]>
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 | ||
645b8934 | 71 | /* XXX: mempool size should be based on system resources. */ |
8a9562d2 PS |
72 | #define NB_MBUF (4096 * 64) |
73 | #define MP_CACHE_SZ (256 * 2) | |
74 | #define SOCKET0 0 | |
75 | ||
79f5354c PM |
76 | #define NIC_PORT_RX_Q_SIZE 2048 /* Size of Physical NIC RX Queue, Max (n+32<=4096)*/ |
77 | #define NIC_PORT_TX_Q_SIZE 2048 /* Size of Physical NIC TX Queue, Max (n+32<=4096)*/ | |
78 | ||
645b8934 | 79 | /* XXX: Needs per NIC value for these constants. */ |
8a9562d2 PS |
80 | #define RX_PTHRESH 32 /* Default values of RX prefetch threshold reg. */ |
81 | #define RX_HTHRESH 32 /* Default values of RX host threshold reg. */ | |
82 | #define RX_WTHRESH 16 /* Default values of RX write-back threshold reg. */ | |
83 | ||
84 | #define TX_PTHRESH 36 /* Default values of TX prefetch threshold reg. */ | |
85 | #define TX_HTHRESH 0 /* Default values of TX host threshold reg. */ | |
86 | #define TX_WTHRESH 0 /* Default values of TX write-back threshold reg. */ | |
87 | ||
58397e6c KT |
88 | #define MAX_PKT_BURST 32 /* Max burst size for RX/TX */ |
89 | ||
90 | /* Character device cuse_dev_name. */ | |
91 | char *cuse_dev_name = NULL; | |
92 | ||
8a9562d2 | 93 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
94 | .rxmode = { |
95 | .mq_mode = ETH_MQ_RX_RSS, | |
96 | .split_hdr_size = 0, | |
97 | .header_split = 0, /* Header Split disabled */ | |
98 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
99 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
100 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
101 | .hw_strip_crc = 0, | |
102 | }, | |
103 | .rx_adv_conf = { | |
104 | .rss_conf = { | |
105 | .rss_key = NULL, | |
61a2647e DDP |
106 | .rss_hf = ETH_RSS_IPV4_TCP | ETH_RSS_IPV4 | ETH_RSS_IPV6 |
107 | | ETH_RSS_IPV4_UDP | ETH_RSS_IPV6_TCP | ETH_RSS_IPV6_UDP, | |
8a9562d2 | 108 | }, |
a28ddd11 DDP |
109 | }, |
110 | .txmode = { | |
111 | .mq_mode = ETH_MQ_TX_NONE, | |
112 | }, | |
8a9562d2 PS |
113 | }; |
114 | ||
115 | static const struct rte_eth_rxconf rx_conf = { | |
a28ddd11 DDP |
116 | .rx_thresh = { |
117 | .pthresh = RX_PTHRESH, | |
118 | .hthresh = RX_HTHRESH, | |
119 | .wthresh = RX_WTHRESH, | |
120 | }, | |
8a9562d2 PS |
121 | }; |
122 | ||
123 | static const struct rte_eth_txconf tx_conf = { | |
a28ddd11 DDP |
124 | .tx_thresh = { |
125 | .pthresh = TX_PTHRESH, | |
126 | .hthresh = TX_HTHRESH, | |
127 | .wthresh = TX_WTHRESH, | |
128 | }, | |
129 | .tx_free_thresh = 0, | |
130 | .tx_rs_thresh = 0, | |
94777510 | 131 | .txq_flags = ETH_TXQ_FLAGS_NOMULTSEGS|ETH_TXQ_FLAGS_NOOFFLOADS, |
8a9562d2 PS |
132 | }; |
133 | ||
3a100265 DDP |
134 | enum { MAX_RX_QUEUE_LEN = 192 }; |
135 | enum { MAX_TX_QUEUE_LEN = 384 }; | |
58f7c37b DDP |
136 | enum { DPDK_RING_SIZE = 256 }; |
137 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
138 | enum { DRAIN_TSC = 200000ULL }; |
139 | ||
58397e6c KT |
140 | enum dpdk_dev_type { |
141 | DPDK_DEV_ETH = 0, | |
142 | DPDK_DEV_VHOST = 1 | |
143 | }; | |
144 | ||
8a9562d2 PS |
145 | static int rte_eal_init_ret = ENODEV; |
146 | ||
147 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; | |
148 | ||
149 | /* Contains all 'struct dpdk_dev's. */ | |
ca6ba700 | 150 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 151 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 152 | |
ca6ba700 | 153 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 154 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 155 | |
db73f716 DDP |
156 | /* This mutex must be used by non pmd threads when allocating or freeing |
157 | * mbufs through mempools. Since dpdk_queue_pkts() and dpdk_queue_flush() may | |
158 | * use mempools, a non pmd thread should hold this mutex while calling them */ | |
159 | struct ovs_mutex nonpmd_mempool_mutex = OVS_MUTEX_INITIALIZER; | |
160 | ||
8a9562d2 PS |
161 | struct dpdk_mp { |
162 | struct rte_mempool *mp; | |
163 | int mtu; | |
164 | int socket_id; | |
165 | int refcount; | |
ca6ba700 | 166 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
167 | }; |
168 | ||
5a034064 AW |
169 | /* There should be one 'struct dpdk_tx_queue' created for |
170 | * each cpu core. */ | |
8a9562d2 | 171 | struct dpdk_tx_queue { |
94143fc4 AW |
172 | bool flush_tx; /* Set to true to flush queue everytime */ |
173 | /* pkts are queued. */ | |
8a9562d2 PS |
174 | int count; |
175 | uint64_t tsc; | |
176 | struct rte_mbuf *burst_pkts[MAX_TX_QUEUE_LEN]; | |
177 | }; | |
178 | ||
95fb793a | 179 | /* dpdk has no way to remove dpdk ring ethernet devices |
180 | so we have to keep them around once they've been created | |
181 | */ | |
182 | ||
ca6ba700 | 183 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 184 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 185 | |
186 | struct dpdk_ring { | |
187 | /* For the client rings */ | |
188 | struct rte_ring *cring_tx; | |
189 | struct rte_ring *cring_rx; | |
190 | int user_port_id; /* User given port no, parsed from port name */ | |
191 | int eth_port_id; /* ethernet device port id */ | |
ca6ba700 | 192 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 193 | }; |
194 | ||
8a9562d2 PS |
195 | struct netdev_dpdk { |
196 | struct netdev up; | |
197 | int port_id; | |
198 | int max_packet_len; | |
58397e6c | 199 | enum dpdk_dev_type type; |
8a9562d2 | 200 | |
5a034064 | 201 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
202 | |
203 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
204 | ||
205 | struct dpdk_mp *dpdk_mp; | |
206 | int mtu; | |
207 | int socket_id; | |
208 | int buf_size; | |
8a9562d2 PS |
209 | struct netdev_stats stats; |
210 | ||
211 | uint8_t hwaddr[ETH_ADDR_LEN]; | |
212 | enum netdev_flags flags; | |
213 | ||
214 | struct rte_eth_link link; | |
215 | int link_reset_cnt; | |
216 | ||
58397e6c KT |
217 | /* virtio-net structure for vhost device */ |
218 | OVSRCU_TYPE(struct virtio_net *) virtio_dev; | |
219 | ||
8a9562d2 | 220 | /* In dpdk_list. */ |
ca6ba700 | 221 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
58397e6c | 222 | rte_spinlock_t txq_lock; |
8a9562d2 PS |
223 | }; |
224 | ||
225 | struct netdev_rxq_dpdk { | |
226 | struct netdev_rxq up; | |
227 | int port_id; | |
228 | }; | |
229 | ||
db73f716 DDP |
230 | static bool thread_is_pmd(void); |
231 | ||
8a9562d2 PS |
232 | static int netdev_dpdk_construct(struct netdev *); |
233 | ||
58397e6c KT |
234 | struct virtio_net * netdev_dpdk_get_virtio(const struct netdev_dpdk *dev); |
235 | ||
8a9562d2 PS |
236 | static bool |
237 | is_dpdk_class(const struct netdev_class *class) | |
238 | { | |
239 | return class->construct == netdev_dpdk_construct; | |
240 | } | |
241 | ||
58397e6c KT |
242 | /* XXX: use dpdk malloc for entire OVS. in fact huge page should be used |
243 | * for all other segments data, bss and text. */ | |
8a9562d2 PS |
244 | |
245 | static void * | |
246 | dpdk_rte_mzalloc(size_t sz) | |
247 | { | |
248 | void *ptr; | |
249 | ||
250 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
251 | if (ptr == NULL) { | |
252 | out_of_memory(); | |
253 | } | |
254 | return ptr; | |
255 | } | |
256 | ||
db73f716 DDP |
257 | /* XXX this function should be called only by pmd threads (or by non pmd |
258 | * threads holding the nonpmd_mempool_mutex) */ | |
8a9562d2 | 259 | void |
e14deea0 | 260 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 261 | { |
db73f716 | 262 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 263 | |
db73f716 | 264 | rte_pktmbuf_free_seg(pkt); |
8a9562d2 PS |
265 | } |
266 | ||
b3cd9f9d PS |
267 | static void |
268 | __rte_pktmbuf_init(struct rte_mempool *mp, | |
269 | void *opaque_arg OVS_UNUSED, | |
270 | void *_m, | |
271 | unsigned i OVS_UNUSED) | |
272 | { | |
273 | struct rte_mbuf *m = _m; | |
e14deea0 | 274 | uint32_t buf_len = mp->elt_size - sizeof(struct dp_packet); |
b3cd9f9d | 275 | |
e14deea0 | 276 | RTE_MBUF_ASSERT(mp->elt_size >= sizeof(struct dp_packet)); |
b3cd9f9d PS |
277 | |
278 | memset(m, 0, mp->elt_size); | |
279 | ||
280 | /* start of buffer is just after mbuf structure */ | |
e14deea0 | 281 | m->buf_addr = (char *)m + sizeof(struct dp_packet); |
b3cd9f9d | 282 | m->buf_physaddr = rte_mempool_virt2phy(mp, m) + |
e14deea0 | 283 | sizeof(struct dp_packet); |
b3cd9f9d PS |
284 | m->buf_len = (uint16_t)buf_len; |
285 | ||
286 | /* keep some headroom between start of buffer and data */ | |
b8e57534 | 287 | m->data_off = RTE_MIN(RTE_PKTMBUF_HEADROOM, m->buf_len); |
b3cd9f9d PS |
288 | |
289 | /* init some constant fields */ | |
b3cd9f9d | 290 | m->pool = mp; |
b8e57534 MK |
291 | m->nb_segs = 1; |
292 | m->port = 0xff; | |
b3cd9f9d PS |
293 | } |
294 | ||
295 | static void | |
296 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
297 | void *opaque_arg OVS_UNUSED, | |
298 | void *_m, | |
299 | unsigned i OVS_UNUSED) | |
300 | { | |
301 | struct rte_mbuf *m = _m; | |
302 | ||
303 | __rte_pktmbuf_init(mp, opaque_arg, _m, i); | |
304 | ||
cf62fa4c | 305 | dp_packet_init_dpdk((struct dp_packet *) m, m->buf_len); |
b3cd9f9d PS |
306 | } |
307 | ||
8a9562d2 PS |
308 | static struct dpdk_mp * |
309 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
310 | { | |
311 | struct dpdk_mp *dmp = NULL; | |
312 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
313 | ||
314 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
315 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
316 | dmp->refcount++; | |
317 | return dmp; | |
318 | } | |
319 | } | |
320 | ||
321 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
322 | dmp->socket_id = socket_id; | |
323 | dmp->mtu = mtu; | |
324 | dmp->refcount = 1; | |
325 | ||
34631d72 AW |
326 | if (snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d_%d", dmp->mtu, |
327 | dmp->socket_id) < 0) { | |
95fb793a | 328 | return NULL; |
329 | } | |
330 | ||
8a9562d2 PS |
331 | dmp->mp = rte_mempool_create(mp_name, NB_MBUF, MBUF_SIZE(mtu), |
332 | MP_CACHE_SZ, | |
333 | sizeof(struct rte_pktmbuf_pool_private), | |
334 | rte_pktmbuf_pool_init, NULL, | |
b3cd9f9d | 335 | ovs_rte_pktmbuf_init, NULL, |
8a9562d2 PS |
336 | socket_id, 0); |
337 | ||
338 | if (dmp->mp == NULL) { | |
339 | return NULL; | |
340 | } | |
341 | ||
342 | list_push_back(&dpdk_mp_list, &dmp->list_node); | |
343 | return dmp; | |
344 | } | |
345 | ||
346 | static void | |
347 | dpdk_mp_put(struct dpdk_mp *dmp) | |
348 | { | |
349 | ||
350 | if (!dmp) { | |
351 | return; | |
352 | } | |
353 | ||
354 | dmp->refcount--; | |
355 | ovs_assert(dmp->refcount >= 0); | |
356 | ||
357 | #if 0 | |
358 | /* I could not find any API to destroy mp. */ | |
359 | if (dmp->refcount == 0) { | |
360 | list_delete(dmp->list_node); | |
361 | /* destroy mp-pool. */ | |
362 | } | |
363 | #endif | |
364 | } | |
365 | ||
366 | static void | |
367 | check_link_status(struct netdev_dpdk *dev) | |
368 | { | |
369 | struct rte_eth_link link; | |
370 | ||
371 | rte_eth_link_get_nowait(dev->port_id, &link); | |
372 | ||
373 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 374 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
375 | |
376 | dev->link_reset_cnt++; | |
377 | dev->link = link; | |
378 | if (dev->link.link_status) { | |
379 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
380 | dev->port_id, (unsigned)dev->link.link_speed, | |
381 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
382 | ("full-duplex") : ("half-duplex")); | |
383 | } else { | |
384 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
385 | } | |
386 | } | |
387 | } | |
388 | ||
389 | static void * | |
390 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
391 | { | |
392 | struct netdev_dpdk *dev; | |
393 | ||
394 | pthread_detach(pthread_self()); | |
395 | ||
396 | for (;;) { | |
397 | ovs_mutex_lock(&dpdk_mutex); | |
398 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
399 | ovs_mutex_lock(&dev->mutex); | |
400 | check_link_status(dev); | |
401 | ovs_mutex_unlock(&dev->mutex); | |
402 | } | |
403 | ovs_mutex_unlock(&dpdk_mutex); | |
404 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
405 | } | |
406 | ||
407 | return NULL; | |
408 | } | |
409 | ||
410 | static int | |
411 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
412 | { | |
413 | struct rte_pktmbuf_pool_private *mbp_priv; | |
414 | struct ether_addr eth_addr; | |
415 | int diag; | |
416 | int i; | |
417 | ||
418 | if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) { | |
95fb793a | 419 | return ENODEV; |
8a9562d2 PS |
420 | } |
421 | ||
5496878c AW |
422 | diag = rte_eth_dev_configure(dev->port_id, dev->up.n_rxq, dev->up.n_txq, |
423 | &port_conf); | |
8a9562d2 PS |
424 | if (diag) { |
425 | VLOG_ERR("eth dev config error %d",diag); | |
95fb793a | 426 | return -diag; |
8a9562d2 PS |
427 | } |
428 | ||
5496878c | 429 | for (i = 0; i < dev->up.n_txq; i++) { |
79f5354c | 430 | diag = rte_eth_tx_queue_setup(dev->port_id, i, NIC_PORT_TX_Q_SIZE, |
d221ffa1 | 431 | dev->socket_id, &tx_conf); |
8a9562d2 PS |
432 | if (diag) { |
433 | VLOG_ERR("eth dev tx queue setup error %d",diag); | |
95fb793a | 434 | return -diag; |
8a9562d2 PS |
435 | } |
436 | } | |
437 | ||
5496878c | 438 | for (i = 0; i < dev->up.n_rxq; i++) { |
79f5354c | 439 | diag = rte_eth_rx_queue_setup(dev->port_id, i, NIC_PORT_RX_Q_SIZE, |
d221ffa1 | 440 | dev->socket_id, |
a715f600 | 441 | &rx_conf, dev->dpdk_mp->mp); |
8a9562d2 PS |
442 | if (diag) { |
443 | VLOG_ERR("eth dev rx queue setup error %d",diag); | |
95fb793a | 444 | return -diag; |
8a9562d2 PS |
445 | } |
446 | } | |
447 | ||
448 | diag = rte_eth_dev_start(dev->port_id); | |
449 | if (diag) { | |
450 | VLOG_ERR("eth dev start error %d",diag); | |
95fb793a | 451 | return -diag; |
8a9562d2 PS |
452 | } |
453 | ||
454 | rte_eth_promiscuous_enable(dev->port_id); | |
455 | rte_eth_allmulticast_enable(dev->port_id); | |
456 | ||
457 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
458 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
459 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
460 | dev->port_id, ETH_ADDR_ARGS(eth_addr.addr_bytes)); | |
461 | ||
462 | memcpy(dev->hwaddr, eth_addr.addr_bytes, ETH_ADDR_LEN); | |
463 | rte_eth_link_get_nowait(dev->port_id, &dev->link); | |
464 | ||
465 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
466 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
467 | ||
468 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
469 | return 0; | |
470 | } | |
471 | ||
472 | static struct netdev_dpdk * | |
473 | netdev_dpdk_cast(const struct netdev *netdev) | |
474 | { | |
475 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
476 | } | |
477 | ||
478 | static struct netdev * | |
479 | netdev_dpdk_alloc(void) | |
480 | { | |
481 | struct netdev_dpdk *netdev = dpdk_rte_mzalloc(sizeof *netdev); | |
482 | return &netdev->up; | |
483 | } | |
484 | ||
5a034064 | 485 | static void |
91968eb0 | 486 | netdev_dpdk_alloc_txq(struct netdev_dpdk *netdev, unsigned int n_txqs) |
5a034064 AW |
487 | { |
488 | int i; | |
489 | ||
490 | netdev->tx_q = dpdk_rte_mzalloc(n_txqs * sizeof *netdev->tx_q); | |
94143fc4 AW |
491 | /* Each index is considered as a cpu core id, since there should |
492 | * be one tx queue for each cpu core. */ | |
5a034064 | 493 | for (i = 0; i < n_txqs; i++) { |
ba0358a1 | 494 | int numa_id = ovs_numa_get_numa_id(i); |
94143fc4 | 495 | |
94143fc4 AW |
496 | /* If the corresponding core is not on the same numa node |
497 | * as 'netdev', flags the 'flush_tx'. */ | |
ba0358a1 | 498 | netdev->tx_q[i].flush_tx = netdev->socket_id == numa_id; |
5a034064 AW |
499 | } |
500 | } | |
501 | ||
8a9562d2 | 502 | static int |
58397e6c KT |
503 | netdev_dpdk_init(struct netdev *netdev_, unsigned int port_no, |
504 | enum dpdk_dev_type type) | |
5a034064 | 505 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 PS |
506 | { |
507 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1b7a04e0 | 508 | int sid; |
95fb793a | 509 | int err = 0; |
8a9562d2 | 510 | |
95fb793a | 511 | ovs_mutex_init(&netdev->mutex); |
95fb793a | 512 | ovs_mutex_lock(&netdev->mutex); |
8a9562d2 | 513 | |
1b7a04e0 AW |
514 | /* If the 'sid' is negative, it means that the kernel fails |
515 | * to obtain the pci numa info. In that situation, always | |
516 | * use 'SOCKET0'. */ | |
58397e6c KT |
517 | if (type == DPDK_DEV_ETH) { |
518 | sid = rte_eth_dev_socket_id(port_no); | |
519 | } else { | |
520 | sid = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
521 | } | |
522 | ||
1b7a04e0 | 523 | netdev->socket_id = sid < 0 ? SOCKET0 : sid; |
95fb793a | 524 | netdev->port_id = port_no; |
58397e6c | 525 | netdev->type = type; |
8a9562d2 | 526 | netdev->flags = 0; |
8a9562d2 PS |
527 | netdev->mtu = ETHER_MTU; |
528 | netdev->max_packet_len = MTU_TO_MAX_LEN(netdev->mtu); | |
58397e6c | 529 | rte_spinlock_init(&netdev->txq_lock); |
8a9562d2 | 530 | |
8a9562d2 PS |
531 | netdev->dpdk_mp = dpdk_mp_get(netdev->socket_id, netdev->mtu); |
532 | if (!netdev->dpdk_mp) { | |
533 | err = ENOMEM; | |
95fb793a | 534 | goto unlock; |
8a9562d2 PS |
535 | } |
536 | ||
5496878c AW |
537 | netdev_->n_txq = NR_QUEUE; |
538 | netdev_->n_rxq = NR_QUEUE; | |
58397e6c KT |
539 | |
540 | if (type == DPDK_DEV_ETH) { | |
541 | netdev_dpdk_alloc_txq(netdev, NR_QUEUE); | |
542 | err = dpdk_eth_dev_init(netdev); | |
543 | if (err) { | |
544 | goto unlock; | |
545 | } | |
8a9562d2 | 546 | } |
8a9562d2 PS |
547 | |
548 | list_push_back(&dpdk_list, &netdev->list_node); | |
549 | ||
95fb793a | 550 | unlock: |
5a034064 AW |
551 | if (err) { |
552 | rte_free(netdev->tx_q); | |
553 | } | |
8a9562d2 | 554 | ovs_mutex_unlock(&netdev->mutex); |
95fb793a | 555 | return err; |
556 | } | |
557 | ||
558 | static int | |
559 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
560 | unsigned int *port_no) | |
561 | { | |
562 | const char *cport; | |
563 | ||
564 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
565 | return ENODEV; | |
566 | } | |
567 | ||
568 | cport = dev_name + strlen(prefix); | |
569 | *port_no = strtol(cport, 0, 0); /* string must be null terminated */ | |
570 | return 0; | |
571 | } | |
572 | ||
58397e6c KT |
573 | static int |
574 | netdev_dpdk_vhost_construct(struct netdev *netdev_) | |
575 | { | |
576 | int err; | |
577 | ||
578 | if (rte_eal_init_ret) { | |
579 | return rte_eal_init_ret; | |
580 | } | |
581 | ||
582 | ovs_mutex_lock(&dpdk_mutex); | |
583 | err = netdev_dpdk_init(netdev_, -1, DPDK_DEV_VHOST); | |
584 | ovs_mutex_unlock(&dpdk_mutex); | |
585 | ||
586 | return err; | |
587 | } | |
588 | ||
95fb793a | 589 | static int |
590 | netdev_dpdk_construct(struct netdev *netdev) | |
591 | { | |
592 | unsigned int port_no; | |
593 | int err; | |
594 | ||
595 | if (rte_eal_init_ret) { | |
596 | return rte_eal_init_ret; | |
597 | } | |
598 | ||
599 | /* Names always start with "dpdk" */ | |
600 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
601 | if (err) { | |
602 | return err; | |
603 | } | |
604 | ||
605 | ovs_mutex_lock(&dpdk_mutex); | |
58397e6c | 606 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
8a9562d2 PS |
607 | ovs_mutex_unlock(&dpdk_mutex); |
608 | return err; | |
609 | } | |
610 | ||
611 | static void | |
612 | netdev_dpdk_destruct(struct netdev *netdev_) | |
613 | { | |
614 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
615 | ||
616 | ovs_mutex_lock(&dev->mutex); | |
617 | rte_eth_dev_stop(dev->port_id); | |
618 | ovs_mutex_unlock(&dev->mutex); | |
619 | ||
620 | ovs_mutex_lock(&dpdk_mutex); | |
5a034064 | 621 | rte_free(dev->tx_q); |
8a9562d2 PS |
622 | list_remove(&dev->list_node); |
623 | dpdk_mp_put(dev->dpdk_mp); | |
624 | ovs_mutex_unlock(&dpdk_mutex); | |
58397e6c | 625 | } |
8a9562d2 | 626 | |
58397e6c KT |
627 | static void |
628 | netdev_dpdk_vhost_destruct(struct netdev *netdev_) | |
629 | { | |
630 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
631 | ||
632 | /* Can't remove a port while a guest is attached to it. */ | |
633 | if (netdev_dpdk_get_virtio(dev) != NULL) { | |
634 | VLOG_ERR("Can not remove port, vhost device still attached"); | |
635 | return; | |
636 | } | |
637 | ||
638 | ovs_mutex_lock(&dpdk_mutex); | |
639 | list_remove(&dev->list_node); | |
640 | dpdk_mp_put(dev->dpdk_mp); | |
641 | ovs_mutex_unlock(&dpdk_mutex); | |
8a9562d2 PS |
642 | } |
643 | ||
644 | static void | |
645 | netdev_dpdk_dealloc(struct netdev *netdev_) | |
646 | { | |
647 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
648 | ||
649 | rte_free(netdev); | |
650 | } | |
651 | ||
652 | static int | |
653 | netdev_dpdk_get_config(const struct netdev *netdev_, struct smap *args) | |
654 | { | |
655 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
656 | ||
657 | ovs_mutex_lock(&dev->mutex); | |
658 | ||
47659062 AW |
659 | smap_add_format(args, "configured_rx_queues", "%d", netdev_->n_rxq); |
660 | smap_add_format(args, "configured_tx_queues", "%d", netdev_->n_txq); | |
8a9562d2 PS |
661 | ovs_mutex_unlock(&dev->mutex); |
662 | ||
663 | return 0; | |
664 | } | |
665 | ||
7dec44fe AW |
666 | static int |
667 | netdev_dpdk_get_numa_id(const struct netdev *netdev_) | |
668 | { | |
669 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
670 | ||
671 | return netdev->socket_id; | |
672 | } | |
673 | ||
5496878c AW |
674 | /* Sets the number of tx queues and rx queues for the dpdk interface. |
675 | * If the configuration fails, do not try restoring its old configuration | |
676 | * and just returns the error. */ | |
677 | static int | |
678 | netdev_dpdk_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
679 | unsigned int n_rxq) | |
680 | { | |
681 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
682 | int err = 0; | |
683 | ||
684 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
685 | return err; | |
686 | } | |
687 | ||
b7ccaf67 | 688 | ovs_mutex_lock(&dpdk_mutex); |
5496878c | 689 | ovs_mutex_lock(&netdev->mutex); |
91968eb0 | 690 | |
5496878c | 691 | rte_eth_dev_stop(netdev->port_id); |
91968eb0 | 692 | |
5496878c AW |
693 | netdev->up.n_txq = n_txq; |
694 | netdev->up.n_rxq = n_rxq; | |
58397e6c | 695 | |
91968eb0 AW |
696 | rte_free(netdev->tx_q); |
697 | netdev_dpdk_alloc_txq(netdev, n_txq); | |
5496878c | 698 | err = dpdk_eth_dev_init(netdev); |
91968eb0 | 699 | |
5496878c | 700 | ovs_mutex_unlock(&netdev->mutex); |
b7ccaf67 | 701 | ovs_mutex_unlock(&dpdk_mutex); |
5496878c AW |
702 | |
703 | return err; | |
704 | } | |
705 | ||
58397e6c KT |
706 | static int |
707 | netdev_dpdk_vhost_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
708 | unsigned int n_rxq) | |
709 | { | |
710 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
711 | int err = 0; | |
712 | ||
713 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
714 | return err; | |
715 | } | |
716 | ||
717 | ovs_mutex_lock(&dpdk_mutex); | |
718 | ovs_mutex_lock(&netdev->mutex); | |
719 | ||
720 | netdev->up.n_txq = n_txq; | |
721 | netdev->up.n_rxq = n_rxq; | |
722 | ||
723 | ovs_mutex_unlock(&netdev->mutex); | |
724 | ovs_mutex_unlock(&dpdk_mutex); | |
725 | ||
726 | return err; | |
727 | } | |
728 | ||
8a9562d2 PS |
729 | static struct netdev_rxq * |
730 | netdev_dpdk_rxq_alloc(void) | |
731 | { | |
732 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
733 | ||
734 | return &rx->up; | |
735 | } | |
736 | ||
737 | static struct netdev_rxq_dpdk * | |
738 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rx) | |
739 | { | |
740 | return CONTAINER_OF(rx, struct netdev_rxq_dpdk, up); | |
741 | } | |
742 | ||
743 | static int | |
744 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq_) | |
745 | { | |
746 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
747 | struct netdev_dpdk *netdev = netdev_dpdk_cast(rx->up.netdev); | |
748 | ||
749 | ovs_mutex_lock(&netdev->mutex); | |
750 | rx->port_id = netdev->port_id; | |
751 | ovs_mutex_unlock(&netdev->mutex); | |
752 | ||
753 | return 0; | |
754 | } | |
755 | ||
756 | static void | |
757 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq_ OVS_UNUSED) | |
758 | { | |
759 | } | |
760 | ||
761 | static void | |
762 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq_) | |
763 | { | |
764 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
765 | ||
766 | rte_free(rx); | |
767 | } | |
768 | ||
b170db2a RW |
769 | static inline void |
770 | dpdk_queue_flush__(struct netdev_dpdk *dev, int qid) | |
8a9562d2 PS |
771 | { |
772 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1304f1f8 DDP |
773 | uint32_t nb_tx = 0; |
774 | ||
775 | while (nb_tx != txq->count) { | |
776 | uint32_t ret; | |
777 | ||
778 | ret = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts + nb_tx, | |
779 | txq->count - nb_tx); | |
780 | if (!ret) { | |
781 | break; | |
782 | } | |
783 | ||
784 | nb_tx += ret; | |
785 | } | |
8a9562d2 | 786 | |
b170db2a | 787 | if (OVS_UNLIKELY(nb_tx != txq->count)) { |
db73f716 DDP |
788 | /* free buffers, which we couldn't transmit, one at a time (each |
789 | * packet could come from a different mempool) */ | |
790 | int i; | |
791 | ||
792 | for (i = nb_tx; i < txq->count; i++) { | |
793 | rte_pktmbuf_free_seg(txq->burst_pkts[i]); | |
794 | } | |
1304f1f8 DDP |
795 | ovs_mutex_lock(&dev->mutex); |
796 | dev->stats.tx_dropped += txq->count-nb_tx; | |
797 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 | 798 | } |
1304f1f8 | 799 | |
8a9562d2 | 800 | txq->count = 0; |
844f2d74 | 801 | txq->tsc = rte_get_timer_cycles(); |
b170db2a RW |
802 | } |
803 | ||
804 | static inline void | |
805 | dpdk_queue_flush(struct netdev_dpdk *dev, int qid) | |
806 | { | |
807 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
808 | ||
809 | if (txq->count == 0) { | |
810 | return; | |
811 | } | |
b170db2a | 812 | dpdk_queue_flush__(dev, qid); |
8a9562d2 PS |
813 | } |
814 | ||
58397e6c KT |
815 | static bool |
816 | is_vhost_running(struct virtio_net *dev) | |
817 | { | |
818 | return (dev != NULL && (dev->flags & VIRTIO_DEV_RUNNING)); | |
819 | } | |
820 | ||
821 | /* | |
822 | * The receive path for the vhost port is the TX path out from guest. | |
823 | */ | |
824 | static int | |
825 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq_, | |
826 | struct dp_packet **packets, int *c) | |
827 | { | |
828 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
829 | struct netdev *netdev = rx->up.netdev; | |
830 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
831 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
832 | int qid = 1; | |
833 | uint16_t nb_rx = 0; | |
834 | ||
835 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
836 | return EAGAIN; | |
837 | } | |
838 | ||
839 | nb_rx = rte_vhost_dequeue_burst(virtio_dev, qid, | |
840 | vhost_dev->dpdk_mp->mp, | |
841 | (struct rte_mbuf **)packets, | |
842 | MAX_PKT_BURST); | |
843 | if (!nb_rx) { | |
844 | return EAGAIN; | |
845 | } | |
846 | ||
847 | vhost_dev->stats.rx_packets += (uint64_t)nb_rx; | |
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 DDP |
868 | (struct rte_mbuf **) packets, |
869 | MIN((int)NETDEV_MAX_RX_BATCH, | |
870 | (int)MAX_RX_QUEUE_LEN)); | |
8a9562d2 PS |
871 | if (!nb_rx) { |
872 | return EAGAIN; | |
873 | } | |
874 | ||
8a9562d2 PS |
875 | *c = nb_rx; |
876 | ||
877 | return 0; | |
878 | } | |
879 | ||
58397e6c KT |
880 | static void |
881 | __netdev_dpdk_vhost_send(struct netdev *netdev, struct dp_packet **pkts, | |
882 | int cnt, bool may_steal) | |
883 | { | |
884 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
885 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
886 | int tx_pkts, i; | |
887 | ||
888 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
889 | ovs_mutex_lock(&vhost_dev->mutex); | |
890 | vhost_dev->stats.tx_dropped+= cnt; | |
891 | ovs_mutex_unlock(&vhost_dev->mutex); | |
892 | goto out; | |
893 | } | |
894 | ||
895 | /* There is vHost TX single queue, So we need to lock it for TX. */ | |
896 | rte_spinlock_lock(&vhost_dev->txq_lock); | |
897 | tx_pkts = rte_vhost_enqueue_burst(virtio_dev, VIRTIO_RXQ, | |
898 | (struct rte_mbuf **)pkts, cnt); | |
899 | ||
900 | vhost_dev->stats.tx_packets += tx_pkts; | |
901 | vhost_dev->stats.tx_dropped += (cnt - tx_pkts); | |
902 | rte_spinlock_unlock(&vhost_dev->txq_lock); | |
903 | ||
904 | out: | |
905 | if (may_steal) { | |
906 | for (i = 0; i < cnt; i++) { | |
907 | dp_packet_delete(pkts[i]); | |
908 | } | |
909 | } | |
910 | } | |
911 | ||
8a9562d2 | 912 | inline static void |
f4fd623c DDP |
913 | dpdk_queue_pkts(struct netdev_dpdk *dev, int qid, |
914 | struct rte_mbuf **pkts, int cnt) | |
8a9562d2 PS |
915 | { |
916 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
917 | uint64_t diff_tsc; | |
8a9562d2 | 918 | |
f4fd623c DDP |
919 | int i = 0; |
920 | ||
f4fd623c DDP |
921 | while (i < cnt) { |
922 | int freeslots = MAX_TX_QUEUE_LEN - txq->count; | |
923 | int tocopy = MIN(freeslots, cnt-i); | |
8a9562d2 | 924 | |
f4fd623c DDP |
925 | memcpy(&txq->burst_pkts[txq->count], &pkts[i], |
926 | tocopy * sizeof (struct rte_mbuf *)); | |
927 | ||
928 | txq->count += tocopy; | |
929 | i += tocopy; | |
930 | ||
94143fc4 | 931 | if (txq->count == MAX_TX_QUEUE_LEN || txq->flush_tx) { |
b170db2a | 932 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 933 | } |
844f2d74 | 934 | diff_tsc = rte_get_timer_cycles() - txq->tsc; |
f4fd623c | 935 | if (diff_tsc >= DRAIN_TSC) { |
b170db2a | 936 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 937 | } |
8a9562d2 | 938 | } |
8a9562d2 PS |
939 | } |
940 | ||
941 | /* Tx function. Transmit packets indefinitely */ | |
942 | static void | |
58397e6c | 943 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet **pkts, |
2654cc33 | 944 | int cnt) |
db73f716 | 945 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 PS |
946 | { |
947 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
f4fd623c | 948 | struct rte_mbuf *mbufs[cnt]; |
175cf4de RW |
949 | int dropped = 0; |
950 | int newcnt = 0; | |
951 | int i; | |
8a9562d2 | 952 | |
db73f716 DDP |
953 | /* If we are on a non pmd thread we have to use the mempool mutex, because |
954 | * every non pmd thread shares the same mempool cache */ | |
955 | ||
956 | if (!thread_is_pmd()) { | |
957 | ovs_mutex_lock(&nonpmd_mempool_mutex); | |
958 | } | |
959 | ||
f4fd623c | 960 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 961 | int size = dp_packet_size(pkts[i]); |
95fb793a | 962 | |
f98d7864 | 963 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c DDP |
964 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
965 | (int)size , dev->max_packet_len); | |
966 | ||
175cf4de | 967 | dropped++; |
f4fd623c DDP |
968 | continue; |
969 | } | |
8a9562d2 | 970 | |
f4fd623c | 971 | mbufs[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 972 | |
f4fd623c | 973 | if (!mbufs[newcnt]) { |
175cf4de RW |
974 | dropped += cnt - i; |
975 | break; | |
f4fd623c DDP |
976 | } |
977 | ||
978 | /* We have to do a copy for now */ | |
b8e57534 | 979 | memcpy(rte_pktmbuf_mtod(mbufs[newcnt], void *), dp_packet_data(pkts[i]), size); |
f4fd623c DDP |
980 | |
981 | rte_pktmbuf_data_len(mbufs[newcnt]) = size; | |
982 | rte_pktmbuf_pkt_len(mbufs[newcnt]) = size; | |
983 | ||
984 | newcnt++; | |
985 | } | |
8a9562d2 | 986 | |
f98d7864 | 987 | if (OVS_UNLIKELY(dropped)) { |
175cf4de RW |
988 | ovs_mutex_lock(&dev->mutex); |
989 | dev->stats.tx_dropped += dropped; | |
990 | ovs_mutex_unlock(&dev->mutex); | |
991 | } | |
992 | ||
58397e6c KT |
993 | if (dev->type == DPDK_DEV_VHOST) { |
994 | __netdev_dpdk_vhost_send(netdev, (struct dp_packet **) mbufs, newcnt, true); | |
995 | } else { | |
996 | dpdk_queue_pkts(dev, qid, mbufs, newcnt); | |
997 | dpdk_queue_flush(dev, qid); | |
998 | } | |
db73f716 DDP |
999 | |
1000 | if (!thread_is_pmd()) { | |
1001 | ovs_mutex_unlock(&nonpmd_mempool_mutex); | |
1002 | } | |
8a9562d2 PS |
1003 | } |
1004 | ||
58397e6c KT |
1005 | static int |
1006 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid OVS_UNUSED, struct dp_packet **pkts, | |
1007 | int cnt, bool may_steal) | |
1008 | { | |
1009 | if (OVS_UNLIKELY(pkts[0]->source != DPBUF_DPDK)) { | |
1010 | int i; | |
1011 | ||
1012 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); | |
1013 | if (may_steal) { | |
1014 | for (i = 0; i < cnt; i++) { | |
1015 | dp_packet_delete(pkts[i]); | |
1016 | } | |
1017 | } | |
1018 | } else { | |
1019 | __netdev_dpdk_vhost_send(netdev, pkts, cnt, may_steal); | |
1020 | } | |
1021 | return 0; | |
1022 | } | |
1023 | ||
7251515e DV |
1024 | static inline void |
1025 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
e14deea0 | 1026 | struct dp_packet **pkts, int cnt, bool may_steal) |
8a9562d2 | 1027 | { |
f4fd623c | 1028 | int i; |
8a9562d2 | 1029 | |
7251515e | 1030 | if (OVS_UNLIKELY(!may_steal || |
cf62fa4c | 1031 | pkts[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1032 | struct netdev *netdev = &dev->up; |
1033 | ||
2654cc33 | 1034 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); |
b3cd9f9d PS |
1035 | |
1036 | if (may_steal) { | |
f4fd623c | 1037 | for (i = 0; i < cnt; i++) { |
e14deea0 | 1038 | dp_packet_delete(pkts[i]); |
f4fd623c | 1039 | } |
b3cd9f9d | 1040 | } |
8a9562d2 | 1041 | } else { |
f4fd623c DDP |
1042 | int next_tx_idx = 0; |
1043 | int dropped = 0; | |
8a9562d2 | 1044 | |
f4fd623c | 1045 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1046 | int size = dp_packet_size(pkts[i]); |
f4fd623c DDP |
1047 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
1048 | if (next_tx_idx != i) { | |
1049 | dpdk_queue_pkts(dev, qid, | |
1050 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1051 | i-next_tx_idx); | |
1ebfe1ac | 1052 | } |
f4fd623c | 1053 | |
1ebfe1ac DDP |
1054 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1055 | (int)size , dev->max_packet_len); | |
f4fd623c | 1056 | |
e14deea0 | 1057 | dp_packet_delete(pkts[i]); |
1ebfe1ac | 1058 | dropped++; |
f4fd623c DDP |
1059 | next_tx_idx = i + 1; |
1060 | } | |
1061 | } | |
1062 | if (next_tx_idx != cnt) { | |
1063 | dpdk_queue_pkts(dev, qid, | |
1064 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1065 | cnt-next_tx_idx); | |
1066 | } | |
8a9562d2 | 1067 | |
f4fd623c DDP |
1068 | if (OVS_UNLIKELY(dropped)) { |
1069 | ovs_mutex_lock(&dev->mutex); | |
1070 | dev->stats.tx_dropped += dropped; | |
1071 | ovs_mutex_unlock(&dev->mutex); | |
1072 | } | |
8a9562d2 | 1073 | } |
7251515e DV |
1074 | } |
1075 | ||
1076 | static int | |
1077 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
e14deea0 | 1078 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1079 | { |
1080 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1081 | |
7251515e DV |
1082 | netdev_dpdk_send__(dev, qid, pkts, cnt, may_steal); |
1083 | return 0; | |
8a9562d2 PS |
1084 | } |
1085 | ||
1086 | static int | |
1087 | netdev_dpdk_set_etheraddr(struct netdev *netdev, | |
1088 | const uint8_t mac[ETH_ADDR_LEN]) | |
1089 | { | |
1090 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1091 | ||
1092 | ovs_mutex_lock(&dev->mutex); | |
1093 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
1094 | memcpy(dev->hwaddr, mac, ETH_ADDR_LEN); | |
045c0d1a | 1095 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1096 | } |
1097 | ovs_mutex_unlock(&dev->mutex); | |
1098 | ||
1099 | return 0; | |
1100 | } | |
1101 | ||
1102 | static int | |
1103 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, | |
1104 | uint8_t mac[ETH_ADDR_LEN]) | |
1105 | { | |
1106 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1107 | ||
1108 | ovs_mutex_lock(&dev->mutex); | |
1109 | memcpy(mac, dev->hwaddr, ETH_ADDR_LEN); | |
1110 | ovs_mutex_unlock(&dev->mutex); | |
1111 | ||
1112 | return 0; | |
1113 | } | |
1114 | ||
1115 | static int | |
1116 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1117 | { | |
1118 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1119 | ||
1120 | ovs_mutex_lock(&dev->mutex); | |
1121 | *mtup = dev->mtu; | |
1122 | ovs_mutex_unlock(&dev->mutex); | |
1123 | ||
1124 | return 0; | |
1125 | } | |
1126 | ||
1127 | static int | |
1128 | netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu) | |
1129 | { | |
1130 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1131 | int old_mtu, err; | |
1132 | struct dpdk_mp *old_mp; | |
1133 | struct dpdk_mp *mp; | |
1134 | ||
1135 | ovs_mutex_lock(&dpdk_mutex); | |
1136 | ovs_mutex_lock(&dev->mutex); | |
1137 | if (dev->mtu == mtu) { | |
1138 | err = 0; | |
1139 | goto out; | |
1140 | } | |
1141 | ||
1142 | mp = dpdk_mp_get(dev->socket_id, dev->mtu); | |
1143 | if (!mp) { | |
1144 | err = ENOMEM; | |
1145 | goto out; | |
1146 | } | |
1147 | ||
1148 | rte_eth_dev_stop(dev->port_id); | |
1149 | ||
1150 | old_mtu = dev->mtu; | |
1151 | old_mp = dev->dpdk_mp; | |
1152 | dev->dpdk_mp = mp; | |
1153 | dev->mtu = mtu; | |
1154 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
1155 | ||
1156 | err = dpdk_eth_dev_init(dev); | |
1157 | if (err) { | |
8a9562d2 PS |
1158 | dpdk_mp_put(mp); |
1159 | dev->mtu = old_mtu; | |
1160 | dev->dpdk_mp = old_mp; | |
1161 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
1162 | dpdk_eth_dev_init(dev); | |
1163 | goto out; | |
1164 | } | |
1165 | ||
1166 | dpdk_mp_put(old_mp); | |
045c0d1a | 1167 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1168 | out: |
1169 | ovs_mutex_unlock(&dev->mutex); | |
1170 | ovs_mutex_unlock(&dpdk_mutex); | |
1171 | return err; | |
1172 | } | |
1173 | ||
1174 | static int | |
1175 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier); | |
1176 | ||
58397e6c KT |
1177 | static int |
1178 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1179 | struct netdev_stats *stats) | |
1180 | { | |
1181 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1182 | ||
1183 | ovs_mutex_lock(&dev->mutex); | |
1184 | memset(stats, 0, sizeof(*stats)); | |
1185 | /* Unsupported Stats */ | |
1186 | stats->rx_errors = UINT64_MAX; | |
1187 | stats->tx_errors = UINT64_MAX; | |
1188 | stats->multicast = UINT64_MAX; | |
1189 | stats->collisions = UINT64_MAX; | |
1190 | stats->rx_crc_errors = UINT64_MAX; | |
1191 | stats->rx_fifo_errors = UINT64_MAX; | |
1192 | stats->rx_frame_errors = UINT64_MAX; | |
1193 | stats->rx_length_errors = UINT64_MAX; | |
1194 | stats->rx_missed_errors = UINT64_MAX; | |
1195 | stats->rx_over_errors = UINT64_MAX; | |
1196 | stats->tx_aborted_errors = UINT64_MAX; | |
1197 | stats->tx_carrier_errors = UINT64_MAX; | |
1198 | stats->tx_errors = UINT64_MAX; | |
1199 | stats->tx_fifo_errors = UINT64_MAX; | |
1200 | stats->tx_heartbeat_errors = UINT64_MAX; | |
1201 | stats->tx_window_errors = UINT64_MAX; | |
1202 | stats->rx_bytes += UINT64_MAX; | |
1203 | stats->rx_dropped += UINT64_MAX; | |
1204 | stats->tx_bytes += UINT64_MAX; | |
1205 | ||
1206 | /* Supported Stats */ | |
1207 | stats->rx_packets += dev->stats.rx_packets; | |
1208 | stats->tx_packets += dev->stats.tx_packets; | |
1209 | stats->tx_dropped += dev->stats.tx_dropped; | |
1210 | ovs_mutex_unlock(&dev->mutex); | |
1211 | ||
1212 | return 0; | |
1213 | } | |
1214 | ||
8a9562d2 PS |
1215 | static int |
1216 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1217 | { | |
1218 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1219 | struct rte_eth_stats rte_stats; | |
1220 | bool gg; | |
1221 | ||
1222 | netdev_dpdk_get_carrier(netdev, &gg); | |
1223 | ovs_mutex_lock(&dev->mutex); | |
1224 | rte_eth_stats_get(dev->port_id, &rte_stats); | |
1225 | ||
2f9dd77f | 1226 | memset(stats, 0, sizeof(*stats)); |
8a9562d2 | 1227 | |
2f9dd77f PS |
1228 | stats->rx_packets = rte_stats.ipackets; |
1229 | stats->tx_packets = rte_stats.opackets; | |
1230 | stats->rx_bytes = rte_stats.ibytes; | |
1231 | stats->tx_bytes = rte_stats.obytes; | |
1232 | stats->rx_errors = rte_stats.ierrors; | |
1233 | stats->tx_errors = rte_stats.oerrors; | |
1234 | stats->multicast = rte_stats.imcasts; | |
8a9562d2 | 1235 | |
2f9dd77f | 1236 | stats->tx_dropped = dev->stats.tx_dropped; |
8a9562d2 PS |
1237 | ovs_mutex_unlock(&dev->mutex); |
1238 | ||
1239 | return 0; | |
1240 | } | |
1241 | ||
1242 | static int | |
1243 | netdev_dpdk_get_features(const struct netdev *netdev_, | |
1244 | enum netdev_features *current, | |
1245 | enum netdev_features *advertised OVS_UNUSED, | |
1246 | enum netdev_features *supported OVS_UNUSED, | |
1247 | enum netdev_features *peer OVS_UNUSED) | |
1248 | { | |
1249 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1250 | struct rte_eth_link link; | |
1251 | ||
1252 | ovs_mutex_lock(&dev->mutex); | |
1253 | link = dev->link; | |
1254 | ovs_mutex_unlock(&dev->mutex); | |
1255 | ||
1256 | if (link.link_duplex == ETH_LINK_AUTONEG_DUPLEX) { | |
1257 | if (link.link_speed == ETH_LINK_SPEED_AUTONEG) { | |
1258 | *current = NETDEV_F_AUTONEG; | |
1259 | } | |
1260 | } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { | |
1261 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1262 | *current = NETDEV_F_10MB_HD; | |
1263 | } | |
1264 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1265 | *current = NETDEV_F_100MB_HD; | |
1266 | } | |
1267 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1268 | *current = NETDEV_F_1GB_HD; | |
1269 | } | |
1270 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
1271 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1272 | *current = NETDEV_F_10MB_FD; | |
1273 | } | |
1274 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1275 | *current = NETDEV_F_100MB_FD; | |
1276 | } | |
1277 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1278 | *current = NETDEV_F_1GB_FD; | |
1279 | } | |
1280 | if (link.link_speed == ETH_LINK_SPEED_10000) { | |
1281 | *current = NETDEV_F_10GB_FD; | |
1282 | } | |
1283 | } | |
1284 | ||
1285 | return 0; | |
1286 | } | |
1287 | ||
1288 | static int | |
1289 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
1290 | { | |
1291 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1292 | int ifindex; | |
1293 | ||
1294 | ovs_mutex_lock(&dev->mutex); | |
1295 | ifindex = dev->port_id; | |
1296 | ovs_mutex_unlock(&dev->mutex); | |
1297 | ||
1298 | return ifindex; | |
1299 | } | |
1300 | ||
1301 | static int | |
1302 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1303 | { | |
1304 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1305 | ||
1306 | ovs_mutex_lock(&dev->mutex); | |
1307 | check_link_status(dev); | |
1308 | *carrier = dev->link.link_status; | |
58397e6c KT |
1309 | |
1310 | ovs_mutex_unlock(&dev->mutex); | |
1311 | ||
1312 | return 0; | |
1313 | } | |
1314 | ||
1315 | static int | |
1316 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1317 | { | |
1318 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1319 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(dev); | |
1320 | ||
1321 | ovs_mutex_lock(&dev->mutex); | |
1322 | ||
1323 | if (is_vhost_running(virtio_dev)) { | |
1324 | *carrier = 1; | |
1325 | } else { | |
1326 | *carrier = 0; | |
1327 | } | |
1328 | ||
8a9562d2 PS |
1329 | ovs_mutex_unlock(&dev->mutex); |
1330 | ||
1331 | return 0; | |
1332 | } | |
1333 | ||
1334 | static long long int | |
1335 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev_) | |
1336 | { | |
1337 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1338 | long long int carrier_resets; | |
1339 | ||
1340 | ovs_mutex_lock(&dev->mutex); | |
1341 | carrier_resets = dev->link_reset_cnt; | |
1342 | ovs_mutex_unlock(&dev->mutex); | |
1343 | ||
1344 | return carrier_resets; | |
1345 | } | |
1346 | ||
1347 | static int | |
1348 | netdev_dpdk_set_miimon(struct netdev *netdev_ OVS_UNUSED, | |
1349 | long long int interval OVS_UNUSED) | |
1350 | { | |
ee32150e | 1351 | return EOPNOTSUPP; |
8a9562d2 PS |
1352 | } |
1353 | ||
1354 | static int | |
1355 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
1356 | enum netdev_flags off, enum netdev_flags on, | |
95fb793a | 1357 | enum netdev_flags *old_flagsp) OVS_REQUIRES(dev->mutex) |
8a9562d2 PS |
1358 | { |
1359 | int err; | |
1360 | ||
1361 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
1362 | return EINVAL; | |
1363 | } | |
1364 | ||
1365 | *old_flagsp = dev->flags; | |
1366 | dev->flags |= on; | |
1367 | dev->flags &= ~off; | |
1368 | ||
1369 | if (dev->flags == *old_flagsp) { | |
1370 | return 0; | |
1371 | } | |
1372 | ||
58397e6c KT |
1373 | if (dev->type == DPDK_DEV_ETH) { |
1374 | if (dev->flags & NETDEV_UP) { | |
1375 | err = rte_eth_dev_start(dev->port_id); | |
1376 | if (err) | |
1377 | return -err; | |
1378 | } | |
8a9562d2 | 1379 | |
58397e6c KT |
1380 | if (dev->flags & NETDEV_PROMISC) { |
1381 | rte_eth_promiscuous_enable(dev->port_id); | |
1382 | } | |
8a9562d2 | 1383 | |
58397e6c KT |
1384 | if (!(dev->flags & NETDEV_UP)) { |
1385 | rte_eth_dev_stop(dev->port_id); | |
1386 | } | |
8a9562d2 PS |
1387 | } |
1388 | ||
1389 | return 0; | |
1390 | } | |
1391 | ||
1392 | static int | |
1393 | netdev_dpdk_update_flags(struct netdev *netdev_, | |
1394 | enum netdev_flags off, enum netdev_flags on, | |
1395 | enum netdev_flags *old_flagsp) | |
1396 | { | |
1397 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1398 | int error; | |
1399 | ||
1400 | ovs_mutex_lock(&netdev->mutex); | |
1401 | error = netdev_dpdk_update_flags__(netdev, off, on, old_flagsp); | |
1402 | ovs_mutex_unlock(&netdev->mutex); | |
1403 | ||
1404 | return error; | |
1405 | } | |
1406 | ||
1407 | static int | |
1408 | netdev_dpdk_get_status(const struct netdev *netdev_, struct smap *args) | |
1409 | { | |
1410 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1411 | struct rte_eth_dev_info dev_info; | |
1412 | ||
e0a801c7 | 1413 | if (dev->port_id < 0) |
8a9562d2 PS |
1414 | return ENODEV; |
1415 | ||
1416 | ovs_mutex_lock(&dev->mutex); | |
1417 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
1418 | ovs_mutex_unlock(&dev->mutex); | |
1419 | ||
1420 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1421 | ||
95fb793a | 1422 | smap_add_format(args, "port_no", "%d", dev->port_id); |
8a9562d2 PS |
1423 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); |
1424 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1425 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
1426 | smap_add_format(args, "max_rx_pktlen", "%u", dev_info.max_rx_pktlen); | |
1427 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); | |
1428 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
1429 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
1430 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
1431 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
1432 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
1433 | ||
1434 | smap_add_format(args, "pci-vendor_id", "0x%u", dev_info.pci_dev->id.vendor_id); | |
1435 | smap_add_format(args, "pci-device_id", "0x%x", dev_info.pci_dev->id.device_id); | |
1436 | ||
1437 | return 0; | |
1438 | } | |
1439 | ||
1440 | static void | |
1441 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
1442 | OVS_REQUIRES(dev->mutex) | |
1443 | { | |
1444 | enum netdev_flags old_flags; | |
1445 | ||
1446 | if (admin_state) { | |
1447 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
1448 | } else { | |
1449 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
1450 | } | |
1451 | } | |
1452 | ||
1453 | static void | |
1454 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
1455 | const char *argv[], void *aux OVS_UNUSED) | |
1456 | { | |
1457 | bool up; | |
1458 | ||
1459 | if (!strcasecmp(argv[argc - 1], "up")) { | |
1460 | up = true; | |
1461 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
1462 | up = false; | |
1463 | } else { | |
1464 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
1465 | return; | |
1466 | } | |
1467 | ||
1468 | if (argc > 2) { | |
1469 | struct netdev *netdev = netdev_from_name(argv[1]); | |
1470 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
1471 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
1472 | ||
1473 | ovs_mutex_lock(&dpdk_dev->mutex); | |
1474 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
1475 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
1476 | ||
1477 | netdev_close(netdev); | |
1478 | } else { | |
1479 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
1480 | netdev_close(netdev); | |
1481 | return; | |
1482 | } | |
1483 | } else { | |
1484 | struct netdev_dpdk *netdev; | |
1485 | ||
1486 | ovs_mutex_lock(&dpdk_mutex); | |
1487 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
1488 | ovs_mutex_lock(&netdev->mutex); | |
1489 | netdev_dpdk_set_admin_state__(netdev, up); | |
1490 | ovs_mutex_unlock(&netdev->mutex); | |
1491 | } | |
1492 | ovs_mutex_unlock(&dpdk_mutex); | |
1493 | } | |
1494 | unixctl_command_reply(conn, "OK"); | |
1495 | } | |
1496 | ||
58397e6c KT |
1497 | /* |
1498 | * Set virtqueue flags so that we do not receive interrupts. | |
1499 | */ | |
1500 | static void | |
1501 | set_irq_status(struct virtio_net *dev) | |
1502 | { | |
1503 | dev->virtqueue[VIRTIO_RXQ]->used->flags = VRING_USED_F_NO_NOTIFY; | |
1504 | dev->virtqueue[VIRTIO_TXQ]->used->flags = VRING_USED_F_NO_NOTIFY; | |
1505 | } | |
1506 | ||
1507 | /* | |
1508 | * A new virtio-net device is added to a vhost port. | |
1509 | */ | |
1510 | static int | |
1511 | new_device(struct virtio_net *dev) | |
1512 | { | |
1513 | struct netdev_dpdk *netdev; | |
1514 | bool exists = false; | |
1515 | ||
1516 | ovs_mutex_lock(&dpdk_mutex); | |
1517 | /* Add device to the vhost port with the same name as that passed down. */ | |
1518 | LIST_FOR_EACH(netdev, list_node, &dpdk_list) { | |
1519 | if (strncmp(dev->ifname, netdev->up.name, IFNAMSIZ) == 0) { | |
1520 | ovs_mutex_lock(&netdev->mutex); | |
1521 | ovsrcu_set(&netdev->virtio_dev, dev); | |
1522 | ovs_mutex_unlock(&netdev->mutex); | |
1523 | exists = true; | |
1524 | dev->flags |= VIRTIO_DEV_RUNNING; | |
1525 | /* Disable notifications. */ | |
1526 | set_irq_status(dev); | |
1527 | break; | |
1528 | } | |
1529 | } | |
1530 | ovs_mutex_unlock(&dpdk_mutex); | |
1531 | ||
1532 | if (!exists) { | |
1533 | VLOG_INFO("vHost Device '%s' (%ld) can't be added - name not found", | |
1534 | dev->ifname, dev->device_fh); | |
1535 | ||
1536 | return -1; | |
1537 | } | |
1538 | ||
1539 | VLOG_INFO("vHost Device '%s' (%ld) has been added", | |
1540 | dev->ifname, dev->device_fh); | |
1541 | return 0; | |
1542 | } | |
1543 | ||
1544 | /* | |
1545 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
1546 | * flag to stop any more packets from being sent or received to/from a VM and | |
1547 | * ensure all currently queued packets have been sent/received before removing | |
1548 | * the device. | |
1549 | */ | |
1550 | static void | |
1551 | destroy_device(volatile struct virtio_net *dev) | |
1552 | { | |
1553 | struct netdev_dpdk *vhost_dev; | |
1554 | ||
1555 | ovs_mutex_lock(&dpdk_mutex); | |
1556 | LIST_FOR_EACH (vhost_dev, list_node, &dpdk_list) { | |
1557 | if (netdev_dpdk_get_virtio(vhost_dev) == dev) { | |
1558 | ||
1559 | ovs_mutex_lock(&vhost_dev->mutex); | |
1560 | dev->flags &= ~VIRTIO_DEV_RUNNING; | |
1561 | ovsrcu_set(&vhost_dev->virtio_dev, NULL); | |
1562 | ovs_mutex_unlock(&vhost_dev->mutex); | |
1563 | ||
1564 | /* | |
1565 | * Wait for other threads to quiesce before | |
1566 | * setting the virtio_dev to NULL. | |
1567 | */ | |
1568 | ovsrcu_synchronize(); | |
1569 | } | |
1570 | } | |
1571 | ovs_mutex_unlock(&dpdk_mutex); | |
1572 | ||
1573 | VLOG_INFO("vHost Device '%s' (%ld) has been removed", | |
1574 | dev->ifname, dev->device_fh); | |
1575 | } | |
1576 | ||
1577 | struct virtio_net * | |
1578 | netdev_dpdk_get_virtio(const struct netdev_dpdk *dev) | |
1579 | { | |
1580 | return ovsrcu_get(struct virtio_net *, &dev->virtio_dev); | |
1581 | } | |
1582 | ||
1583 | /* | |
1584 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
1585 | * configuration has been fully complete. | |
1586 | */ | |
1587 | const struct virtio_net_device_ops virtio_net_device_ops = | |
1588 | { | |
1589 | .new_device = new_device, | |
1590 | .destroy_device = destroy_device, | |
1591 | }; | |
1592 | ||
1593 | static void * | |
1594 | start_cuse_session_loop(void *dummy OVS_UNUSED) | |
1595 | { | |
1596 | pthread_detach(pthread_self()); | |
1597 | rte_vhost_driver_session_start(); | |
1598 | return NULL; | |
1599 | } | |
1600 | ||
1601 | static int | |
1602 | dpdk_vhost_class_init(void) | |
1603 | { | |
1604 | pthread_t thread; | |
1605 | int err = -1; | |
1606 | ||
1607 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
1608 | ||
1609 | /* Register CUSE device to handle IOCTLs. | |
1610 | * Unless otherwise specified on the vswitchd command line, cuse_dev_name | |
1611 | * is set to vhost-net. | |
1612 | */ | |
1613 | err = rte_vhost_driver_register(cuse_dev_name); | |
1614 | ||
1615 | if (err != 0) { | |
1616 | VLOG_ERR("CUSE device setup failure."); | |
1617 | return -1; | |
1618 | } | |
1619 | ||
1620 | /* start_cuse_session_loop blocks OVS RCU quiescent state, so directly use | |
1621 | * pthread API. */ | |
1622 | return pthread_create(&thread, NULL, start_cuse_session_loop, NULL); | |
1623 | } | |
1624 | ||
033e9df2 DDP |
1625 | static void |
1626 | dpdk_common_init(void) | |
1627 | { | |
1628 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
1629 | "[netdev] up|down", 1, 2, | |
1630 | netdev_dpdk_set_admin_state, NULL); | |
1631 | ||
1632 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
1633 | } | |
1634 | ||
95fb793a | 1635 | /* Client Rings */ |
1636 | ||
95fb793a | 1637 | static int |
1638 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
1639 | unsigned int *eth_port_id) | |
1640 | { | |
1641 | struct dpdk_ring *ivshmem; | |
1642 | char ring_name[10]; | |
1643 | int err; | |
1644 | ||
1645 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
1646 | if (ivshmem == NULL) { | |
1647 | return ENOMEM; | |
1648 | } | |
1649 | ||
7251515e | 1650 | /* XXX: Add support for multiquque ring. */ |
95fb793a | 1651 | err = snprintf(ring_name, 10, "%s_tx", dev_name); |
1652 | if (err < 0) { | |
1653 | return -err; | |
1654 | } | |
1655 | ||
7251515e DV |
1656 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1657 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1658 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1659 | if (ivshmem->cring_tx == NULL) { |
1660 | rte_free(ivshmem); | |
1661 | return ENOMEM; | |
1662 | } | |
1663 | ||
1664 | err = snprintf(ring_name, 10, "%s_rx", dev_name); | |
1665 | if (err < 0) { | |
1666 | return -err; | |
1667 | } | |
1668 | ||
7251515e DV |
1669 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1670 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1671 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1672 | if (ivshmem->cring_rx == NULL) { |
1673 | rte_free(ivshmem); | |
1674 | return ENOMEM; | |
1675 | } | |
1676 | ||
d7310583 DDP |
1677 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
1678 | &ivshmem->cring_tx, 1, SOCKET0); | |
1679 | ||
95fb793a | 1680 | if (err < 0) { |
1681 | rte_free(ivshmem); | |
1682 | return ENODEV; | |
1683 | } | |
1684 | ||
1685 | ivshmem->user_port_id = port_no; | |
1686 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
1687 | list_push_back(&dpdk_ring_list, &ivshmem->list_node); | |
1688 | ||
1689 | *eth_port_id = ivshmem->eth_port_id; | |
1690 | return 0; | |
1691 | } | |
1692 | ||
1693 | static int | |
1694 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) OVS_REQUIRES(dpdk_mutex) | |
1695 | { | |
1696 | struct dpdk_ring *ivshmem; | |
1697 | unsigned int port_no; | |
1698 | int err = 0; | |
1699 | ||
1700 | /* Names always start with "dpdkr" */ | |
1701 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
1702 | if (err) { | |
1703 | return err; | |
1704 | } | |
1705 | ||
1706 | /* look through our list to find the device */ | |
1707 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { | |
1708 | if (ivshmem->user_port_id == port_no) { | |
58397e6c | 1709 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
95fb793a | 1710 | *eth_port_id = ivshmem->eth_port_id; /* really all that is needed */ |
1711 | return 0; | |
1712 | } | |
1713 | } | |
1714 | /* Need to create the device rings */ | |
1715 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
1716 | } | |
1717 | ||
7251515e DV |
1718 | static int |
1719 | netdev_dpdk_ring_send(struct netdev *netdev, int qid OVS_UNUSED, | |
e14deea0 | 1720 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1721 | { |
1722 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1723 | ||
1724 | /* DPDK Rings have a single TX queue, Therefore needs locking. */ | |
58397e6c | 1725 | rte_spinlock_lock(&dev->txq_lock); |
7251515e | 1726 | netdev_dpdk_send__(dev, 0, pkts, cnt, may_steal); |
58397e6c | 1727 | rte_spinlock_unlock(&dev->txq_lock); |
7251515e DV |
1728 | return 0; |
1729 | } | |
1730 | ||
95fb793a | 1731 | static int |
1732 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
1733 | { | |
1734 | unsigned int port_no = 0; | |
1735 | int err = 0; | |
1736 | ||
1737 | if (rte_eal_init_ret) { | |
1738 | return rte_eal_init_ret; | |
1739 | } | |
1740 | ||
1741 | ovs_mutex_lock(&dpdk_mutex); | |
1742 | ||
1743 | err = dpdk_ring_open(netdev->name, &port_no); | |
1744 | if (err) { | |
1745 | goto unlock_dpdk; | |
1746 | } | |
1747 | ||
58397e6c | 1748 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
95fb793a | 1749 | |
1750 | unlock_dpdk: | |
1751 | ovs_mutex_unlock(&dpdk_mutex); | |
1752 | return err; | |
1753 | } | |
1754 | ||
58397e6c KT |
1755 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, MULTIQ, SEND, \ |
1756 | GET_CARRIER, GET_STATS, GET_FEATURES, GET_STATUS, RXQ_RECV) \ | |
95fb793a | 1757 | { \ |
1758 | NAME, \ | |
1759 | INIT, /* init */ \ | |
1760 | NULL, /* netdev_dpdk_run */ \ | |
1761 | NULL, /* netdev_dpdk_wait */ \ | |
1762 | \ | |
1763 | netdev_dpdk_alloc, \ | |
1764 | CONSTRUCT, \ | |
58397e6c | 1765 | DESTRUCT, \ |
95fb793a | 1766 | netdev_dpdk_dealloc, \ |
1767 | netdev_dpdk_get_config, \ | |
1768 | NULL, /* netdev_dpdk_set_config */ \ | |
1769 | NULL, /* get_tunnel_config */ \ | |
58397e6c KT |
1770 | NULL, /* build header */ \ |
1771 | NULL, /* push header */ \ | |
1772 | NULL, /* pop header */ \ | |
7dec44fe | 1773 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
5496878c | 1774 | MULTIQ, /* set_multiq */ \ |
95fb793a | 1775 | \ |
7251515e | 1776 | SEND, /* send */ \ |
95fb793a | 1777 | NULL, /* send_wait */ \ |
1778 | \ | |
1779 | netdev_dpdk_set_etheraddr, \ | |
1780 | netdev_dpdk_get_etheraddr, \ | |
1781 | netdev_dpdk_get_mtu, \ | |
1782 | netdev_dpdk_set_mtu, \ | |
1783 | netdev_dpdk_get_ifindex, \ | |
58397e6c | 1784 | GET_CARRIER, \ |
95fb793a | 1785 | netdev_dpdk_get_carrier_resets, \ |
1786 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
1787 | GET_STATS, \ |
1788 | GET_FEATURES, \ | |
95fb793a | 1789 | NULL, /* set_advertisements */ \ |
1790 | \ | |
1791 | NULL, /* set_policing */ \ | |
1792 | NULL, /* get_qos_types */ \ | |
1793 | NULL, /* get_qos_capabilities */ \ | |
1794 | NULL, /* get_qos */ \ | |
1795 | NULL, /* set_qos */ \ | |
1796 | NULL, /* get_queue */ \ | |
1797 | NULL, /* set_queue */ \ | |
1798 | NULL, /* delete_queue */ \ | |
1799 | NULL, /* get_queue_stats */ \ | |
1800 | NULL, /* queue_dump_start */ \ | |
1801 | NULL, /* queue_dump_next */ \ | |
1802 | NULL, /* queue_dump_done */ \ | |
1803 | NULL, /* dump_queue_stats */ \ | |
1804 | \ | |
1805 | NULL, /* get_in4 */ \ | |
1806 | NULL, /* set_in4 */ \ | |
1807 | NULL, /* get_in6 */ \ | |
1808 | NULL, /* add_router */ \ | |
1809 | NULL, /* get_next_hop */ \ | |
58397e6c | 1810 | GET_STATUS, \ |
95fb793a | 1811 | NULL, /* arp_lookup */ \ |
1812 | \ | |
1813 | netdev_dpdk_update_flags, \ | |
1814 | \ | |
1815 | netdev_dpdk_rxq_alloc, \ | |
1816 | netdev_dpdk_rxq_construct, \ | |
1817 | netdev_dpdk_rxq_destruct, \ | |
1818 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 1819 | RXQ_RECV, \ |
95fb793a | 1820 | NULL, /* rx_wait */ \ |
1821 | NULL, /* rxq_drain */ \ | |
1822 | } | |
8a9562d2 PS |
1823 | |
1824 | int | |
1825 | dpdk_init(int argc, char **argv) | |
1826 | { | |
1827 | int result; | |
58397e6c KT |
1828 | int base = 0; |
1829 | char *pragram_name = argv[0]; | |
8a9562d2 | 1830 | |
9441caf3 | 1831 | if (argc < 2 || strcmp(argv[1], "--dpdk")) |
8a9562d2 PS |
1832 | return 0; |
1833 | ||
58397e6c | 1834 | /* Remove the --dpdk argument from arg list.*/ |
8a9562d2 PS |
1835 | argc--; |
1836 | argv++; | |
1837 | ||
58397e6c KT |
1838 | /* If the cuse_dev_name parameter has been provided, set 'cuse_dev_name' to |
1839 | * this string if it meets the correct criteria. Otherwise, set it to the | |
1840 | * default (vhost-net). | |
1841 | */ | |
1842 | if (!strcmp(argv[1], "--cuse_dev_name") && | |
1843 | (strlen(argv[2]) <= NAME_MAX)) { | |
1844 | ||
1845 | cuse_dev_name = strdup(argv[2]); | |
1846 | ||
1847 | /* Remove the cuse_dev_name configuration parameters from the argument | |
1848 | * list, so that the correct elements are passed to the DPDK | |
1849 | * initialization function | |
1850 | */ | |
1851 | argc -= 2; | |
1852 | argv += 2; /* Increment by two to bypass the cuse_dev_name arguments */ | |
1853 | base = 2; | |
1854 | ||
1855 | VLOG_ERR("User-provided cuse_dev_name in use: /dev/%s", cuse_dev_name); | |
1856 | } else { | |
1857 | cuse_dev_name = "vhost-net"; | |
1858 | VLOG_INFO("No cuse_dev_name provided - defaulting to /dev/vhost-net"); | |
1859 | } | |
1860 | ||
1861 | /* Keep the program name argument as this is needed for call to | |
1862 | * rte_eal_init() | |
1863 | */ | |
1864 | argv[0] = pragram_name; | |
1865 | ||
8a9562d2 PS |
1866 | /* Make sure things are initialized ... */ |
1867 | result = rte_eal_init(argc, argv); | |
451450fa | 1868 | if (result < 0) { |
58397e6c | 1869 | ovs_abort(result, "Cannot init EAL"); |
451450fa | 1870 | } |
8a9562d2 | 1871 | |
d7310583 | 1872 | rte_memzone_dump(stdout); |
8a9562d2 PS |
1873 | rte_eal_init_ret = 0; |
1874 | ||
451450fa | 1875 | if (argc > result) { |
9441caf3 | 1876 | argv[result] = argv[0]; |
451450fa | 1877 | } |
9441caf3 | 1878 | |
db73f716 DDP |
1879 | /* We are called from the main thread here */ |
1880 | thread_set_nonpmd(); | |
1881 | ||
58397e6c | 1882 | return result + 1 + base; |
8a9562d2 PS |
1883 | } |
1884 | ||
95fb793a | 1885 | const struct netdev_class dpdk_class = |
1886 | NETDEV_DPDK_CLASS( | |
1887 | "dpdk", | |
b8e57534 | 1888 | NULL, |
5496878c | 1889 | netdev_dpdk_construct, |
58397e6c | 1890 | netdev_dpdk_destruct, |
7251515e | 1891 | netdev_dpdk_set_multiq, |
58397e6c KT |
1892 | netdev_dpdk_eth_send, |
1893 | netdev_dpdk_get_carrier, | |
1894 | netdev_dpdk_get_stats, | |
1895 | netdev_dpdk_get_features, | |
1896 | netdev_dpdk_get_status, | |
1897 | netdev_dpdk_rxq_recv); | |
95fb793a | 1898 | |
1899 | const struct netdev_class dpdk_ring_class = | |
1900 | NETDEV_DPDK_CLASS( | |
1901 | "dpdkr", | |
033e9df2 | 1902 | NULL, |
5496878c | 1903 | netdev_dpdk_ring_construct, |
58397e6c KT |
1904 | netdev_dpdk_destruct, |
1905 | NULL, | |
1906 | netdev_dpdk_ring_send, | |
1907 | netdev_dpdk_get_carrier, | |
1908 | netdev_dpdk_get_stats, | |
1909 | netdev_dpdk_get_features, | |
1910 | netdev_dpdk_get_status, | |
1911 | netdev_dpdk_rxq_recv); | |
1912 | ||
1913 | const struct netdev_class dpdk_vhost_class = | |
1914 | NETDEV_DPDK_CLASS( | |
1915 | "dpdkvhost", | |
1916 | dpdk_vhost_class_init, | |
1917 | netdev_dpdk_vhost_construct, | |
1918 | netdev_dpdk_vhost_destruct, | |
1919 | netdev_dpdk_vhost_set_multiq, | |
1920 | netdev_dpdk_vhost_send, | |
1921 | netdev_dpdk_vhost_get_carrier, | |
1922 | netdev_dpdk_vhost_get_stats, | |
1923 | NULL, | |
7251515e | 1924 | NULL, |
58397e6c | 1925 | netdev_dpdk_vhost_rxq_recv); |
95fb793a | 1926 | |
8a9562d2 PS |
1927 | void |
1928 | netdev_dpdk_register(void) | |
1929 | { | |
95fb793a | 1930 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
1931 | ||
033e9df2 DDP |
1932 | if (rte_eal_init_ret) { |
1933 | return; | |
1934 | } | |
1935 | ||
95fb793a | 1936 | if (ovsthread_once_start(&once)) { |
033e9df2 | 1937 | dpdk_common_init(); |
95fb793a | 1938 | netdev_register_provider(&dpdk_class); |
1939 | netdev_register_provider(&dpdk_ring_class); | |
58397e6c | 1940 | netdev_register_provider(&dpdk_vhost_class); |
95fb793a | 1941 | ovsthread_once_done(&once); |
1942 | } | |
8a9562d2 | 1943 | } |
8617afff PS |
1944 | |
1945 | int | |
1946 | pmd_thread_setaffinity_cpu(int cpu) | |
1947 | { | |
1948 | cpu_set_t cpuset; | |
1949 | int err; | |
1950 | ||
1951 | CPU_ZERO(&cpuset); | |
1952 | CPU_SET(cpu, &cpuset); | |
1953 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset); | |
1954 | if (err) { | |
1955 | VLOG_ERR("Thread affinity error %d",err); | |
1956 | return err; | |
1957 | } | |
abb5943d AW |
1958 | /* NON_PMD_CORE_ID is reserved for use by non pmd threads. */ |
1959 | ovs_assert(cpu != NON_PMD_CORE_ID); | |
65f13b50 | 1960 | RTE_PER_LCORE(_lcore_id) = cpu; |
8617afff PS |
1961 | |
1962 | return 0; | |
1963 | } | |
db73f716 DDP |
1964 | |
1965 | void | |
1966 | thread_set_nonpmd(void) | |
1967 | { | |
abb5943d AW |
1968 | /* We have to use NON_PMD_CORE_ID to allow non-pmd threads to perform |
1969 | * certain DPDK operations, like rte_eth_dev_configure(). */ | |
1970 | RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID; | |
db73f716 DDP |
1971 | } |
1972 | ||
1973 | static bool | |
1974 | thread_is_pmd(void) | |
1975 | { | |
abb5943d | 1976 | return rte_lcore_id() != NON_PMD_CORE_ID; |
db73f716 | 1977 | } |