]>
Commit | Line | Data |
---|---|---|
8a9562d2 PS |
1 | /* |
2 | * Copyright (c) 2014 Nicira, Inc. | |
3 | * | |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
19 | #include <stdio.h> | |
20 | #include <string.h> | |
21 | #include <signal.h> | |
22 | #include <stdlib.h> | |
23 | #include <pthread.h> | |
24 | #include <config.h> | |
25 | #include <errno.h> | |
26 | #include <sched.h> | |
27 | #include <stdlib.h> | |
28 | #include <unistd.h> | |
29 | #include <stdio.h> | |
30 | ||
e14deea0 | 31 | #include "dp-packet.h" |
8a9562d2 PS |
32 | #include "dpif-netdev.h" |
33 | #include "list.h" | |
34 | #include "netdev-dpdk.h" | |
35 | #include "netdev-provider.h" | |
36 | #include "netdev-vport.h" | |
37 | #include "odp-util.h" | |
38 | #include "ofp-print.h" | |
94143fc4 | 39 | #include "ovs-numa.h" |
8a9562d2 PS |
40 | #include "ovs-thread.h" |
41 | #include "ovs-rcu.h" | |
42 | #include "packets.h" | |
43 | #include "shash.h" | |
8a9562d2 PS |
44 | #include "sset.h" |
45 | #include "unaligned.h" | |
46 | #include "timeval.h" | |
47 | #include "unixctl.h" | |
e6211adc | 48 | #include "openvswitch/vlog.h" |
8a9562d2 | 49 | |
b8e57534 MK |
50 | #include "rte_config.h" |
51 | #include "rte_mbuf.h" | |
58397e6c | 52 | #include "rte_virtio_net.h" |
b8e57534 | 53 | |
8a9562d2 PS |
54 | VLOG_DEFINE_THIS_MODULE(dpdk); |
55 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
56 | ||
57 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
58 | ||
59 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
60 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
61 | ||
62 | /* | |
63 | * need to reserve tons of extra space in the mbufs so we can align the | |
64 | * DMA addresses to 4KB. | |
65 | */ | |
66 | ||
67 | #define MTU_TO_MAX_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) | |
68 | #define MBUF_SIZE(mtu) (MTU_TO_MAX_LEN(mtu) + (512) + \ | |
69 | sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM) | |
70 | ||
da79ce2b DDP |
71 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
72 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
73 | * enough hugepages) we keep halving the number until the allocation succeeds | |
74 | * or we reach MIN_NB_MBUF */ | |
75 | ||
76 | #define MAX_NB_MBUF (4096 * 64) | |
77 | #define MIN_NB_MBUF (4096 * 4) | |
78 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
79 | ||
80 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
81 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
82 | ||
83 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
84 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
85 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
86 | % MP_CACHE_SZ == 0); | |
87 | ||
8a9562d2 PS |
88 | #define SOCKET0 0 |
89 | ||
79f5354c PM |
90 | #define NIC_PORT_RX_Q_SIZE 2048 /* Size of Physical NIC RX Queue, Max (n+32<=4096)*/ |
91 | #define NIC_PORT_TX_Q_SIZE 2048 /* Size of Physical NIC TX Queue, Max (n+32<=4096)*/ | |
92 | ||
58397e6c | 93 | /* Character device cuse_dev_name. */ |
bce01e3a | 94 | static char *cuse_dev_name = NULL; |
58397e6c | 95 | |
95e9881f KT |
96 | /* |
97 | * Maximum amount of time in micro seconds to try and enqueue to vhost. | |
98 | */ | |
99 | #define VHOST_ENQ_RETRY_USECS 100 | |
100 | ||
8a9562d2 | 101 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
102 | .rxmode = { |
103 | .mq_mode = ETH_MQ_RX_RSS, | |
104 | .split_hdr_size = 0, | |
105 | .header_split = 0, /* Header Split disabled */ | |
106 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
107 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
108 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
109 | .hw_strip_crc = 0, | |
110 | }, | |
111 | .rx_adv_conf = { | |
112 | .rss_conf = { | |
113 | .rss_key = NULL, | |
543342a4 | 114 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 115 | }, |
a28ddd11 DDP |
116 | }, |
117 | .txmode = { | |
118 | .mq_mode = ETH_MQ_TX_NONE, | |
119 | }, | |
8a9562d2 PS |
120 | }; |
121 | ||
3a100265 | 122 | enum { MAX_TX_QUEUE_LEN = 384 }; |
58f7c37b DDP |
123 | enum { DPDK_RING_SIZE = 256 }; |
124 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
125 | enum { DRAIN_TSC = 200000ULL }; |
126 | ||
58397e6c KT |
127 | enum dpdk_dev_type { |
128 | DPDK_DEV_ETH = 0, | |
129 | DPDK_DEV_VHOST = 1 | |
130 | }; | |
131 | ||
8a9562d2 PS |
132 | static int rte_eal_init_ret = ENODEV; |
133 | ||
134 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; | |
135 | ||
136 | /* Contains all 'struct dpdk_dev's. */ | |
ca6ba700 | 137 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 138 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 139 | |
ca6ba700 | 140 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 141 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 142 | |
db73f716 DDP |
143 | /* This mutex must be used by non pmd threads when allocating or freeing |
144 | * mbufs through mempools. Since dpdk_queue_pkts() and dpdk_queue_flush() may | |
145 | * use mempools, a non pmd thread should hold this mutex while calling them */ | |
bce01e3a | 146 | static struct ovs_mutex nonpmd_mempool_mutex = OVS_MUTEX_INITIALIZER; |
db73f716 | 147 | |
8a9562d2 PS |
148 | struct dpdk_mp { |
149 | struct rte_mempool *mp; | |
150 | int mtu; | |
151 | int socket_id; | |
152 | int refcount; | |
ca6ba700 | 153 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
154 | }; |
155 | ||
5a034064 AW |
156 | /* There should be one 'struct dpdk_tx_queue' created for |
157 | * each cpu core. */ | |
8a9562d2 | 158 | struct dpdk_tx_queue { |
94143fc4 AW |
159 | bool flush_tx; /* Set to true to flush queue everytime */ |
160 | /* pkts are queued. */ | |
8a9562d2 PS |
161 | int count; |
162 | uint64_t tsc; | |
163 | struct rte_mbuf *burst_pkts[MAX_TX_QUEUE_LEN]; | |
164 | }; | |
165 | ||
95fb793a | 166 | /* dpdk has no way to remove dpdk ring ethernet devices |
167 | so we have to keep them around once they've been created | |
168 | */ | |
169 | ||
ca6ba700 | 170 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 171 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 172 | |
173 | struct dpdk_ring { | |
174 | /* For the client rings */ | |
175 | struct rte_ring *cring_tx; | |
176 | struct rte_ring *cring_rx; | |
177 | int user_port_id; /* User given port no, parsed from port name */ | |
178 | int eth_port_id; /* ethernet device port id */ | |
ca6ba700 | 179 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 180 | }; |
181 | ||
8a9562d2 PS |
182 | struct netdev_dpdk { |
183 | struct netdev up; | |
184 | int port_id; | |
185 | int max_packet_len; | |
58397e6c | 186 | enum dpdk_dev_type type; |
8a9562d2 | 187 | |
5a034064 | 188 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
189 | |
190 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
191 | ||
192 | struct dpdk_mp *dpdk_mp; | |
193 | int mtu; | |
194 | int socket_id; | |
195 | int buf_size; | |
8a9562d2 PS |
196 | struct netdev_stats stats; |
197 | ||
198 | uint8_t hwaddr[ETH_ADDR_LEN]; | |
199 | enum netdev_flags flags; | |
200 | ||
201 | struct rte_eth_link link; | |
202 | int link_reset_cnt; | |
203 | ||
58397e6c KT |
204 | /* virtio-net structure for vhost device */ |
205 | OVSRCU_TYPE(struct virtio_net *) virtio_dev; | |
206 | ||
8a9562d2 | 207 | /* In dpdk_list. */ |
ca6ba700 | 208 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
58397e6c | 209 | rte_spinlock_t txq_lock; |
8a9562d2 PS |
210 | }; |
211 | ||
212 | struct netdev_rxq_dpdk { | |
213 | struct netdev_rxq up; | |
214 | int port_id; | |
215 | }; | |
216 | ||
db73f716 DDP |
217 | static bool thread_is_pmd(void); |
218 | ||
8a9562d2 PS |
219 | static int netdev_dpdk_construct(struct netdev *); |
220 | ||
58397e6c KT |
221 | struct virtio_net * netdev_dpdk_get_virtio(const struct netdev_dpdk *dev); |
222 | ||
8a9562d2 PS |
223 | static bool |
224 | is_dpdk_class(const struct netdev_class *class) | |
225 | { | |
226 | return class->construct == netdev_dpdk_construct; | |
227 | } | |
228 | ||
58397e6c KT |
229 | /* XXX: use dpdk malloc for entire OVS. in fact huge page should be used |
230 | * for all other segments data, bss and text. */ | |
8a9562d2 PS |
231 | |
232 | static void * | |
233 | dpdk_rte_mzalloc(size_t sz) | |
234 | { | |
235 | void *ptr; | |
236 | ||
237 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
238 | if (ptr == NULL) { | |
239 | out_of_memory(); | |
240 | } | |
241 | return ptr; | |
242 | } | |
243 | ||
db73f716 DDP |
244 | /* XXX this function should be called only by pmd threads (or by non pmd |
245 | * threads holding the nonpmd_mempool_mutex) */ | |
8a9562d2 | 246 | void |
e14deea0 | 247 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 248 | { |
db73f716 | 249 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 250 | |
db73f716 | 251 | rte_pktmbuf_free_seg(pkt); |
8a9562d2 PS |
252 | } |
253 | ||
b3cd9f9d PS |
254 | static void |
255 | __rte_pktmbuf_init(struct rte_mempool *mp, | |
256 | void *opaque_arg OVS_UNUSED, | |
257 | void *_m, | |
258 | unsigned i OVS_UNUSED) | |
259 | { | |
260 | struct rte_mbuf *m = _m; | |
e14deea0 | 261 | uint32_t buf_len = mp->elt_size - sizeof(struct dp_packet); |
b3cd9f9d | 262 | |
e14deea0 | 263 | RTE_MBUF_ASSERT(mp->elt_size >= sizeof(struct dp_packet)); |
b3cd9f9d PS |
264 | |
265 | memset(m, 0, mp->elt_size); | |
266 | ||
267 | /* start of buffer is just after mbuf structure */ | |
e14deea0 | 268 | m->buf_addr = (char *)m + sizeof(struct dp_packet); |
b3cd9f9d | 269 | m->buf_physaddr = rte_mempool_virt2phy(mp, m) + |
e14deea0 | 270 | sizeof(struct dp_packet); |
b3cd9f9d PS |
271 | m->buf_len = (uint16_t)buf_len; |
272 | ||
273 | /* keep some headroom between start of buffer and data */ | |
b8e57534 | 274 | m->data_off = RTE_MIN(RTE_PKTMBUF_HEADROOM, m->buf_len); |
b3cd9f9d PS |
275 | |
276 | /* init some constant fields */ | |
b3cd9f9d | 277 | m->pool = mp; |
b8e57534 MK |
278 | m->nb_segs = 1; |
279 | m->port = 0xff; | |
b3cd9f9d PS |
280 | } |
281 | ||
282 | static void | |
283 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
284 | void *opaque_arg OVS_UNUSED, | |
285 | void *_m, | |
286 | unsigned i OVS_UNUSED) | |
287 | { | |
288 | struct rte_mbuf *m = _m; | |
289 | ||
290 | __rte_pktmbuf_init(mp, opaque_arg, _m, i); | |
291 | ||
cf62fa4c | 292 | dp_packet_init_dpdk((struct dp_packet *) m, m->buf_len); |
b3cd9f9d PS |
293 | } |
294 | ||
8a9562d2 PS |
295 | static struct dpdk_mp * |
296 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
297 | { | |
298 | struct dpdk_mp *dmp = NULL; | |
299 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
da79ce2b | 300 | unsigned mp_size; |
8a9562d2 PS |
301 | |
302 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
303 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
304 | dmp->refcount++; | |
305 | return dmp; | |
306 | } | |
307 | } | |
308 | ||
309 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
310 | dmp->socket_id = socket_id; | |
311 | dmp->mtu = mtu; | |
312 | dmp->refcount = 1; | |
313 | ||
da79ce2b DDP |
314 | mp_size = MAX_NB_MBUF; |
315 | do { | |
316 | if (snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d_%d_%u", | |
317 | dmp->mtu, dmp->socket_id, mp_size) < 0) { | |
318 | return NULL; | |
319 | } | |
95fb793a | 320 | |
da79ce2b DDP |
321 | dmp->mp = rte_mempool_create(mp_name, mp_size, MBUF_SIZE(mtu), |
322 | MP_CACHE_SZ, | |
323 | sizeof(struct rte_pktmbuf_pool_private), | |
324 | rte_pktmbuf_pool_init, NULL, | |
325 | ovs_rte_pktmbuf_init, NULL, | |
326 | socket_id, 0); | |
327 | } while (!dmp->mp && rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
8a9562d2 PS |
328 | |
329 | if (dmp->mp == NULL) { | |
330 | return NULL; | |
da79ce2b DDP |
331 | } else { |
332 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", mp_name, mp_size ); | |
8a9562d2 PS |
333 | } |
334 | ||
335 | list_push_back(&dpdk_mp_list, &dmp->list_node); | |
336 | return dmp; | |
337 | } | |
338 | ||
339 | static void | |
340 | dpdk_mp_put(struct dpdk_mp *dmp) | |
341 | { | |
342 | ||
343 | if (!dmp) { | |
344 | return; | |
345 | } | |
346 | ||
347 | dmp->refcount--; | |
348 | ovs_assert(dmp->refcount >= 0); | |
349 | ||
350 | #if 0 | |
351 | /* I could not find any API to destroy mp. */ | |
352 | if (dmp->refcount == 0) { | |
353 | list_delete(dmp->list_node); | |
354 | /* destroy mp-pool. */ | |
355 | } | |
356 | #endif | |
357 | } | |
358 | ||
359 | static void | |
360 | check_link_status(struct netdev_dpdk *dev) | |
361 | { | |
362 | struct rte_eth_link link; | |
363 | ||
364 | rte_eth_link_get_nowait(dev->port_id, &link); | |
365 | ||
366 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 367 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
368 | |
369 | dev->link_reset_cnt++; | |
370 | dev->link = link; | |
371 | if (dev->link.link_status) { | |
372 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
373 | dev->port_id, (unsigned)dev->link.link_speed, | |
374 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
375 | ("full-duplex") : ("half-duplex")); | |
376 | } else { | |
377 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
378 | } | |
379 | } | |
380 | } | |
381 | ||
382 | static void * | |
383 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
384 | { | |
385 | struct netdev_dpdk *dev; | |
386 | ||
387 | pthread_detach(pthread_self()); | |
388 | ||
389 | for (;;) { | |
390 | ovs_mutex_lock(&dpdk_mutex); | |
391 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
392 | ovs_mutex_lock(&dev->mutex); | |
393 | check_link_status(dev); | |
394 | ovs_mutex_unlock(&dev->mutex); | |
395 | } | |
396 | ovs_mutex_unlock(&dpdk_mutex); | |
397 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
398 | } | |
399 | ||
400 | return NULL; | |
401 | } | |
402 | ||
403 | static int | |
404 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
405 | { | |
406 | struct rte_pktmbuf_pool_private *mbp_priv; | |
407 | struct ether_addr eth_addr; | |
408 | int diag; | |
409 | int i; | |
410 | ||
411 | if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) { | |
95fb793a | 412 | return ENODEV; |
8a9562d2 PS |
413 | } |
414 | ||
5496878c AW |
415 | diag = rte_eth_dev_configure(dev->port_id, dev->up.n_rxq, dev->up.n_txq, |
416 | &port_conf); | |
8a9562d2 PS |
417 | if (diag) { |
418 | VLOG_ERR("eth dev config error %d",diag); | |
95fb793a | 419 | return -diag; |
8a9562d2 PS |
420 | } |
421 | ||
5496878c | 422 | for (i = 0; i < dev->up.n_txq; i++) { |
79f5354c | 423 | diag = rte_eth_tx_queue_setup(dev->port_id, i, NIC_PORT_TX_Q_SIZE, |
9154f798 | 424 | dev->socket_id, NULL); |
8a9562d2 PS |
425 | if (diag) { |
426 | VLOG_ERR("eth dev tx queue setup error %d",diag); | |
95fb793a | 427 | return -diag; |
8a9562d2 PS |
428 | } |
429 | } | |
430 | ||
5496878c | 431 | for (i = 0; i < dev->up.n_rxq; i++) { |
79f5354c | 432 | diag = rte_eth_rx_queue_setup(dev->port_id, i, NIC_PORT_RX_Q_SIZE, |
d221ffa1 | 433 | dev->socket_id, |
9154f798 | 434 | NULL, dev->dpdk_mp->mp); |
8a9562d2 PS |
435 | if (diag) { |
436 | VLOG_ERR("eth dev rx queue setup error %d",diag); | |
95fb793a | 437 | return -diag; |
8a9562d2 PS |
438 | } |
439 | } | |
440 | ||
441 | diag = rte_eth_dev_start(dev->port_id); | |
442 | if (diag) { | |
443 | VLOG_ERR("eth dev start error %d",diag); | |
95fb793a | 444 | return -diag; |
8a9562d2 PS |
445 | } |
446 | ||
447 | rte_eth_promiscuous_enable(dev->port_id); | |
448 | rte_eth_allmulticast_enable(dev->port_id); | |
449 | ||
450 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
451 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
452 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
453 | dev->port_id, ETH_ADDR_ARGS(eth_addr.addr_bytes)); | |
454 | ||
455 | memcpy(dev->hwaddr, eth_addr.addr_bytes, ETH_ADDR_LEN); | |
456 | rte_eth_link_get_nowait(dev->port_id, &dev->link); | |
457 | ||
458 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
459 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
460 | ||
461 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
462 | return 0; | |
463 | } | |
464 | ||
465 | static struct netdev_dpdk * | |
466 | netdev_dpdk_cast(const struct netdev *netdev) | |
467 | { | |
468 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
469 | } | |
470 | ||
471 | static struct netdev * | |
472 | netdev_dpdk_alloc(void) | |
473 | { | |
474 | struct netdev_dpdk *netdev = dpdk_rte_mzalloc(sizeof *netdev); | |
475 | return &netdev->up; | |
476 | } | |
477 | ||
5a034064 | 478 | static void |
91968eb0 | 479 | netdev_dpdk_alloc_txq(struct netdev_dpdk *netdev, unsigned int n_txqs) |
5a034064 | 480 | { |
bd5131ba | 481 | unsigned i; |
5a034064 AW |
482 | |
483 | netdev->tx_q = dpdk_rte_mzalloc(n_txqs * sizeof *netdev->tx_q); | |
94143fc4 AW |
484 | /* Each index is considered as a cpu core id, since there should |
485 | * be one tx queue for each cpu core. */ | |
5a034064 | 486 | for (i = 0; i < n_txqs; i++) { |
ba0358a1 | 487 | int numa_id = ovs_numa_get_numa_id(i); |
94143fc4 | 488 | |
94143fc4 AW |
489 | /* If the corresponding core is not on the same numa node |
490 | * as 'netdev', flags the 'flush_tx'. */ | |
ba0358a1 | 491 | netdev->tx_q[i].flush_tx = netdev->socket_id == numa_id; |
5a034064 AW |
492 | } |
493 | } | |
494 | ||
8a9562d2 | 495 | static int |
58397e6c KT |
496 | netdev_dpdk_init(struct netdev *netdev_, unsigned int port_no, |
497 | enum dpdk_dev_type type) | |
5a034064 | 498 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 PS |
499 | { |
500 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1b7a04e0 | 501 | int sid; |
95fb793a | 502 | int err = 0; |
8a9562d2 | 503 | |
95fb793a | 504 | ovs_mutex_init(&netdev->mutex); |
95fb793a | 505 | ovs_mutex_lock(&netdev->mutex); |
8a9562d2 | 506 | |
1b7a04e0 AW |
507 | /* If the 'sid' is negative, it means that the kernel fails |
508 | * to obtain the pci numa info. In that situation, always | |
509 | * use 'SOCKET0'. */ | |
58397e6c KT |
510 | if (type == DPDK_DEV_ETH) { |
511 | sid = rte_eth_dev_socket_id(port_no); | |
512 | } else { | |
513 | sid = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
514 | } | |
515 | ||
1b7a04e0 | 516 | netdev->socket_id = sid < 0 ? SOCKET0 : sid; |
95fb793a | 517 | netdev->port_id = port_no; |
58397e6c | 518 | netdev->type = type; |
8a9562d2 | 519 | netdev->flags = 0; |
8a9562d2 PS |
520 | netdev->mtu = ETHER_MTU; |
521 | netdev->max_packet_len = MTU_TO_MAX_LEN(netdev->mtu); | |
58397e6c | 522 | rte_spinlock_init(&netdev->txq_lock); |
8a9562d2 | 523 | |
8a9562d2 PS |
524 | netdev->dpdk_mp = dpdk_mp_get(netdev->socket_id, netdev->mtu); |
525 | if (!netdev->dpdk_mp) { | |
526 | err = ENOMEM; | |
95fb793a | 527 | goto unlock; |
8a9562d2 PS |
528 | } |
529 | ||
5496878c AW |
530 | netdev_->n_txq = NR_QUEUE; |
531 | netdev_->n_rxq = NR_QUEUE; | |
58397e6c KT |
532 | |
533 | if (type == DPDK_DEV_ETH) { | |
1b99bb05 MG |
534 | netdev_dpdk_alloc_txq(netdev, NR_QUEUE); |
535 | err = dpdk_eth_dev_init(netdev); | |
536 | if (err) { | |
537 | goto unlock; | |
538 | } | |
8a9562d2 | 539 | } |
8a9562d2 PS |
540 | |
541 | list_push_back(&dpdk_list, &netdev->list_node); | |
542 | ||
95fb793a | 543 | unlock: |
5a034064 AW |
544 | if (err) { |
545 | rte_free(netdev->tx_q); | |
546 | } | |
8a9562d2 | 547 | ovs_mutex_unlock(&netdev->mutex); |
95fb793a | 548 | return err; |
549 | } | |
550 | ||
551 | static int | |
552 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
553 | unsigned int *port_no) | |
554 | { | |
555 | const char *cport; | |
556 | ||
557 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
558 | return ENODEV; | |
559 | } | |
560 | ||
561 | cport = dev_name + strlen(prefix); | |
bce01e3a | 562 | *port_no = strtol(cport, NULL, 0); /* string must be null terminated */ |
95fb793a | 563 | return 0; |
564 | } | |
565 | ||
58397e6c KT |
566 | static int |
567 | netdev_dpdk_vhost_construct(struct netdev *netdev_) | |
568 | { | |
569 | int err; | |
570 | ||
571 | if (rte_eal_init_ret) { | |
572 | return rte_eal_init_ret; | |
573 | } | |
574 | ||
575 | ovs_mutex_lock(&dpdk_mutex); | |
576 | err = netdev_dpdk_init(netdev_, -1, DPDK_DEV_VHOST); | |
577 | ovs_mutex_unlock(&dpdk_mutex); | |
578 | ||
579 | return err; | |
580 | } | |
581 | ||
95fb793a | 582 | static int |
583 | netdev_dpdk_construct(struct netdev *netdev) | |
584 | { | |
585 | unsigned int port_no; | |
586 | int err; | |
587 | ||
588 | if (rte_eal_init_ret) { | |
589 | return rte_eal_init_ret; | |
590 | } | |
591 | ||
592 | /* Names always start with "dpdk" */ | |
593 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
594 | if (err) { | |
595 | return err; | |
596 | } | |
597 | ||
598 | ovs_mutex_lock(&dpdk_mutex); | |
58397e6c | 599 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
8a9562d2 PS |
600 | ovs_mutex_unlock(&dpdk_mutex); |
601 | return err; | |
602 | } | |
603 | ||
604 | static void | |
605 | netdev_dpdk_destruct(struct netdev *netdev_) | |
606 | { | |
607 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
608 | ||
609 | ovs_mutex_lock(&dev->mutex); | |
610 | rte_eth_dev_stop(dev->port_id); | |
611 | ovs_mutex_unlock(&dev->mutex); | |
612 | ||
613 | ovs_mutex_lock(&dpdk_mutex); | |
5a034064 | 614 | rte_free(dev->tx_q); |
8a9562d2 PS |
615 | list_remove(&dev->list_node); |
616 | dpdk_mp_put(dev->dpdk_mp); | |
617 | ovs_mutex_unlock(&dpdk_mutex); | |
58397e6c | 618 | } |
8a9562d2 | 619 | |
58397e6c KT |
620 | static void |
621 | netdev_dpdk_vhost_destruct(struct netdev *netdev_) | |
622 | { | |
623 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
624 | ||
625 | /* Can't remove a port while a guest is attached to it. */ | |
626 | if (netdev_dpdk_get_virtio(dev) != NULL) { | |
627 | VLOG_ERR("Can not remove port, vhost device still attached"); | |
628 | return; | |
629 | } | |
630 | ||
631 | ovs_mutex_lock(&dpdk_mutex); | |
632 | list_remove(&dev->list_node); | |
633 | dpdk_mp_put(dev->dpdk_mp); | |
634 | ovs_mutex_unlock(&dpdk_mutex); | |
8a9562d2 PS |
635 | } |
636 | ||
637 | static void | |
638 | netdev_dpdk_dealloc(struct netdev *netdev_) | |
639 | { | |
640 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
641 | ||
642 | rte_free(netdev); | |
643 | } | |
644 | ||
645 | static int | |
646 | netdev_dpdk_get_config(const struct netdev *netdev_, struct smap *args) | |
647 | { | |
648 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
649 | ||
650 | ovs_mutex_lock(&dev->mutex); | |
651 | ||
47659062 AW |
652 | smap_add_format(args, "configured_rx_queues", "%d", netdev_->n_rxq); |
653 | smap_add_format(args, "configured_tx_queues", "%d", netdev_->n_txq); | |
8a9562d2 PS |
654 | ovs_mutex_unlock(&dev->mutex); |
655 | ||
656 | return 0; | |
657 | } | |
658 | ||
7dec44fe AW |
659 | static int |
660 | netdev_dpdk_get_numa_id(const struct netdev *netdev_) | |
661 | { | |
662 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
663 | ||
664 | return netdev->socket_id; | |
665 | } | |
666 | ||
5496878c AW |
667 | /* Sets the number of tx queues and rx queues for the dpdk interface. |
668 | * If the configuration fails, do not try restoring its old configuration | |
669 | * and just returns the error. */ | |
670 | static int | |
671 | netdev_dpdk_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
672 | unsigned int n_rxq) | |
673 | { | |
674 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
675 | int err = 0; | |
676 | ||
677 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
678 | return err; | |
679 | } | |
680 | ||
b7ccaf67 | 681 | ovs_mutex_lock(&dpdk_mutex); |
5496878c | 682 | ovs_mutex_lock(&netdev->mutex); |
91968eb0 | 683 | |
5496878c | 684 | rte_eth_dev_stop(netdev->port_id); |
91968eb0 | 685 | |
5496878c AW |
686 | netdev->up.n_txq = n_txq; |
687 | netdev->up.n_rxq = n_rxq; | |
58397e6c | 688 | |
91968eb0 AW |
689 | rte_free(netdev->tx_q); |
690 | netdev_dpdk_alloc_txq(netdev, n_txq); | |
5496878c | 691 | err = dpdk_eth_dev_init(netdev); |
91968eb0 | 692 | |
5496878c | 693 | ovs_mutex_unlock(&netdev->mutex); |
b7ccaf67 | 694 | ovs_mutex_unlock(&dpdk_mutex); |
5496878c AW |
695 | |
696 | return err; | |
697 | } | |
698 | ||
58397e6c KT |
699 | static int |
700 | netdev_dpdk_vhost_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
701 | unsigned int n_rxq) | |
702 | { | |
703 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
704 | int err = 0; | |
705 | ||
706 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
707 | return err; | |
708 | } | |
709 | ||
710 | ovs_mutex_lock(&dpdk_mutex); | |
711 | ovs_mutex_lock(&netdev->mutex); | |
712 | ||
713 | netdev->up.n_txq = n_txq; | |
714 | netdev->up.n_rxq = n_rxq; | |
715 | ||
716 | ovs_mutex_unlock(&netdev->mutex); | |
717 | ovs_mutex_unlock(&dpdk_mutex); | |
718 | ||
719 | return err; | |
720 | } | |
721 | ||
8a9562d2 PS |
722 | static struct netdev_rxq * |
723 | netdev_dpdk_rxq_alloc(void) | |
724 | { | |
725 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
726 | ||
727 | return &rx->up; | |
728 | } | |
729 | ||
730 | static struct netdev_rxq_dpdk * | |
731 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rx) | |
732 | { | |
733 | return CONTAINER_OF(rx, struct netdev_rxq_dpdk, up); | |
734 | } | |
735 | ||
736 | static int | |
737 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq_) | |
738 | { | |
739 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
740 | struct netdev_dpdk *netdev = netdev_dpdk_cast(rx->up.netdev); | |
741 | ||
742 | ovs_mutex_lock(&netdev->mutex); | |
743 | rx->port_id = netdev->port_id; | |
744 | ovs_mutex_unlock(&netdev->mutex); | |
745 | ||
746 | return 0; | |
747 | } | |
748 | ||
749 | static void | |
750 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq_ OVS_UNUSED) | |
751 | { | |
752 | } | |
753 | ||
754 | static void | |
755 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq_) | |
756 | { | |
757 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
758 | ||
759 | rte_free(rx); | |
760 | } | |
761 | ||
b170db2a RW |
762 | static inline void |
763 | dpdk_queue_flush__(struct netdev_dpdk *dev, int qid) | |
8a9562d2 PS |
764 | { |
765 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1304f1f8 DDP |
766 | uint32_t nb_tx = 0; |
767 | ||
768 | while (nb_tx != txq->count) { | |
769 | uint32_t ret; | |
770 | ||
771 | ret = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts + nb_tx, | |
772 | txq->count - nb_tx); | |
773 | if (!ret) { | |
774 | break; | |
775 | } | |
776 | ||
777 | nb_tx += ret; | |
778 | } | |
8a9562d2 | 779 | |
b170db2a | 780 | if (OVS_UNLIKELY(nb_tx != txq->count)) { |
db73f716 DDP |
781 | /* free buffers, which we couldn't transmit, one at a time (each |
782 | * packet could come from a different mempool) */ | |
783 | int i; | |
784 | ||
785 | for (i = nb_tx; i < txq->count; i++) { | |
786 | rte_pktmbuf_free_seg(txq->burst_pkts[i]); | |
787 | } | |
1304f1f8 DDP |
788 | ovs_mutex_lock(&dev->mutex); |
789 | dev->stats.tx_dropped += txq->count-nb_tx; | |
790 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 | 791 | } |
1304f1f8 | 792 | |
8a9562d2 | 793 | txq->count = 0; |
844f2d74 | 794 | txq->tsc = rte_get_timer_cycles(); |
b170db2a RW |
795 | } |
796 | ||
797 | static inline void | |
798 | dpdk_queue_flush(struct netdev_dpdk *dev, int qid) | |
799 | { | |
800 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
801 | ||
802 | if (txq->count == 0) { | |
803 | return; | |
804 | } | |
b170db2a | 805 | dpdk_queue_flush__(dev, qid); |
8a9562d2 PS |
806 | } |
807 | ||
58397e6c KT |
808 | static bool |
809 | is_vhost_running(struct virtio_net *dev) | |
810 | { | |
811 | return (dev != NULL && (dev->flags & VIRTIO_DEV_RUNNING)); | |
812 | } | |
813 | ||
814 | /* | |
815 | * The receive path for the vhost port is the TX path out from guest. | |
816 | */ | |
817 | static int | |
818 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq_, | |
819 | struct dp_packet **packets, int *c) | |
820 | { | |
821 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
822 | struct netdev *netdev = rx->up.netdev; | |
823 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
824 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
825 | int qid = 1; | |
826 | uint16_t nb_rx = 0; | |
827 | ||
828 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
829 | return EAGAIN; | |
830 | } | |
831 | ||
832 | nb_rx = rte_vhost_dequeue_burst(virtio_dev, qid, | |
833 | vhost_dev->dpdk_mp->mp, | |
834 | (struct rte_mbuf **)packets, | |
cd159f1a | 835 | NETDEV_MAX_BURST); |
58397e6c KT |
836 | if (!nb_rx) { |
837 | return EAGAIN; | |
838 | } | |
839 | ||
840 | vhost_dev->stats.rx_packets += (uint64_t)nb_rx; | |
841 | *c = (int) nb_rx; | |
842 | return 0; | |
843 | } | |
844 | ||
8a9562d2 | 845 | static int |
e14deea0 | 846 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet **packets, |
91088554 | 847 | int *c) |
8a9562d2 PS |
848 | { |
849 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
850 | struct netdev *netdev = rx->up.netdev; | |
851 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 852 | int nb_rx; |
8a9562d2 | 853 | |
5496878c AW |
854 | /* There is only one tx queue for this core. Do not flush other |
855 | * queueus. */ | |
856 | if (rxq_->queue_id == rte_lcore_id()) { | |
857 | dpdk_queue_flush(dev, rxq_->queue_id); | |
858 | } | |
8a9562d2 PS |
859 | |
860 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq_->queue_id, | |
7d08d53e | 861 | (struct rte_mbuf **) packets, |
cd159f1a | 862 | NETDEV_MAX_BURST); |
8a9562d2 PS |
863 | if (!nb_rx) { |
864 | return EAGAIN; | |
865 | } | |
866 | ||
8a9562d2 PS |
867 | *c = nb_rx; |
868 | ||
869 | return 0; | |
870 | } | |
871 | ||
58397e6c KT |
872 | static void |
873 | __netdev_dpdk_vhost_send(struct netdev *netdev, struct dp_packet **pkts, | |
874 | int cnt, bool may_steal) | |
875 | { | |
876 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
877 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
95e9881f KT |
878 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
879 | unsigned int total_pkts = cnt; | |
880 | uint64_t start = 0; | |
58397e6c KT |
881 | |
882 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
1b99bb05 MG |
883 | ovs_mutex_lock(&vhost_dev->mutex); |
884 | vhost_dev->stats.tx_dropped+= cnt; | |
885 | ovs_mutex_unlock(&vhost_dev->mutex); | |
886 | goto out; | |
58397e6c KT |
887 | } |
888 | ||
889 | /* There is vHost TX single queue, So we need to lock it for TX. */ | |
890 | rte_spinlock_lock(&vhost_dev->txq_lock); | |
58397e6c | 891 | |
95e9881f KT |
892 | do { |
893 | unsigned int tx_pkts; | |
894 | ||
895 | tx_pkts = rte_vhost_enqueue_burst(virtio_dev, VIRTIO_RXQ, | |
896 | cur_pkts, cnt); | |
897 | if (OVS_LIKELY(tx_pkts)) { | |
898 | /* Packets have been sent.*/ | |
899 | cnt -= tx_pkts; | |
900 | /* Prepare for possible next iteration.*/ | |
901 | cur_pkts = &cur_pkts[tx_pkts]; | |
902 | } else { | |
903 | uint64_t timeout = VHOST_ENQ_RETRY_USECS * rte_get_timer_hz() / 1E6; | |
904 | unsigned int expired = 0; | |
905 | ||
906 | if (!start) { | |
907 | start = rte_get_timer_cycles(); | |
908 | } | |
909 | ||
910 | /* | |
911 | * Unable to enqueue packets to vhost interface. | |
912 | * Check available entries before retrying. | |
913 | */ | |
914 | while (!rte_vring_available_entries(virtio_dev, VIRTIO_RXQ)) { | |
915 | if (OVS_UNLIKELY((rte_get_timer_cycles() - start) > timeout)) { | |
916 | expired = 1; | |
917 | break; | |
918 | } | |
919 | } | |
920 | if (expired) { | |
921 | /* break out of main loop. */ | |
922 | break; | |
923 | } | |
924 | } | |
925 | } while (cnt); | |
926 | ||
927 | vhost_dev->stats.tx_packets += (total_pkts - cnt); | |
928 | vhost_dev->stats.tx_dropped += cnt; | |
58397e6c KT |
929 | rte_spinlock_unlock(&vhost_dev->txq_lock); |
930 | ||
931 | out: | |
932 | if (may_steal) { | |
95e9881f KT |
933 | int i; |
934 | ||
935 | for (i = 0; i < total_pkts; i++) { | |
1b99bb05 MG |
936 | dp_packet_delete(pkts[i]); |
937 | } | |
58397e6c KT |
938 | } |
939 | } | |
940 | ||
8a9562d2 | 941 | inline static void |
f4fd623c DDP |
942 | dpdk_queue_pkts(struct netdev_dpdk *dev, int qid, |
943 | struct rte_mbuf **pkts, int cnt) | |
8a9562d2 PS |
944 | { |
945 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
946 | uint64_t diff_tsc; | |
8a9562d2 | 947 | |
f4fd623c DDP |
948 | int i = 0; |
949 | ||
f4fd623c DDP |
950 | while (i < cnt) { |
951 | int freeslots = MAX_TX_QUEUE_LEN - txq->count; | |
952 | int tocopy = MIN(freeslots, cnt-i); | |
8a9562d2 | 953 | |
f4fd623c DDP |
954 | memcpy(&txq->burst_pkts[txq->count], &pkts[i], |
955 | tocopy * sizeof (struct rte_mbuf *)); | |
956 | ||
957 | txq->count += tocopy; | |
958 | i += tocopy; | |
959 | ||
94143fc4 | 960 | if (txq->count == MAX_TX_QUEUE_LEN || txq->flush_tx) { |
b170db2a | 961 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 962 | } |
844f2d74 | 963 | diff_tsc = rte_get_timer_cycles() - txq->tsc; |
f4fd623c | 964 | if (diff_tsc >= DRAIN_TSC) { |
b170db2a | 965 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 966 | } |
8a9562d2 | 967 | } |
8a9562d2 PS |
968 | } |
969 | ||
970 | /* Tx function. Transmit packets indefinitely */ | |
971 | static void | |
58397e6c | 972 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet **pkts, |
2654cc33 | 973 | int cnt) |
db73f716 | 974 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 975 | { |
bce01e3a EJ |
976 | #if !defined(__CHECKER__) && !defined(_WIN32) |
977 | const size_t PKT_ARRAY_SIZE = cnt; | |
978 | #else | |
979 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 980 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 981 | #endif |
8a9562d2 | 982 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
bce01e3a | 983 | struct rte_mbuf *mbufs[PKT_ARRAY_SIZE]; |
175cf4de RW |
984 | int dropped = 0; |
985 | int newcnt = 0; | |
986 | int i; | |
8a9562d2 | 987 | |
db73f716 DDP |
988 | /* If we are on a non pmd thread we have to use the mempool mutex, because |
989 | * every non pmd thread shares the same mempool cache */ | |
990 | ||
991 | if (!thread_is_pmd()) { | |
992 | ovs_mutex_lock(&nonpmd_mempool_mutex); | |
993 | } | |
994 | ||
f4fd623c | 995 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 996 | int size = dp_packet_size(pkts[i]); |
95fb793a | 997 | |
f98d7864 | 998 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c DDP |
999 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1000 | (int)size , dev->max_packet_len); | |
1001 | ||
175cf4de | 1002 | dropped++; |
f4fd623c DDP |
1003 | continue; |
1004 | } | |
8a9562d2 | 1005 | |
f4fd623c | 1006 | mbufs[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1007 | |
f4fd623c | 1008 | if (!mbufs[newcnt]) { |
175cf4de RW |
1009 | dropped += cnt - i; |
1010 | break; | |
f4fd623c DDP |
1011 | } |
1012 | ||
1013 | /* We have to do a copy for now */ | |
b8e57534 | 1014 | memcpy(rte_pktmbuf_mtod(mbufs[newcnt], void *), dp_packet_data(pkts[i]), size); |
f4fd623c DDP |
1015 | |
1016 | rte_pktmbuf_data_len(mbufs[newcnt]) = size; | |
1017 | rte_pktmbuf_pkt_len(mbufs[newcnt]) = size; | |
1018 | ||
1019 | newcnt++; | |
1020 | } | |
8a9562d2 | 1021 | |
f98d7864 | 1022 | if (OVS_UNLIKELY(dropped)) { |
175cf4de RW |
1023 | ovs_mutex_lock(&dev->mutex); |
1024 | dev->stats.tx_dropped += dropped; | |
1025 | ovs_mutex_unlock(&dev->mutex); | |
1026 | } | |
1027 | ||
58397e6c KT |
1028 | if (dev->type == DPDK_DEV_VHOST) { |
1029 | __netdev_dpdk_vhost_send(netdev, (struct dp_packet **) mbufs, newcnt, true); | |
1030 | } else { | |
1031 | dpdk_queue_pkts(dev, qid, mbufs, newcnt); | |
1032 | dpdk_queue_flush(dev, qid); | |
1033 | } | |
db73f716 DDP |
1034 | |
1035 | if (!thread_is_pmd()) { | |
1036 | ovs_mutex_unlock(&nonpmd_mempool_mutex); | |
1037 | } | |
8a9562d2 PS |
1038 | } |
1039 | ||
58397e6c KT |
1040 | static int |
1041 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid OVS_UNUSED, struct dp_packet **pkts, | |
1042 | int cnt, bool may_steal) | |
1043 | { | |
1044 | if (OVS_UNLIKELY(pkts[0]->source != DPBUF_DPDK)) { | |
1045 | int i; | |
1046 | ||
1047 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); | |
1048 | if (may_steal) { | |
1049 | for (i = 0; i < cnt; i++) { | |
1050 | dp_packet_delete(pkts[i]); | |
1051 | } | |
1052 | } | |
1053 | } else { | |
1054 | __netdev_dpdk_vhost_send(netdev, pkts, cnt, may_steal); | |
1055 | } | |
1056 | return 0; | |
1057 | } | |
1058 | ||
7251515e DV |
1059 | static inline void |
1060 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
e14deea0 | 1061 | struct dp_packet **pkts, int cnt, bool may_steal) |
8a9562d2 | 1062 | { |
f4fd623c | 1063 | int i; |
8a9562d2 | 1064 | |
7251515e | 1065 | if (OVS_UNLIKELY(!may_steal || |
cf62fa4c | 1066 | pkts[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1067 | struct netdev *netdev = &dev->up; |
1068 | ||
2654cc33 | 1069 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); |
b3cd9f9d PS |
1070 | |
1071 | if (may_steal) { | |
f4fd623c | 1072 | for (i = 0; i < cnt; i++) { |
e14deea0 | 1073 | dp_packet_delete(pkts[i]); |
f4fd623c | 1074 | } |
b3cd9f9d | 1075 | } |
8a9562d2 | 1076 | } else { |
f4fd623c DDP |
1077 | int next_tx_idx = 0; |
1078 | int dropped = 0; | |
8a9562d2 | 1079 | |
f4fd623c | 1080 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1081 | int size = dp_packet_size(pkts[i]); |
1b99bb05 | 1082 | |
f4fd623c DDP |
1083 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
1084 | if (next_tx_idx != i) { | |
1085 | dpdk_queue_pkts(dev, qid, | |
1086 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1087 | i-next_tx_idx); | |
1ebfe1ac | 1088 | } |
f4fd623c | 1089 | |
1ebfe1ac DDP |
1090 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1091 | (int)size , dev->max_packet_len); | |
f4fd623c | 1092 | |
e14deea0 | 1093 | dp_packet_delete(pkts[i]); |
1ebfe1ac | 1094 | dropped++; |
f4fd623c DDP |
1095 | next_tx_idx = i + 1; |
1096 | } | |
1097 | } | |
1098 | if (next_tx_idx != cnt) { | |
1099 | dpdk_queue_pkts(dev, qid, | |
1100 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1101 | cnt-next_tx_idx); | |
1102 | } | |
8a9562d2 | 1103 | |
f4fd623c DDP |
1104 | if (OVS_UNLIKELY(dropped)) { |
1105 | ovs_mutex_lock(&dev->mutex); | |
1106 | dev->stats.tx_dropped += dropped; | |
1107 | ovs_mutex_unlock(&dev->mutex); | |
1108 | } | |
8a9562d2 | 1109 | } |
7251515e DV |
1110 | } |
1111 | ||
1112 | static int | |
1113 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
e14deea0 | 1114 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1115 | { |
1116 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1117 | |
7251515e DV |
1118 | netdev_dpdk_send__(dev, qid, pkts, cnt, may_steal); |
1119 | return 0; | |
8a9562d2 PS |
1120 | } |
1121 | ||
1122 | static int | |
1123 | netdev_dpdk_set_etheraddr(struct netdev *netdev, | |
1124 | const uint8_t mac[ETH_ADDR_LEN]) | |
1125 | { | |
1126 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1127 | ||
1128 | ovs_mutex_lock(&dev->mutex); | |
1129 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
1130 | memcpy(dev->hwaddr, mac, ETH_ADDR_LEN); | |
045c0d1a | 1131 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1132 | } |
1133 | ovs_mutex_unlock(&dev->mutex); | |
1134 | ||
1135 | return 0; | |
1136 | } | |
1137 | ||
1138 | static int | |
1139 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, | |
1140 | uint8_t mac[ETH_ADDR_LEN]) | |
1141 | { | |
1142 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1143 | ||
1144 | ovs_mutex_lock(&dev->mutex); | |
1145 | memcpy(mac, dev->hwaddr, ETH_ADDR_LEN); | |
1146 | ovs_mutex_unlock(&dev->mutex); | |
1147 | ||
1148 | return 0; | |
1149 | } | |
1150 | ||
1151 | static int | |
1152 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1153 | { | |
1154 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1155 | ||
1156 | ovs_mutex_lock(&dev->mutex); | |
1157 | *mtup = dev->mtu; | |
1158 | ovs_mutex_unlock(&dev->mutex); | |
1159 | ||
1160 | return 0; | |
1161 | } | |
1162 | ||
1163 | static int | |
1164 | netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu) | |
1165 | { | |
1166 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1167 | int old_mtu, err; | |
1168 | struct dpdk_mp *old_mp; | |
1169 | struct dpdk_mp *mp; | |
1170 | ||
1171 | ovs_mutex_lock(&dpdk_mutex); | |
1172 | ovs_mutex_lock(&dev->mutex); | |
1173 | if (dev->mtu == mtu) { | |
1174 | err = 0; | |
1175 | goto out; | |
1176 | } | |
1177 | ||
1178 | mp = dpdk_mp_get(dev->socket_id, dev->mtu); | |
1179 | if (!mp) { | |
1180 | err = ENOMEM; | |
1181 | goto out; | |
1182 | } | |
1183 | ||
1184 | rte_eth_dev_stop(dev->port_id); | |
1185 | ||
1186 | old_mtu = dev->mtu; | |
1187 | old_mp = dev->dpdk_mp; | |
1188 | dev->dpdk_mp = mp; | |
1189 | dev->mtu = mtu; | |
1190 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
1191 | ||
1192 | err = dpdk_eth_dev_init(dev); | |
1193 | if (err) { | |
8a9562d2 PS |
1194 | dpdk_mp_put(mp); |
1195 | dev->mtu = old_mtu; | |
1196 | dev->dpdk_mp = old_mp; | |
1197 | dev->max_packet_len = MTU_TO_MAX_LEN(dev->mtu); | |
1198 | dpdk_eth_dev_init(dev); | |
1199 | goto out; | |
1200 | } | |
1201 | ||
1202 | dpdk_mp_put(old_mp); | |
045c0d1a | 1203 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1204 | out: |
1205 | ovs_mutex_unlock(&dev->mutex); | |
1206 | ovs_mutex_unlock(&dpdk_mutex); | |
1207 | return err; | |
1208 | } | |
1209 | ||
1210 | static int | |
1211 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier); | |
1212 | ||
58397e6c KT |
1213 | static int |
1214 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1215 | struct netdev_stats *stats) | |
1216 | { | |
1217 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1218 | ||
1219 | ovs_mutex_lock(&dev->mutex); | |
1220 | memset(stats, 0, sizeof(*stats)); | |
1221 | /* Unsupported Stats */ | |
1222 | stats->rx_errors = UINT64_MAX; | |
1223 | stats->tx_errors = UINT64_MAX; | |
1224 | stats->multicast = UINT64_MAX; | |
1225 | stats->collisions = UINT64_MAX; | |
1226 | stats->rx_crc_errors = UINT64_MAX; | |
1227 | stats->rx_fifo_errors = UINT64_MAX; | |
1228 | stats->rx_frame_errors = UINT64_MAX; | |
1229 | stats->rx_length_errors = UINT64_MAX; | |
1230 | stats->rx_missed_errors = UINT64_MAX; | |
1231 | stats->rx_over_errors = UINT64_MAX; | |
1232 | stats->tx_aborted_errors = UINT64_MAX; | |
1233 | stats->tx_carrier_errors = UINT64_MAX; | |
1234 | stats->tx_errors = UINT64_MAX; | |
1235 | stats->tx_fifo_errors = UINT64_MAX; | |
1236 | stats->tx_heartbeat_errors = UINT64_MAX; | |
1237 | stats->tx_window_errors = UINT64_MAX; | |
1238 | stats->rx_bytes += UINT64_MAX; | |
1239 | stats->rx_dropped += UINT64_MAX; | |
1240 | stats->tx_bytes += UINT64_MAX; | |
1241 | ||
1242 | /* Supported Stats */ | |
1243 | stats->rx_packets += dev->stats.rx_packets; | |
1244 | stats->tx_packets += dev->stats.tx_packets; | |
1245 | stats->tx_dropped += dev->stats.tx_dropped; | |
1246 | ovs_mutex_unlock(&dev->mutex); | |
1247 | ||
1248 | return 0; | |
1249 | } | |
1250 | ||
8a9562d2 PS |
1251 | static int |
1252 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1253 | { | |
1254 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1255 | struct rte_eth_stats rte_stats; | |
1256 | bool gg; | |
1257 | ||
1258 | netdev_dpdk_get_carrier(netdev, &gg); | |
1259 | ovs_mutex_lock(&dev->mutex); | |
1260 | rte_eth_stats_get(dev->port_id, &rte_stats); | |
1261 | ||
2f9dd77f | 1262 | memset(stats, 0, sizeof(*stats)); |
8a9562d2 | 1263 | |
2f9dd77f PS |
1264 | stats->rx_packets = rte_stats.ipackets; |
1265 | stats->tx_packets = rte_stats.opackets; | |
1266 | stats->rx_bytes = rte_stats.ibytes; | |
1267 | stats->tx_bytes = rte_stats.obytes; | |
1268 | stats->rx_errors = rte_stats.ierrors; | |
1269 | stats->tx_errors = rte_stats.oerrors; | |
1270 | stats->multicast = rte_stats.imcasts; | |
8a9562d2 | 1271 | |
2f9dd77f | 1272 | stats->tx_dropped = dev->stats.tx_dropped; |
8a9562d2 PS |
1273 | ovs_mutex_unlock(&dev->mutex); |
1274 | ||
1275 | return 0; | |
1276 | } | |
1277 | ||
1278 | static int | |
1279 | netdev_dpdk_get_features(const struct netdev *netdev_, | |
1280 | enum netdev_features *current, | |
1281 | enum netdev_features *advertised OVS_UNUSED, | |
1282 | enum netdev_features *supported OVS_UNUSED, | |
1283 | enum netdev_features *peer OVS_UNUSED) | |
1284 | { | |
1285 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1286 | struct rte_eth_link link; | |
1287 | ||
1288 | ovs_mutex_lock(&dev->mutex); | |
1289 | link = dev->link; | |
1290 | ovs_mutex_unlock(&dev->mutex); | |
1291 | ||
1292 | if (link.link_duplex == ETH_LINK_AUTONEG_DUPLEX) { | |
1293 | if (link.link_speed == ETH_LINK_SPEED_AUTONEG) { | |
1294 | *current = NETDEV_F_AUTONEG; | |
1295 | } | |
1296 | } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { | |
1297 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1298 | *current = NETDEV_F_10MB_HD; | |
1299 | } | |
1300 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1301 | *current = NETDEV_F_100MB_HD; | |
1302 | } | |
1303 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1304 | *current = NETDEV_F_1GB_HD; | |
1305 | } | |
1306 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
1307 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1308 | *current = NETDEV_F_10MB_FD; | |
1309 | } | |
1310 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1311 | *current = NETDEV_F_100MB_FD; | |
1312 | } | |
1313 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1314 | *current = NETDEV_F_1GB_FD; | |
1315 | } | |
1316 | if (link.link_speed == ETH_LINK_SPEED_10000) { | |
1317 | *current = NETDEV_F_10GB_FD; | |
1318 | } | |
1319 | } | |
1320 | ||
1321 | return 0; | |
1322 | } | |
1323 | ||
1324 | static int | |
1325 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
1326 | { | |
1327 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1328 | int ifindex; | |
1329 | ||
1330 | ovs_mutex_lock(&dev->mutex); | |
1331 | ifindex = dev->port_id; | |
1332 | ovs_mutex_unlock(&dev->mutex); | |
1333 | ||
1334 | return ifindex; | |
1335 | } | |
1336 | ||
1337 | static int | |
1338 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1339 | { | |
1340 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1341 | ||
1342 | ovs_mutex_lock(&dev->mutex); | |
1343 | check_link_status(dev); | |
1344 | *carrier = dev->link.link_status; | |
58397e6c KT |
1345 | |
1346 | ovs_mutex_unlock(&dev->mutex); | |
1347 | ||
1348 | return 0; | |
1349 | } | |
1350 | ||
1351 | static int | |
1352 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1353 | { | |
1354 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1355 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(dev); | |
1356 | ||
1357 | ovs_mutex_lock(&dev->mutex); | |
1358 | ||
1359 | if (is_vhost_running(virtio_dev)) { | |
1360 | *carrier = 1; | |
1361 | } else { | |
1362 | *carrier = 0; | |
1363 | } | |
1364 | ||
8a9562d2 PS |
1365 | ovs_mutex_unlock(&dev->mutex); |
1366 | ||
1367 | return 0; | |
1368 | } | |
1369 | ||
1370 | static long long int | |
1371 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev_) | |
1372 | { | |
1373 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1374 | long long int carrier_resets; | |
1375 | ||
1376 | ovs_mutex_lock(&dev->mutex); | |
1377 | carrier_resets = dev->link_reset_cnt; | |
1378 | ovs_mutex_unlock(&dev->mutex); | |
1379 | ||
1380 | return carrier_resets; | |
1381 | } | |
1382 | ||
1383 | static int | |
1384 | netdev_dpdk_set_miimon(struct netdev *netdev_ OVS_UNUSED, | |
1385 | long long int interval OVS_UNUSED) | |
1386 | { | |
ee32150e | 1387 | return EOPNOTSUPP; |
8a9562d2 PS |
1388 | } |
1389 | ||
1390 | static int | |
1391 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
1392 | enum netdev_flags off, enum netdev_flags on, | |
95fb793a | 1393 | enum netdev_flags *old_flagsp) OVS_REQUIRES(dev->mutex) |
8a9562d2 PS |
1394 | { |
1395 | int err; | |
1396 | ||
1397 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
1398 | return EINVAL; | |
1399 | } | |
1400 | ||
1401 | *old_flagsp = dev->flags; | |
1402 | dev->flags |= on; | |
1403 | dev->flags &= ~off; | |
1404 | ||
1405 | if (dev->flags == *old_flagsp) { | |
1406 | return 0; | |
1407 | } | |
1408 | ||
58397e6c KT |
1409 | if (dev->type == DPDK_DEV_ETH) { |
1410 | if (dev->flags & NETDEV_UP) { | |
1411 | err = rte_eth_dev_start(dev->port_id); | |
1412 | if (err) | |
1413 | return -err; | |
1414 | } | |
8a9562d2 | 1415 | |
58397e6c KT |
1416 | if (dev->flags & NETDEV_PROMISC) { |
1417 | rte_eth_promiscuous_enable(dev->port_id); | |
1418 | } | |
8a9562d2 | 1419 | |
58397e6c KT |
1420 | if (!(dev->flags & NETDEV_UP)) { |
1421 | rte_eth_dev_stop(dev->port_id); | |
1422 | } | |
8a9562d2 PS |
1423 | } |
1424 | ||
1425 | return 0; | |
1426 | } | |
1427 | ||
1428 | static int | |
1429 | netdev_dpdk_update_flags(struct netdev *netdev_, | |
1430 | enum netdev_flags off, enum netdev_flags on, | |
1431 | enum netdev_flags *old_flagsp) | |
1432 | { | |
1433 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1434 | int error; | |
1435 | ||
1436 | ovs_mutex_lock(&netdev->mutex); | |
1437 | error = netdev_dpdk_update_flags__(netdev, off, on, old_flagsp); | |
1438 | ovs_mutex_unlock(&netdev->mutex); | |
1439 | ||
1440 | return error; | |
1441 | } | |
1442 | ||
1443 | static int | |
1444 | netdev_dpdk_get_status(const struct netdev *netdev_, struct smap *args) | |
1445 | { | |
1446 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1447 | struct rte_eth_dev_info dev_info; | |
1448 | ||
e0a801c7 | 1449 | if (dev->port_id < 0) |
8a9562d2 PS |
1450 | return ENODEV; |
1451 | ||
1452 | ovs_mutex_lock(&dev->mutex); | |
1453 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
1454 | ovs_mutex_unlock(&dev->mutex); | |
1455 | ||
1456 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1457 | ||
95fb793a | 1458 | smap_add_format(args, "port_no", "%d", dev->port_id); |
8a9562d2 PS |
1459 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); |
1460 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1461 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
1462 | smap_add_format(args, "max_rx_pktlen", "%u", dev_info.max_rx_pktlen); | |
1463 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); | |
1464 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
1465 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
1466 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
1467 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
1468 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
1469 | ||
1470 | smap_add_format(args, "pci-vendor_id", "0x%u", dev_info.pci_dev->id.vendor_id); | |
1471 | smap_add_format(args, "pci-device_id", "0x%x", dev_info.pci_dev->id.device_id); | |
1472 | ||
1473 | return 0; | |
1474 | } | |
1475 | ||
1476 | static void | |
1477 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
1478 | OVS_REQUIRES(dev->mutex) | |
1479 | { | |
1480 | enum netdev_flags old_flags; | |
1481 | ||
1482 | if (admin_state) { | |
1483 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
1484 | } else { | |
1485 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
1486 | } | |
1487 | } | |
1488 | ||
1489 | static void | |
1490 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
1491 | const char *argv[], void *aux OVS_UNUSED) | |
1492 | { | |
1493 | bool up; | |
1494 | ||
1495 | if (!strcasecmp(argv[argc - 1], "up")) { | |
1496 | up = true; | |
1497 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
1498 | up = false; | |
1499 | } else { | |
1500 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
1501 | return; | |
1502 | } | |
1503 | ||
1504 | if (argc > 2) { | |
1505 | struct netdev *netdev = netdev_from_name(argv[1]); | |
1506 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
1507 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
1508 | ||
1509 | ovs_mutex_lock(&dpdk_dev->mutex); | |
1510 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
1511 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
1512 | ||
1513 | netdev_close(netdev); | |
1514 | } else { | |
1515 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
1516 | netdev_close(netdev); | |
1517 | return; | |
1518 | } | |
1519 | } else { | |
1520 | struct netdev_dpdk *netdev; | |
1521 | ||
1522 | ovs_mutex_lock(&dpdk_mutex); | |
1523 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
1524 | ovs_mutex_lock(&netdev->mutex); | |
1525 | netdev_dpdk_set_admin_state__(netdev, up); | |
1526 | ovs_mutex_unlock(&netdev->mutex); | |
1527 | } | |
1528 | ovs_mutex_unlock(&dpdk_mutex); | |
1529 | } | |
1530 | unixctl_command_reply(conn, "OK"); | |
1531 | } | |
1532 | ||
58397e6c KT |
1533 | /* |
1534 | * Set virtqueue flags so that we do not receive interrupts. | |
1535 | */ | |
1536 | static void | |
1537 | set_irq_status(struct virtio_net *dev) | |
1538 | { | |
1539 | dev->virtqueue[VIRTIO_RXQ]->used->flags = VRING_USED_F_NO_NOTIFY; | |
1540 | dev->virtqueue[VIRTIO_TXQ]->used->flags = VRING_USED_F_NO_NOTIFY; | |
1541 | } | |
1542 | ||
1543 | /* | |
1544 | * A new virtio-net device is added to a vhost port. | |
1545 | */ | |
1546 | static int | |
1547 | new_device(struct virtio_net *dev) | |
1548 | { | |
1549 | struct netdev_dpdk *netdev; | |
1550 | bool exists = false; | |
1551 | ||
1552 | ovs_mutex_lock(&dpdk_mutex); | |
1553 | /* Add device to the vhost port with the same name as that passed down. */ | |
1554 | LIST_FOR_EACH(netdev, list_node, &dpdk_list) { | |
1555 | if (strncmp(dev->ifname, netdev->up.name, IFNAMSIZ) == 0) { | |
1556 | ovs_mutex_lock(&netdev->mutex); | |
1557 | ovsrcu_set(&netdev->virtio_dev, dev); | |
1558 | ovs_mutex_unlock(&netdev->mutex); | |
1559 | exists = true; | |
1560 | dev->flags |= VIRTIO_DEV_RUNNING; | |
1561 | /* Disable notifications. */ | |
1562 | set_irq_status(dev); | |
1563 | break; | |
1564 | } | |
1565 | } | |
1566 | ovs_mutex_unlock(&dpdk_mutex); | |
1567 | ||
1568 | if (!exists) { | |
1569 | VLOG_INFO("vHost Device '%s' (%ld) can't be added - name not found", | |
1570 | dev->ifname, dev->device_fh); | |
1571 | ||
1572 | return -1; | |
1573 | } | |
1574 | ||
1575 | VLOG_INFO("vHost Device '%s' (%ld) has been added", | |
1576 | dev->ifname, dev->device_fh); | |
1577 | return 0; | |
1578 | } | |
1579 | ||
1580 | /* | |
1581 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
1582 | * flag to stop any more packets from being sent or received to/from a VM and | |
1583 | * ensure all currently queued packets have been sent/received before removing | |
1584 | * the device. | |
1585 | */ | |
1586 | static void | |
1587 | destroy_device(volatile struct virtio_net *dev) | |
1588 | { | |
1589 | struct netdev_dpdk *vhost_dev; | |
1590 | ||
1591 | ovs_mutex_lock(&dpdk_mutex); | |
1592 | LIST_FOR_EACH (vhost_dev, list_node, &dpdk_list) { | |
1593 | if (netdev_dpdk_get_virtio(vhost_dev) == dev) { | |
1594 | ||
1595 | ovs_mutex_lock(&vhost_dev->mutex); | |
1596 | dev->flags &= ~VIRTIO_DEV_RUNNING; | |
1597 | ovsrcu_set(&vhost_dev->virtio_dev, NULL); | |
1598 | ovs_mutex_unlock(&vhost_dev->mutex); | |
1599 | ||
1600 | /* | |
1601 | * Wait for other threads to quiesce before | |
1602 | * setting the virtio_dev to NULL. | |
1603 | */ | |
1604 | ovsrcu_synchronize(); | |
618f44f7 KT |
1605 | /* |
1606 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
1607 | * put thread back into quiescent state before returning. | |
1608 | */ | |
1609 | ovsrcu_quiesce_start(); | |
58397e6c KT |
1610 | } |
1611 | } | |
1612 | ovs_mutex_unlock(&dpdk_mutex); | |
1613 | ||
1614 | VLOG_INFO("vHost Device '%s' (%ld) has been removed", | |
1615 | dev->ifname, dev->device_fh); | |
1616 | } | |
1617 | ||
1618 | struct virtio_net * | |
1619 | netdev_dpdk_get_virtio(const struct netdev_dpdk *dev) | |
1620 | { | |
1621 | return ovsrcu_get(struct virtio_net *, &dev->virtio_dev); | |
1622 | } | |
1623 | ||
1624 | /* | |
1625 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
1626 | * configuration has been fully complete. | |
1627 | */ | |
bce01e3a | 1628 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
1629 | { |
1630 | .new_device = new_device, | |
1631 | .destroy_device = destroy_device, | |
1632 | }; | |
1633 | ||
1634 | static void * | |
1635 | start_cuse_session_loop(void *dummy OVS_UNUSED) | |
1636 | { | |
1637 | pthread_detach(pthread_self()); | |
618f44f7 KT |
1638 | /* Put the cuse thread into quiescent state. */ |
1639 | ovsrcu_quiesce_start(); | |
58397e6c KT |
1640 | rte_vhost_driver_session_start(); |
1641 | return NULL; | |
1642 | } | |
1643 | ||
1644 | static int | |
1645 | dpdk_vhost_class_init(void) | |
1646 | { | |
58397e6c KT |
1647 | int err = -1; |
1648 | ||
1649 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
1650 | ||
1651 | /* Register CUSE device to handle IOCTLs. | |
1652 | * Unless otherwise specified on the vswitchd command line, cuse_dev_name | |
1653 | * is set to vhost-net. | |
1654 | */ | |
1655 | err = rte_vhost_driver_register(cuse_dev_name); | |
1656 | ||
1657 | if (err != 0) { | |
1658 | VLOG_ERR("CUSE device setup failure."); | |
1659 | return -1; | |
1660 | } | |
1661 | ||
618f44f7 KT |
1662 | ovs_thread_create("cuse_thread", start_cuse_session_loop, NULL); |
1663 | return 0; | |
58397e6c KT |
1664 | } |
1665 | ||
033e9df2 DDP |
1666 | static void |
1667 | dpdk_common_init(void) | |
1668 | { | |
1669 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
1670 | "[netdev] up|down", 1, 2, | |
1671 | netdev_dpdk_set_admin_state, NULL); | |
1672 | ||
1673 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
1674 | } | |
1675 | ||
95fb793a | 1676 | /* Client Rings */ |
1677 | ||
95fb793a | 1678 | static int |
1679 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
1680 | unsigned int *eth_port_id) | |
1681 | { | |
1682 | struct dpdk_ring *ivshmem; | |
1683 | char ring_name[10]; | |
1684 | int err; | |
1685 | ||
1686 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
1687 | if (ivshmem == NULL) { | |
1688 | return ENOMEM; | |
1689 | } | |
1690 | ||
7251515e | 1691 | /* XXX: Add support for multiquque ring. */ |
95fb793a | 1692 | err = snprintf(ring_name, 10, "%s_tx", dev_name); |
1693 | if (err < 0) { | |
1694 | return -err; | |
1695 | } | |
1696 | ||
7251515e DV |
1697 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1698 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1699 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1700 | if (ivshmem->cring_tx == NULL) { |
1701 | rte_free(ivshmem); | |
1702 | return ENOMEM; | |
1703 | } | |
1704 | ||
1705 | err = snprintf(ring_name, 10, "%s_rx", dev_name); | |
1706 | if (err < 0) { | |
1707 | return -err; | |
1708 | } | |
1709 | ||
7251515e DV |
1710 | /* Create single consumer/producer rings, netdev does explicit locking. */ |
1711 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, | |
1712 | RING_F_SP_ENQ | RING_F_SC_DEQ); | |
95fb793a | 1713 | if (ivshmem->cring_rx == NULL) { |
1714 | rte_free(ivshmem); | |
1715 | return ENOMEM; | |
1716 | } | |
1717 | ||
d7310583 DDP |
1718 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
1719 | &ivshmem->cring_tx, 1, SOCKET0); | |
1720 | ||
95fb793a | 1721 | if (err < 0) { |
1722 | rte_free(ivshmem); | |
1723 | return ENODEV; | |
1724 | } | |
1725 | ||
1726 | ivshmem->user_port_id = port_no; | |
1727 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
1728 | list_push_back(&dpdk_ring_list, &ivshmem->list_node); | |
1729 | ||
1730 | *eth_port_id = ivshmem->eth_port_id; | |
1731 | return 0; | |
1732 | } | |
1733 | ||
1734 | static int | |
1735 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) OVS_REQUIRES(dpdk_mutex) | |
1736 | { | |
1737 | struct dpdk_ring *ivshmem; | |
1738 | unsigned int port_no; | |
1739 | int err = 0; | |
1740 | ||
1741 | /* Names always start with "dpdkr" */ | |
1742 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
1743 | if (err) { | |
1744 | return err; | |
1745 | } | |
1746 | ||
1747 | /* look through our list to find the device */ | |
1748 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { | |
1749 | if (ivshmem->user_port_id == port_no) { | |
58397e6c | 1750 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
95fb793a | 1751 | *eth_port_id = ivshmem->eth_port_id; /* really all that is needed */ |
1752 | return 0; | |
1753 | } | |
1754 | } | |
1755 | /* Need to create the device rings */ | |
1756 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
1757 | } | |
1758 | ||
7251515e DV |
1759 | static int |
1760 | netdev_dpdk_ring_send(struct netdev *netdev, int qid OVS_UNUSED, | |
e14deea0 | 1761 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1762 | { |
1763 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1b99bb05 MG |
1764 | unsigned i; |
1765 | ||
1766 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that the | |
1767 | * rss hash field is clear. This is because the same mbuf may be modified by | |
1768 | * the consumer of the ring and return into the datapath without recalculating | |
1769 | * the RSS hash. */ | |
1770 | for (i = 0; i < cnt; i++) { | |
1771 | dp_packet_set_rss_hash(pkts[i], 0); | |
1772 | } | |
7251515e DV |
1773 | |
1774 | /* DPDK Rings have a single TX queue, Therefore needs locking. */ | |
58397e6c | 1775 | rte_spinlock_lock(&dev->txq_lock); |
7251515e | 1776 | netdev_dpdk_send__(dev, 0, pkts, cnt, may_steal); |
58397e6c | 1777 | rte_spinlock_unlock(&dev->txq_lock); |
7251515e DV |
1778 | return 0; |
1779 | } | |
1780 | ||
95fb793a | 1781 | static int |
1782 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
1783 | { | |
1784 | unsigned int port_no = 0; | |
1785 | int err = 0; | |
1786 | ||
1787 | if (rte_eal_init_ret) { | |
1788 | return rte_eal_init_ret; | |
1789 | } | |
1790 | ||
1791 | ovs_mutex_lock(&dpdk_mutex); | |
1792 | ||
1793 | err = dpdk_ring_open(netdev->name, &port_no); | |
1794 | if (err) { | |
1795 | goto unlock_dpdk; | |
1796 | } | |
1797 | ||
58397e6c | 1798 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
95fb793a | 1799 | |
1800 | unlock_dpdk: | |
1801 | ovs_mutex_unlock(&dpdk_mutex); | |
1802 | return err; | |
1803 | } | |
1804 | ||
58397e6c KT |
1805 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, MULTIQ, SEND, \ |
1806 | GET_CARRIER, GET_STATS, GET_FEATURES, GET_STATUS, RXQ_RECV) \ | |
95fb793a | 1807 | { \ |
1808 | NAME, \ | |
1809 | INIT, /* init */ \ | |
1810 | NULL, /* netdev_dpdk_run */ \ | |
1811 | NULL, /* netdev_dpdk_wait */ \ | |
1812 | \ | |
1813 | netdev_dpdk_alloc, \ | |
1814 | CONSTRUCT, \ | |
58397e6c | 1815 | DESTRUCT, \ |
95fb793a | 1816 | netdev_dpdk_dealloc, \ |
1817 | netdev_dpdk_get_config, \ | |
1818 | NULL, /* netdev_dpdk_set_config */ \ | |
1819 | NULL, /* get_tunnel_config */ \ | |
58397e6c KT |
1820 | NULL, /* build header */ \ |
1821 | NULL, /* push header */ \ | |
1822 | NULL, /* pop header */ \ | |
7dec44fe | 1823 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
5496878c | 1824 | MULTIQ, /* set_multiq */ \ |
95fb793a | 1825 | \ |
7251515e | 1826 | SEND, /* send */ \ |
95fb793a | 1827 | NULL, /* send_wait */ \ |
1828 | \ | |
1829 | netdev_dpdk_set_etheraddr, \ | |
1830 | netdev_dpdk_get_etheraddr, \ | |
1831 | netdev_dpdk_get_mtu, \ | |
1832 | netdev_dpdk_set_mtu, \ | |
1833 | netdev_dpdk_get_ifindex, \ | |
58397e6c | 1834 | GET_CARRIER, \ |
95fb793a | 1835 | netdev_dpdk_get_carrier_resets, \ |
1836 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
1837 | GET_STATS, \ |
1838 | GET_FEATURES, \ | |
95fb793a | 1839 | NULL, /* set_advertisements */ \ |
1840 | \ | |
1841 | NULL, /* set_policing */ \ | |
1842 | NULL, /* get_qos_types */ \ | |
1843 | NULL, /* get_qos_capabilities */ \ | |
1844 | NULL, /* get_qos */ \ | |
1845 | NULL, /* set_qos */ \ | |
1846 | NULL, /* get_queue */ \ | |
1847 | NULL, /* set_queue */ \ | |
1848 | NULL, /* delete_queue */ \ | |
1849 | NULL, /* get_queue_stats */ \ | |
1850 | NULL, /* queue_dump_start */ \ | |
1851 | NULL, /* queue_dump_next */ \ | |
1852 | NULL, /* queue_dump_done */ \ | |
1853 | NULL, /* dump_queue_stats */ \ | |
1854 | \ | |
1855 | NULL, /* get_in4 */ \ | |
1856 | NULL, /* set_in4 */ \ | |
1857 | NULL, /* get_in6 */ \ | |
1858 | NULL, /* add_router */ \ | |
1859 | NULL, /* get_next_hop */ \ | |
58397e6c | 1860 | GET_STATUS, \ |
95fb793a | 1861 | NULL, /* arp_lookup */ \ |
1862 | \ | |
1863 | netdev_dpdk_update_flags, \ | |
1864 | \ | |
1865 | netdev_dpdk_rxq_alloc, \ | |
1866 | netdev_dpdk_rxq_construct, \ | |
1867 | netdev_dpdk_rxq_destruct, \ | |
1868 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 1869 | RXQ_RECV, \ |
95fb793a | 1870 | NULL, /* rx_wait */ \ |
1871 | NULL, /* rxq_drain */ \ | |
1872 | } | |
8a9562d2 PS |
1873 | |
1874 | int | |
1875 | dpdk_init(int argc, char **argv) | |
1876 | { | |
1877 | int result; | |
58397e6c KT |
1878 | int base = 0; |
1879 | char *pragram_name = argv[0]; | |
8a9562d2 | 1880 | |
9441caf3 | 1881 | if (argc < 2 || strcmp(argv[1], "--dpdk")) |
8a9562d2 PS |
1882 | return 0; |
1883 | ||
58397e6c | 1884 | /* Remove the --dpdk argument from arg list.*/ |
8a9562d2 PS |
1885 | argc--; |
1886 | argv++; | |
1887 | ||
58397e6c KT |
1888 | /* If the cuse_dev_name parameter has been provided, set 'cuse_dev_name' to |
1889 | * this string if it meets the correct criteria. Otherwise, set it to the | |
1890 | * default (vhost-net). | |
1891 | */ | |
1892 | if (!strcmp(argv[1], "--cuse_dev_name") && | |
1893 | (strlen(argv[2]) <= NAME_MAX)) { | |
1894 | ||
1895 | cuse_dev_name = strdup(argv[2]); | |
1896 | ||
1897 | /* Remove the cuse_dev_name configuration parameters from the argument | |
1898 | * list, so that the correct elements are passed to the DPDK | |
1899 | * initialization function | |
1900 | */ | |
1901 | argc -= 2; | |
1902 | argv += 2; /* Increment by two to bypass the cuse_dev_name arguments */ | |
1903 | base = 2; | |
1904 | ||
1905 | VLOG_ERR("User-provided cuse_dev_name in use: /dev/%s", cuse_dev_name); | |
1906 | } else { | |
1907 | cuse_dev_name = "vhost-net"; | |
1908 | VLOG_INFO("No cuse_dev_name provided - defaulting to /dev/vhost-net"); | |
1909 | } | |
1910 | ||
1911 | /* Keep the program name argument as this is needed for call to | |
1912 | * rte_eal_init() | |
1913 | */ | |
1914 | argv[0] = pragram_name; | |
1915 | ||
8a9562d2 PS |
1916 | /* Make sure things are initialized ... */ |
1917 | result = rte_eal_init(argc, argv); | |
451450fa | 1918 | if (result < 0) { |
58397e6c | 1919 | ovs_abort(result, "Cannot init EAL"); |
451450fa | 1920 | } |
8a9562d2 | 1921 | |
d7310583 | 1922 | rte_memzone_dump(stdout); |
8a9562d2 PS |
1923 | rte_eal_init_ret = 0; |
1924 | ||
451450fa | 1925 | if (argc > result) { |
9441caf3 | 1926 | argv[result] = argv[0]; |
451450fa | 1927 | } |
9441caf3 | 1928 | |
db73f716 DDP |
1929 | /* We are called from the main thread here */ |
1930 | thread_set_nonpmd(); | |
1931 | ||
58397e6c | 1932 | return result + 1 + base; |
8a9562d2 PS |
1933 | } |
1934 | ||
bce01e3a | 1935 | static const struct netdev_class dpdk_class = |
95fb793a | 1936 | NETDEV_DPDK_CLASS( |
1937 | "dpdk", | |
b8e57534 | 1938 | NULL, |
5496878c | 1939 | netdev_dpdk_construct, |
58397e6c | 1940 | netdev_dpdk_destruct, |
7251515e | 1941 | netdev_dpdk_set_multiq, |
58397e6c KT |
1942 | netdev_dpdk_eth_send, |
1943 | netdev_dpdk_get_carrier, | |
1944 | netdev_dpdk_get_stats, | |
1945 | netdev_dpdk_get_features, | |
1946 | netdev_dpdk_get_status, | |
1947 | netdev_dpdk_rxq_recv); | |
95fb793a | 1948 | |
bce01e3a | 1949 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 1950 | NETDEV_DPDK_CLASS( |
1951 | "dpdkr", | |
033e9df2 | 1952 | NULL, |
5496878c | 1953 | netdev_dpdk_ring_construct, |
58397e6c KT |
1954 | netdev_dpdk_destruct, |
1955 | NULL, | |
1956 | netdev_dpdk_ring_send, | |
1957 | netdev_dpdk_get_carrier, | |
1958 | netdev_dpdk_get_stats, | |
1959 | netdev_dpdk_get_features, | |
1960 | netdev_dpdk_get_status, | |
1961 | netdev_dpdk_rxq_recv); | |
1962 | ||
bce01e3a | 1963 | static const struct netdev_class dpdk_vhost_class = |
58397e6c KT |
1964 | NETDEV_DPDK_CLASS( |
1965 | "dpdkvhost", | |
1966 | dpdk_vhost_class_init, | |
1967 | netdev_dpdk_vhost_construct, | |
1968 | netdev_dpdk_vhost_destruct, | |
1969 | netdev_dpdk_vhost_set_multiq, | |
1970 | netdev_dpdk_vhost_send, | |
1971 | netdev_dpdk_vhost_get_carrier, | |
1972 | netdev_dpdk_vhost_get_stats, | |
1973 | NULL, | |
7251515e | 1974 | NULL, |
58397e6c | 1975 | netdev_dpdk_vhost_rxq_recv); |
95fb793a | 1976 | |
8a9562d2 PS |
1977 | void |
1978 | netdev_dpdk_register(void) | |
1979 | { | |
95fb793a | 1980 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
1981 | ||
033e9df2 DDP |
1982 | if (rte_eal_init_ret) { |
1983 | return; | |
1984 | } | |
1985 | ||
95fb793a | 1986 | if (ovsthread_once_start(&once)) { |
033e9df2 | 1987 | dpdk_common_init(); |
95fb793a | 1988 | netdev_register_provider(&dpdk_class); |
1989 | netdev_register_provider(&dpdk_ring_class); | |
58397e6c | 1990 | netdev_register_provider(&dpdk_vhost_class); |
95fb793a | 1991 | ovsthread_once_done(&once); |
1992 | } | |
8a9562d2 | 1993 | } |
8617afff PS |
1994 | |
1995 | int | |
bd5131ba | 1996 | pmd_thread_setaffinity_cpu(unsigned cpu) |
8617afff PS |
1997 | { |
1998 | cpu_set_t cpuset; | |
1999 | int err; | |
2000 | ||
2001 | CPU_ZERO(&cpuset); | |
2002 | CPU_SET(cpu, &cpuset); | |
2003 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset); | |
2004 | if (err) { | |
2005 | VLOG_ERR("Thread affinity error %d",err); | |
2006 | return err; | |
2007 | } | |
abb5943d AW |
2008 | /* NON_PMD_CORE_ID is reserved for use by non pmd threads. */ |
2009 | ovs_assert(cpu != NON_PMD_CORE_ID); | |
65f13b50 | 2010 | RTE_PER_LCORE(_lcore_id) = cpu; |
8617afff PS |
2011 | |
2012 | return 0; | |
2013 | } | |
db73f716 DDP |
2014 | |
2015 | void | |
2016 | thread_set_nonpmd(void) | |
2017 | { | |
abb5943d AW |
2018 | /* We have to use NON_PMD_CORE_ID to allow non-pmd threads to perform |
2019 | * certain DPDK operations, like rte_eth_dev_configure(). */ | |
2020 | RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID; | |
db73f716 DDP |
2021 | } |
2022 | ||
2023 | static bool | |
2024 | thread_is_pmd(void) | |
2025 | { | |
abb5943d | 2026 | return rte_lcore_id() != NON_PMD_CORE_ID; |
db73f716 | 2027 | } |