]>
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 | ||
645b8934 | 93 | /* XXX: Needs per NIC value for these constants. */ |
8a9562d2 PS |
94 | #define RX_PTHRESH 32 /* Default values of RX prefetch threshold reg. */ |
95 | #define RX_HTHRESH 32 /* Default values of RX host threshold reg. */ | |
96 | #define RX_WTHRESH 16 /* Default values of RX write-back threshold reg. */ | |
97 | ||
98 | #define TX_PTHRESH 36 /* Default values of TX prefetch threshold reg. */ | |
99 | #define TX_HTHRESH 0 /* Default values of TX host threshold reg. */ | |
100 | #define TX_WTHRESH 0 /* Default values of TX write-back threshold reg. */ | |
101 | ||
58397e6c KT |
102 | #define MAX_PKT_BURST 32 /* Max burst size for RX/TX */ |
103 | ||
104 | /* Character device cuse_dev_name. */ | |
105 | char *cuse_dev_name = NULL; | |
106 | ||
95e9881f KT |
107 | /* |
108 | * Maximum amount of time in micro seconds to try and enqueue to vhost. | |
109 | */ | |
110 | #define VHOST_ENQ_RETRY_USECS 100 | |
111 | ||
8a9562d2 | 112 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
113 | .rxmode = { |
114 | .mq_mode = ETH_MQ_RX_RSS, | |
115 | .split_hdr_size = 0, | |
116 | .header_split = 0, /* Header Split disabled */ | |
117 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
118 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
119 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
120 | .hw_strip_crc = 0, | |
121 | }, | |
122 | .rx_adv_conf = { | |
123 | .rss_conf = { | |
124 | .rss_key = NULL, | |
543342a4 | 125 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 126 | }, |
a28ddd11 DDP |
127 | }, |
128 | .txmode = { | |
129 | .mq_mode = ETH_MQ_TX_NONE, | |
130 | }, | |
8a9562d2 PS |
131 | }; |
132 | ||
133 | static const struct rte_eth_rxconf rx_conf = { | |
a28ddd11 DDP |
134 | .rx_thresh = { |
135 | .pthresh = RX_PTHRESH, | |
136 | .hthresh = RX_HTHRESH, | |
137 | .wthresh = RX_WTHRESH, | |
138 | }, | |
8a9562d2 PS |
139 | }; |
140 | ||
141 | static const struct rte_eth_txconf tx_conf = { | |
a28ddd11 DDP |
142 | .tx_thresh = { |
143 | .pthresh = TX_PTHRESH, | |
144 | .hthresh = TX_HTHRESH, | |
145 | .wthresh = TX_WTHRESH, | |
146 | }, | |
147 | .tx_free_thresh = 0, | |
148 | .tx_rs_thresh = 0, | |
94777510 | 149 | .txq_flags = ETH_TXQ_FLAGS_NOMULTSEGS|ETH_TXQ_FLAGS_NOOFFLOADS, |
8a9562d2 PS |
150 | }; |
151 | ||
3a100265 | 152 | enum { MAX_TX_QUEUE_LEN = 384 }; |
58f7c37b DDP |
153 | enum { DPDK_RING_SIZE = 256 }; |
154 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
155 | enum { DRAIN_TSC = 200000ULL }; |
156 | ||
58397e6c KT |
157 | enum dpdk_dev_type { |
158 | DPDK_DEV_ETH = 0, | |
159 | DPDK_DEV_VHOST = 1 | |
160 | }; | |
161 | ||
8a9562d2 PS |
162 | static int rte_eal_init_ret = ENODEV; |
163 | ||
164 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; | |
165 | ||
166 | /* Contains all 'struct dpdk_dev's. */ | |
ca6ba700 | 167 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 168 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 169 | |
ca6ba700 | 170 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 171 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 172 | |
db73f716 DDP |
173 | /* This mutex must be used by non pmd threads when allocating or freeing |
174 | * mbufs through mempools. Since dpdk_queue_pkts() and dpdk_queue_flush() may | |
175 | * use mempools, a non pmd thread should hold this mutex while calling them */ | |
176 | struct ovs_mutex nonpmd_mempool_mutex = OVS_MUTEX_INITIALIZER; | |
177 | ||
8a9562d2 PS |
178 | struct dpdk_mp { |
179 | struct rte_mempool *mp; | |
180 | int mtu; | |
181 | int socket_id; | |
182 | int refcount; | |
ca6ba700 | 183 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
184 | }; |
185 | ||
5a034064 AW |
186 | /* There should be one 'struct dpdk_tx_queue' created for |
187 | * each cpu core. */ | |
8a9562d2 | 188 | struct dpdk_tx_queue { |
94143fc4 AW |
189 | bool flush_tx; /* Set to true to flush queue everytime */ |
190 | /* pkts are queued. */ | |
8a9562d2 PS |
191 | int count; |
192 | uint64_t tsc; | |
193 | struct rte_mbuf *burst_pkts[MAX_TX_QUEUE_LEN]; | |
194 | }; | |
195 | ||
95fb793a | 196 | /* dpdk has no way to remove dpdk ring ethernet devices |
197 | so we have to keep them around once they've been created | |
198 | */ | |
199 | ||
ca6ba700 | 200 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 201 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 202 | |
203 | struct dpdk_ring { | |
204 | /* For the client rings */ | |
205 | struct rte_ring *cring_tx; | |
206 | struct rte_ring *cring_rx; | |
207 | int user_port_id; /* User given port no, parsed from port name */ | |
208 | int eth_port_id; /* ethernet device port id */ | |
ca6ba700 | 209 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 210 | }; |
211 | ||
8a9562d2 PS |
212 | struct netdev_dpdk { |
213 | struct netdev up; | |
214 | int port_id; | |
215 | int max_packet_len; | |
58397e6c | 216 | enum dpdk_dev_type type; |
8a9562d2 | 217 | |
5a034064 | 218 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
219 | |
220 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
221 | ||
222 | struct dpdk_mp *dpdk_mp; | |
223 | int mtu; | |
224 | int socket_id; | |
225 | int buf_size; | |
8a9562d2 PS |
226 | struct netdev_stats stats; |
227 | ||
228 | uint8_t hwaddr[ETH_ADDR_LEN]; | |
229 | enum netdev_flags flags; | |
230 | ||
231 | struct rte_eth_link link; | |
232 | int link_reset_cnt; | |
233 | ||
58397e6c KT |
234 | /* virtio-net structure for vhost device */ |
235 | OVSRCU_TYPE(struct virtio_net *) virtio_dev; | |
236 | ||
8a9562d2 | 237 | /* In dpdk_list. */ |
ca6ba700 | 238 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
58397e6c | 239 | rte_spinlock_t txq_lock; |
8a9562d2 PS |
240 | }; |
241 | ||
242 | struct netdev_rxq_dpdk { | |
243 | struct netdev_rxq up; | |
244 | int port_id; | |
245 | }; | |
246 | ||
db73f716 DDP |
247 | static bool thread_is_pmd(void); |
248 | ||
8a9562d2 PS |
249 | static int netdev_dpdk_construct(struct netdev *); |
250 | ||
58397e6c KT |
251 | struct virtio_net * netdev_dpdk_get_virtio(const struct netdev_dpdk *dev); |
252 | ||
8a9562d2 PS |
253 | static bool |
254 | is_dpdk_class(const struct netdev_class *class) | |
255 | { | |
256 | return class->construct == netdev_dpdk_construct; | |
257 | } | |
258 | ||
58397e6c KT |
259 | /* XXX: use dpdk malloc for entire OVS. in fact huge page should be used |
260 | * for all other segments data, bss and text. */ | |
8a9562d2 PS |
261 | |
262 | static void * | |
263 | dpdk_rte_mzalloc(size_t sz) | |
264 | { | |
265 | void *ptr; | |
266 | ||
267 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
268 | if (ptr == NULL) { | |
269 | out_of_memory(); | |
270 | } | |
271 | return ptr; | |
272 | } | |
273 | ||
db73f716 DDP |
274 | /* XXX this function should be called only by pmd threads (or by non pmd |
275 | * threads holding the nonpmd_mempool_mutex) */ | |
8a9562d2 | 276 | void |
e14deea0 | 277 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 278 | { |
db73f716 | 279 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 280 | |
db73f716 | 281 | rte_pktmbuf_free_seg(pkt); |
8a9562d2 PS |
282 | } |
283 | ||
b3cd9f9d PS |
284 | static void |
285 | __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; | |
e14deea0 | 291 | uint32_t buf_len = mp->elt_size - sizeof(struct dp_packet); |
b3cd9f9d | 292 | |
e14deea0 | 293 | RTE_MBUF_ASSERT(mp->elt_size >= sizeof(struct dp_packet)); |
b3cd9f9d PS |
294 | |
295 | memset(m, 0, mp->elt_size); | |
296 | ||
297 | /* start of buffer is just after mbuf structure */ | |
e14deea0 | 298 | m->buf_addr = (char *)m + sizeof(struct dp_packet); |
b3cd9f9d | 299 | m->buf_physaddr = rte_mempool_virt2phy(mp, m) + |
e14deea0 | 300 | sizeof(struct dp_packet); |
b3cd9f9d PS |
301 | m->buf_len = (uint16_t)buf_len; |
302 | ||
303 | /* keep some headroom between start of buffer and data */ | |
b8e57534 | 304 | m->data_off = RTE_MIN(RTE_PKTMBUF_HEADROOM, m->buf_len); |
b3cd9f9d PS |
305 | |
306 | /* init some constant fields */ | |
b3cd9f9d | 307 | m->pool = mp; |
b8e57534 MK |
308 | m->nb_segs = 1; |
309 | m->port = 0xff; | |
b3cd9f9d PS |
310 | } |
311 | ||
312 | static void | |
313 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
314 | void *opaque_arg OVS_UNUSED, | |
315 | void *_m, | |
316 | unsigned i OVS_UNUSED) | |
317 | { | |
318 | struct rte_mbuf *m = _m; | |
319 | ||
320 | __rte_pktmbuf_init(mp, opaque_arg, _m, i); | |
321 | ||
cf62fa4c | 322 | dp_packet_init_dpdk((struct dp_packet *) m, m->buf_len); |
b3cd9f9d PS |
323 | } |
324 | ||
8a9562d2 PS |
325 | static struct dpdk_mp * |
326 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
327 | { | |
328 | struct dpdk_mp *dmp = NULL; | |
329 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
da79ce2b | 330 | unsigned mp_size; |
8a9562d2 PS |
331 | |
332 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
333 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
334 | dmp->refcount++; | |
335 | return dmp; | |
336 | } | |
337 | } | |
338 | ||
339 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
340 | dmp->socket_id = socket_id; | |
341 | dmp->mtu = mtu; | |
342 | dmp->refcount = 1; | |
343 | ||
da79ce2b DDP |
344 | mp_size = MAX_NB_MBUF; |
345 | do { | |
346 | if (snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d_%d_%u", | |
347 | dmp->mtu, dmp->socket_id, mp_size) < 0) { | |
348 | return NULL; | |
349 | } | |
95fb793a | 350 | |
da79ce2b DDP |
351 | dmp->mp = rte_mempool_create(mp_name, mp_size, MBUF_SIZE(mtu), |
352 | MP_CACHE_SZ, | |
353 | sizeof(struct rte_pktmbuf_pool_private), | |
354 | rte_pktmbuf_pool_init, NULL, | |
355 | ovs_rte_pktmbuf_init, NULL, | |
356 | socket_id, 0); | |
357 | } while (!dmp->mp && rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
8a9562d2 PS |
358 | |
359 | if (dmp->mp == NULL) { | |
360 | return NULL; | |
da79ce2b DDP |
361 | } else { |
362 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", mp_name, mp_size ); | |
8a9562d2 PS |
363 | } |
364 | ||
365 | list_push_back(&dpdk_mp_list, &dmp->list_node); | |
366 | return dmp; | |
367 | } | |
368 | ||
369 | static void | |
370 | dpdk_mp_put(struct dpdk_mp *dmp) | |
371 | { | |
372 | ||
373 | if (!dmp) { | |
374 | return; | |
375 | } | |
376 | ||
377 | dmp->refcount--; | |
378 | ovs_assert(dmp->refcount >= 0); | |
379 | ||
380 | #if 0 | |
381 | /* I could not find any API to destroy mp. */ | |
382 | if (dmp->refcount == 0) { | |
383 | list_delete(dmp->list_node); | |
384 | /* destroy mp-pool. */ | |
385 | } | |
386 | #endif | |
387 | } | |
388 | ||
389 | static void | |
390 | check_link_status(struct netdev_dpdk *dev) | |
391 | { | |
392 | struct rte_eth_link link; | |
393 | ||
394 | rte_eth_link_get_nowait(dev->port_id, &link); | |
395 | ||
396 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 397 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
398 | |
399 | dev->link_reset_cnt++; | |
400 | dev->link = link; | |
401 | if (dev->link.link_status) { | |
402 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
403 | dev->port_id, (unsigned)dev->link.link_speed, | |
404 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
405 | ("full-duplex") : ("half-duplex")); | |
406 | } else { | |
407 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
408 | } | |
409 | } | |
410 | } | |
411 | ||
412 | static void * | |
413 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
414 | { | |
415 | struct netdev_dpdk *dev; | |
416 | ||
417 | pthread_detach(pthread_self()); | |
418 | ||
419 | for (;;) { | |
420 | ovs_mutex_lock(&dpdk_mutex); | |
421 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
422 | ovs_mutex_lock(&dev->mutex); | |
423 | check_link_status(dev); | |
424 | ovs_mutex_unlock(&dev->mutex); | |
425 | } | |
426 | ovs_mutex_unlock(&dpdk_mutex); | |
427 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
428 | } | |
429 | ||
430 | return NULL; | |
431 | } | |
432 | ||
433 | static int | |
434 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
435 | { | |
436 | struct rte_pktmbuf_pool_private *mbp_priv; | |
437 | struct ether_addr eth_addr; | |
438 | int diag; | |
439 | int i; | |
440 | ||
441 | if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) { | |
95fb793a | 442 | return ENODEV; |
8a9562d2 PS |
443 | } |
444 | ||
5496878c AW |
445 | diag = rte_eth_dev_configure(dev->port_id, dev->up.n_rxq, dev->up.n_txq, |
446 | &port_conf); | |
8a9562d2 PS |
447 | if (diag) { |
448 | VLOG_ERR("eth dev config error %d",diag); | |
95fb793a | 449 | return -diag; |
8a9562d2 PS |
450 | } |
451 | ||
5496878c | 452 | for (i = 0; i < dev->up.n_txq; i++) { |
79f5354c | 453 | diag = rte_eth_tx_queue_setup(dev->port_id, i, NIC_PORT_TX_Q_SIZE, |
d221ffa1 | 454 | dev->socket_id, &tx_conf); |
8a9562d2 PS |
455 | if (diag) { |
456 | VLOG_ERR("eth dev tx queue setup error %d",diag); | |
95fb793a | 457 | return -diag; |
8a9562d2 PS |
458 | } |
459 | } | |
460 | ||
5496878c | 461 | for (i = 0; i < dev->up.n_rxq; i++) { |
79f5354c | 462 | diag = rte_eth_rx_queue_setup(dev->port_id, i, NIC_PORT_RX_Q_SIZE, |
d221ffa1 | 463 | dev->socket_id, |
a715f600 | 464 | &rx_conf, dev->dpdk_mp->mp); |
8a9562d2 PS |
465 | if (diag) { |
466 | VLOG_ERR("eth dev rx queue setup error %d",diag); | |
95fb793a | 467 | return -diag; |
8a9562d2 PS |
468 | } |
469 | } | |
470 | ||
471 | diag = rte_eth_dev_start(dev->port_id); | |
472 | if (diag) { | |
473 | VLOG_ERR("eth dev start error %d",diag); | |
95fb793a | 474 | return -diag; |
8a9562d2 PS |
475 | } |
476 | ||
477 | rte_eth_promiscuous_enable(dev->port_id); | |
478 | rte_eth_allmulticast_enable(dev->port_id); | |
479 | ||
480 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
481 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
482 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
483 | dev->port_id, ETH_ADDR_ARGS(eth_addr.addr_bytes)); | |
484 | ||
485 | memcpy(dev->hwaddr, eth_addr.addr_bytes, ETH_ADDR_LEN); | |
486 | rte_eth_link_get_nowait(dev->port_id, &dev->link); | |
487 | ||
488 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
489 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
490 | ||
491 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
492 | return 0; | |
493 | } | |
494 | ||
495 | static struct netdev_dpdk * | |
496 | netdev_dpdk_cast(const struct netdev *netdev) | |
497 | { | |
498 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
499 | } | |
500 | ||
501 | static struct netdev * | |
502 | netdev_dpdk_alloc(void) | |
503 | { | |
504 | struct netdev_dpdk *netdev = dpdk_rte_mzalloc(sizeof *netdev); | |
505 | return &netdev->up; | |
506 | } | |
507 | ||
5a034064 | 508 | static void |
91968eb0 | 509 | netdev_dpdk_alloc_txq(struct netdev_dpdk *netdev, unsigned int n_txqs) |
5a034064 AW |
510 | { |
511 | int i; | |
512 | ||
513 | netdev->tx_q = dpdk_rte_mzalloc(n_txqs * sizeof *netdev->tx_q); | |
94143fc4 AW |
514 | /* Each index is considered as a cpu core id, since there should |
515 | * be one tx queue for each cpu core. */ | |
5a034064 | 516 | for (i = 0; i < n_txqs; i++) { |
ba0358a1 | 517 | int numa_id = ovs_numa_get_numa_id(i); |
94143fc4 | 518 | |
94143fc4 AW |
519 | /* If the corresponding core is not on the same numa node |
520 | * as 'netdev', flags the 'flush_tx'. */ | |
ba0358a1 | 521 | netdev->tx_q[i].flush_tx = netdev->socket_id == numa_id; |
5a034064 AW |
522 | } |
523 | } | |
524 | ||
8a9562d2 | 525 | static int |
58397e6c KT |
526 | netdev_dpdk_init(struct netdev *netdev_, unsigned int port_no, |
527 | enum dpdk_dev_type type) | |
5a034064 | 528 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 PS |
529 | { |
530 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1b7a04e0 | 531 | int sid; |
95fb793a | 532 | int err = 0; |
8a9562d2 | 533 | |
95fb793a | 534 | ovs_mutex_init(&netdev->mutex); |
95fb793a | 535 | ovs_mutex_lock(&netdev->mutex); |
8a9562d2 | 536 | |
1b7a04e0 AW |
537 | /* If the 'sid' is negative, it means that the kernel fails |
538 | * to obtain the pci numa info. In that situation, always | |
539 | * use 'SOCKET0'. */ | |
58397e6c KT |
540 | if (type == DPDK_DEV_ETH) { |
541 | sid = rte_eth_dev_socket_id(port_no); | |
542 | } else { | |
543 | sid = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
544 | } | |
545 | ||
1b7a04e0 | 546 | netdev->socket_id = sid < 0 ? SOCKET0 : sid; |
95fb793a | 547 | netdev->port_id = port_no; |
58397e6c | 548 | netdev->type = type; |
8a9562d2 | 549 | netdev->flags = 0; |
8a9562d2 PS |
550 | netdev->mtu = ETHER_MTU; |
551 | netdev->max_packet_len = MTU_TO_MAX_LEN(netdev->mtu); | |
58397e6c | 552 | rte_spinlock_init(&netdev->txq_lock); |
8a9562d2 | 553 | |
8a9562d2 PS |
554 | netdev->dpdk_mp = dpdk_mp_get(netdev->socket_id, netdev->mtu); |
555 | if (!netdev->dpdk_mp) { | |
556 | err = ENOMEM; | |
95fb793a | 557 | goto unlock; |
8a9562d2 PS |
558 | } |
559 | ||
5496878c AW |
560 | netdev_->n_txq = NR_QUEUE; |
561 | netdev_->n_rxq = NR_QUEUE; | |
58397e6c KT |
562 | |
563 | if (type == DPDK_DEV_ETH) { | |
1b99bb05 MG |
564 | netdev_dpdk_alloc_txq(netdev, NR_QUEUE); |
565 | err = dpdk_eth_dev_init(netdev); | |
566 | if (err) { | |
567 | goto unlock; | |
568 | } | |
8a9562d2 | 569 | } |
8a9562d2 PS |
570 | |
571 | list_push_back(&dpdk_list, &netdev->list_node); | |
572 | ||
95fb793a | 573 | unlock: |
5a034064 AW |
574 | if (err) { |
575 | rte_free(netdev->tx_q); | |
576 | } | |
8a9562d2 | 577 | ovs_mutex_unlock(&netdev->mutex); |
95fb793a | 578 | return err; |
579 | } | |
580 | ||
581 | static int | |
582 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
583 | unsigned int *port_no) | |
584 | { | |
585 | const char *cport; | |
586 | ||
587 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
588 | return ENODEV; | |
589 | } | |
590 | ||
591 | cport = dev_name + strlen(prefix); | |
592 | *port_no = strtol(cport, 0, 0); /* string must be null terminated */ | |
593 | return 0; | |
594 | } | |
595 | ||
58397e6c KT |
596 | static int |
597 | netdev_dpdk_vhost_construct(struct netdev *netdev_) | |
598 | { | |
599 | int err; | |
600 | ||
601 | if (rte_eal_init_ret) { | |
602 | return rte_eal_init_ret; | |
603 | } | |
604 | ||
605 | ovs_mutex_lock(&dpdk_mutex); | |
606 | err = netdev_dpdk_init(netdev_, -1, DPDK_DEV_VHOST); | |
607 | ovs_mutex_unlock(&dpdk_mutex); | |
608 | ||
609 | return err; | |
610 | } | |
611 | ||
95fb793a | 612 | static int |
613 | netdev_dpdk_construct(struct netdev *netdev) | |
614 | { | |
615 | unsigned int port_no; | |
616 | int err; | |
617 | ||
618 | if (rte_eal_init_ret) { | |
619 | return rte_eal_init_ret; | |
620 | } | |
621 | ||
622 | /* Names always start with "dpdk" */ | |
623 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
624 | if (err) { | |
625 | return err; | |
626 | } | |
627 | ||
628 | ovs_mutex_lock(&dpdk_mutex); | |
58397e6c | 629 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
8a9562d2 PS |
630 | ovs_mutex_unlock(&dpdk_mutex); |
631 | return err; | |
632 | } | |
633 | ||
634 | static void | |
635 | netdev_dpdk_destruct(struct netdev *netdev_) | |
636 | { | |
637 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
638 | ||
639 | ovs_mutex_lock(&dev->mutex); | |
640 | rte_eth_dev_stop(dev->port_id); | |
641 | ovs_mutex_unlock(&dev->mutex); | |
642 | ||
643 | ovs_mutex_lock(&dpdk_mutex); | |
5a034064 | 644 | rte_free(dev->tx_q); |
8a9562d2 PS |
645 | list_remove(&dev->list_node); |
646 | dpdk_mp_put(dev->dpdk_mp); | |
647 | ovs_mutex_unlock(&dpdk_mutex); | |
58397e6c | 648 | } |
8a9562d2 | 649 | |
58397e6c KT |
650 | static void |
651 | netdev_dpdk_vhost_destruct(struct netdev *netdev_) | |
652 | { | |
653 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
654 | ||
655 | /* Can't remove a port while a guest is attached to it. */ | |
656 | if (netdev_dpdk_get_virtio(dev) != NULL) { | |
657 | VLOG_ERR("Can not remove port, vhost device still attached"); | |
658 | return; | |
659 | } | |
660 | ||
661 | ovs_mutex_lock(&dpdk_mutex); | |
662 | list_remove(&dev->list_node); | |
663 | dpdk_mp_put(dev->dpdk_mp); | |
664 | ovs_mutex_unlock(&dpdk_mutex); | |
8a9562d2 PS |
665 | } |
666 | ||
667 | static void | |
668 | netdev_dpdk_dealloc(struct netdev *netdev_) | |
669 | { | |
670 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
671 | ||
672 | rte_free(netdev); | |
673 | } | |
674 | ||
675 | static int | |
676 | netdev_dpdk_get_config(const struct netdev *netdev_, struct smap *args) | |
677 | { | |
678 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
679 | ||
680 | ovs_mutex_lock(&dev->mutex); | |
681 | ||
47659062 AW |
682 | smap_add_format(args, "configured_rx_queues", "%d", netdev_->n_rxq); |
683 | smap_add_format(args, "configured_tx_queues", "%d", netdev_->n_txq); | |
8a9562d2 PS |
684 | ovs_mutex_unlock(&dev->mutex); |
685 | ||
686 | return 0; | |
687 | } | |
688 | ||
7dec44fe AW |
689 | static int |
690 | netdev_dpdk_get_numa_id(const struct netdev *netdev_) | |
691 | { | |
692 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
693 | ||
694 | return netdev->socket_id; | |
695 | } | |
696 | ||
5496878c AW |
697 | /* Sets the number of tx queues and rx queues for the dpdk interface. |
698 | * If the configuration fails, do not try restoring its old configuration | |
699 | * and just returns the error. */ | |
700 | static int | |
701 | netdev_dpdk_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
702 | unsigned int n_rxq) | |
703 | { | |
704 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
705 | int err = 0; | |
706 | ||
707 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
708 | return err; | |
709 | } | |
710 | ||
b7ccaf67 | 711 | ovs_mutex_lock(&dpdk_mutex); |
5496878c | 712 | ovs_mutex_lock(&netdev->mutex); |
91968eb0 | 713 | |
5496878c | 714 | rte_eth_dev_stop(netdev->port_id); |
91968eb0 | 715 | |
5496878c AW |
716 | netdev->up.n_txq = n_txq; |
717 | netdev->up.n_rxq = n_rxq; | |
58397e6c | 718 | |
91968eb0 AW |
719 | rte_free(netdev->tx_q); |
720 | netdev_dpdk_alloc_txq(netdev, n_txq); | |
5496878c | 721 | err = dpdk_eth_dev_init(netdev); |
91968eb0 | 722 | |
5496878c | 723 | ovs_mutex_unlock(&netdev->mutex); |
b7ccaf67 | 724 | ovs_mutex_unlock(&dpdk_mutex); |
5496878c AW |
725 | |
726 | return err; | |
727 | } | |
728 | ||
58397e6c KT |
729 | static int |
730 | netdev_dpdk_vhost_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
731 | unsigned int n_rxq) | |
732 | { | |
733 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
734 | int err = 0; | |
735 | ||
736 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
737 | return err; | |
738 | } | |
739 | ||
740 | ovs_mutex_lock(&dpdk_mutex); | |
741 | ovs_mutex_lock(&netdev->mutex); | |
742 | ||
743 | netdev->up.n_txq = n_txq; | |
744 | netdev->up.n_rxq = n_rxq; | |
745 | ||
746 | ovs_mutex_unlock(&netdev->mutex); | |
747 | ovs_mutex_unlock(&dpdk_mutex); | |
748 | ||
749 | return err; | |
750 | } | |
751 | ||
8a9562d2 PS |
752 | static struct netdev_rxq * |
753 | netdev_dpdk_rxq_alloc(void) | |
754 | { | |
755 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
756 | ||
757 | return &rx->up; | |
758 | } | |
759 | ||
760 | static struct netdev_rxq_dpdk * | |
761 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rx) | |
762 | { | |
763 | return CONTAINER_OF(rx, struct netdev_rxq_dpdk, up); | |
764 | } | |
765 | ||
766 | static int | |
767 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq_) | |
768 | { | |
769 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
770 | struct netdev_dpdk *netdev = netdev_dpdk_cast(rx->up.netdev); | |
771 | ||
772 | ovs_mutex_lock(&netdev->mutex); | |
773 | rx->port_id = netdev->port_id; | |
774 | ovs_mutex_unlock(&netdev->mutex); | |
775 | ||
776 | return 0; | |
777 | } | |
778 | ||
779 | static void | |
780 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq_ OVS_UNUSED) | |
781 | { | |
782 | } | |
783 | ||
784 | static void | |
785 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq_) | |
786 | { | |
787 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
788 | ||
789 | rte_free(rx); | |
790 | } | |
791 | ||
b170db2a RW |
792 | static inline void |
793 | dpdk_queue_flush__(struct netdev_dpdk *dev, int qid) | |
8a9562d2 PS |
794 | { |
795 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1304f1f8 DDP |
796 | uint32_t nb_tx = 0; |
797 | ||
798 | while (nb_tx != txq->count) { | |
799 | uint32_t ret; | |
800 | ||
801 | ret = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts + nb_tx, | |
802 | txq->count - nb_tx); | |
803 | if (!ret) { | |
804 | break; | |
805 | } | |
806 | ||
807 | nb_tx += ret; | |
808 | } | |
8a9562d2 | 809 | |
b170db2a | 810 | if (OVS_UNLIKELY(nb_tx != txq->count)) { |
db73f716 DDP |
811 | /* free buffers, which we couldn't transmit, one at a time (each |
812 | * packet could come from a different mempool) */ | |
813 | int i; | |
814 | ||
815 | for (i = nb_tx; i < txq->count; i++) { | |
816 | rte_pktmbuf_free_seg(txq->burst_pkts[i]); | |
817 | } | |
1304f1f8 DDP |
818 | ovs_mutex_lock(&dev->mutex); |
819 | dev->stats.tx_dropped += txq->count-nb_tx; | |
820 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 | 821 | } |
1304f1f8 | 822 | |
8a9562d2 | 823 | txq->count = 0; |
844f2d74 | 824 | txq->tsc = rte_get_timer_cycles(); |
b170db2a RW |
825 | } |
826 | ||
827 | static inline void | |
828 | dpdk_queue_flush(struct netdev_dpdk *dev, int qid) | |
829 | { | |
830 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
831 | ||
832 | if (txq->count == 0) { | |
833 | return; | |
834 | } | |
b170db2a | 835 | dpdk_queue_flush__(dev, qid); |
8a9562d2 PS |
836 | } |
837 | ||
58397e6c KT |
838 | static bool |
839 | is_vhost_running(struct virtio_net *dev) | |
840 | { | |
841 | return (dev != NULL && (dev->flags & VIRTIO_DEV_RUNNING)); | |
842 | } | |
843 | ||
844 | /* | |
845 | * The receive path for the vhost port is the TX path out from guest. | |
846 | */ | |
847 | static int | |
848 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq_, | |
849 | struct dp_packet **packets, int *c) | |
850 | { | |
851 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
852 | struct netdev *netdev = rx->up.netdev; | |
853 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
854 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
855 | int qid = 1; | |
856 | uint16_t nb_rx = 0; | |
857 | ||
858 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
859 | return EAGAIN; | |
860 | } | |
861 | ||
862 | nb_rx = rte_vhost_dequeue_burst(virtio_dev, qid, | |
863 | vhost_dev->dpdk_mp->mp, | |
864 | (struct rte_mbuf **)packets, | |
865 | MAX_PKT_BURST); | |
866 | if (!nb_rx) { | |
867 | return EAGAIN; | |
868 | } | |
869 | ||
870 | vhost_dev->stats.rx_packets += (uint64_t)nb_rx; | |
871 | *c = (int) nb_rx; | |
872 | return 0; | |
873 | } | |
874 | ||
8a9562d2 | 875 | static int |
e14deea0 | 876 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet **packets, |
91088554 | 877 | int *c) |
8a9562d2 PS |
878 | { |
879 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
880 | struct netdev *netdev = rx->up.netdev; | |
881 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 882 | int nb_rx; |
8a9562d2 | 883 | |
5496878c AW |
884 | /* There is only one tx queue for this core. Do not flush other |
885 | * queueus. */ | |
886 | if (rxq_->queue_id == rte_lcore_id()) { | |
887 | dpdk_queue_flush(dev, rxq_->queue_id); | |
888 | } | |
8a9562d2 PS |
889 | |
890 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq_->queue_id, | |
7d08d53e | 891 | (struct rte_mbuf **) packets, |
4345e1b5 KT |
892 | MIN((int) NETDEV_MAX_RX_BATCH, |
893 | (int) MAX_PKT_BURST)); | |
8a9562d2 PS |
894 | if (!nb_rx) { |
895 | return EAGAIN; | |
896 | } | |
897 | ||
8a9562d2 PS |
898 | *c = nb_rx; |
899 | ||
900 | return 0; | |
901 | } | |
902 | ||
58397e6c KT |
903 | static void |
904 | __netdev_dpdk_vhost_send(struct netdev *netdev, struct dp_packet **pkts, | |
905 | int cnt, bool may_steal) | |
906 | { | |
907 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
908 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
95e9881f KT |
909 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
910 | unsigned int total_pkts = cnt; | |
911 | uint64_t start = 0; | |
58397e6c KT |
912 | |
913 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
1b99bb05 MG |
914 | ovs_mutex_lock(&vhost_dev->mutex); |
915 | vhost_dev->stats.tx_dropped+= cnt; | |
916 | ovs_mutex_unlock(&vhost_dev->mutex); | |
917 | goto out; | |
58397e6c KT |
918 | } |
919 | ||
920 | /* There is vHost TX single queue, So we need to lock it for TX. */ | |
921 | rte_spinlock_lock(&vhost_dev->txq_lock); | |
58397e6c | 922 | |
95e9881f KT |
923 | do { |
924 | unsigned int tx_pkts; | |
925 | ||
926 | tx_pkts = rte_vhost_enqueue_burst(virtio_dev, VIRTIO_RXQ, | |
927 | cur_pkts, cnt); | |
928 | if (OVS_LIKELY(tx_pkts)) { | |
929 | /* Packets have been sent.*/ | |
930 | cnt -= tx_pkts; | |
931 | /* Prepare for possible next iteration.*/ | |
932 | cur_pkts = &cur_pkts[tx_pkts]; | |
933 | } else { | |
934 | uint64_t timeout = VHOST_ENQ_RETRY_USECS * rte_get_timer_hz() / 1E6; | |
935 | unsigned int expired = 0; | |
936 | ||
937 | if (!start) { | |
938 | start = rte_get_timer_cycles(); | |
939 | } | |
940 | ||
941 | /* | |
942 | * Unable to enqueue packets to vhost interface. | |
943 | * Check available entries before retrying. | |
944 | */ | |
945 | while (!rte_vring_available_entries(virtio_dev, VIRTIO_RXQ)) { | |
946 | if (OVS_UNLIKELY((rte_get_timer_cycles() - start) > timeout)) { | |
947 | expired = 1; | |
948 | break; | |
949 | } | |
950 | } | |
951 | if (expired) { | |
952 | /* break out of main loop. */ | |
953 | break; | |
954 | } | |
955 | } | |
956 | } while (cnt); | |
957 | ||
958 | vhost_dev->stats.tx_packets += (total_pkts - cnt); | |
959 | vhost_dev->stats.tx_dropped += cnt; | |
58397e6c KT |
960 | rte_spinlock_unlock(&vhost_dev->txq_lock); |
961 | ||
962 | out: | |
963 | if (may_steal) { | |
95e9881f KT |
964 | int i; |
965 | ||
966 | for (i = 0; i < total_pkts; i++) { | |
1b99bb05 MG |
967 | dp_packet_delete(pkts[i]); |
968 | } | |
58397e6c KT |
969 | } |
970 | } | |
971 | ||
8a9562d2 | 972 | inline static void |
f4fd623c DDP |
973 | dpdk_queue_pkts(struct netdev_dpdk *dev, int qid, |
974 | struct rte_mbuf **pkts, int cnt) | |
8a9562d2 PS |
975 | { |
976 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
977 | uint64_t diff_tsc; | |
8a9562d2 | 978 | |
f4fd623c DDP |
979 | int i = 0; |
980 | ||
f4fd623c DDP |
981 | while (i < cnt) { |
982 | int freeslots = MAX_TX_QUEUE_LEN - txq->count; | |
983 | int tocopy = MIN(freeslots, cnt-i); | |
8a9562d2 | 984 | |
f4fd623c DDP |
985 | memcpy(&txq->burst_pkts[txq->count], &pkts[i], |
986 | tocopy * sizeof (struct rte_mbuf *)); | |
987 | ||
988 | txq->count += tocopy; | |
989 | i += tocopy; | |
990 | ||
94143fc4 | 991 | if (txq->count == MAX_TX_QUEUE_LEN || txq->flush_tx) { |
b170db2a | 992 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 993 | } |
844f2d74 | 994 | diff_tsc = rte_get_timer_cycles() - txq->tsc; |
f4fd623c | 995 | if (diff_tsc >= DRAIN_TSC) { |
b170db2a | 996 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 997 | } |
8a9562d2 | 998 | } |
8a9562d2 PS |
999 | } |
1000 | ||
1001 | /* Tx function. Transmit packets indefinitely */ | |
1002 | static void | |
58397e6c | 1003 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet **pkts, |
2654cc33 | 1004 | int cnt) |
db73f716 | 1005 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 PS |
1006 | { |
1007 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
f4fd623c | 1008 | struct rte_mbuf *mbufs[cnt]; |
175cf4de RW |
1009 | int dropped = 0; |
1010 | int newcnt = 0; | |
1011 | int i; | |
8a9562d2 | 1012 | |
db73f716 DDP |
1013 | /* If we are on a non pmd thread we have to use the mempool mutex, because |
1014 | * every non pmd thread shares the same mempool cache */ | |
1015 | ||
1016 | if (!thread_is_pmd()) { | |
1017 | ovs_mutex_lock(&nonpmd_mempool_mutex); | |
1018 | } | |
1019 | ||
f4fd623c | 1020 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1021 | int size = dp_packet_size(pkts[i]); |
95fb793a | 1022 | |
f98d7864 | 1023 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c DDP |
1024 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1025 | (int)size , dev->max_packet_len); | |
1026 | ||
175cf4de | 1027 | dropped++; |
f4fd623c DDP |
1028 | continue; |
1029 | } | |
8a9562d2 | 1030 | |
f4fd623c | 1031 | mbufs[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1032 | |
f4fd623c | 1033 | if (!mbufs[newcnt]) { |
175cf4de RW |
1034 | dropped += cnt - i; |
1035 | break; | |
f4fd623c DDP |
1036 | } |
1037 | ||
1038 | /* We have to do a copy for now */ | |
b8e57534 | 1039 | memcpy(rte_pktmbuf_mtod(mbufs[newcnt], void *), dp_packet_data(pkts[i]), size); |
f4fd623c DDP |
1040 | |
1041 | rte_pktmbuf_data_len(mbufs[newcnt]) = size; | |
1042 | rte_pktmbuf_pkt_len(mbufs[newcnt]) = size; | |
1043 | ||
1044 | newcnt++; | |
1045 | } | |
8a9562d2 | 1046 | |
f98d7864 | 1047 | if (OVS_UNLIKELY(dropped)) { |
175cf4de RW |
1048 | ovs_mutex_lock(&dev->mutex); |
1049 | dev->stats.tx_dropped += dropped; | |
1050 | ovs_mutex_unlock(&dev->mutex); | |
1051 | } | |
1052 | ||
58397e6c KT |
1053 | if (dev->type == DPDK_DEV_VHOST) { |
1054 | __netdev_dpdk_vhost_send(netdev, (struct dp_packet **) mbufs, newcnt, true); | |
1055 | } else { | |
1056 | dpdk_queue_pkts(dev, qid, mbufs, newcnt); | |
1057 | dpdk_queue_flush(dev, qid); | |
1058 | } | |
db73f716 DDP |
1059 | |
1060 | if (!thread_is_pmd()) { | |
1061 | ovs_mutex_unlock(&nonpmd_mempool_mutex); | |
1062 | } | |
8a9562d2 PS |
1063 | } |
1064 | ||
58397e6c KT |
1065 | static int |
1066 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid OVS_UNUSED, struct dp_packet **pkts, | |
1067 | int cnt, bool may_steal) | |
1068 | { | |
1069 | if (OVS_UNLIKELY(pkts[0]->source != DPBUF_DPDK)) { | |
1070 | int i; | |
1071 | ||
1072 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); | |
1073 | if (may_steal) { | |
1074 | for (i = 0; i < cnt; i++) { | |
1075 | dp_packet_delete(pkts[i]); | |
1076 | } | |
1077 | } | |
1078 | } else { | |
1079 | __netdev_dpdk_vhost_send(netdev, pkts, cnt, may_steal); | |
1080 | } | |
1081 | return 0; | |
1082 | } | |
1083 | ||
7251515e DV |
1084 | static inline void |
1085 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
e14deea0 | 1086 | struct dp_packet **pkts, int cnt, bool may_steal) |
8a9562d2 | 1087 | { |
f4fd623c | 1088 | int i; |
8a9562d2 | 1089 | |
7251515e | 1090 | if (OVS_UNLIKELY(!may_steal || |
cf62fa4c | 1091 | pkts[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1092 | struct netdev *netdev = &dev->up; |
1093 | ||
2654cc33 | 1094 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); |
b3cd9f9d PS |
1095 | |
1096 | if (may_steal) { | |
f4fd623c | 1097 | for (i = 0; i < cnt; i++) { |
e14deea0 | 1098 | dp_packet_delete(pkts[i]); |
f4fd623c | 1099 | } |
b3cd9f9d | 1100 | } |
8a9562d2 | 1101 | } else { |
f4fd623c DDP |
1102 | int next_tx_idx = 0; |
1103 | int dropped = 0; | |
8a9562d2 | 1104 | |
f4fd623c | 1105 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1106 | int size = dp_packet_size(pkts[i]); |
1b99bb05 | 1107 | |
f4fd623c DDP |
1108 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
1109 | if (next_tx_idx != i) { | |
1110 | dpdk_queue_pkts(dev, qid, | |
1111 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1112 | i-next_tx_idx); | |
1ebfe1ac | 1113 | } |
f4fd623c | 1114 | |
1ebfe1ac DDP |
1115 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1116 | (int)size , dev->max_packet_len); | |
f4fd623c | 1117 | |
e14deea0 | 1118 | dp_packet_delete(pkts[i]); |
1ebfe1ac | 1119 | dropped++; |
f4fd623c DDP |
1120 | next_tx_idx = i + 1; |
1121 | } | |
1122 | } | |
1123 | if (next_tx_idx != cnt) { | |
1124 | dpdk_queue_pkts(dev, qid, | |
1125 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1126 | cnt-next_tx_idx); | |
1127 | } | |
8a9562d2 | 1128 | |
f4fd623c DDP |
1129 | if (OVS_UNLIKELY(dropped)) { |
1130 | ovs_mutex_lock(&dev->mutex); | |
1131 | dev->stats.tx_dropped += dropped; | |
1132 | ovs_mutex_unlock(&dev->mutex); | |
1133 | } | |
8a9562d2 | 1134 | } |
7251515e DV |
1135 | } |
1136 | ||
1137 | static int | |
1138 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
e14deea0 | 1139 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1140 | { |
1141 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1142 | |
7251515e DV |
1143 | netdev_dpdk_send__(dev, qid, pkts, cnt, may_steal); |
1144 | return 0; | |
8a9562d2 PS |
1145 | } |
1146 | ||
1147 | static int | |
1148 | netdev_dpdk_set_etheraddr(struct netdev *netdev, | |
1149 | const uint8_t mac[ETH_ADDR_LEN]) | |
1150 | { | |
1151 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1152 | ||
1153 | ovs_mutex_lock(&dev->mutex); | |
1154 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
1155 | memcpy(dev->hwaddr, mac, ETH_ADDR_LEN); | |
045c0d1a | 1156 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1157 | } |
1158 | ovs_mutex_unlock(&dev->mutex); | |
1159 | ||
1160 | return 0; | |
1161 | } | |
1162 | ||
1163 | static int | |
1164 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, | |
1165 | uint8_t mac[ETH_ADDR_LEN]) | |
1166 | { | |
1167 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1168 | ||
1169 | ovs_mutex_lock(&dev->mutex); | |
1170 | memcpy(mac, dev->hwaddr, ETH_ADDR_LEN); | |
1171 | ovs_mutex_unlock(&dev->mutex); | |
1172 | ||
1173 | return 0; | |
1174 | } | |
1175 | ||
1176 | static int | |
1177 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1178 | { | |
1179 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1180 | ||
1181 | ovs_mutex_lock(&dev->mutex); | |
1182 | *mtup = dev->mtu; | |
1183 | ovs_mutex_unlock(&dev->mutex); | |
1184 | ||
1185 | return 0; | |
1186 | } | |
1187 | ||
1188 | static int | |
1189 | netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu) | |
1190 | { | |
1191 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1192 | int old_mtu, err; | |
1193 | struct dpdk_mp *old_mp; | |
1194 | struct dpdk_mp *mp; | |
1195 | ||
1196 | ovs_mutex_lock(&dpdk_mutex); | |
1197 | ovs_mutex_lock(&dev->mutex); | |
1198 | if (dev->mtu == mtu) { | |
1199 | err = 0; | |
1200 | goto out; | |
1201 | } | |
1202 | ||
1203 | mp = dpdk_mp_get(dev->socket_id, dev->mtu); | |
1204 | if (!mp) { | |
1205 | err = ENOMEM; | |
1206 | goto out; | |
1207 | } | |
1208 | ||
1209 | rte_eth_dev_stop(dev->port_id); | |
1210 | ||
1211 | old_mtu = dev->mtu; | |
1212 | old_mp = dev->dpdk_mp; | |
1213 | dev->dpdk_mp = mp; | |
1214 | dev->mtu = mtu; | |
1215 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
1216 | ||
1217 | err = dpdk_eth_dev_init(dev); | |
1218 | if (err) { | |
8a9562d2 PS |
1219 | dpdk_mp_put(mp); |
1220 | dev->mtu = old_mtu; | |
1221 | dev->dpdk_mp = old_mp; | |
1222 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
1223 | dpdk_eth_dev_init(dev); | |
1224 | goto out; | |
1225 | } | |
1226 | ||
1227 | dpdk_mp_put(old_mp); | |
045c0d1a | 1228 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1229 | out: |
1230 | ovs_mutex_unlock(&dev->mutex); | |
1231 | ovs_mutex_unlock(&dpdk_mutex); | |
1232 | return err; | |
1233 | } | |
1234 | ||
1235 | static int | |
1236 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier); | |
1237 | ||
58397e6c KT |
1238 | static int |
1239 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1240 | struct netdev_stats *stats) | |
1241 | { | |
1242 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1243 | ||
1244 | ovs_mutex_lock(&dev->mutex); | |
1245 | memset(stats, 0, sizeof(*stats)); | |
1246 | /* Unsupported Stats */ | |
1247 | stats->rx_errors = UINT64_MAX; | |
1248 | stats->tx_errors = UINT64_MAX; | |
1249 | stats->multicast = UINT64_MAX; | |
1250 | stats->collisions = UINT64_MAX; | |
1251 | stats->rx_crc_errors = UINT64_MAX; | |
1252 | stats->rx_fifo_errors = UINT64_MAX; | |
1253 | stats->rx_frame_errors = UINT64_MAX; | |
1254 | stats->rx_length_errors = UINT64_MAX; | |
1255 | stats->rx_missed_errors = UINT64_MAX; | |
1256 | stats->rx_over_errors = UINT64_MAX; | |
1257 | stats->tx_aborted_errors = UINT64_MAX; | |
1258 | stats->tx_carrier_errors = UINT64_MAX; | |
1259 | stats->tx_errors = UINT64_MAX; | |
1260 | stats->tx_fifo_errors = UINT64_MAX; | |
1261 | stats->tx_heartbeat_errors = UINT64_MAX; | |
1262 | stats->tx_window_errors = UINT64_MAX; | |
1263 | stats->rx_bytes += UINT64_MAX; | |
1264 | stats->rx_dropped += UINT64_MAX; | |
1265 | stats->tx_bytes += UINT64_MAX; | |
1266 | ||
1267 | /* Supported Stats */ | |
1268 | stats->rx_packets += dev->stats.rx_packets; | |
1269 | stats->tx_packets += dev->stats.tx_packets; | |
1270 | stats->tx_dropped += dev->stats.tx_dropped; | |
1271 | ovs_mutex_unlock(&dev->mutex); | |
1272 | ||
1273 | return 0; | |
1274 | } | |
1275 | ||
8a9562d2 PS |
1276 | static int |
1277 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1278 | { | |
1279 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1280 | struct rte_eth_stats rte_stats; | |
1281 | bool gg; | |
1282 | ||
1283 | netdev_dpdk_get_carrier(netdev, &gg); | |
1284 | ovs_mutex_lock(&dev->mutex); | |
1285 | rte_eth_stats_get(dev->port_id, &rte_stats); | |
1286 | ||
2f9dd77f | 1287 | memset(stats, 0, sizeof(*stats)); |
8a9562d2 | 1288 | |
2f9dd77f PS |
1289 | stats->rx_packets = rte_stats.ipackets; |
1290 | stats->tx_packets = rte_stats.opackets; | |
1291 | stats->rx_bytes = rte_stats.ibytes; | |
1292 | stats->tx_bytes = rte_stats.obytes; | |
1293 | stats->rx_errors = rte_stats.ierrors; | |
1294 | stats->tx_errors = rte_stats.oerrors; | |
1295 | stats->multicast = rte_stats.imcasts; | |
8a9562d2 | 1296 | |
2f9dd77f | 1297 | stats->tx_dropped = dev->stats.tx_dropped; |
8a9562d2 PS |
1298 | ovs_mutex_unlock(&dev->mutex); |
1299 | ||
1300 | return 0; | |
1301 | } | |
1302 | ||
1303 | static int | |
1304 | netdev_dpdk_get_features(const struct netdev *netdev_, | |
1305 | enum netdev_features *current, | |
1306 | enum netdev_features *advertised OVS_UNUSED, | |
1307 | enum netdev_features *supported OVS_UNUSED, | |
1308 | enum netdev_features *peer OVS_UNUSED) | |
1309 | { | |
1310 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1311 | struct rte_eth_link link; | |
1312 | ||
1313 | ovs_mutex_lock(&dev->mutex); | |
1314 | link = dev->link; | |
1315 | ovs_mutex_unlock(&dev->mutex); | |
1316 | ||
1317 | if (link.link_duplex == ETH_LINK_AUTONEG_DUPLEX) { | |
1318 | if (link.link_speed == ETH_LINK_SPEED_AUTONEG) { | |
1319 | *current = NETDEV_F_AUTONEG; | |
1320 | } | |
1321 | } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { | |
1322 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1323 | *current = NETDEV_F_10MB_HD; | |
1324 | } | |
1325 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1326 | *current = NETDEV_F_100MB_HD; | |
1327 | } | |
1328 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1329 | *current = NETDEV_F_1GB_HD; | |
1330 | } | |
1331 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
1332 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1333 | *current = NETDEV_F_10MB_FD; | |
1334 | } | |
1335 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1336 | *current = NETDEV_F_100MB_FD; | |
1337 | } | |
1338 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1339 | *current = NETDEV_F_1GB_FD; | |
1340 | } | |
1341 | if (link.link_speed == ETH_LINK_SPEED_10000) { | |
1342 | *current = NETDEV_F_10GB_FD; | |
1343 | } | |
1344 | } | |
1345 | ||
1346 | return 0; | |
1347 | } | |
1348 | ||
1349 | static int | |
1350 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
1351 | { | |
1352 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1353 | int ifindex; | |
1354 | ||
1355 | ovs_mutex_lock(&dev->mutex); | |
1356 | ifindex = dev->port_id; | |
1357 | ovs_mutex_unlock(&dev->mutex); | |
1358 | ||
1359 | return ifindex; | |
1360 | } | |
1361 | ||
1362 | static int | |
1363 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1364 | { | |
1365 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1366 | ||
1367 | ovs_mutex_lock(&dev->mutex); | |
1368 | check_link_status(dev); | |
1369 | *carrier = dev->link.link_status; | |
58397e6c KT |
1370 | |
1371 | ovs_mutex_unlock(&dev->mutex); | |
1372 | ||
1373 | return 0; | |
1374 | } | |
1375 | ||
1376 | static int | |
1377 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1378 | { | |
1379 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1380 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(dev); | |
1381 | ||
1382 | ovs_mutex_lock(&dev->mutex); | |
1383 | ||
1384 | if (is_vhost_running(virtio_dev)) { | |
1385 | *carrier = 1; | |
1386 | } else { | |
1387 | *carrier = 0; | |
1388 | } | |
1389 | ||
8a9562d2 PS |
1390 | ovs_mutex_unlock(&dev->mutex); |
1391 | ||
1392 | return 0; | |
1393 | } | |
1394 | ||
1395 | static long long int | |
1396 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev_) | |
1397 | { | |
1398 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1399 | long long int carrier_resets; | |
1400 | ||
1401 | ovs_mutex_lock(&dev->mutex); | |
1402 | carrier_resets = dev->link_reset_cnt; | |
1403 | ovs_mutex_unlock(&dev->mutex); | |
1404 | ||
1405 | return carrier_resets; | |
1406 | } | |
1407 | ||
1408 | static int | |
1409 | netdev_dpdk_set_miimon(struct netdev *netdev_ OVS_UNUSED, | |
1410 | long long int interval OVS_UNUSED) | |
1411 | { | |
ee32150e | 1412 | return EOPNOTSUPP; |
8a9562d2 PS |
1413 | } |
1414 | ||
1415 | static int | |
1416 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
1417 | enum netdev_flags off, enum netdev_flags on, | |
95fb793a | 1418 | enum netdev_flags *old_flagsp) OVS_REQUIRES(dev->mutex) |
8a9562d2 PS |
1419 | { |
1420 | int err; | |
1421 | ||
1422 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
1423 | return EINVAL; | |
1424 | } | |
1425 | ||
1426 | *old_flagsp = dev->flags; | |
1427 | dev->flags |= on; | |
1428 | dev->flags &= ~off; | |
1429 | ||
1430 | if (dev->flags == *old_flagsp) { | |
1431 | return 0; | |
1432 | } | |
1433 | ||
58397e6c KT |
1434 | if (dev->type == DPDK_DEV_ETH) { |
1435 | if (dev->flags & NETDEV_UP) { | |
1436 | err = rte_eth_dev_start(dev->port_id); | |
1437 | if (err) | |
1438 | return -err; | |
1439 | } | |
8a9562d2 | 1440 | |
58397e6c KT |
1441 | if (dev->flags & NETDEV_PROMISC) { |
1442 | rte_eth_promiscuous_enable(dev->port_id); | |
1443 | } | |
8a9562d2 | 1444 | |
58397e6c KT |
1445 | if (!(dev->flags & NETDEV_UP)) { |
1446 | rte_eth_dev_stop(dev->port_id); | |
1447 | } | |
8a9562d2 PS |
1448 | } |
1449 | ||
1450 | return 0; | |
1451 | } | |
1452 | ||
1453 | static int | |
1454 | netdev_dpdk_update_flags(struct netdev *netdev_, | |
1455 | enum netdev_flags off, enum netdev_flags on, | |
1456 | enum netdev_flags *old_flagsp) | |
1457 | { | |
1458 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1459 | int error; | |
1460 | ||
1461 | ovs_mutex_lock(&netdev->mutex); | |
1462 | error = netdev_dpdk_update_flags__(netdev, off, on, old_flagsp); | |
1463 | ovs_mutex_unlock(&netdev->mutex); | |
1464 | ||
1465 | return error; | |
1466 | } | |
1467 | ||
1468 | static int | |
1469 | netdev_dpdk_get_status(const struct netdev *netdev_, struct smap *args) | |
1470 | { | |
1471 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1472 | struct rte_eth_dev_info dev_info; | |
1473 | ||
e0a801c7 | 1474 | if (dev->port_id < 0) |
8a9562d2 PS |
1475 | return ENODEV; |
1476 | ||
1477 | ovs_mutex_lock(&dev->mutex); | |
1478 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
1479 | ovs_mutex_unlock(&dev->mutex); | |
1480 | ||
1481 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1482 | ||
95fb793a | 1483 | smap_add_format(args, "port_no", "%d", dev->port_id); |
8a9562d2 PS |
1484 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); |
1485 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1486 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
1487 | smap_add_format(args, "max_rx_pktlen", "%u", dev_info.max_rx_pktlen); | |
1488 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); | |
1489 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
1490 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
1491 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
1492 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
1493 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
1494 | ||
1495 | smap_add_format(args, "pci-vendor_id", "0x%u", dev_info.pci_dev->id.vendor_id); | |
1496 | smap_add_format(args, "pci-device_id", "0x%x", dev_info.pci_dev->id.device_id); | |
1497 | ||
1498 | return 0; | |
1499 | } | |
1500 | ||
1501 | static void | |
1502 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
1503 | OVS_REQUIRES(dev->mutex) | |
1504 | { | |
1505 | enum netdev_flags old_flags; | |
1506 | ||
1507 | if (admin_state) { | |
1508 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
1509 | } else { | |
1510 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
1511 | } | |
1512 | } | |
1513 | ||
1514 | static void | |
1515 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
1516 | const char *argv[], void *aux OVS_UNUSED) | |
1517 | { | |
1518 | bool up; | |
1519 | ||
1520 | if (!strcasecmp(argv[argc - 1], "up")) { | |
1521 | up = true; | |
1522 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
1523 | up = false; | |
1524 | } else { | |
1525 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
1526 | return; | |
1527 | } | |
1528 | ||
1529 | if (argc > 2) { | |
1530 | struct netdev *netdev = netdev_from_name(argv[1]); | |
1531 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
1532 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
1533 | ||
1534 | ovs_mutex_lock(&dpdk_dev->mutex); | |
1535 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
1536 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
1537 | ||
1538 | netdev_close(netdev); | |
1539 | } else { | |
1540 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
1541 | netdev_close(netdev); | |
1542 | return; | |
1543 | } | |
1544 | } else { | |
1545 | struct netdev_dpdk *netdev; | |
1546 | ||
1547 | ovs_mutex_lock(&dpdk_mutex); | |
1548 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
1549 | ovs_mutex_lock(&netdev->mutex); | |
1550 | netdev_dpdk_set_admin_state__(netdev, up); | |
1551 | ovs_mutex_unlock(&netdev->mutex); | |
1552 | } | |
1553 | ovs_mutex_unlock(&dpdk_mutex); | |
1554 | } | |
1555 | unixctl_command_reply(conn, "OK"); | |
1556 | } | |
1557 | ||
58397e6c KT |
1558 | /* |
1559 | * Set virtqueue flags so that we do not receive interrupts. | |
1560 | */ | |
1561 | static void | |
1562 | set_irq_status(struct virtio_net *dev) | |
1563 | { | |
1564 | dev->virtqueue[VIRTIO_RXQ]->used->flags = VRING_USED_F_NO_NOTIFY; | |
1565 | dev->virtqueue[VIRTIO_TXQ]->used->flags = VRING_USED_F_NO_NOTIFY; | |
1566 | } | |
1567 | ||
1568 | /* | |
1569 | * A new virtio-net device is added to a vhost port. | |
1570 | */ | |
1571 | static int | |
1572 | new_device(struct virtio_net *dev) | |
1573 | { | |
1574 | struct netdev_dpdk *netdev; | |
1575 | bool exists = false; | |
1576 | ||
1577 | ovs_mutex_lock(&dpdk_mutex); | |
1578 | /* Add device to the vhost port with the same name as that passed down. */ | |
1579 | LIST_FOR_EACH(netdev, list_node, &dpdk_list) { | |
1580 | if (strncmp(dev->ifname, netdev->up.name, IFNAMSIZ) == 0) { | |
1581 | ovs_mutex_lock(&netdev->mutex); | |
1582 | ovsrcu_set(&netdev->virtio_dev, dev); | |
1583 | ovs_mutex_unlock(&netdev->mutex); | |
1584 | exists = true; | |
1585 | dev->flags |= VIRTIO_DEV_RUNNING; | |
1586 | /* Disable notifications. */ | |
1587 | set_irq_status(dev); | |
1588 | break; | |
1589 | } | |
1590 | } | |
1591 | ovs_mutex_unlock(&dpdk_mutex); | |
1592 | ||
1593 | if (!exists) { | |
1594 | VLOG_INFO("vHost Device '%s' (%ld) can't be added - name not found", | |
1595 | dev->ifname, dev->device_fh); | |
1596 | ||
1597 | return -1; | |
1598 | } | |
1599 | ||
1600 | VLOG_INFO("vHost Device '%s' (%ld) has been added", | |
1601 | dev->ifname, dev->device_fh); | |
1602 | return 0; | |
1603 | } | |
1604 | ||
1605 | /* | |
1606 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
1607 | * flag to stop any more packets from being sent or received to/from a VM and | |
1608 | * ensure all currently queued packets have been sent/received before removing | |
1609 | * the device. | |
1610 | */ | |
1611 | static void | |
1612 | destroy_device(volatile struct virtio_net *dev) | |
1613 | { | |
1614 | struct netdev_dpdk *vhost_dev; | |
1615 | ||
1616 | ovs_mutex_lock(&dpdk_mutex); | |
1617 | LIST_FOR_EACH (vhost_dev, list_node, &dpdk_list) { | |
1618 | if (netdev_dpdk_get_virtio(vhost_dev) == dev) { | |
1619 | ||
1620 | ovs_mutex_lock(&vhost_dev->mutex); | |
1621 | dev->flags &= ~VIRTIO_DEV_RUNNING; | |
1622 | ovsrcu_set(&vhost_dev->virtio_dev, NULL); | |
1623 | ovs_mutex_unlock(&vhost_dev->mutex); | |
1624 | ||
1625 | /* | |
1626 | * Wait for other threads to quiesce before | |
1627 | * setting the virtio_dev to NULL. | |
1628 | */ | |
1629 | ovsrcu_synchronize(); | |
618f44f7 KT |
1630 | /* |
1631 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
1632 | * put thread back into quiescent state before returning. | |
1633 | */ | |
1634 | ovsrcu_quiesce_start(); | |
58397e6c KT |
1635 | } |
1636 | } | |
1637 | ovs_mutex_unlock(&dpdk_mutex); | |
1638 | ||
1639 | VLOG_INFO("vHost Device '%s' (%ld) has been removed", | |
1640 | dev->ifname, dev->device_fh); | |
1641 | } | |
1642 | ||
1643 | struct virtio_net * | |
1644 | netdev_dpdk_get_virtio(const struct netdev_dpdk *dev) | |
1645 | { | |
1646 | return ovsrcu_get(struct virtio_net *, &dev->virtio_dev); | |
1647 | } | |
1648 | ||
1649 | /* | |
1650 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
1651 | * configuration has been fully complete. | |
1652 | */ | |
1653 | const struct virtio_net_device_ops virtio_net_device_ops = | |
1654 | { | |
1655 | .new_device = new_device, | |
1656 | .destroy_device = destroy_device, | |
1657 | }; | |
1658 | ||
1659 | static void * | |
1660 | start_cuse_session_loop(void *dummy OVS_UNUSED) | |
1661 | { | |
1662 | pthread_detach(pthread_self()); | |
618f44f7 KT |
1663 | /* Put the cuse thread into quiescent state. */ |
1664 | ovsrcu_quiesce_start(); | |
58397e6c KT |
1665 | rte_vhost_driver_session_start(); |
1666 | return NULL; | |
1667 | } | |
1668 | ||
1669 | static int | |
1670 | dpdk_vhost_class_init(void) | |
1671 | { | |
58397e6c KT |
1672 | int err = -1; |
1673 | ||
1674 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
1675 | ||
1676 | /* Register CUSE device to handle IOCTLs. | |
1677 | * Unless otherwise specified on the vswitchd command line, cuse_dev_name | |
1678 | * is set to vhost-net. | |
1679 | */ | |
1680 | err = rte_vhost_driver_register(cuse_dev_name); | |
1681 | ||
1682 | if (err != 0) { | |
1683 | VLOG_ERR("CUSE device setup failure."); | |
1684 | return -1; | |
1685 | } | |
1686 | ||
618f44f7 KT |
1687 | ovs_thread_create("cuse_thread", start_cuse_session_loop, NULL); |
1688 | return 0; | |
58397e6c KT |
1689 | } |
1690 | ||
033e9df2 DDP |
1691 | static void |
1692 | dpdk_common_init(void) | |
1693 | { | |
1694 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
1695 | "[netdev] up|down", 1, 2, | |
1696 | netdev_dpdk_set_admin_state, NULL); | |
1697 | ||
1698 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
1699 | } | |
1700 | ||
95fb793a | 1701 | /* Client Rings */ |
1702 | ||
95fb793a | 1703 | static int |
1704 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
1705 | unsigned int *eth_port_id) | |
1706 | { | |
1707 | struct dpdk_ring *ivshmem; | |
1708 | char ring_name[10]; | |
1709 | int err; | |
1710 | ||
1711 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
1712 | if (ivshmem == NULL) { | |
1713 | return ENOMEM; | |
1714 | } | |
1715 | ||
7251515e | 1716 | /* XXX: Add support for multiquque ring. */ |
95fb793a | 1717 | err = snprintf(ring_name, 10, "%s_tx", dev_name); |
1718 | if (err < 0) { | |
1719 | return -err; | |
1720 | } | |
1721 | ||
7251515e DV |
1722 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1723 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1724 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1725 | if (ivshmem->cring_tx == NULL) { |
1726 | rte_free(ivshmem); | |
1727 | return ENOMEM; | |
1728 | } | |
1729 | ||
1730 | err = snprintf(ring_name, 10, "%s_rx", dev_name); | |
1731 | if (err < 0) { | |
1732 | return -err; | |
1733 | } | |
1734 | ||
7251515e DV |
1735 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1736 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1737 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1738 | if (ivshmem->cring_rx == NULL) { |
1739 | rte_free(ivshmem); | |
1740 | return ENOMEM; | |
1741 | } | |
1742 | ||
d7310583 DDP |
1743 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
1744 | &ivshmem->cring_tx, 1, SOCKET0); | |
1745 | ||
95fb793a | 1746 | if (err < 0) { |
1747 | rte_free(ivshmem); | |
1748 | return ENODEV; | |
1749 | } | |
1750 | ||
1751 | ivshmem->user_port_id = port_no; | |
1752 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
1753 | list_push_back(&dpdk_ring_list, &ivshmem->list_node); | |
1754 | ||
1755 | *eth_port_id = ivshmem->eth_port_id; | |
1756 | return 0; | |
1757 | } | |
1758 | ||
1759 | static int | |
1760 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) OVS_REQUIRES(dpdk_mutex) | |
1761 | { | |
1762 | struct dpdk_ring *ivshmem; | |
1763 | unsigned int port_no; | |
1764 | int err = 0; | |
1765 | ||
1766 | /* Names always start with "dpdkr" */ | |
1767 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
1768 | if (err) { | |
1769 | return err; | |
1770 | } | |
1771 | ||
1772 | /* look through our list to find the device */ | |
1773 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { | |
1774 | if (ivshmem->user_port_id == port_no) { | |
58397e6c | 1775 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
95fb793a | 1776 | *eth_port_id = ivshmem->eth_port_id; /* really all that is needed */ |
1777 | return 0; | |
1778 | } | |
1779 | } | |
1780 | /* Need to create the device rings */ | |
1781 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
1782 | } | |
1783 | ||
7251515e DV |
1784 | static int |
1785 | netdev_dpdk_ring_send(struct netdev *netdev, int qid OVS_UNUSED, | |
e14deea0 | 1786 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1787 | { |
1788 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1b99bb05 MG |
1789 | unsigned i; |
1790 | ||
1791 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that the | |
1792 | * rss hash field is clear. This is because the same mbuf may be modified by | |
1793 | * the consumer of the ring and return into the datapath without recalculating | |
1794 | * the RSS hash. */ | |
1795 | for (i = 0; i < cnt; i++) { | |
1796 | dp_packet_set_rss_hash(pkts[i], 0); | |
1797 | } | |
7251515e DV |
1798 | |
1799 | /* DPDK Rings have a single TX queue, Therefore needs locking. */ | |
58397e6c | 1800 | rte_spinlock_lock(&dev->txq_lock); |
7251515e | 1801 | netdev_dpdk_send__(dev, 0, pkts, cnt, may_steal); |
58397e6c | 1802 | rte_spinlock_unlock(&dev->txq_lock); |
7251515e DV |
1803 | return 0; |
1804 | } | |
1805 | ||
95fb793a | 1806 | static int |
1807 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
1808 | { | |
1809 | unsigned int port_no = 0; | |
1810 | int err = 0; | |
1811 | ||
1812 | if (rte_eal_init_ret) { | |
1813 | return rte_eal_init_ret; | |
1814 | } | |
1815 | ||
1816 | ovs_mutex_lock(&dpdk_mutex); | |
1817 | ||
1818 | err = dpdk_ring_open(netdev->name, &port_no); | |
1819 | if (err) { | |
1820 | goto unlock_dpdk; | |
1821 | } | |
1822 | ||
58397e6c | 1823 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
95fb793a | 1824 | |
1825 | unlock_dpdk: | |
1826 | ovs_mutex_unlock(&dpdk_mutex); | |
1827 | return err; | |
1828 | } | |
1829 | ||
58397e6c KT |
1830 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, MULTIQ, SEND, \ |
1831 | GET_CARRIER, GET_STATS, GET_FEATURES, GET_STATUS, RXQ_RECV) \ | |
95fb793a | 1832 | { \ |
1833 | NAME, \ | |
1834 | INIT, /* init */ \ | |
1835 | NULL, /* netdev_dpdk_run */ \ | |
1836 | NULL, /* netdev_dpdk_wait */ \ | |
1837 | \ | |
1838 | netdev_dpdk_alloc, \ | |
1839 | CONSTRUCT, \ | |
58397e6c | 1840 | DESTRUCT, \ |
95fb793a | 1841 | netdev_dpdk_dealloc, \ |
1842 | netdev_dpdk_get_config, \ | |
1843 | NULL, /* netdev_dpdk_set_config */ \ | |
1844 | NULL, /* get_tunnel_config */ \ | |
58397e6c KT |
1845 | NULL, /* build header */ \ |
1846 | NULL, /* push header */ \ | |
1847 | NULL, /* pop header */ \ | |
7dec44fe | 1848 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
5496878c | 1849 | MULTIQ, /* set_multiq */ \ |
95fb793a | 1850 | \ |
7251515e | 1851 | SEND, /* send */ \ |
95fb793a | 1852 | NULL, /* send_wait */ \ |
1853 | \ | |
1854 | netdev_dpdk_set_etheraddr, \ | |
1855 | netdev_dpdk_get_etheraddr, \ | |
1856 | netdev_dpdk_get_mtu, \ | |
1857 | netdev_dpdk_set_mtu, \ | |
1858 | netdev_dpdk_get_ifindex, \ | |
58397e6c | 1859 | GET_CARRIER, \ |
95fb793a | 1860 | netdev_dpdk_get_carrier_resets, \ |
1861 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
1862 | GET_STATS, \ |
1863 | GET_FEATURES, \ | |
95fb793a | 1864 | NULL, /* set_advertisements */ \ |
1865 | \ | |
1866 | NULL, /* set_policing */ \ | |
1867 | NULL, /* get_qos_types */ \ | |
1868 | NULL, /* get_qos_capabilities */ \ | |
1869 | NULL, /* get_qos */ \ | |
1870 | NULL, /* set_qos */ \ | |
1871 | NULL, /* get_queue */ \ | |
1872 | NULL, /* set_queue */ \ | |
1873 | NULL, /* delete_queue */ \ | |
1874 | NULL, /* get_queue_stats */ \ | |
1875 | NULL, /* queue_dump_start */ \ | |
1876 | NULL, /* queue_dump_next */ \ | |
1877 | NULL, /* queue_dump_done */ \ | |
1878 | NULL, /* dump_queue_stats */ \ | |
1879 | \ | |
1880 | NULL, /* get_in4 */ \ | |
1881 | NULL, /* set_in4 */ \ | |
1882 | NULL, /* get_in6 */ \ | |
1883 | NULL, /* add_router */ \ | |
1884 | NULL, /* get_next_hop */ \ | |
58397e6c | 1885 | GET_STATUS, \ |
95fb793a | 1886 | NULL, /* arp_lookup */ \ |
1887 | \ | |
1888 | netdev_dpdk_update_flags, \ | |
1889 | \ | |
1890 | netdev_dpdk_rxq_alloc, \ | |
1891 | netdev_dpdk_rxq_construct, \ | |
1892 | netdev_dpdk_rxq_destruct, \ | |
1893 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 1894 | RXQ_RECV, \ |
95fb793a | 1895 | NULL, /* rx_wait */ \ |
1896 | NULL, /* rxq_drain */ \ | |
1897 | } | |
8a9562d2 PS |
1898 | |
1899 | int | |
1900 | dpdk_init(int argc, char **argv) | |
1901 | { | |
1902 | int result; | |
58397e6c KT |
1903 | int base = 0; |
1904 | char *pragram_name = argv[0]; | |
8a9562d2 | 1905 | |
9441caf3 | 1906 | if (argc < 2 || strcmp(argv[1], "--dpdk")) |
8a9562d2 PS |
1907 | return 0; |
1908 | ||
58397e6c | 1909 | /* Remove the --dpdk argument from arg list.*/ |
8a9562d2 PS |
1910 | argc--; |
1911 | argv++; | |
1912 | ||
58397e6c KT |
1913 | /* If the cuse_dev_name parameter has been provided, set 'cuse_dev_name' to |
1914 | * this string if it meets the correct criteria. Otherwise, set it to the | |
1915 | * default (vhost-net). | |
1916 | */ | |
1917 | if (!strcmp(argv[1], "--cuse_dev_name") && | |
1918 | (strlen(argv[2]) <= NAME_MAX)) { | |
1919 | ||
1920 | cuse_dev_name = strdup(argv[2]); | |
1921 | ||
1922 | /* Remove the cuse_dev_name configuration parameters from the argument | |
1923 | * list, so that the correct elements are passed to the DPDK | |
1924 | * initialization function | |
1925 | */ | |
1926 | argc -= 2; | |
1927 | argv += 2; /* Increment by two to bypass the cuse_dev_name arguments */ | |
1928 | base = 2; | |
1929 | ||
1930 | VLOG_ERR("User-provided cuse_dev_name in use: /dev/%s", cuse_dev_name); | |
1931 | } else { | |
1932 | cuse_dev_name = "vhost-net"; | |
1933 | VLOG_INFO("No cuse_dev_name provided - defaulting to /dev/vhost-net"); | |
1934 | } | |
1935 | ||
1936 | /* Keep the program name argument as this is needed for call to | |
1937 | * rte_eal_init() | |
1938 | */ | |
1939 | argv[0] = pragram_name; | |
1940 | ||
8a9562d2 PS |
1941 | /* Make sure things are initialized ... */ |
1942 | result = rte_eal_init(argc, argv); | |
451450fa | 1943 | if (result < 0) { |
58397e6c | 1944 | ovs_abort(result, "Cannot init EAL"); |
451450fa | 1945 | } |
8a9562d2 | 1946 | |
d7310583 | 1947 | rte_memzone_dump(stdout); |
8a9562d2 PS |
1948 | rte_eal_init_ret = 0; |
1949 | ||
451450fa | 1950 | if (argc > result) { |
9441caf3 | 1951 | argv[result] = argv[0]; |
451450fa | 1952 | } |
9441caf3 | 1953 | |
db73f716 DDP |
1954 | /* We are called from the main thread here */ |
1955 | thread_set_nonpmd(); | |
1956 | ||
58397e6c | 1957 | return result + 1 + base; |
8a9562d2 PS |
1958 | } |
1959 | ||
95fb793a | 1960 | const struct netdev_class dpdk_class = |
1961 | NETDEV_DPDK_CLASS( | |
1962 | "dpdk", | |
b8e57534 | 1963 | NULL, |
5496878c | 1964 | netdev_dpdk_construct, |
58397e6c | 1965 | netdev_dpdk_destruct, |
7251515e | 1966 | netdev_dpdk_set_multiq, |
58397e6c KT |
1967 | netdev_dpdk_eth_send, |
1968 | netdev_dpdk_get_carrier, | |
1969 | netdev_dpdk_get_stats, | |
1970 | netdev_dpdk_get_features, | |
1971 | netdev_dpdk_get_status, | |
1972 | netdev_dpdk_rxq_recv); | |
95fb793a | 1973 | |
1974 | const struct netdev_class dpdk_ring_class = | |
1975 | NETDEV_DPDK_CLASS( | |
1976 | "dpdkr", | |
033e9df2 | 1977 | NULL, |
5496878c | 1978 | netdev_dpdk_ring_construct, |
58397e6c KT |
1979 | netdev_dpdk_destruct, |
1980 | NULL, | |
1981 | netdev_dpdk_ring_send, | |
1982 | netdev_dpdk_get_carrier, | |
1983 | netdev_dpdk_get_stats, | |
1984 | netdev_dpdk_get_features, | |
1985 | netdev_dpdk_get_status, | |
1986 | netdev_dpdk_rxq_recv); | |
1987 | ||
1988 | const struct netdev_class dpdk_vhost_class = | |
1989 | NETDEV_DPDK_CLASS( | |
1990 | "dpdkvhost", | |
1991 | dpdk_vhost_class_init, | |
1992 | netdev_dpdk_vhost_construct, | |
1993 | netdev_dpdk_vhost_destruct, | |
1994 | netdev_dpdk_vhost_set_multiq, | |
1995 | netdev_dpdk_vhost_send, | |
1996 | netdev_dpdk_vhost_get_carrier, | |
1997 | netdev_dpdk_vhost_get_stats, | |
1998 | NULL, | |
7251515e | 1999 | NULL, |
58397e6c | 2000 | netdev_dpdk_vhost_rxq_recv); |
95fb793a | 2001 | |
8a9562d2 PS |
2002 | void |
2003 | netdev_dpdk_register(void) | |
2004 | { | |
95fb793a | 2005 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2006 | ||
033e9df2 DDP |
2007 | if (rte_eal_init_ret) { |
2008 | return; | |
2009 | } | |
2010 | ||
95fb793a | 2011 | if (ovsthread_once_start(&once)) { |
033e9df2 | 2012 | dpdk_common_init(); |
95fb793a | 2013 | netdev_register_provider(&dpdk_class); |
2014 | netdev_register_provider(&dpdk_ring_class); | |
58397e6c | 2015 | netdev_register_provider(&dpdk_vhost_class); |
95fb793a | 2016 | ovsthread_once_done(&once); |
2017 | } | |
8a9562d2 | 2018 | } |
8617afff PS |
2019 | |
2020 | int | |
2021 | pmd_thread_setaffinity_cpu(int cpu) | |
2022 | { | |
2023 | cpu_set_t cpuset; | |
2024 | int err; | |
2025 | ||
2026 | CPU_ZERO(&cpuset); | |
2027 | CPU_SET(cpu, &cpuset); | |
2028 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset); | |
2029 | if (err) { | |
2030 | VLOG_ERR("Thread affinity error %d",err); | |
2031 | return err; | |
2032 | } | |
abb5943d AW |
2033 | /* NON_PMD_CORE_ID is reserved for use by non pmd threads. */ |
2034 | ovs_assert(cpu != NON_PMD_CORE_ID); | |
65f13b50 | 2035 | RTE_PER_LCORE(_lcore_id) = cpu; |
8617afff PS |
2036 | |
2037 | return 0; | |
2038 | } | |
db73f716 DDP |
2039 | |
2040 | void | |
2041 | thread_set_nonpmd(void) | |
2042 | { | |
abb5943d AW |
2043 | /* We have to use NON_PMD_CORE_ID to allow non-pmd threads to perform |
2044 | * certain DPDK operations, like rte_eth_dev_configure(). */ | |
2045 | RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID; | |
db73f716 DDP |
2046 | } |
2047 | ||
2048 | static bool | |
2049 | thread_is_pmd(void) | |
2050 | { | |
abb5943d | 2051 | return rte_lcore_id() != NON_PMD_CORE_ID; |
db73f716 | 2052 | } |