]>
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 | ||
8a9562d2 PS |
19 | #include <string.h> |
20 | #include <signal.h> | |
21 | #include <stdlib.h> | |
22 | #include <pthread.h> | |
23 | #include <config.h> | |
24 | #include <errno.h> | |
25 | #include <sched.h> | |
26 | #include <stdlib.h> | |
27 | #include <unistd.h> | |
7d1ced01 | 28 | #include <sys/stat.h> |
8a9562d2 | 29 | #include <stdio.h> |
7d1ced01 CL |
30 | #include <sys/types.h> |
31 | #include <sys/stat.h> | |
8a9562d2 | 32 | |
7d1ced01 | 33 | #include "dirs.h" |
e14deea0 | 34 | #include "dp-packet.h" |
8a9562d2 PS |
35 | #include "dpif-netdev.h" |
36 | #include "list.h" | |
37 | #include "netdev-dpdk.h" | |
38 | #include "netdev-provider.h" | |
39 | #include "netdev-vport.h" | |
40 | #include "odp-util.h" | |
41 | #include "ofp-print.h" | |
94143fc4 | 42 | #include "ovs-numa.h" |
8a9562d2 PS |
43 | #include "ovs-thread.h" |
44 | #include "ovs-rcu.h" | |
45 | #include "packets.h" | |
46 | #include "shash.h" | |
8a9562d2 PS |
47 | #include "sset.h" |
48 | #include "unaligned.h" | |
49 | #include "timeval.h" | |
50 | #include "unixctl.h" | |
e6211adc | 51 | #include "openvswitch/vlog.h" |
8a9562d2 | 52 | |
b8e57534 MK |
53 | #include "rte_config.h" |
54 | #include "rte_mbuf.h" | |
58397e6c | 55 | #include "rte_virtio_net.h" |
b8e57534 | 56 | |
8a9562d2 PS |
57 | VLOG_DEFINE_THIS_MODULE(dpdk); |
58 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
59 | ||
60 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
61 | ||
62 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
63 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
64 | ||
65 | /* | |
66 | * need to reserve tons of extra space in the mbufs so we can align the | |
67 | * DMA addresses to 4KB. | |
68 | */ | |
69 | ||
70 | #define MTU_TO_MAX_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) | |
71 | #define MBUF_SIZE(mtu) (MTU_TO_MAX_LEN(mtu) + (512) + \ | |
72 | sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM) | |
73 | ||
da79ce2b DDP |
74 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
75 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
76 | * enough hugepages) we keep halving the number until the allocation succeeds | |
77 | * or we reach MIN_NB_MBUF */ | |
78 | ||
79 | #define MAX_NB_MBUF (4096 * 64) | |
80 | #define MIN_NB_MBUF (4096 * 4) | |
81 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
82 | ||
83 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
84 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
85 | ||
86 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
87 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
88 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
89 | % MP_CACHE_SZ == 0); | |
90 | ||
8a9562d2 PS |
91 | #define SOCKET0 0 |
92 | ||
79f5354c PM |
93 | #define NIC_PORT_RX_Q_SIZE 2048 /* Size of Physical NIC RX Queue, Max (n+32<=4096)*/ |
94 | #define NIC_PORT_TX_Q_SIZE 2048 /* Size of Physical NIC TX Queue, Max (n+32<=4096)*/ | |
95 | ||
7d1ced01 CL |
96 | char *cuse_dev_name = NULL; /* Character device cuse_dev_name. */ |
97 | char *vhost_sock_dir = NULL; /* Location of vhost-user sockets */ | |
58397e6c | 98 | |
95e9881f KT |
99 | /* |
100 | * Maximum amount of time in micro seconds to try and enqueue to vhost. | |
101 | */ | |
102 | #define VHOST_ENQ_RETRY_USECS 100 | |
103 | ||
8a9562d2 | 104 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
105 | .rxmode = { |
106 | .mq_mode = ETH_MQ_RX_RSS, | |
107 | .split_hdr_size = 0, | |
108 | .header_split = 0, /* Header Split disabled */ | |
109 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
110 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
111 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
112 | .hw_strip_crc = 0, | |
113 | }, | |
114 | .rx_adv_conf = { | |
115 | .rss_conf = { | |
116 | .rss_key = NULL, | |
543342a4 | 117 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 118 | }, |
a28ddd11 DDP |
119 | }, |
120 | .txmode = { | |
121 | .mq_mode = ETH_MQ_TX_NONE, | |
122 | }, | |
8a9562d2 PS |
123 | }; |
124 | ||
3a100265 | 125 | enum { MAX_TX_QUEUE_LEN = 384 }; |
58f7c37b DDP |
126 | enum { DPDK_RING_SIZE = 256 }; |
127 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
128 | enum { DRAIN_TSC = 200000ULL }; |
129 | ||
58397e6c KT |
130 | enum dpdk_dev_type { |
131 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 132 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
133 | }; |
134 | ||
8a9562d2 PS |
135 | static int rte_eal_init_ret = ENODEV; |
136 | ||
137 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; | |
138 | ||
139 | /* Contains all 'struct dpdk_dev's. */ | |
ca6ba700 | 140 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 141 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 142 | |
ca6ba700 | 143 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 144 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 145 | |
db73f716 DDP |
146 | /* This mutex must be used by non pmd threads when allocating or freeing |
147 | * mbufs through mempools. Since dpdk_queue_pkts() and dpdk_queue_flush() may | |
148 | * use mempools, a non pmd thread should hold this mutex while calling them */ | |
bce01e3a | 149 | static struct ovs_mutex nonpmd_mempool_mutex = OVS_MUTEX_INITIALIZER; |
db73f716 | 150 | |
8a9562d2 PS |
151 | struct dpdk_mp { |
152 | struct rte_mempool *mp; | |
153 | int mtu; | |
154 | int socket_id; | |
155 | int refcount; | |
ca6ba700 | 156 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
157 | }; |
158 | ||
5a034064 AW |
159 | /* There should be one 'struct dpdk_tx_queue' created for |
160 | * each cpu core. */ | |
8a9562d2 | 161 | struct dpdk_tx_queue { |
94143fc4 AW |
162 | bool flush_tx; /* Set to true to flush queue everytime */ |
163 | /* pkts are queued. */ | |
8a9562d2 | 164 | int count; |
a0cb2d66 DDP |
165 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
166 | * from concurrent access. It is used only | |
167 | * if the queue is shared among different | |
168 | * pmd threads (see 'txq_needs_locking'). */ | |
8a9562d2 PS |
169 | uint64_t tsc; |
170 | struct rte_mbuf *burst_pkts[MAX_TX_QUEUE_LEN]; | |
171 | }; | |
172 | ||
95fb793a | 173 | /* dpdk has no way to remove dpdk ring ethernet devices |
174 | so we have to keep them around once they've been created | |
175 | */ | |
176 | ||
ca6ba700 | 177 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 178 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 179 | |
180 | struct dpdk_ring { | |
181 | /* For the client rings */ | |
182 | struct rte_ring *cring_tx; | |
183 | struct rte_ring *cring_rx; | |
184 | int user_port_id; /* User given port no, parsed from port name */ | |
185 | int eth_port_id; /* ethernet device port id */ | |
ca6ba700 | 186 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 187 | }; |
188 | ||
8a9562d2 PS |
189 | struct netdev_dpdk { |
190 | struct netdev up; | |
191 | int port_id; | |
192 | int max_packet_len; | |
58397e6c | 193 | enum dpdk_dev_type type; |
8a9562d2 | 194 | |
5a034064 | 195 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
196 | |
197 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
198 | ||
199 | struct dpdk_mp *dpdk_mp; | |
200 | int mtu; | |
201 | int socket_id; | |
202 | int buf_size; | |
8a9562d2 | 203 | struct netdev_stats stats; |
45d947c4 DDP |
204 | /* Protects stats */ |
205 | rte_spinlock_t stats_lock; | |
8a9562d2 PS |
206 | |
207 | uint8_t hwaddr[ETH_ADDR_LEN]; | |
208 | enum netdev_flags flags; | |
209 | ||
210 | struct rte_eth_link link; | |
211 | int link_reset_cnt; | |
212 | ||
a0cb2d66 DDP |
213 | /* The user might request more txqs than the NIC has. We remap those |
214 | * ('up.n_txq') on these ('real_n_txq'). | |
215 | * If the numbers match, 'txq_needs_locking' is false, otherwise it is | |
216 | * true and we will take a spinlock on transmission */ | |
217 | int real_n_txq; | |
218 | bool txq_needs_locking; | |
219 | ||
220 | /* Spinlock for vhost transmission. Other DPDK devices use spinlocks in | |
221 | * dpdk_tx_queue */ | |
222 | rte_spinlock_t vhost_tx_lock; | |
223 | ||
58397e6c KT |
224 | /* virtio-net structure for vhost device */ |
225 | OVSRCU_TYPE(struct virtio_net *) virtio_dev; | |
226 | ||
7d1ced01 CL |
227 | /* Identifier used to distinguish vhost devices from each other */ |
228 | char vhost_id[PATH_MAX]; | |
229 | ||
8a9562d2 | 230 | /* In dpdk_list. */ |
ca6ba700 | 231 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
232 | }; |
233 | ||
234 | struct netdev_rxq_dpdk { | |
235 | struct netdev_rxq up; | |
236 | int port_id; | |
237 | }; | |
238 | ||
db73f716 DDP |
239 | static bool thread_is_pmd(void); |
240 | ||
8a9562d2 PS |
241 | static int netdev_dpdk_construct(struct netdev *); |
242 | ||
58397e6c KT |
243 | struct virtio_net * netdev_dpdk_get_virtio(const struct netdev_dpdk *dev); |
244 | ||
8a9562d2 PS |
245 | static bool |
246 | is_dpdk_class(const struct netdev_class *class) | |
247 | { | |
248 | return class->construct == netdev_dpdk_construct; | |
249 | } | |
250 | ||
58397e6c KT |
251 | /* XXX: use dpdk malloc for entire OVS. in fact huge page should be used |
252 | * for all other segments data, bss and text. */ | |
8a9562d2 PS |
253 | |
254 | static void * | |
255 | dpdk_rte_mzalloc(size_t sz) | |
256 | { | |
257 | void *ptr; | |
258 | ||
259 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
260 | if (ptr == NULL) { | |
261 | out_of_memory(); | |
262 | } | |
263 | return ptr; | |
264 | } | |
265 | ||
db73f716 DDP |
266 | /* XXX this function should be called only by pmd threads (or by non pmd |
267 | * threads holding the nonpmd_mempool_mutex) */ | |
8a9562d2 | 268 | void |
e14deea0 | 269 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 270 | { |
db73f716 | 271 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 272 | |
db73f716 | 273 | rte_pktmbuf_free_seg(pkt); |
8a9562d2 PS |
274 | } |
275 | ||
b3cd9f9d PS |
276 | static void |
277 | __rte_pktmbuf_init(struct rte_mempool *mp, | |
278 | void *opaque_arg OVS_UNUSED, | |
279 | void *_m, | |
280 | unsigned i OVS_UNUSED) | |
281 | { | |
282 | struct rte_mbuf *m = _m; | |
e14deea0 | 283 | uint32_t buf_len = mp->elt_size - sizeof(struct dp_packet); |
b3cd9f9d | 284 | |
e14deea0 | 285 | RTE_MBUF_ASSERT(mp->elt_size >= sizeof(struct dp_packet)); |
b3cd9f9d PS |
286 | |
287 | memset(m, 0, mp->elt_size); | |
288 | ||
289 | /* start of buffer is just after mbuf structure */ | |
e14deea0 | 290 | m->buf_addr = (char *)m + sizeof(struct dp_packet); |
b3cd9f9d | 291 | m->buf_physaddr = rte_mempool_virt2phy(mp, m) + |
e14deea0 | 292 | sizeof(struct dp_packet); |
b3cd9f9d PS |
293 | m->buf_len = (uint16_t)buf_len; |
294 | ||
295 | /* keep some headroom between start of buffer and data */ | |
b8e57534 | 296 | m->data_off = RTE_MIN(RTE_PKTMBUF_HEADROOM, m->buf_len); |
b3cd9f9d PS |
297 | |
298 | /* init some constant fields */ | |
b3cd9f9d | 299 | m->pool = mp; |
b8e57534 MK |
300 | m->nb_segs = 1; |
301 | m->port = 0xff; | |
b3cd9f9d PS |
302 | } |
303 | ||
304 | static void | |
305 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
306 | void *opaque_arg OVS_UNUSED, | |
307 | void *_m, | |
308 | unsigned i OVS_UNUSED) | |
309 | { | |
310 | struct rte_mbuf *m = _m; | |
311 | ||
312 | __rte_pktmbuf_init(mp, opaque_arg, _m, i); | |
313 | ||
cf62fa4c | 314 | dp_packet_init_dpdk((struct dp_packet *) m, m->buf_len); |
b3cd9f9d PS |
315 | } |
316 | ||
8a9562d2 PS |
317 | static struct dpdk_mp * |
318 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
319 | { | |
320 | struct dpdk_mp *dmp = NULL; | |
321 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
da79ce2b | 322 | unsigned mp_size; |
8a9562d2 PS |
323 | |
324 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
325 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
326 | dmp->refcount++; | |
327 | return dmp; | |
328 | } | |
329 | } | |
330 | ||
331 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
332 | dmp->socket_id = socket_id; | |
333 | dmp->mtu = mtu; | |
334 | dmp->refcount = 1; | |
335 | ||
da79ce2b DDP |
336 | mp_size = MAX_NB_MBUF; |
337 | do { | |
338 | if (snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d_%d_%u", | |
339 | dmp->mtu, dmp->socket_id, mp_size) < 0) { | |
340 | return NULL; | |
341 | } | |
95fb793a | 342 | |
da79ce2b DDP |
343 | dmp->mp = rte_mempool_create(mp_name, mp_size, MBUF_SIZE(mtu), |
344 | MP_CACHE_SZ, | |
345 | sizeof(struct rte_pktmbuf_pool_private), | |
346 | rte_pktmbuf_pool_init, NULL, | |
347 | ovs_rte_pktmbuf_init, NULL, | |
348 | socket_id, 0); | |
349 | } while (!dmp->mp && rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
8a9562d2 PS |
350 | |
351 | if (dmp->mp == NULL) { | |
352 | return NULL; | |
da79ce2b DDP |
353 | } else { |
354 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", mp_name, mp_size ); | |
8a9562d2 PS |
355 | } |
356 | ||
357 | list_push_back(&dpdk_mp_list, &dmp->list_node); | |
358 | return dmp; | |
359 | } | |
360 | ||
361 | static void | |
362 | dpdk_mp_put(struct dpdk_mp *dmp) | |
363 | { | |
364 | ||
365 | if (!dmp) { | |
366 | return; | |
367 | } | |
368 | ||
369 | dmp->refcount--; | |
370 | ovs_assert(dmp->refcount >= 0); | |
371 | ||
372 | #if 0 | |
373 | /* I could not find any API to destroy mp. */ | |
374 | if (dmp->refcount == 0) { | |
375 | list_delete(dmp->list_node); | |
376 | /* destroy mp-pool. */ | |
377 | } | |
378 | #endif | |
379 | } | |
380 | ||
381 | static void | |
382 | check_link_status(struct netdev_dpdk *dev) | |
383 | { | |
384 | struct rte_eth_link link; | |
385 | ||
386 | rte_eth_link_get_nowait(dev->port_id, &link); | |
387 | ||
388 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 389 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
390 | |
391 | dev->link_reset_cnt++; | |
392 | dev->link = link; | |
393 | if (dev->link.link_status) { | |
394 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
395 | dev->port_id, (unsigned)dev->link.link_speed, | |
396 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
397 | ("full-duplex") : ("half-duplex")); | |
398 | } else { | |
399 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
400 | } | |
401 | } | |
402 | } | |
403 | ||
404 | static void * | |
405 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
406 | { | |
407 | struct netdev_dpdk *dev; | |
408 | ||
409 | pthread_detach(pthread_self()); | |
410 | ||
411 | for (;;) { | |
412 | ovs_mutex_lock(&dpdk_mutex); | |
413 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
414 | ovs_mutex_lock(&dev->mutex); | |
415 | check_link_status(dev); | |
416 | ovs_mutex_unlock(&dev->mutex); | |
417 | } | |
418 | ovs_mutex_unlock(&dpdk_mutex); | |
419 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
420 | } | |
421 | ||
422 | return NULL; | |
423 | } | |
424 | ||
425 | static int | |
426 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
427 | { | |
428 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 429 | struct rte_eth_dev_info info; |
8a9562d2 PS |
430 | struct ether_addr eth_addr; |
431 | int diag; | |
432 | int i; | |
433 | ||
434 | if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) { | |
95fb793a | 435 | return ENODEV; |
8a9562d2 PS |
436 | } |
437 | ||
a0cb2d66 DDP |
438 | rte_eth_dev_info_get(dev->port_id, &info); |
439 | dev->up.n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); | |
440 | dev->real_n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
441 | ||
442 | diag = rte_eth_dev_configure(dev->port_id, dev->up.n_rxq, dev->real_n_txq, | |
5496878c | 443 | &port_conf); |
8a9562d2 | 444 | if (diag) { |
a0cb2d66 DDP |
445 | VLOG_ERR("eth dev config error %d. rxq:%d txq:%d", diag, dev->up.n_rxq, |
446 | dev->real_n_txq); | |
95fb793a | 447 | return -diag; |
8a9562d2 PS |
448 | } |
449 | ||
a0cb2d66 | 450 | for (i = 0; i < dev->real_n_txq; i++) { |
79f5354c | 451 | diag = rte_eth_tx_queue_setup(dev->port_id, i, NIC_PORT_TX_Q_SIZE, |
9154f798 | 452 | dev->socket_id, NULL); |
8a9562d2 PS |
453 | if (diag) { |
454 | VLOG_ERR("eth dev tx queue setup error %d",diag); | |
95fb793a | 455 | return -diag; |
8a9562d2 PS |
456 | } |
457 | } | |
458 | ||
5496878c | 459 | for (i = 0; i < dev->up.n_rxq; i++) { |
79f5354c | 460 | diag = rte_eth_rx_queue_setup(dev->port_id, i, NIC_PORT_RX_Q_SIZE, |
d221ffa1 | 461 | dev->socket_id, |
9154f798 | 462 | NULL, dev->dpdk_mp->mp); |
8a9562d2 PS |
463 | if (diag) { |
464 | VLOG_ERR("eth dev rx queue setup error %d",diag); | |
95fb793a | 465 | return -diag; |
8a9562d2 PS |
466 | } |
467 | } | |
468 | ||
469 | diag = rte_eth_dev_start(dev->port_id); | |
470 | if (diag) { | |
471 | VLOG_ERR("eth dev start error %d",diag); | |
95fb793a | 472 | return -diag; |
8a9562d2 PS |
473 | } |
474 | ||
475 | rte_eth_promiscuous_enable(dev->port_id); | |
476 | rte_eth_allmulticast_enable(dev->port_id); | |
477 | ||
478 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
479 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
480 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
481 | dev->port_id, ETH_ADDR_ARGS(eth_addr.addr_bytes)); | |
482 | ||
483 | memcpy(dev->hwaddr, eth_addr.addr_bytes, ETH_ADDR_LEN); | |
484 | rte_eth_link_get_nowait(dev->port_id, &dev->link); | |
485 | ||
486 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
487 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
488 | ||
489 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
490 | return 0; | |
491 | } | |
492 | ||
493 | static struct netdev_dpdk * | |
494 | netdev_dpdk_cast(const struct netdev *netdev) | |
495 | { | |
496 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
497 | } | |
498 | ||
499 | static struct netdev * | |
500 | netdev_dpdk_alloc(void) | |
501 | { | |
502 | struct netdev_dpdk *netdev = dpdk_rte_mzalloc(sizeof *netdev); | |
503 | return &netdev->up; | |
504 | } | |
505 | ||
5a034064 | 506 | static void |
91968eb0 | 507 | netdev_dpdk_alloc_txq(struct netdev_dpdk *netdev, unsigned int n_txqs) |
5a034064 | 508 | { |
bd5131ba | 509 | unsigned i; |
5a034064 AW |
510 | |
511 | netdev->tx_q = dpdk_rte_mzalloc(n_txqs * sizeof *netdev->tx_q); | |
512 | for (i = 0; i < n_txqs; i++) { | |
ba0358a1 | 513 | int numa_id = ovs_numa_get_numa_id(i); |
94143fc4 | 514 | |
a0cb2d66 DDP |
515 | if (!netdev->txq_needs_locking) { |
516 | /* Each index is considered as a cpu core id, since there should | |
517 | * be one tx queue for each cpu core. If the corresponding core | |
518 | * is not on the same numa node as 'netdev', flags the | |
519 | * 'flush_tx'. */ | |
520 | netdev->tx_q[i].flush_tx = netdev->socket_id == numa_id; | |
521 | } else { | |
522 | /* Queues are shared among CPUs. Always flush */ | |
523 | netdev->tx_q[i].flush_tx = true; | |
524 | } | |
525 | rte_spinlock_init(&netdev->tx_q[i].tx_lock); | |
5a034064 AW |
526 | } |
527 | } | |
528 | ||
8a9562d2 | 529 | static int |
58397e6c KT |
530 | netdev_dpdk_init(struct netdev *netdev_, unsigned int port_no, |
531 | enum dpdk_dev_type type) | |
5a034064 | 532 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 PS |
533 | { |
534 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1b7a04e0 | 535 | int sid; |
95fb793a | 536 | int err = 0; |
8a9562d2 | 537 | |
95fb793a | 538 | ovs_mutex_init(&netdev->mutex); |
95fb793a | 539 | ovs_mutex_lock(&netdev->mutex); |
8a9562d2 | 540 | |
45d947c4 DDP |
541 | rte_spinlock_init(&netdev->stats_lock); |
542 | ||
1b7a04e0 AW |
543 | /* If the 'sid' is negative, it means that the kernel fails |
544 | * to obtain the pci numa info. In that situation, always | |
545 | * use 'SOCKET0'. */ | |
58397e6c KT |
546 | if (type == DPDK_DEV_ETH) { |
547 | sid = rte_eth_dev_socket_id(port_no); | |
548 | } else { | |
549 | sid = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
550 | } | |
551 | ||
1b7a04e0 | 552 | netdev->socket_id = sid < 0 ? SOCKET0 : sid; |
95fb793a | 553 | netdev->port_id = port_no; |
58397e6c | 554 | netdev->type = type; |
8a9562d2 | 555 | netdev->flags = 0; |
8a9562d2 PS |
556 | netdev->mtu = ETHER_MTU; |
557 | netdev->max_packet_len = MTU_TO_MAX_LEN(netdev->mtu); | |
558 | ||
8a9562d2 PS |
559 | netdev->dpdk_mp = dpdk_mp_get(netdev->socket_id, netdev->mtu); |
560 | if (!netdev->dpdk_mp) { | |
561 | err = ENOMEM; | |
95fb793a | 562 | goto unlock; |
8a9562d2 PS |
563 | } |
564 | ||
5496878c AW |
565 | netdev_->n_txq = NR_QUEUE; |
566 | netdev_->n_rxq = NR_QUEUE; | |
a0cb2d66 | 567 | netdev->real_n_txq = NR_QUEUE; |
58397e6c KT |
568 | |
569 | if (type == DPDK_DEV_ETH) { | |
1b99bb05 MG |
570 | netdev_dpdk_alloc_txq(netdev, NR_QUEUE); |
571 | err = dpdk_eth_dev_init(netdev); | |
572 | if (err) { | |
573 | goto unlock; | |
574 | } | |
8a9562d2 | 575 | } |
8a9562d2 PS |
576 | |
577 | list_push_back(&dpdk_list, &netdev->list_node); | |
578 | ||
95fb793a | 579 | unlock: |
5a034064 AW |
580 | if (err) { |
581 | rte_free(netdev->tx_q); | |
582 | } | |
8a9562d2 | 583 | ovs_mutex_unlock(&netdev->mutex); |
95fb793a | 584 | return err; |
585 | } | |
586 | ||
587 | static int | |
588 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
589 | unsigned int *port_no) | |
590 | { | |
591 | const char *cport; | |
592 | ||
593 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
594 | return ENODEV; | |
595 | } | |
596 | ||
597 | cport = dev_name + strlen(prefix); | |
bce01e3a | 598 | *port_no = strtol(cport, NULL, 0); /* string must be null terminated */ |
95fb793a | 599 | return 0; |
600 | } | |
601 | ||
58397e6c | 602 | static int |
7d1ced01 | 603 | vhost_construct_helper(struct netdev *netdev_) |
58397e6c | 604 | { |
a0cb2d66 | 605 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); |
58397e6c KT |
606 | |
607 | if (rte_eal_init_ret) { | |
608 | return rte_eal_init_ret; | |
609 | } | |
610 | ||
7d1ced01 CL |
611 | rte_spinlock_init(&netdev->vhost_tx_lock); |
612 | return netdev_dpdk_init(netdev_, -1, DPDK_DEV_VHOST); | |
613 | } | |
614 | ||
615 | static int | |
616 | netdev_dpdk_vhost_cuse_construct(struct netdev *netdev_) | |
617 | { | |
618 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
619 | int err; | |
620 | ||
58397e6c | 621 | ovs_mutex_lock(&dpdk_mutex); |
7d1ced01 CL |
622 | strncpy(netdev->vhost_id, netdev->up.name, sizeof(netdev->vhost_id)); |
623 | err = vhost_construct_helper(netdev_); | |
58397e6c | 624 | ovs_mutex_unlock(&dpdk_mutex); |
7d1ced01 CL |
625 | return err; |
626 | } | |
58397e6c | 627 | |
7d1ced01 CL |
628 | static int |
629 | netdev_dpdk_vhost_user_construct(struct netdev *netdev_) | |
630 | { | |
631 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
632 | int err; | |
a0cb2d66 | 633 | |
7d1ced01 CL |
634 | ovs_mutex_lock(&dpdk_mutex); |
635 | /* Take the name of the vhost-user port and append it to the location where | |
636 | * the socket is to be created, then register the socket. | |
637 | */ | |
638 | snprintf(netdev->vhost_id, sizeof(netdev->vhost_id), "%s/%s", | |
639 | vhost_sock_dir, netdev_->name); | |
640 | err = rte_vhost_driver_register(netdev->vhost_id); | |
641 | if (err) { | |
642 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
643 | netdev->vhost_id); | |
644 | } | |
645 | VLOG_INFO("Socket %s created for vhost-user port %s\n", netdev->vhost_id, netdev_->name); | |
646 | err = vhost_construct_helper(netdev_); | |
647 | ovs_mutex_unlock(&dpdk_mutex); | |
58397e6c KT |
648 | return err; |
649 | } | |
650 | ||
95fb793a | 651 | static int |
652 | netdev_dpdk_construct(struct netdev *netdev) | |
653 | { | |
654 | unsigned int port_no; | |
655 | int err; | |
656 | ||
657 | if (rte_eal_init_ret) { | |
658 | return rte_eal_init_ret; | |
659 | } | |
660 | ||
661 | /* Names always start with "dpdk" */ | |
662 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
663 | if (err) { | |
664 | return err; | |
665 | } | |
666 | ||
667 | ovs_mutex_lock(&dpdk_mutex); | |
58397e6c | 668 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
8a9562d2 PS |
669 | ovs_mutex_unlock(&dpdk_mutex); |
670 | return err; | |
671 | } | |
672 | ||
673 | static void | |
674 | netdev_dpdk_destruct(struct netdev *netdev_) | |
675 | { | |
676 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
677 | ||
678 | ovs_mutex_lock(&dev->mutex); | |
679 | rte_eth_dev_stop(dev->port_id); | |
680 | ovs_mutex_unlock(&dev->mutex); | |
681 | ||
682 | ovs_mutex_lock(&dpdk_mutex); | |
5a034064 | 683 | rte_free(dev->tx_q); |
8a9562d2 PS |
684 | list_remove(&dev->list_node); |
685 | dpdk_mp_put(dev->dpdk_mp); | |
686 | ovs_mutex_unlock(&dpdk_mutex); | |
58397e6c | 687 | } |
8a9562d2 | 688 | |
58397e6c KT |
689 | static void |
690 | netdev_dpdk_vhost_destruct(struct netdev *netdev_) | |
691 | { | |
692 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
693 | ||
694 | /* Can't remove a port while a guest is attached to it. */ | |
695 | if (netdev_dpdk_get_virtio(dev) != NULL) { | |
696 | VLOG_ERR("Can not remove port, vhost device still attached"); | |
697 | return; | |
698 | } | |
699 | ||
700 | ovs_mutex_lock(&dpdk_mutex); | |
701 | list_remove(&dev->list_node); | |
702 | dpdk_mp_put(dev->dpdk_mp); | |
703 | ovs_mutex_unlock(&dpdk_mutex); | |
8a9562d2 PS |
704 | } |
705 | ||
706 | static void | |
707 | netdev_dpdk_dealloc(struct netdev *netdev_) | |
708 | { | |
709 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
710 | ||
711 | rte_free(netdev); | |
712 | } | |
713 | ||
714 | static int | |
715 | netdev_dpdk_get_config(const struct netdev *netdev_, struct smap *args) | |
716 | { | |
717 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
718 | ||
719 | ovs_mutex_lock(&dev->mutex); | |
720 | ||
47659062 | 721 | smap_add_format(args, "configured_rx_queues", "%d", netdev_->n_rxq); |
a0cb2d66 DDP |
722 | smap_add_format(args, "requested_tx_queues", "%d", netdev_->n_txq); |
723 | smap_add_format(args, "configured_tx_queues", "%d", dev->real_n_txq); | |
8a9562d2 PS |
724 | ovs_mutex_unlock(&dev->mutex); |
725 | ||
726 | return 0; | |
727 | } | |
728 | ||
7dec44fe AW |
729 | static int |
730 | netdev_dpdk_get_numa_id(const struct netdev *netdev_) | |
731 | { | |
732 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
733 | ||
734 | return netdev->socket_id; | |
735 | } | |
736 | ||
5496878c AW |
737 | /* Sets the number of tx queues and rx queues for the dpdk interface. |
738 | * If the configuration fails, do not try restoring its old configuration | |
739 | * and just returns the error. */ | |
740 | static int | |
741 | netdev_dpdk_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
742 | unsigned int n_rxq) | |
743 | { | |
744 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
745 | int err = 0; | |
746 | ||
747 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
748 | return err; | |
749 | } | |
750 | ||
b7ccaf67 | 751 | ovs_mutex_lock(&dpdk_mutex); |
5496878c | 752 | ovs_mutex_lock(&netdev->mutex); |
91968eb0 | 753 | |
5496878c | 754 | rte_eth_dev_stop(netdev->port_id); |
91968eb0 | 755 | |
5496878c AW |
756 | netdev->up.n_txq = n_txq; |
757 | netdev->up.n_rxq = n_rxq; | |
58397e6c | 758 | |
91968eb0 | 759 | rte_free(netdev->tx_q); |
5496878c | 760 | err = dpdk_eth_dev_init(netdev); |
a0cb2d66 DDP |
761 | netdev_dpdk_alloc_txq(netdev, netdev->real_n_txq); |
762 | ||
763 | netdev->txq_needs_locking = netdev->real_n_txq != netdev->up.n_txq; | |
91968eb0 | 764 | |
5496878c | 765 | ovs_mutex_unlock(&netdev->mutex); |
b7ccaf67 | 766 | ovs_mutex_unlock(&dpdk_mutex); |
5496878c AW |
767 | |
768 | return err; | |
769 | } | |
770 | ||
58397e6c KT |
771 | static int |
772 | netdev_dpdk_vhost_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
a0cb2d66 | 773 | unsigned int n_rxq) |
58397e6c KT |
774 | { |
775 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
776 | int err = 0; | |
777 | ||
778 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
779 | return err; | |
780 | } | |
781 | ||
782 | ovs_mutex_lock(&dpdk_mutex); | |
783 | ovs_mutex_lock(&netdev->mutex); | |
784 | ||
785 | netdev->up.n_txq = n_txq; | |
a0cb2d66 DDP |
786 | netdev->real_n_txq = 1; |
787 | netdev->up.n_rxq = 1; | |
58397e6c KT |
788 | |
789 | ovs_mutex_unlock(&netdev->mutex); | |
790 | ovs_mutex_unlock(&dpdk_mutex); | |
791 | ||
792 | return err; | |
793 | } | |
794 | ||
8a9562d2 PS |
795 | static struct netdev_rxq * |
796 | netdev_dpdk_rxq_alloc(void) | |
797 | { | |
798 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
799 | ||
800 | return &rx->up; | |
801 | } | |
802 | ||
803 | static struct netdev_rxq_dpdk * | |
804 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rx) | |
805 | { | |
806 | return CONTAINER_OF(rx, struct netdev_rxq_dpdk, up); | |
807 | } | |
808 | ||
809 | static int | |
810 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq_) | |
811 | { | |
812 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
813 | struct netdev_dpdk *netdev = netdev_dpdk_cast(rx->up.netdev); | |
814 | ||
815 | ovs_mutex_lock(&netdev->mutex); | |
816 | rx->port_id = netdev->port_id; | |
817 | ovs_mutex_unlock(&netdev->mutex); | |
818 | ||
819 | return 0; | |
820 | } | |
821 | ||
822 | static void | |
823 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq_ OVS_UNUSED) | |
824 | { | |
825 | } | |
826 | ||
827 | static void | |
828 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq_) | |
829 | { | |
830 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
831 | ||
832 | rte_free(rx); | |
833 | } | |
834 | ||
b170db2a RW |
835 | static inline void |
836 | dpdk_queue_flush__(struct netdev_dpdk *dev, int qid) | |
8a9562d2 PS |
837 | { |
838 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1304f1f8 DDP |
839 | uint32_t nb_tx = 0; |
840 | ||
841 | while (nb_tx != txq->count) { | |
842 | uint32_t ret; | |
843 | ||
844 | ret = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts + nb_tx, | |
845 | txq->count - nb_tx); | |
846 | if (!ret) { | |
847 | break; | |
848 | } | |
849 | ||
850 | nb_tx += ret; | |
851 | } | |
8a9562d2 | 852 | |
b170db2a | 853 | if (OVS_UNLIKELY(nb_tx != txq->count)) { |
db73f716 DDP |
854 | /* free buffers, which we couldn't transmit, one at a time (each |
855 | * packet could come from a different mempool) */ | |
856 | int i; | |
857 | ||
858 | for (i = nb_tx; i < txq->count; i++) { | |
859 | rte_pktmbuf_free_seg(txq->burst_pkts[i]); | |
860 | } | |
45d947c4 | 861 | rte_spinlock_lock(&dev->stats_lock); |
1304f1f8 | 862 | dev->stats.tx_dropped += txq->count-nb_tx; |
45d947c4 | 863 | rte_spinlock_unlock(&dev->stats_lock); |
8a9562d2 | 864 | } |
1304f1f8 | 865 | |
8a9562d2 | 866 | txq->count = 0; |
844f2d74 | 867 | txq->tsc = rte_get_timer_cycles(); |
b170db2a RW |
868 | } |
869 | ||
870 | static inline void | |
871 | dpdk_queue_flush(struct netdev_dpdk *dev, int qid) | |
872 | { | |
873 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
874 | ||
875 | if (txq->count == 0) { | |
876 | return; | |
877 | } | |
b170db2a | 878 | dpdk_queue_flush__(dev, qid); |
8a9562d2 PS |
879 | } |
880 | ||
58397e6c KT |
881 | static bool |
882 | is_vhost_running(struct virtio_net *dev) | |
883 | { | |
884 | return (dev != NULL && (dev->flags & VIRTIO_DEV_RUNNING)); | |
885 | } | |
886 | ||
887 | /* | |
888 | * The receive path for the vhost port is the TX path out from guest. | |
889 | */ | |
890 | static int | |
891 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq_, | |
892 | struct dp_packet **packets, int *c) | |
893 | { | |
894 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
895 | struct netdev *netdev = rx->up.netdev; | |
896 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
897 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
898 | int qid = 1; | |
899 | uint16_t nb_rx = 0; | |
900 | ||
901 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
902 | return EAGAIN; | |
903 | } | |
904 | ||
905 | nb_rx = rte_vhost_dequeue_burst(virtio_dev, qid, | |
906 | vhost_dev->dpdk_mp->mp, | |
907 | (struct rte_mbuf **)packets, | |
cd159f1a | 908 | NETDEV_MAX_BURST); |
58397e6c KT |
909 | if (!nb_rx) { |
910 | return EAGAIN; | |
911 | } | |
912 | ||
45d947c4 | 913 | rte_spinlock_lock(&vhost_dev->stats_lock); |
58397e6c | 914 | vhost_dev->stats.rx_packets += (uint64_t)nb_rx; |
45d947c4 DDP |
915 | rte_spinlock_unlock(&vhost_dev->stats_lock); |
916 | ||
58397e6c KT |
917 | *c = (int) nb_rx; |
918 | return 0; | |
919 | } | |
920 | ||
8a9562d2 | 921 | static int |
e14deea0 | 922 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet **packets, |
91088554 | 923 | int *c) |
8a9562d2 PS |
924 | { |
925 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
926 | struct netdev *netdev = rx->up.netdev; | |
927 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 928 | int nb_rx; |
8a9562d2 | 929 | |
5496878c AW |
930 | /* There is only one tx queue for this core. Do not flush other |
931 | * queueus. */ | |
932 | if (rxq_->queue_id == rte_lcore_id()) { | |
933 | dpdk_queue_flush(dev, rxq_->queue_id); | |
934 | } | |
8a9562d2 PS |
935 | |
936 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq_->queue_id, | |
7d08d53e | 937 | (struct rte_mbuf **) packets, |
cd159f1a | 938 | NETDEV_MAX_BURST); |
8a9562d2 PS |
939 | if (!nb_rx) { |
940 | return EAGAIN; | |
941 | } | |
942 | ||
8a9562d2 PS |
943 | *c = nb_rx; |
944 | ||
945 | return 0; | |
946 | } | |
947 | ||
58397e6c KT |
948 | static void |
949 | __netdev_dpdk_vhost_send(struct netdev *netdev, struct dp_packet **pkts, | |
950 | int cnt, bool may_steal) | |
951 | { | |
952 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
953 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
95e9881f KT |
954 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
955 | unsigned int total_pkts = cnt; | |
956 | uint64_t start = 0; | |
58397e6c KT |
957 | |
958 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
45d947c4 | 959 | rte_spinlock_lock(&vhost_dev->stats_lock); |
1b99bb05 | 960 | vhost_dev->stats.tx_dropped+= cnt; |
45d947c4 | 961 | rte_spinlock_unlock(&vhost_dev->stats_lock); |
1b99bb05 | 962 | goto out; |
58397e6c KT |
963 | } |
964 | ||
965 | /* There is vHost TX single queue, So we need to lock it for TX. */ | |
a0cb2d66 | 966 | rte_spinlock_lock(&vhost_dev->vhost_tx_lock); |
58397e6c | 967 | |
95e9881f KT |
968 | do { |
969 | unsigned int tx_pkts; | |
970 | ||
971 | tx_pkts = rte_vhost_enqueue_burst(virtio_dev, VIRTIO_RXQ, | |
972 | cur_pkts, cnt); | |
973 | if (OVS_LIKELY(tx_pkts)) { | |
974 | /* Packets have been sent.*/ | |
975 | cnt -= tx_pkts; | |
976 | /* Prepare for possible next iteration.*/ | |
977 | cur_pkts = &cur_pkts[tx_pkts]; | |
978 | } else { | |
979 | uint64_t timeout = VHOST_ENQ_RETRY_USECS * rte_get_timer_hz() / 1E6; | |
980 | unsigned int expired = 0; | |
981 | ||
982 | if (!start) { | |
983 | start = rte_get_timer_cycles(); | |
984 | } | |
985 | ||
986 | /* | |
987 | * Unable to enqueue packets to vhost interface. | |
988 | * Check available entries before retrying. | |
989 | */ | |
990 | while (!rte_vring_available_entries(virtio_dev, VIRTIO_RXQ)) { | |
991 | if (OVS_UNLIKELY((rte_get_timer_cycles() - start) > timeout)) { | |
992 | expired = 1; | |
993 | break; | |
994 | } | |
995 | } | |
996 | if (expired) { | |
997 | /* break out of main loop. */ | |
998 | break; | |
999 | } | |
1000 | } | |
1001 | } while (cnt); | |
a0cb2d66 | 1002 | rte_spinlock_unlock(&vhost_dev->vhost_tx_lock); |
95e9881f | 1003 | |
45d947c4 | 1004 | rte_spinlock_lock(&vhost_dev->stats_lock); |
95e9881f KT |
1005 | vhost_dev->stats.tx_packets += (total_pkts - cnt); |
1006 | vhost_dev->stats.tx_dropped += cnt; | |
45d947c4 | 1007 | rte_spinlock_unlock(&vhost_dev->stats_lock); |
58397e6c KT |
1008 | |
1009 | out: | |
1010 | if (may_steal) { | |
95e9881f KT |
1011 | int i; |
1012 | ||
1013 | for (i = 0; i < total_pkts; i++) { | |
1b99bb05 MG |
1014 | dp_packet_delete(pkts[i]); |
1015 | } | |
58397e6c KT |
1016 | } |
1017 | } | |
1018 | ||
8a9562d2 | 1019 | inline static void |
f4fd623c DDP |
1020 | dpdk_queue_pkts(struct netdev_dpdk *dev, int qid, |
1021 | struct rte_mbuf **pkts, int cnt) | |
8a9562d2 PS |
1022 | { |
1023 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1024 | uint64_t diff_tsc; | |
8a9562d2 | 1025 | |
f4fd623c DDP |
1026 | int i = 0; |
1027 | ||
f4fd623c DDP |
1028 | while (i < cnt) { |
1029 | int freeslots = MAX_TX_QUEUE_LEN - txq->count; | |
1030 | int tocopy = MIN(freeslots, cnt-i); | |
8a9562d2 | 1031 | |
f4fd623c DDP |
1032 | memcpy(&txq->burst_pkts[txq->count], &pkts[i], |
1033 | tocopy * sizeof (struct rte_mbuf *)); | |
1034 | ||
1035 | txq->count += tocopy; | |
1036 | i += tocopy; | |
1037 | ||
94143fc4 | 1038 | if (txq->count == MAX_TX_QUEUE_LEN || txq->flush_tx) { |
b170db2a | 1039 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 1040 | } |
844f2d74 | 1041 | diff_tsc = rte_get_timer_cycles() - txq->tsc; |
f4fd623c | 1042 | if (diff_tsc >= DRAIN_TSC) { |
b170db2a | 1043 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 1044 | } |
8a9562d2 | 1045 | } |
8a9562d2 PS |
1046 | } |
1047 | ||
1048 | /* Tx function. Transmit packets indefinitely */ | |
1049 | static void | |
58397e6c | 1050 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet **pkts, |
2654cc33 | 1051 | int cnt) |
db73f716 | 1052 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1053 | { |
bce01e3a EJ |
1054 | #if !defined(__CHECKER__) && !defined(_WIN32) |
1055 | const size_t PKT_ARRAY_SIZE = cnt; | |
1056 | #else | |
1057 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1058 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1059 | #endif |
8a9562d2 | 1060 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
bce01e3a | 1061 | struct rte_mbuf *mbufs[PKT_ARRAY_SIZE]; |
175cf4de RW |
1062 | int dropped = 0; |
1063 | int newcnt = 0; | |
1064 | int i; | |
8a9562d2 | 1065 | |
db73f716 DDP |
1066 | /* If we are on a non pmd thread we have to use the mempool mutex, because |
1067 | * every non pmd thread shares the same mempool cache */ | |
1068 | ||
1069 | if (!thread_is_pmd()) { | |
1070 | ovs_mutex_lock(&nonpmd_mempool_mutex); | |
1071 | } | |
1072 | ||
f4fd623c | 1073 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1074 | int size = dp_packet_size(pkts[i]); |
95fb793a | 1075 | |
f98d7864 | 1076 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c DDP |
1077 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1078 | (int)size , dev->max_packet_len); | |
1079 | ||
175cf4de | 1080 | dropped++; |
f4fd623c DDP |
1081 | continue; |
1082 | } | |
8a9562d2 | 1083 | |
f4fd623c | 1084 | mbufs[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1085 | |
f4fd623c | 1086 | if (!mbufs[newcnt]) { |
175cf4de RW |
1087 | dropped += cnt - i; |
1088 | break; | |
f4fd623c DDP |
1089 | } |
1090 | ||
1091 | /* We have to do a copy for now */ | |
b8e57534 | 1092 | memcpy(rte_pktmbuf_mtod(mbufs[newcnt], void *), dp_packet_data(pkts[i]), size); |
f4fd623c DDP |
1093 | |
1094 | rte_pktmbuf_data_len(mbufs[newcnt]) = size; | |
1095 | rte_pktmbuf_pkt_len(mbufs[newcnt]) = size; | |
1096 | ||
1097 | newcnt++; | |
1098 | } | |
8a9562d2 | 1099 | |
f98d7864 | 1100 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1101 | rte_spinlock_lock(&dev->stats_lock); |
175cf4de | 1102 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1103 | rte_spinlock_unlock(&dev->stats_lock); |
175cf4de RW |
1104 | } |
1105 | ||
58397e6c KT |
1106 | if (dev->type == DPDK_DEV_VHOST) { |
1107 | __netdev_dpdk_vhost_send(netdev, (struct dp_packet **) mbufs, newcnt, true); | |
1108 | } else { | |
1109 | dpdk_queue_pkts(dev, qid, mbufs, newcnt); | |
1110 | dpdk_queue_flush(dev, qid); | |
1111 | } | |
db73f716 DDP |
1112 | |
1113 | if (!thread_is_pmd()) { | |
1114 | ovs_mutex_unlock(&nonpmd_mempool_mutex); | |
1115 | } | |
8a9562d2 PS |
1116 | } |
1117 | ||
58397e6c KT |
1118 | static int |
1119 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid OVS_UNUSED, struct dp_packet **pkts, | |
1120 | int cnt, bool may_steal) | |
1121 | { | |
1122 | if (OVS_UNLIKELY(pkts[0]->source != DPBUF_DPDK)) { | |
1123 | int i; | |
1124 | ||
1125 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); | |
1126 | if (may_steal) { | |
1127 | for (i = 0; i < cnt; i++) { | |
1128 | dp_packet_delete(pkts[i]); | |
1129 | } | |
1130 | } | |
1131 | } else { | |
1132 | __netdev_dpdk_vhost_send(netdev, pkts, cnt, may_steal); | |
1133 | } | |
1134 | return 0; | |
1135 | } | |
1136 | ||
7251515e DV |
1137 | static inline void |
1138 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
e14deea0 | 1139 | struct dp_packet **pkts, int cnt, bool may_steal) |
8a9562d2 | 1140 | { |
f4fd623c | 1141 | int i; |
8a9562d2 | 1142 | |
a0cb2d66 DDP |
1143 | if (OVS_UNLIKELY(dev->txq_needs_locking)) { |
1144 | qid = qid % dev->real_n_txq; | |
1145 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); | |
1146 | } | |
1147 | ||
7251515e | 1148 | if (OVS_UNLIKELY(!may_steal || |
cf62fa4c | 1149 | pkts[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1150 | struct netdev *netdev = &dev->up; |
1151 | ||
2654cc33 | 1152 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); |
b3cd9f9d PS |
1153 | |
1154 | if (may_steal) { | |
f4fd623c | 1155 | for (i = 0; i < cnt; i++) { |
e14deea0 | 1156 | dp_packet_delete(pkts[i]); |
f4fd623c | 1157 | } |
b3cd9f9d | 1158 | } |
8a9562d2 | 1159 | } else { |
f4fd623c DDP |
1160 | int next_tx_idx = 0; |
1161 | int dropped = 0; | |
8a9562d2 | 1162 | |
f4fd623c | 1163 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1164 | int size = dp_packet_size(pkts[i]); |
1b99bb05 | 1165 | |
f4fd623c DDP |
1166 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
1167 | if (next_tx_idx != i) { | |
1168 | dpdk_queue_pkts(dev, qid, | |
1169 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1170 | i-next_tx_idx); | |
1ebfe1ac | 1171 | } |
f4fd623c | 1172 | |
1ebfe1ac DDP |
1173 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1174 | (int)size , dev->max_packet_len); | |
f4fd623c | 1175 | |
e14deea0 | 1176 | dp_packet_delete(pkts[i]); |
1ebfe1ac | 1177 | dropped++; |
f4fd623c DDP |
1178 | next_tx_idx = i + 1; |
1179 | } | |
1180 | } | |
1181 | if (next_tx_idx != cnt) { | |
1182 | dpdk_queue_pkts(dev, qid, | |
1183 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1184 | cnt-next_tx_idx); | |
1185 | } | |
8a9562d2 | 1186 | |
f4fd623c | 1187 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1188 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1189 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1190 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1191 | } |
8a9562d2 | 1192 | } |
a0cb2d66 DDP |
1193 | |
1194 | if (OVS_UNLIKELY(dev->txq_needs_locking)) { | |
1195 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); | |
1196 | } | |
7251515e DV |
1197 | } |
1198 | ||
1199 | static int | |
1200 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
e14deea0 | 1201 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1202 | { |
1203 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1204 | |
7251515e DV |
1205 | netdev_dpdk_send__(dev, qid, pkts, cnt, may_steal); |
1206 | return 0; | |
8a9562d2 PS |
1207 | } |
1208 | ||
1209 | static int | |
1210 | netdev_dpdk_set_etheraddr(struct netdev *netdev, | |
1211 | const uint8_t mac[ETH_ADDR_LEN]) | |
1212 | { | |
1213 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1214 | ||
1215 | ovs_mutex_lock(&dev->mutex); | |
1216 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
1217 | memcpy(dev->hwaddr, mac, ETH_ADDR_LEN); | |
045c0d1a | 1218 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1219 | } |
1220 | ovs_mutex_unlock(&dev->mutex); | |
1221 | ||
1222 | return 0; | |
1223 | } | |
1224 | ||
1225 | static int | |
1226 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, | |
1227 | uint8_t mac[ETH_ADDR_LEN]) | |
1228 | { | |
1229 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1230 | ||
1231 | ovs_mutex_lock(&dev->mutex); | |
1232 | memcpy(mac, dev->hwaddr, ETH_ADDR_LEN); | |
1233 | ovs_mutex_unlock(&dev->mutex); | |
1234 | ||
1235 | return 0; | |
1236 | } | |
1237 | ||
1238 | static int | |
1239 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1240 | { | |
1241 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1242 | ||
1243 | ovs_mutex_lock(&dev->mutex); | |
1244 | *mtup = dev->mtu; | |
1245 | ovs_mutex_unlock(&dev->mutex); | |
1246 | ||
1247 | return 0; | |
1248 | } | |
1249 | ||
1250 | static int | |
1251 | netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu) | |
1252 | { | |
1253 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1254 | int old_mtu, err; | |
1255 | struct dpdk_mp *old_mp; | |
1256 | struct dpdk_mp *mp; | |
1257 | ||
1258 | ovs_mutex_lock(&dpdk_mutex); | |
1259 | ovs_mutex_lock(&dev->mutex); | |
1260 | if (dev->mtu == mtu) { | |
1261 | err = 0; | |
1262 | goto out; | |
1263 | } | |
1264 | ||
1265 | mp = dpdk_mp_get(dev->socket_id, dev->mtu); | |
1266 | if (!mp) { | |
1267 | err = ENOMEM; | |
1268 | goto out; | |
1269 | } | |
1270 | ||
1271 | rte_eth_dev_stop(dev->port_id); | |
1272 | ||
1273 | old_mtu = dev->mtu; | |
1274 | old_mp = dev->dpdk_mp; | |
1275 | dev->dpdk_mp = mp; | |
1276 | dev->mtu = mtu; | |
1277 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
1278 | ||
1279 | err = dpdk_eth_dev_init(dev); | |
1280 | if (err) { | |
8a9562d2 PS |
1281 | dpdk_mp_put(mp); |
1282 | dev->mtu = old_mtu; | |
1283 | dev->dpdk_mp = old_mp; | |
1284 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
1285 | dpdk_eth_dev_init(dev); | |
1286 | goto out; | |
1287 | } | |
1288 | ||
1289 | dpdk_mp_put(old_mp); | |
045c0d1a | 1290 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1291 | out: |
1292 | ovs_mutex_unlock(&dev->mutex); | |
1293 | ovs_mutex_unlock(&dpdk_mutex); | |
1294 | return err; | |
1295 | } | |
1296 | ||
1297 | static int | |
1298 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier); | |
1299 | ||
58397e6c KT |
1300 | static int |
1301 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1302 | struct netdev_stats *stats) | |
1303 | { | |
1304 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1305 | ||
1306 | ovs_mutex_lock(&dev->mutex); | |
1307 | memset(stats, 0, sizeof(*stats)); | |
1308 | /* Unsupported Stats */ | |
1309 | stats->rx_errors = UINT64_MAX; | |
1310 | stats->tx_errors = UINT64_MAX; | |
1311 | stats->multicast = UINT64_MAX; | |
1312 | stats->collisions = UINT64_MAX; | |
1313 | stats->rx_crc_errors = UINT64_MAX; | |
1314 | stats->rx_fifo_errors = UINT64_MAX; | |
1315 | stats->rx_frame_errors = UINT64_MAX; | |
1316 | stats->rx_length_errors = UINT64_MAX; | |
1317 | stats->rx_missed_errors = UINT64_MAX; | |
1318 | stats->rx_over_errors = UINT64_MAX; | |
1319 | stats->tx_aborted_errors = UINT64_MAX; | |
1320 | stats->tx_carrier_errors = UINT64_MAX; | |
1321 | stats->tx_errors = UINT64_MAX; | |
1322 | stats->tx_fifo_errors = UINT64_MAX; | |
1323 | stats->tx_heartbeat_errors = UINT64_MAX; | |
1324 | stats->tx_window_errors = UINT64_MAX; | |
1325 | stats->rx_bytes += UINT64_MAX; | |
1326 | stats->rx_dropped += UINT64_MAX; | |
1327 | stats->tx_bytes += UINT64_MAX; | |
1328 | ||
45d947c4 | 1329 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
1330 | /* Supported Stats */ |
1331 | stats->rx_packets += dev->stats.rx_packets; | |
1332 | stats->tx_packets += dev->stats.tx_packets; | |
1333 | stats->tx_dropped += dev->stats.tx_dropped; | |
45d947c4 | 1334 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1335 | ovs_mutex_unlock(&dev->mutex); |
1336 | ||
1337 | return 0; | |
1338 | } | |
1339 | ||
8a9562d2 PS |
1340 | static int |
1341 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1342 | { | |
1343 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1344 | struct rte_eth_stats rte_stats; | |
1345 | bool gg; | |
1346 | ||
1347 | netdev_dpdk_get_carrier(netdev, &gg); | |
1348 | ovs_mutex_lock(&dev->mutex); | |
1349 | rte_eth_stats_get(dev->port_id, &rte_stats); | |
1350 | ||
2f9dd77f | 1351 | memset(stats, 0, sizeof(*stats)); |
8a9562d2 | 1352 | |
2f9dd77f PS |
1353 | stats->rx_packets = rte_stats.ipackets; |
1354 | stats->tx_packets = rte_stats.opackets; | |
1355 | stats->rx_bytes = rte_stats.ibytes; | |
1356 | stats->tx_bytes = rte_stats.obytes; | |
1357 | stats->rx_errors = rte_stats.ierrors; | |
1358 | stats->tx_errors = rte_stats.oerrors; | |
1359 | stats->multicast = rte_stats.imcasts; | |
8a9562d2 | 1360 | |
45d947c4 | 1361 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 1362 | stats->tx_dropped = dev->stats.tx_dropped; |
45d947c4 | 1363 | rte_spinlock_unlock(&dev->stats_lock); |
8a9562d2 PS |
1364 | ovs_mutex_unlock(&dev->mutex); |
1365 | ||
1366 | return 0; | |
1367 | } | |
1368 | ||
1369 | static int | |
1370 | netdev_dpdk_get_features(const struct netdev *netdev_, | |
1371 | enum netdev_features *current, | |
1372 | enum netdev_features *advertised OVS_UNUSED, | |
1373 | enum netdev_features *supported OVS_UNUSED, | |
1374 | enum netdev_features *peer OVS_UNUSED) | |
1375 | { | |
1376 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1377 | struct rte_eth_link link; | |
1378 | ||
1379 | ovs_mutex_lock(&dev->mutex); | |
1380 | link = dev->link; | |
1381 | ovs_mutex_unlock(&dev->mutex); | |
1382 | ||
1383 | if (link.link_duplex == ETH_LINK_AUTONEG_DUPLEX) { | |
1384 | if (link.link_speed == ETH_LINK_SPEED_AUTONEG) { | |
1385 | *current = NETDEV_F_AUTONEG; | |
1386 | } | |
1387 | } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { | |
1388 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1389 | *current = NETDEV_F_10MB_HD; | |
1390 | } | |
1391 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1392 | *current = NETDEV_F_100MB_HD; | |
1393 | } | |
1394 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1395 | *current = NETDEV_F_1GB_HD; | |
1396 | } | |
1397 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
1398 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1399 | *current = NETDEV_F_10MB_FD; | |
1400 | } | |
1401 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1402 | *current = NETDEV_F_100MB_FD; | |
1403 | } | |
1404 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1405 | *current = NETDEV_F_1GB_FD; | |
1406 | } | |
1407 | if (link.link_speed == ETH_LINK_SPEED_10000) { | |
1408 | *current = NETDEV_F_10GB_FD; | |
1409 | } | |
1410 | } | |
1411 | ||
1412 | return 0; | |
1413 | } | |
1414 | ||
1415 | static int | |
1416 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
1417 | { | |
1418 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1419 | int ifindex; | |
1420 | ||
1421 | ovs_mutex_lock(&dev->mutex); | |
1422 | ifindex = dev->port_id; | |
1423 | ovs_mutex_unlock(&dev->mutex); | |
1424 | ||
1425 | return ifindex; | |
1426 | } | |
1427 | ||
1428 | static int | |
1429 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1430 | { | |
1431 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1432 | ||
1433 | ovs_mutex_lock(&dev->mutex); | |
1434 | check_link_status(dev); | |
1435 | *carrier = dev->link.link_status; | |
58397e6c KT |
1436 | |
1437 | ovs_mutex_unlock(&dev->mutex); | |
1438 | ||
1439 | return 0; | |
1440 | } | |
1441 | ||
1442 | static int | |
1443 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1444 | { | |
1445 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1446 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(dev); | |
1447 | ||
1448 | ovs_mutex_lock(&dev->mutex); | |
1449 | ||
1450 | if (is_vhost_running(virtio_dev)) { | |
1451 | *carrier = 1; | |
1452 | } else { | |
1453 | *carrier = 0; | |
1454 | } | |
1455 | ||
8a9562d2 PS |
1456 | ovs_mutex_unlock(&dev->mutex); |
1457 | ||
1458 | return 0; | |
1459 | } | |
1460 | ||
1461 | static long long int | |
1462 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev_) | |
1463 | { | |
1464 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1465 | long long int carrier_resets; | |
1466 | ||
1467 | ovs_mutex_lock(&dev->mutex); | |
1468 | carrier_resets = dev->link_reset_cnt; | |
1469 | ovs_mutex_unlock(&dev->mutex); | |
1470 | ||
1471 | return carrier_resets; | |
1472 | } | |
1473 | ||
1474 | static int | |
1475 | netdev_dpdk_set_miimon(struct netdev *netdev_ OVS_UNUSED, | |
1476 | long long int interval OVS_UNUSED) | |
1477 | { | |
ee32150e | 1478 | return EOPNOTSUPP; |
8a9562d2 PS |
1479 | } |
1480 | ||
1481 | static int | |
1482 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
1483 | enum netdev_flags off, enum netdev_flags on, | |
95fb793a | 1484 | enum netdev_flags *old_flagsp) OVS_REQUIRES(dev->mutex) |
8a9562d2 PS |
1485 | { |
1486 | int err; | |
1487 | ||
1488 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
1489 | return EINVAL; | |
1490 | } | |
1491 | ||
1492 | *old_flagsp = dev->flags; | |
1493 | dev->flags |= on; | |
1494 | dev->flags &= ~off; | |
1495 | ||
1496 | if (dev->flags == *old_flagsp) { | |
1497 | return 0; | |
1498 | } | |
1499 | ||
58397e6c KT |
1500 | if (dev->type == DPDK_DEV_ETH) { |
1501 | if (dev->flags & NETDEV_UP) { | |
1502 | err = rte_eth_dev_start(dev->port_id); | |
1503 | if (err) | |
1504 | return -err; | |
1505 | } | |
8a9562d2 | 1506 | |
58397e6c KT |
1507 | if (dev->flags & NETDEV_PROMISC) { |
1508 | rte_eth_promiscuous_enable(dev->port_id); | |
1509 | } | |
8a9562d2 | 1510 | |
58397e6c KT |
1511 | if (!(dev->flags & NETDEV_UP)) { |
1512 | rte_eth_dev_stop(dev->port_id); | |
1513 | } | |
8a9562d2 PS |
1514 | } |
1515 | ||
1516 | return 0; | |
1517 | } | |
1518 | ||
1519 | static int | |
1520 | netdev_dpdk_update_flags(struct netdev *netdev_, | |
1521 | enum netdev_flags off, enum netdev_flags on, | |
1522 | enum netdev_flags *old_flagsp) | |
1523 | { | |
1524 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1525 | int error; | |
1526 | ||
1527 | ovs_mutex_lock(&netdev->mutex); | |
1528 | error = netdev_dpdk_update_flags__(netdev, off, on, old_flagsp); | |
1529 | ovs_mutex_unlock(&netdev->mutex); | |
1530 | ||
1531 | return error; | |
1532 | } | |
1533 | ||
1534 | static int | |
1535 | netdev_dpdk_get_status(const struct netdev *netdev_, struct smap *args) | |
1536 | { | |
1537 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1538 | struct rte_eth_dev_info dev_info; | |
1539 | ||
e0a801c7 | 1540 | if (dev->port_id < 0) |
8a9562d2 PS |
1541 | return ENODEV; |
1542 | ||
1543 | ovs_mutex_lock(&dev->mutex); | |
1544 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
1545 | ovs_mutex_unlock(&dev->mutex); | |
1546 | ||
1547 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1548 | ||
95fb793a | 1549 | smap_add_format(args, "port_no", "%d", dev->port_id); |
8a9562d2 PS |
1550 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); |
1551 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1552 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
1553 | smap_add_format(args, "max_rx_pktlen", "%u", dev_info.max_rx_pktlen); | |
1554 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); | |
1555 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
1556 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
1557 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
1558 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
1559 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
1560 | ||
1561 | smap_add_format(args, "pci-vendor_id", "0x%u", dev_info.pci_dev->id.vendor_id); | |
1562 | smap_add_format(args, "pci-device_id", "0x%x", dev_info.pci_dev->id.device_id); | |
1563 | ||
1564 | return 0; | |
1565 | } | |
1566 | ||
1567 | static void | |
1568 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
1569 | OVS_REQUIRES(dev->mutex) | |
1570 | { | |
1571 | enum netdev_flags old_flags; | |
1572 | ||
1573 | if (admin_state) { | |
1574 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
1575 | } else { | |
1576 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
1577 | } | |
1578 | } | |
1579 | ||
1580 | static void | |
1581 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
1582 | const char *argv[], void *aux OVS_UNUSED) | |
1583 | { | |
1584 | bool up; | |
1585 | ||
1586 | if (!strcasecmp(argv[argc - 1], "up")) { | |
1587 | up = true; | |
1588 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
1589 | up = false; | |
1590 | } else { | |
1591 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
1592 | return; | |
1593 | } | |
1594 | ||
1595 | if (argc > 2) { | |
1596 | struct netdev *netdev = netdev_from_name(argv[1]); | |
1597 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
1598 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
1599 | ||
1600 | ovs_mutex_lock(&dpdk_dev->mutex); | |
1601 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
1602 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
1603 | ||
1604 | netdev_close(netdev); | |
1605 | } else { | |
1606 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
1607 | netdev_close(netdev); | |
1608 | return; | |
1609 | } | |
1610 | } else { | |
1611 | struct netdev_dpdk *netdev; | |
1612 | ||
1613 | ovs_mutex_lock(&dpdk_mutex); | |
1614 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
1615 | ovs_mutex_lock(&netdev->mutex); | |
1616 | netdev_dpdk_set_admin_state__(netdev, up); | |
1617 | ovs_mutex_unlock(&netdev->mutex); | |
1618 | } | |
1619 | ovs_mutex_unlock(&dpdk_mutex); | |
1620 | } | |
1621 | unixctl_command_reply(conn, "OK"); | |
1622 | } | |
1623 | ||
58397e6c KT |
1624 | /* |
1625 | * Set virtqueue flags so that we do not receive interrupts. | |
1626 | */ | |
1627 | static void | |
1628 | set_irq_status(struct virtio_net *dev) | |
1629 | { | |
1630 | dev->virtqueue[VIRTIO_RXQ]->used->flags = VRING_USED_F_NO_NOTIFY; | |
1631 | dev->virtqueue[VIRTIO_TXQ]->used->flags = VRING_USED_F_NO_NOTIFY; | |
1632 | } | |
1633 | ||
1634 | /* | |
1635 | * A new virtio-net device is added to a vhost port. | |
1636 | */ | |
1637 | static int | |
1638 | new_device(struct virtio_net *dev) | |
1639 | { | |
1640 | struct netdev_dpdk *netdev; | |
1641 | bool exists = false; | |
1642 | ||
1643 | ovs_mutex_lock(&dpdk_mutex); | |
1644 | /* Add device to the vhost port with the same name as that passed down. */ | |
1645 | LIST_FOR_EACH(netdev, list_node, &dpdk_list) { | |
7d1ced01 | 1646 | if (strncmp(dev->ifname, netdev->vhost_id, IF_NAME_SZ) == 0) { |
58397e6c KT |
1647 | ovs_mutex_lock(&netdev->mutex); |
1648 | ovsrcu_set(&netdev->virtio_dev, dev); | |
1649 | ovs_mutex_unlock(&netdev->mutex); | |
1650 | exists = true; | |
1651 | dev->flags |= VIRTIO_DEV_RUNNING; | |
1652 | /* Disable notifications. */ | |
1653 | set_irq_status(dev); | |
1654 | break; | |
1655 | } | |
1656 | } | |
1657 | ovs_mutex_unlock(&dpdk_mutex); | |
1658 | ||
1659 | if (!exists) { | |
1660 | VLOG_INFO("vHost Device '%s' (%ld) can't be added - name not found", | |
1661 | dev->ifname, dev->device_fh); | |
1662 | ||
1663 | return -1; | |
1664 | } | |
1665 | ||
1666 | VLOG_INFO("vHost Device '%s' (%ld) has been added", | |
1667 | dev->ifname, dev->device_fh); | |
1668 | return 0; | |
1669 | } | |
1670 | ||
1671 | /* | |
1672 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
1673 | * flag to stop any more packets from being sent or received to/from a VM and | |
1674 | * ensure all currently queued packets have been sent/received before removing | |
1675 | * the device. | |
1676 | */ | |
1677 | static void | |
1678 | destroy_device(volatile struct virtio_net *dev) | |
1679 | { | |
1680 | struct netdev_dpdk *vhost_dev; | |
1681 | ||
1682 | ovs_mutex_lock(&dpdk_mutex); | |
1683 | LIST_FOR_EACH (vhost_dev, list_node, &dpdk_list) { | |
1684 | if (netdev_dpdk_get_virtio(vhost_dev) == dev) { | |
1685 | ||
1686 | ovs_mutex_lock(&vhost_dev->mutex); | |
1687 | dev->flags &= ~VIRTIO_DEV_RUNNING; | |
1688 | ovsrcu_set(&vhost_dev->virtio_dev, NULL); | |
1689 | ovs_mutex_unlock(&vhost_dev->mutex); | |
1690 | ||
1691 | /* | |
1692 | * Wait for other threads to quiesce before | |
1693 | * setting the virtio_dev to NULL. | |
1694 | */ | |
1695 | ovsrcu_synchronize(); | |
618f44f7 KT |
1696 | /* |
1697 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
1698 | * put thread back into quiescent state before returning. | |
1699 | */ | |
1700 | ovsrcu_quiesce_start(); | |
58397e6c KT |
1701 | } |
1702 | } | |
1703 | ovs_mutex_unlock(&dpdk_mutex); | |
1704 | ||
1705 | VLOG_INFO("vHost Device '%s' (%ld) has been removed", | |
1706 | dev->ifname, dev->device_fh); | |
1707 | } | |
1708 | ||
1709 | struct virtio_net * | |
1710 | netdev_dpdk_get_virtio(const struct netdev_dpdk *dev) | |
1711 | { | |
1712 | return ovsrcu_get(struct virtio_net *, &dev->virtio_dev); | |
1713 | } | |
1714 | ||
1715 | /* | |
1716 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
1717 | * configuration has been fully complete. | |
1718 | */ | |
bce01e3a | 1719 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
1720 | { |
1721 | .new_device = new_device, | |
1722 | .destroy_device = destroy_device, | |
1723 | }; | |
1724 | ||
1725 | static void * | |
7d1ced01 | 1726 | start_vhost_loop(void *dummy OVS_UNUSED) |
58397e6c KT |
1727 | { |
1728 | pthread_detach(pthread_self()); | |
618f44f7 KT |
1729 | /* Put the cuse thread into quiescent state. */ |
1730 | ovsrcu_quiesce_start(); | |
58397e6c KT |
1731 | rte_vhost_driver_session_start(); |
1732 | return NULL; | |
1733 | } | |
1734 | ||
1735 | static int | |
1736 | dpdk_vhost_class_init(void) | |
7d1ced01 CL |
1737 | { |
1738 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
1739 | ovs_thread_create("vhost_thread", start_vhost_loop, NULL); | |
1740 | return 0; | |
1741 | } | |
1742 | ||
1743 | static int | |
1744 | dpdk_vhost_cuse_class_init(void) | |
58397e6c | 1745 | { |
58397e6c KT |
1746 | int err = -1; |
1747 | ||
58397e6c KT |
1748 | |
1749 | /* Register CUSE device to handle IOCTLs. | |
1750 | * Unless otherwise specified on the vswitchd command line, cuse_dev_name | |
1751 | * is set to vhost-net. | |
1752 | */ | |
1753 | err = rte_vhost_driver_register(cuse_dev_name); | |
1754 | ||
1755 | if (err != 0) { | |
1756 | VLOG_ERR("CUSE device setup failure."); | |
1757 | return -1; | |
1758 | } | |
1759 | ||
7d1ced01 CL |
1760 | dpdk_vhost_class_init(); |
1761 | return 0; | |
1762 | } | |
1763 | ||
1764 | static int | |
1765 | dpdk_vhost_user_class_init(void) | |
1766 | { | |
1767 | dpdk_vhost_class_init(); | |
618f44f7 | 1768 | return 0; |
58397e6c KT |
1769 | } |
1770 | ||
033e9df2 DDP |
1771 | static void |
1772 | dpdk_common_init(void) | |
1773 | { | |
1774 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
1775 | "[netdev] up|down", 1, 2, | |
1776 | netdev_dpdk_set_admin_state, NULL); | |
1777 | ||
1778 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
1779 | } | |
1780 | ||
95fb793a | 1781 | /* Client Rings */ |
1782 | ||
95fb793a | 1783 | static int |
1784 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
1785 | unsigned int *eth_port_id) | |
1786 | { | |
1787 | struct dpdk_ring *ivshmem; | |
1788 | char ring_name[10]; | |
1789 | int err; | |
1790 | ||
1791 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
1792 | if (ivshmem == NULL) { | |
1793 | return ENOMEM; | |
1794 | } | |
1795 | ||
7251515e | 1796 | /* XXX: Add support for multiquque ring. */ |
95fb793a | 1797 | err = snprintf(ring_name, 10, "%s_tx", dev_name); |
1798 | if (err < 0) { | |
1799 | return -err; | |
1800 | } | |
1801 | ||
7251515e DV |
1802 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1803 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1804 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1805 | if (ivshmem->cring_tx == NULL) { |
1806 | rte_free(ivshmem); | |
1807 | return ENOMEM; | |
1808 | } | |
1809 | ||
1810 | err = snprintf(ring_name, 10, "%s_rx", dev_name); | |
1811 | if (err < 0) { | |
1812 | return -err; | |
1813 | } | |
1814 | ||
7251515e DV |
1815 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1816 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1817 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1818 | if (ivshmem->cring_rx == NULL) { |
1819 | rte_free(ivshmem); | |
1820 | return ENOMEM; | |
1821 | } | |
1822 | ||
d7310583 DDP |
1823 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
1824 | &ivshmem->cring_tx, 1, SOCKET0); | |
1825 | ||
95fb793a | 1826 | if (err < 0) { |
1827 | rte_free(ivshmem); | |
1828 | return ENODEV; | |
1829 | } | |
1830 | ||
1831 | ivshmem->user_port_id = port_no; | |
1832 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
1833 | list_push_back(&dpdk_ring_list, &ivshmem->list_node); | |
1834 | ||
1835 | *eth_port_id = ivshmem->eth_port_id; | |
1836 | return 0; | |
1837 | } | |
1838 | ||
1839 | static int | |
1840 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) OVS_REQUIRES(dpdk_mutex) | |
1841 | { | |
1842 | struct dpdk_ring *ivshmem; | |
1843 | unsigned int port_no; | |
1844 | int err = 0; | |
1845 | ||
1846 | /* Names always start with "dpdkr" */ | |
1847 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
1848 | if (err) { | |
1849 | return err; | |
1850 | } | |
1851 | ||
1852 | /* look through our list to find the device */ | |
1853 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { | |
1854 | if (ivshmem->user_port_id == port_no) { | |
58397e6c | 1855 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
95fb793a | 1856 | *eth_port_id = ivshmem->eth_port_id; /* really all that is needed */ |
1857 | return 0; | |
1858 | } | |
1859 | } | |
1860 | /* Need to create the device rings */ | |
1861 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
1862 | } | |
1863 | ||
7251515e | 1864 | static int |
a0cb2d66 | 1865 | netdev_dpdk_ring_send(struct netdev *netdev_, int qid, |
e14deea0 | 1866 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e | 1867 | { |
a0cb2d66 | 1868 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); |
1b99bb05 MG |
1869 | unsigned i; |
1870 | ||
1871 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that the | |
1872 | * rss hash field is clear. This is because the same mbuf may be modified by | |
1873 | * the consumer of the ring and return into the datapath without recalculating | |
1874 | * the RSS hash. */ | |
1875 | for (i = 0; i < cnt; i++) { | |
1876 | dp_packet_set_rss_hash(pkts[i], 0); | |
1877 | } | |
7251515e | 1878 | |
a0cb2d66 | 1879 | netdev_dpdk_send__(netdev, qid, pkts, cnt, may_steal); |
7251515e DV |
1880 | return 0; |
1881 | } | |
1882 | ||
95fb793a | 1883 | static int |
1884 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
1885 | { | |
1886 | unsigned int port_no = 0; | |
1887 | int err = 0; | |
1888 | ||
1889 | if (rte_eal_init_ret) { | |
1890 | return rte_eal_init_ret; | |
1891 | } | |
1892 | ||
1893 | ovs_mutex_lock(&dpdk_mutex); | |
1894 | ||
1895 | err = dpdk_ring_open(netdev->name, &port_no); | |
1896 | if (err) { | |
1897 | goto unlock_dpdk; | |
1898 | } | |
1899 | ||
58397e6c | 1900 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
95fb793a | 1901 | |
1902 | unlock_dpdk: | |
1903 | ovs_mutex_unlock(&dpdk_mutex); | |
1904 | return err; | |
1905 | } | |
1906 | ||
58397e6c KT |
1907 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, MULTIQ, SEND, \ |
1908 | GET_CARRIER, GET_STATS, GET_FEATURES, GET_STATUS, RXQ_RECV) \ | |
95fb793a | 1909 | { \ |
1910 | NAME, \ | |
1911 | INIT, /* init */ \ | |
1912 | NULL, /* netdev_dpdk_run */ \ | |
1913 | NULL, /* netdev_dpdk_wait */ \ | |
1914 | \ | |
1915 | netdev_dpdk_alloc, \ | |
1916 | CONSTRUCT, \ | |
58397e6c | 1917 | DESTRUCT, \ |
95fb793a | 1918 | netdev_dpdk_dealloc, \ |
1919 | netdev_dpdk_get_config, \ | |
1920 | NULL, /* netdev_dpdk_set_config */ \ | |
1921 | NULL, /* get_tunnel_config */ \ | |
58397e6c KT |
1922 | NULL, /* build header */ \ |
1923 | NULL, /* push header */ \ | |
1924 | NULL, /* pop header */ \ | |
7dec44fe | 1925 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
5496878c | 1926 | MULTIQ, /* set_multiq */ \ |
95fb793a | 1927 | \ |
7251515e | 1928 | SEND, /* send */ \ |
95fb793a | 1929 | NULL, /* send_wait */ \ |
1930 | \ | |
1931 | netdev_dpdk_set_etheraddr, \ | |
1932 | netdev_dpdk_get_etheraddr, \ | |
1933 | netdev_dpdk_get_mtu, \ | |
1934 | netdev_dpdk_set_mtu, \ | |
1935 | netdev_dpdk_get_ifindex, \ | |
58397e6c | 1936 | GET_CARRIER, \ |
95fb793a | 1937 | netdev_dpdk_get_carrier_resets, \ |
1938 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
1939 | GET_STATS, \ |
1940 | GET_FEATURES, \ | |
95fb793a | 1941 | NULL, /* set_advertisements */ \ |
1942 | \ | |
1943 | NULL, /* set_policing */ \ | |
1944 | NULL, /* get_qos_types */ \ | |
1945 | NULL, /* get_qos_capabilities */ \ | |
1946 | NULL, /* get_qos */ \ | |
1947 | NULL, /* set_qos */ \ | |
1948 | NULL, /* get_queue */ \ | |
1949 | NULL, /* set_queue */ \ | |
1950 | NULL, /* delete_queue */ \ | |
1951 | NULL, /* get_queue_stats */ \ | |
1952 | NULL, /* queue_dump_start */ \ | |
1953 | NULL, /* queue_dump_next */ \ | |
1954 | NULL, /* queue_dump_done */ \ | |
1955 | NULL, /* dump_queue_stats */ \ | |
1956 | \ | |
1957 | NULL, /* get_in4 */ \ | |
1958 | NULL, /* set_in4 */ \ | |
1959 | NULL, /* get_in6 */ \ | |
1960 | NULL, /* add_router */ \ | |
1961 | NULL, /* get_next_hop */ \ | |
58397e6c | 1962 | GET_STATUS, \ |
95fb793a | 1963 | NULL, /* arp_lookup */ \ |
1964 | \ | |
1965 | netdev_dpdk_update_flags, \ | |
1966 | \ | |
1967 | netdev_dpdk_rxq_alloc, \ | |
1968 | netdev_dpdk_rxq_construct, \ | |
1969 | netdev_dpdk_rxq_destruct, \ | |
1970 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 1971 | RXQ_RECV, \ |
95fb793a | 1972 | NULL, /* rx_wait */ \ |
1973 | NULL, /* rxq_drain */ \ | |
1974 | } | |
8a9562d2 | 1975 | |
7d1ced01 CL |
1976 | static int |
1977 | process_vhost_flags(char *flag, char *default_val, int size, | |
1978 | char **argv, char **new_val) | |
1979 | { | |
1980 | int changed = 0; | |
1981 | ||
1982 | /* Depending on which version of vhost is in use, process the vhost-specific | |
1983 | * flag if it is provided on the vswitchd command line, otherwise resort to | |
1984 | * a default value. | |
1985 | * | |
1986 | * For vhost-user: Process "-cuse_dev_name" to set the custom location of | |
1987 | * the vhost-user socket(s). | |
1988 | * For vhost-cuse: Process "-vhost_sock_dir" to set the custom name of the | |
1989 | * vhost-cuse character device. | |
1990 | */ | |
1991 | if (!strcmp(argv[1], flag) && (strlen(argv[2]) <= size)) { | |
1992 | changed = 1; | |
1993 | *new_val = strdup(argv[2]); | |
1994 | VLOG_INFO("User-provided %s in use: %s", flag, *new_val); | |
1995 | } else { | |
1996 | VLOG_INFO("No %s provided - defaulting to %s", flag, default_val); | |
1997 | *new_val = default_val; | |
1998 | } | |
1999 | ||
2000 | return changed; | |
2001 | } | |
2002 | ||
8a9562d2 PS |
2003 | int |
2004 | dpdk_init(int argc, char **argv) | |
2005 | { | |
2006 | int result; | |
58397e6c KT |
2007 | int base = 0; |
2008 | char *pragram_name = argv[0]; | |
8a9562d2 | 2009 | |
9441caf3 | 2010 | if (argc < 2 || strcmp(argv[1], "--dpdk")) |
8a9562d2 PS |
2011 | return 0; |
2012 | ||
58397e6c | 2013 | /* Remove the --dpdk argument from arg list.*/ |
8a9562d2 PS |
2014 | argc--; |
2015 | argv++; | |
2016 | ||
7d1ced01 CL |
2017 | #ifdef VHOST_CUSE |
2018 | if (process_vhost_flags("-cuse_dev_name", strdup("vhost-net"), | |
2019 | PATH_MAX, argv, &cuse_dev_name)) { | |
2020 | #else | |
2021 | if (process_vhost_flags("-vhost_sock_dir", strdup(ovs_rundir()), | |
2022 | NAME_MAX, argv, &vhost_sock_dir)) { | |
2023 | struct stat s; | |
2024 | int err; | |
58397e6c | 2025 | |
7d1ced01 CL |
2026 | err = stat(vhost_sock_dir, &s); |
2027 | if (err) { | |
2028 | VLOG_ERR("vHostUser socket DIR '%s' does not exist.", | |
2029 | vhost_sock_dir); | |
2030 | return err; | |
2031 | } | |
2032 | #endif | |
2033 | /* Remove the vhost flag configuration parameters from the argument | |
58397e6c KT |
2034 | * list, so that the correct elements are passed to the DPDK |
2035 | * initialization function | |
2036 | */ | |
2037 | argc -= 2; | |
7d1ced01 | 2038 | argv += 2; /* Increment by two to bypass the vhost flag arguments */ |
58397e6c | 2039 | base = 2; |
58397e6c KT |
2040 | } |
2041 | ||
2042 | /* Keep the program name argument as this is needed for call to | |
2043 | * rte_eal_init() | |
2044 | */ | |
2045 | argv[0] = pragram_name; | |
2046 | ||
8a9562d2 PS |
2047 | /* Make sure things are initialized ... */ |
2048 | result = rte_eal_init(argc, argv); | |
451450fa | 2049 | if (result < 0) { |
58397e6c | 2050 | ovs_abort(result, "Cannot init EAL"); |
451450fa | 2051 | } |
8a9562d2 | 2052 | |
d7310583 | 2053 | rte_memzone_dump(stdout); |
8a9562d2 PS |
2054 | rte_eal_init_ret = 0; |
2055 | ||
451450fa | 2056 | if (argc > result) { |
9441caf3 | 2057 | argv[result] = argv[0]; |
451450fa | 2058 | } |
9441caf3 | 2059 | |
db73f716 | 2060 | /* We are called from the main thread here */ |
d5c199ea | 2061 | RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID; |
db73f716 | 2062 | |
58397e6c | 2063 | return result + 1 + base; |
8a9562d2 PS |
2064 | } |
2065 | ||
bce01e3a | 2066 | static const struct netdev_class dpdk_class = |
95fb793a | 2067 | NETDEV_DPDK_CLASS( |
2068 | "dpdk", | |
b8e57534 | 2069 | NULL, |
5496878c | 2070 | netdev_dpdk_construct, |
58397e6c | 2071 | netdev_dpdk_destruct, |
7251515e | 2072 | netdev_dpdk_set_multiq, |
58397e6c KT |
2073 | netdev_dpdk_eth_send, |
2074 | netdev_dpdk_get_carrier, | |
2075 | netdev_dpdk_get_stats, | |
2076 | netdev_dpdk_get_features, | |
2077 | netdev_dpdk_get_status, | |
2078 | netdev_dpdk_rxq_recv); | |
95fb793a | 2079 | |
bce01e3a | 2080 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 2081 | NETDEV_DPDK_CLASS( |
2082 | "dpdkr", | |
033e9df2 | 2083 | NULL, |
5496878c | 2084 | netdev_dpdk_ring_construct, |
58397e6c | 2085 | netdev_dpdk_destruct, |
a0cb2d66 | 2086 | netdev_dpdk_set_multiq, |
58397e6c KT |
2087 | netdev_dpdk_ring_send, |
2088 | netdev_dpdk_get_carrier, | |
2089 | netdev_dpdk_get_stats, | |
2090 | netdev_dpdk_get_features, | |
2091 | netdev_dpdk_get_status, | |
2092 | netdev_dpdk_rxq_recv); | |
2093 | ||
7d1ced01 | 2094 | static const struct netdev_class dpdk_vhost_cuse_class = |
58397e6c | 2095 | NETDEV_DPDK_CLASS( |
7d1ced01 CL |
2096 | "dpdkvhostcuse", |
2097 | dpdk_vhost_cuse_class_init, | |
2098 | netdev_dpdk_vhost_cuse_construct, | |
2099 | netdev_dpdk_vhost_destruct, | |
2100 | netdev_dpdk_vhost_set_multiq, | |
2101 | netdev_dpdk_vhost_send, | |
2102 | netdev_dpdk_vhost_get_carrier, | |
2103 | netdev_dpdk_vhost_get_stats, | |
2104 | NULL, | |
2105 | NULL, | |
2106 | netdev_dpdk_vhost_rxq_recv); | |
2107 | ||
2108 | const struct netdev_class dpdk_vhost_user_class = | |
2109 | NETDEV_DPDK_CLASS( | |
2110 | "dpdkvhostuser", | |
2111 | dpdk_vhost_user_class_init, | |
2112 | netdev_dpdk_vhost_user_construct, | |
58397e6c KT |
2113 | netdev_dpdk_vhost_destruct, |
2114 | netdev_dpdk_vhost_set_multiq, | |
2115 | netdev_dpdk_vhost_send, | |
2116 | netdev_dpdk_vhost_get_carrier, | |
2117 | netdev_dpdk_vhost_get_stats, | |
2118 | NULL, | |
7251515e | 2119 | NULL, |
58397e6c | 2120 | netdev_dpdk_vhost_rxq_recv); |
95fb793a | 2121 | |
8a9562d2 PS |
2122 | void |
2123 | netdev_dpdk_register(void) | |
2124 | { | |
95fb793a | 2125 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2126 | ||
033e9df2 DDP |
2127 | if (rte_eal_init_ret) { |
2128 | return; | |
2129 | } | |
2130 | ||
95fb793a | 2131 | if (ovsthread_once_start(&once)) { |
033e9df2 | 2132 | dpdk_common_init(); |
95fb793a | 2133 | netdev_register_provider(&dpdk_class); |
2134 | netdev_register_provider(&dpdk_ring_class); | |
7d1ced01 CL |
2135 | #ifdef VHOST_CUSE |
2136 | netdev_register_provider(&dpdk_vhost_cuse_class); | |
2137 | #else | |
2138 | netdev_register_provider(&dpdk_vhost_user_class); | |
2139 | #endif | |
95fb793a | 2140 | ovsthread_once_done(&once); |
2141 | } | |
8a9562d2 | 2142 | } |
8617afff PS |
2143 | |
2144 | int | |
bd5131ba | 2145 | pmd_thread_setaffinity_cpu(unsigned cpu) |
8617afff PS |
2146 | { |
2147 | cpu_set_t cpuset; | |
2148 | int err; | |
2149 | ||
2150 | CPU_ZERO(&cpuset); | |
2151 | CPU_SET(cpu, &cpuset); | |
2152 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset); | |
2153 | if (err) { | |
2154 | VLOG_ERR("Thread affinity error %d",err); | |
2155 | return err; | |
2156 | } | |
abb5943d AW |
2157 | /* NON_PMD_CORE_ID is reserved for use by non pmd threads. */ |
2158 | ovs_assert(cpu != NON_PMD_CORE_ID); | |
65f13b50 | 2159 | RTE_PER_LCORE(_lcore_id) = cpu; |
8617afff PS |
2160 | |
2161 | return 0; | |
2162 | } | |
db73f716 | 2163 | |
db73f716 DDP |
2164 | static bool |
2165 | thread_is_pmd(void) | |
2166 | { | |
abb5943d | 2167 | return rte_lcore_id() != NON_PMD_CORE_ID; |
db73f716 | 2168 | } |