_carrier, | |
1985 | netdev_dpdk_vhost_get_stats, | |
1986 | NULL, | |
7251515e | 1987 | NULL, |
58397e6c | 1988 | netdev_dpdk_vhost_rxq_recv); |
95fb793a | 1989 | |
8a9562d2 PS |
1990 | void |
1991 | netdev_dpdk_register(void) | |
1992 | { | |
95fb793a | 1993 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
1994 | ||
033e9df2 DDP |
1995 | if (rte_eal_init_ret) { |
1996 | return; | |
1997 | } | |
1998 | ||
95fb793a | 1999 | if (ovsthread_once_start(&once)) { |
033e9df2 | 2000 | dpdk_common_init(); |
95fb793a | 2001 | netdev_register_provider(&dpdk_class); |
2002 | netdev_register_provider(&dpdk_ring_class); | |
58397e6c | 2003 | netdev_register_provider(&dpdk_vhost_class); |
95fb793a | 2004 | ovsthread_once_done(&once); |
2005 | } | |
8a9562d2 | 2006 | } |
8617afff PS |
2007 | |
2008 | int | |
bd5131ba | 2009 | pmd_thread_setaffinity_cpu(unsigned cpu) |
8617afff PS |
2010 | { |
2011 | cpu_set_t cpuset; | |
2012 | int err; | |
2013 | ||
2014 | CPU_ZERO(&cpuset); | |
2015 | CPU_SET(cpu, &cpuset); | |
2016 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset); | |
2017 | if (err) { | |
2018 | VLOG_ERR("Thread affinity error %d",err); | |
2019 | return err; | |
2020 | } | |
abb5943d AW |
2021 | /* NON_PMD_CORE_ID is reserved for use by non pmd threads. */ |
2022 | ovs_assert(cpu != NON_PMD_CORE_ID); | |
65f13b50 | 2023 | RTE_PER_LCORE(_lcore_id) = cpu; |
8617afff PS |
2024 | |
2025 | return 0; | |
2026 | } | |
db73f716 | 2027 | |
db73f716 DDP |
2028 | static bool |
2029 | thread_is_pmd(void) | |
2030 | { | |
abb5943d | 2031 | return rte_lcore_id() != NON_PMD_CORE_ID; |
db73f716 | 2032 | } |