]>
Commit | Line | Data |
---|---|---|
8a9562d2 | 1 | /* |
0d4af148 | 2 | * Copyright (c) 2014, 2015, 2016 Nicira, Inc. |
8a9562d2 PS |
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 | 35 | #include "dpif-netdev.h" |
e5c0f5a4 | 36 | #include "fatal-signal.h" |
8a9562d2 PS |
37 | #include "list.h" |
38 | #include "netdev-dpdk.h" | |
39 | #include "netdev-provider.h" | |
40 | #include "netdev-vport.h" | |
41 | #include "odp-util.h" | |
42 | #include "ofp-print.h" | |
94143fc4 | 43 | #include "ovs-numa.h" |
8a9562d2 PS |
44 | #include "ovs-thread.h" |
45 | #include "ovs-rcu.h" | |
46 | #include "packets.h" | |
47 | #include "shash.h" | |
8a9562d2 PS |
48 | #include "sset.h" |
49 | #include "unaligned.h" | |
50 | #include "timeval.h" | |
51 | #include "unixctl.h" | |
e6211adc | 52 | #include "openvswitch/vlog.h" |
8a9562d2 | 53 | |
b8e57534 MK |
54 | #include "rte_config.h" |
55 | #include "rte_mbuf.h" | |
58397e6c | 56 | #include "rte_virtio_net.h" |
b8e57534 | 57 | |
8a9562d2 PS |
58 | VLOG_DEFINE_THIS_MODULE(dpdk); |
59 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
60 | ||
61 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
62 | ||
63 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
64 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
65 | ||
66 | /* | |
67 | * need to reserve tons of extra space in the mbufs so we can align the | |
68 | * DMA addresses to 4KB. | |
18f777b2 TP |
69 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
70 | * performance for standard Ethernet MTU. | |
8a9562d2 | 71 | */ |
4be4d22c MK |
72 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN + (2 * VLAN_HEADER_LEN)) |
73 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) | |
74 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
75 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len)- ETHER_HDR_LEN - ETHER_CRC_LEN) | |
76 | #define MBUF_SIZE(mtu) ( MTU_TO_MAX_FRAME_LEN(mtu) \ | |
77 | + sizeof(struct dp_packet) \ | |
78 | + RTE_PKTMBUF_HEADROOM) | |
79 | #define NETDEV_DPDK_MBUF_ALIGN 1024 | |
8a9562d2 | 80 | |
da79ce2b DDP |
81 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
82 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
83 | * enough hugepages) we keep halving the number until the allocation succeeds | |
84 | * or we reach MIN_NB_MBUF */ | |
85 | ||
86 | #define MAX_NB_MBUF (4096 * 64) | |
87 | #define MIN_NB_MBUF (4096 * 4) | |
88 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
89 | ||
90 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
91 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
92 | ||
93 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
94 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
95 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
96 | % MP_CACHE_SZ == 0); | |
97 | ||
8a9562d2 PS |
98 | #define SOCKET0 0 |
99 | ||
79f5354c PM |
100 | #define NIC_PORT_RX_Q_SIZE 2048 /* Size of Physical NIC RX Queue, Max (n+32<=4096)*/ |
101 | #define NIC_PORT_TX_Q_SIZE 2048 /* Size of Physical NIC TX Queue, Max (n+32<=4096)*/ | |
102 | ||
585a5bea IM |
103 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
104 | ||
bd8baf47 DDP |
105 | static char *cuse_dev_name = NULL; /* Character device cuse_dev_name. */ |
106 | static char *vhost_sock_dir = NULL; /* Location of vhost-user sockets */ | |
58397e6c | 107 | |
95e9881f KT |
108 | /* |
109 | * Maximum amount of time in micro seconds to try and enqueue to vhost. | |
110 | */ | |
111 | #define VHOST_ENQ_RETRY_USECS 100 | |
112 | ||
8a9562d2 | 113 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
114 | .rxmode = { |
115 | .mq_mode = ETH_MQ_RX_RSS, | |
116 | .split_hdr_size = 0, | |
117 | .header_split = 0, /* Header Split disabled */ | |
118 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
119 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
120 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
121 | .hw_strip_crc = 0, | |
122 | }, | |
123 | .rx_adv_conf = { | |
124 | .rss_conf = { | |
125 | .rss_key = NULL, | |
543342a4 | 126 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 127 | }, |
a28ddd11 DDP |
128 | }, |
129 | .txmode = { | |
130 | .mq_mode = ETH_MQ_TX_NONE, | |
131 | }, | |
8a9562d2 PS |
132 | }; |
133 | ||
3a100265 | 134 | enum { MAX_TX_QUEUE_LEN = 384 }; |
58f7c37b DDP |
135 | enum { DPDK_RING_SIZE = 256 }; |
136 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
137 | enum { DRAIN_TSC = 200000ULL }; |
138 | ||
58397e6c KT |
139 | enum dpdk_dev_type { |
140 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 141 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
142 | }; |
143 | ||
8a9562d2 PS |
144 | static int rte_eal_init_ret = ENODEV; |
145 | ||
146 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; | |
147 | ||
148 | /* Contains all 'struct dpdk_dev's. */ | |
ca6ba700 | 149 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 150 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 151 | |
ca6ba700 | 152 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 153 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 154 | |
db73f716 DDP |
155 | /* This mutex must be used by non pmd threads when allocating or freeing |
156 | * mbufs through mempools. Since dpdk_queue_pkts() and dpdk_queue_flush() may | |
157 | * use mempools, a non pmd thread should hold this mutex while calling them */ | |
bce01e3a | 158 | static struct ovs_mutex nonpmd_mempool_mutex = OVS_MUTEX_INITIALIZER; |
db73f716 | 159 | |
8a9562d2 PS |
160 | struct dpdk_mp { |
161 | struct rte_mempool *mp; | |
162 | int mtu; | |
163 | int socket_id; | |
164 | int refcount; | |
ca6ba700 | 165 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
166 | }; |
167 | ||
5a034064 AW |
168 | /* There should be one 'struct dpdk_tx_queue' created for |
169 | * each cpu core. */ | |
8a9562d2 | 170 | struct dpdk_tx_queue { |
94143fc4 AW |
171 | bool flush_tx; /* Set to true to flush queue everytime */ |
172 | /* pkts are queued. */ | |
8a9562d2 | 173 | int count; |
a0cb2d66 DDP |
174 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
175 | * from concurrent access. It is used only | |
176 | * if the queue is shared among different | |
177 | * pmd threads (see 'txq_needs_locking'). */ | |
585a5bea IM |
178 | int map; /* Mapping of configured vhost-user queues |
179 | * to enabled by guest. */ | |
8a9562d2 PS |
180 | uint64_t tsc; |
181 | struct rte_mbuf *burst_pkts[MAX_TX_QUEUE_LEN]; | |
182 | }; | |
183 | ||
95fb793a | 184 | /* dpdk has no way to remove dpdk ring ethernet devices |
185 | so we have to keep them around once they've been created | |
186 | */ | |
187 | ||
ca6ba700 | 188 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 189 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 190 | |
191 | struct dpdk_ring { | |
192 | /* For the client rings */ | |
193 | struct rte_ring *cring_tx; | |
194 | struct rte_ring *cring_rx; | |
b83a2df1 | 195 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
95fb793a | 196 | int eth_port_id; /* ethernet device port id */ |
ca6ba700 | 197 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 198 | }; |
199 | ||
8a9562d2 PS |
200 | struct netdev_dpdk { |
201 | struct netdev up; | |
202 | int port_id; | |
203 | int max_packet_len; | |
58397e6c | 204 | enum dpdk_dev_type type; |
8a9562d2 | 205 | |
5a034064 | 206 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
207 | |
208 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
209 | ||
210 | struct dpdk_mp *dpdk_mp; | |
211 | int mtu; | |
212 | int socket_id; | |
213 | int buf_size; | |
8a9562d2 | 214 | struct netdev_stats stats; |
45d947c4 DDP |
215 | /* Protects stats */ |
216 | rte_spinlock_t stats_lock; | |
8a9562d2 | 217 | |
74ff3298 | 218 | struct eth_addr hwaddr; |
8a9562d2 PS |
219 | enum netdev_flags flags; |
220 | ||
221 | struct rte_eth_link link; | |
222 | int link_reset_cnt; | |
223 | ||
a0cb2d66 DDP |
224 | /* The user might request more txqs than the NIC has. We remap those |
225 | * ('up.n_txq') on these ('real_n_txq'). | |
226 | * If the numbers match, 'txq_needs_locking' is false, otherwise it is | |
227 | * true and we will take a spinlock on transmission */ | |
228 | int real_n_txq; | |
4573fbd3 | 229 | int real_n_rxq; |
a0cb2d66 DDP |
230 | bool txq_needs_locking; |
231 | ||
58397e6c KT |
232 | /* virtio-net structure for vhost device */ |
233 | OVSRCU_TYPE(struct virtio_net *) virtio_dev; | |
234 | ||
7d1ced01 CL |
235 | /* Identifier used to distinguish vhost devices from each other */ |
236 | char vhost_id[PATH_MAX]; | |
237 | ||
8a9562d2 | 238 | /* In dpdk_list. */ |
ca6ba700 | 239 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
240 | }; |
241 | ||
242 | struct netdev_rxq_dpdk { | |
243 | struct netdev_rxq up; | |
244 | int port_id; | |
245 | }; | |
246 | ||
5f17de68 | 247 | static bool dpdk_thread_is_pmd(void); |
db73f716 | 248 | |
8a9562d2 PS |
249 | static int netdev_dpdk_construct(struct netdev *); |
250 | ||
58397e6c KT |
251 | struct virtio_net * netdev_dpdk_get_virtio(const struct netdev_dpdk *dev); |
252 | ||
8a9562d2 PS |
253 | static bool |
254 | is_dpdk_class(const struct netdev_class *class) | |
255 | { | |
256 | return class->construct == netdev_dpdk_construct; | |
257 | } | |
258 | ||
4be4d22c MK |
259 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
260 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
261 | * value, insufficient buffers are allocated to accomodate the packet in its | |
262 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
263 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
264 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
265 | * behaviour, which reduces performance. To prevent this, use a buffer size that | |
266 | * is closest to 'mtu', but which satisfies the aforementioned criteria. | |
267 | */ | |
268 | static uint32_t | |
269 | dpdk_buf_size(int mtu) | |
270 | { | |
271 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
272 | NETDEV_DPDK_MBUF_ALIGN); | |
273 | } | |
274 | ||
58397e6c KT |
275 | /* XXX: use dpdk malloc for entire OVS. in fact huge page should be used |
276 | * for all other segments data, bss and text. */ | |
8a9562d2 PS |
277 | |
278 | static void * | |
279 | dpdk_rte_mzalloc(size_t sz) | |
280 | { | |
281 | void *ptr; | |
282 | ||
283 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
284 | if (ptr == NULL) { | |
285 | out_of_memory(); | |
286 | } | |
287 | return ptr; | |
288 | } | |
289 | ||
db73f716 DDP |
290 | /* XXX this function should be called only by pmd threads (or by non pmd |
291 | * threads holding the nonpmd_mempool_mutex) */ | |
8a9562d2 | 292 | void |
e14deea0 | 293 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 294 | { |
db73f716 | 295 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 296 | |
db73f716 | 297 | rte_pktmbuf_free_seg(pkt); |
8a9562d2 PS |
298 | } |
299 | ||
b3cd9f9d PS |
300 | static void |
301 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
302 | void *opaque_arg OVS_UNUSED, | |
303 | void *_m, | |
304 | unsigned i OVS_UNUSED) | |
305 | { | |
306 | struct rte_mbuf *m = _m; | |
307 | ||
4be4d22c | 308 | rte_pktmbuf_init(mp, opaque_arg, _m, i); |
b3cd9f9d | 309 | |
cf62fa4c | 310 | dp_packet_init_dpdk((struct dp_packet *) m, m->buf_len); |
b3cd9f9d PS |
311 | } |
312 | ||
8a9562d2 PS |
313 | static struct dpdk_mp * |
314 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
315 | { | |
316 | struct dpdk_mp *dmp = NULL; | |
317 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
da79ce2b | 318 | unsigned mp_size; |
4be4d22c | 319 | struct rte_pktmbuf_pool_private mbp_priv; |
8a9562d2 PS |
320 | |
321 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
322 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
323 | dmp->refcount++; | |
324 | return dmp; | |
325 | } | |
326 | } | |
327 | ||
328 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
329 | dmp->socket_id = socket_id; | |
330 | dmp->mtu = mtu; | |
331 | dmp->refcount = 1; | |
4be4d22c MK |
332 | mbp_priv.mbuf_data_room_size = MBUF_SIZE(mtu) - sizeof(struct dp_packet); |
333 | mbp_priv.mbuf_priv_size = sizeof (struct dp_packet) - sizeof (struct rte_mbuf); | |
8a9562d2 | 334 | |
da79ce2b DDP |
335 | mp_size = MAX_NB_MBUF; |
336 | do { | |
337 | if (snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d_%d_%u", | |
338 | dmp->mtu, dmp->socket_id, mp_size) < 0) { | |
339 | return NULL; | |
340 | } | |
95fb793a | 341 | |
da79ce2b DDP |
342 | dmp->mp = rte_mempool_create(mp_name, mp_size, MBUF_SIZE(mtu), |
343 | MP_CACHE_SZ, | |
344 | sizeof(struct rte_pktmbuf_pool_private), | |
4be4d22c | 345 | rte_pktmbuf_pool_init, &mbp_priv, |
da79ce2b DDP |
346 | ovs_rte_pktmbuf_init, NULL, |
347 | socket_id, 0); | |
348 | } while (!dmp->mp && rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
8a9562d2 PS |
349 | |
350 | if (dmp->mp == NULL) { | |
351 | return NULL; | |
da79ce2b DDP |
352 | } else { |
353 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", mp_name, mp_size ); | |
8a9562d2 PS |
354 | } |
355 | ||
356 | list_push_back(&dpdk_mp_list, &dmp->list_node); | |
357 | return dmp; | |
358 | } | |
359 | ||
360 | static void | |
361 | dpdk_mp_put(struct dpdk_mp *dmp) | |
362 | { | |
363 | ||
364 | if (!dmp) { | |
365 | return; | |
366 | } | |
367 | ||
368 | dmp->refcount--; | |
369 | ovs_assert(dmp->refcount >= 0); | |
370 | ||
371 | #if 0 | |
372 | /* I could not find any API to destroy mp. */ | |
373 | if (dmp->refcount == 0) { | |
374 | list_delete(dmp->list_node); | |
375 | /* destroy mp-pool. */ | |
376 | } | |
377 | #endif | |
378 | } | |
379 | ||
380 | static void | |
381 | check_link_status(struct netdev_dpdk *dev) | |
382 | { | |
383 | struct rte_eth_link link; | |
384 | ||
385 | rte_eth_link_get_nowait(dev->port_id, &link); | |
386 | ||
387 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 388 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
389 | |
390 | dev->link_reset_cnt++; | |
391 | dev->link = link; | |
392 | if (dev->link.link_status) { | |
393 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
394 | dev->port_id, (unsigned)dev->link.link_speed, | |
395 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
396 | ("full-duplex") : ("half-duplex")); | |
397 | } else { | |
398 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
399 | } | |
400 | } | |
401 | } | |
402 | ||
403 | static void * | |
404 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
405 | { | |
406 | struct netdev_dpdk *dev; | |
407 | ||
408 | pthread_detach(pthread_self()); | |
409 | ||
410 | for (;;) { | |
411 | ovs_mutex_lock(&dpdk_mutex); | |
412 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
413 | ovs_mutex_lock(&dev->mutex); | |
414 | check_link_status(dev); | |
415 | ovs_mutex_unlock(&dev->mutex); | |
416 | } | |
417 | ovs_mutex_unlock(&dpdk_mutex); | |
418 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
419 | } | |
420 | ||
421 | return NULL; | |
422 | } | |
423 | ||
b98d7669 DDP |
424 | static int |
425 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
426 | { | |
427 | int diag = 0; | |
428 | int i; | |
429 | ||
430 | /* A device may report more queues than it makes available (this has | |
431 | * been observed for Intel xl710, which reserves some of them for | |
432 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
433 | * available. When this happens we can retry the configuration | |
434 | * and request less queues */ | |
435 | while (n_rxq && n_txq) { | |
436 | if (diag) { | |
437 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
438 | } | |
439 | ||
440 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &port_conf); | |
441 | if (diag) { | |
442 | break; | |
443 | } | |
444 | ||
445 | for (i = 0; i < n_txq; i++) { | |
446 | diag = rte_eth_tx_queue_setup(dev->port_id, i, NIC_PORT_TX_Q_SIZE, | |
447 | dev->socket_id, NULL); | |
448 | if (diag) { | |
449 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
450 | dev->up.name, i, rte_strerror(-diag)); | |
451 | break; | |
452 | } | |
453 | } | |
454 | ||
455 | if (i != n_txq) { | |
456 | /* Retry with less tx queues */ | |
457 | n_txq = i; | |
458 | continue; | |
459 | } | |
460 | ||
461 | for (i = 0; i < n_rxq; i++) { | |
462 | diag = rte_eth_rx_queue_setup(dev->port_id, i, NIC_PORT_RX_Q_SIZE, | |
463 | dev->socket_id, NULL, | |
464 | dev->dpdk_mp->mp); | |
465 | if (diag) { | |
466 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
467 | dev->up.name, i, rte_strerror(-diag)); | |
468 | break; | |
469 | } | |
470 | } | |
471 | ||
472 | if (i != n_rxq) { | |
473 | /* Retry with less rx queues */ | |
474 | n_rxq = i; | |
475 | continue; | |
476 | } | |
477 | ||
478 | dev->up.n_rxq = n_rxq; | |
479 | dev->real_n_txq = n_txq; | |
480 | ||
481 | return 0; | |
482 | } | |
483 | ||
484 | return diag; | |
485 | } | |
486 | ||
487 | ||
8a9562d2 PS |
488 | static int |
489 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
490 | { | |
491 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 492 | struct rte_eth_dev_info info; |
8a9562d2 PS |
493 | struct ether_addr eth_addr; |
494 | int diag; | |
b98d7669 | 495 | int n_rxq, n_txq; |
8a9562d2 PS |
496 | |
497 | if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) { | |
95fb793a | 498 | return ENODEV; |
8a9562d2 PS |
499 | } |
500 | ||
a0cb2d66 | 501 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 502 | |
b98d7669 DDP |
503 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
504 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
505 | ||
506 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 507 | if (diag) { |
b98d7669 DDP |
508 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
509 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 510 | return -diag; |
8a9562d2 PS |
511 | } |
512 | ||
8a9562d2 PS |
513 | diag = rte_eth_dev_start(dev->port_id); |
514 | if (diag) { | |
b98d7669 DDP |
515 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
516 | rte_strerror(-diag)); | |
95fb793a | 517 | return -diag; |
8a9562d2 PS |
518 | } |
519 | ||
520 | rte_eth_promiscuous_enable(dev->port_id); | |
521 | rte_eth_allmulticast_enable(dev->port_id); | |
522 | ||
523 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
524 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
525 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
ca92d173 | 526 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 527 | |
ca92d173 | 528 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
529 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
530 | ||
531 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
532 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
533 | ||
534 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
535 | return 0; | |
536 | } | |
537 | ||
538 | static struct netdev_dpdk * | |
539 | netdev_dpdk_cast(const struct netdev *netdev) | |
540 | { | |
541 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
542 | } | |
543 | ||
544 | static struct netdev * | |
545 | netdev_dpdk_alloc(void) | |
546 | { | |
547 | struct netdev_dpdk *netdev = dpdk_rte_mzalloc(sizeof *netdev); | |
548 | return &netdev->up; | |
549 | } | |
550 | ||
5a034064 | 551 | static void |
91968eb0 | 552 | netdev_dpdk_alloc_txq(struct netdev_dpdk *netdev, unsigned int n_txqs) |
5a034064 | 553 | { |
bd5131ba | 554 | unsigned i; |
5a034064 AW |
555 | |
556 | netdev->tx_q = dpdk_rte_mzalloc(n_txqs * sizeof *netdev->tx_q); | |
557 | for (i = 0; i < n_txqs; i++) { | |
ba0358a1 | 558 | int numa_id = ovs_numa_get_numa_id(i); |
94143fc4 | 559 | |
a0cb2d66 DDP |
560 | if (!netdev->txq_needs_locking) { |
561 | /* Each index is considered as a cpu core id, since there should | |
562 | * be one tx queue for each cpu core. If the corresponding core | |
563 | * is not on the same numa node as 'netdev', flags the | |
564 | * 'flush_tx'. */ | |
565 | netdev->tx_q[i].flush_tx = netdev->socket_id == numa_id; | |
566 | } else { | |
567 | /* Queues are shared among CPUs. Always flush */ | |
568 | netdev->tx_q[i].flush_tx = true; | |
569 | } | |
585a5bea IM |
570 | |
571 | /* Initialize map for vhost devices. */ | |
572 | netdev->tx_q[i].map = -1; | |
a0cb2d66 | 573 | rte_spinlock_init(&netdev->tx_q[i].tx_lock); |
5a034064 AW |
574 | } |
575 | } | |
576 | ||
8a9562d2 | 577 | static int |
58397e6c KT |
578 | netdev_dpdk_init(struct netdev *netdev_, unsigned int port_no, |
579 | enum dpdk_dev_type type) | |
5a034064 | 580 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 PS |
581 | { |
582 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1b7a04e0 | 583 | int sid; |
95fb793a | 584 | int err = 0; |
4be4d22c | 585 | uint32_t buf_size; |
8a9562d2 | 586 | |
95fb793a | 587 | ovs_mutex_init(&netdev->mutex); |
95fb793a | 588 | ovs_mutex_lock(&netdev->mutex); |
8a9562d2 | 589 | |
45d947c4 DDP |
590 | rte_spinlock_init(&netdev->stats_lock); |
591 | ||
1b7a04e0 AW |
592 | /* If the 'sid' is negative, it means that the kernel fails |
593 | * to obtain the pci numa info. In that situation, always | |
594 | * use 'SOCKET0'. */ | |
58397e6c KT |
595 | if (type == DPDK_DEV_ETH) { |
596 | sid = rte_eth_dev_socket_id(port_no); | |
597 | } else { | |
598 | sid = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
599 | } | |
600 | ||
1b7a04e0 | 601 | netdev->socket_id = sid < 0 ? SOCKET0 : sid; |
95fb793a | 602 | netdev->port_id = port_no; |
58397e6c | 603 | netdev->type = type; |
8a9562d2 | 604 | netdev->flags = 0; |
8a9562d2 | 605 | netdev->mtu = ETHER_MTU; |
4be4d22c | 606 | netdev->max_packet_len = MTU_TO_FRAME_LEN(netdev->mtu); |
8a9562d2 | 607 | |
4be4d22c MK |
608 | buf_size = dpdk_buf_size(netdev->mtu); |
609 | netdev->dpdk_mp = dpdk_mp_get(netdev->socket_id, FRAME_LEN_TO_MTU(buf_size)); | |
8a9562d2 PS |
610 | if (!netdev->dpdk_mp) { |
611 | err = ENOMEM; | |
95fb793a | 612 | goto unlock; |
8a9562d2 PS |
613 | } |
614 | ||
5496878c AW |
615 | netdev_->n_txq = NR_QUEUE; |
616 | netdev_->n_rxq = NR_QUEUE; | |
a14b8947 | 617 | netdev_->requested_n_rxq = NR_QUEUE; |
a0cb2d66 | 618 | netdev->real_n_txq = NR_QUEUE; |
58397e6c KT |
619 | |
620 | if (type == DPDK_DEV_ETH) { | |
1b99bb05 MG |
621 | netdev_dpdk_alloc_txq(netdev, NR_QUEUE); |
622 | err = dpdk_eth_dev_init(netdev); | |
623 | if (err) { | |
624 | goto unlock; | |
625 | } | |
585a5bea IM |
626 | } else { |
627 | netdev_dpdk_alloc_txq(netdev, OVS_VHOST_MAX_QUEUE_NUM); | |
8a9562d2 | 628 | } |
8a9562d2 PS |
629 | |
630 | list_push_back(&dpdk_list, &netdev->list_node); | |
631 | ||
95fb793a | 632 | unlock: |
5a034064 AW |
633 | if (err) { |
634 | rte_free(netdev->tx_q); | |
635 | } | |
8a9562d2 | 636 | ovs_mutex_unlock(&netdev->mutex); |
95fb793a | 637 | return err; |
638 | } | |
639 | ||
b83a2df1 MV |
640 | /* dev_name must be the prefix followed by a positive decimal number. |
641 | * (no leading + or - signs are allowed) */ | |
95fb793a | 642 | static int |
643 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
644 | unsigned int *port_no) | |
645 | { | |
646 | const char *cport; | |
647 | ||
648 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
649 | return ENODEV; | |
650 | } | |
651 | ||
652 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
653 | |
654 | if (str_to_uint(cport, 10, port_no)) { | |
655 | return 0; | |
656 | } else { | |
657 | return ENODEV; | |
658 | } | |
95fb793a | 659 | } |
660 | ||
58397e6c | 661 | static int |
bd8baf47 | 662 | vhost_construct_helper(struct netdev *netdev_) OVS_REQUIRES(dpdk_mutex) |
58397e6c | 663 | { |
58397e6c KT |
664 | if (rte_eal_init_ret) { |
665 | return rte_eal_init_ret; | |
666 | } | |
667 | ||
7d1ced01 CL |
668 | return netdev_dpdk_init(netdev_, -1, DPDK_DEV_VHOST); |
669 | } | |
670 | ||
671 | static int | |
672 | netdev_dpdk_vhost_cuse_construct(struct netdev *netdev_) | |
673 | { | |
674 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
675 | int err; | |
676 | ||
58397e6c | 677 | ovs_mutex_lock(&dpdk_mutex); |
7d1ced01 CL |
678 | strncpy(netdev->vhost_id, netdev->up.name, sizeof(netdev->vhost_id)); |
679 | err = vhost_construct_helper(netdev_); | |
58397e6c | 680 | ovs_mutex_unlock(&dpdk_mutex); |
7d1ced01 CL |
681 | return err; |
682 | } | |
58397e6c | 683 | |
7d1ced01 CL |
684 | static int |
685 | netdev_dpdk_vhost_user_construct(struct netdev *netdev_) | |
686 | { | |
687 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1af27e8a | 688 | const char *name = netdev_->name; |
7d1ced01 | 689 | int err; |
a0cb2d66 | 690 | |
1af27e8a DDP |
691 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
692 | * the file system. '/' or '\' would traverse directories, so they're not | |
693 | * acceptable in 'name'. */ | |
694 | if (strchr(name, '/') || strchr(name, '\\')) { | |
695 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
696 | "A valid name must not include '/' or '\\'", | |
697 | name); | |
698 | return EINVAL; | |
699 | } | |
700 | ||
7d1ced01 CL |
701 | ovs_mutex_lock(&dpdk_mutex); |
702 | /* Take the name of the vhost-user port and append it to the location where | |
703 | * the socket is to be created, then register the socket. | |
704 | */ | |
705 | snprintf(netdev->vhost_id, sizeof(netdev->vhost_id), "%s/%s", | |
1af27e8a DDP |
706 | vhost_sock_dir, name); |
707 | ||
7d1ced01 CL |
708 | err = rte_vhost_driver_register(netdev->vhost_id); |
709 | if (err) { | |
710 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
711 | netdev->vhost_id); | |
e5c0f5a4 IM |
712 | } else { |
713 | fatal_signal_add_file_to_unlink(netdev->vhost_id); | |
1a64eb93 | 714 | VLOG_INFO("Socket %s created for vhost-user port %s\n", |
1af27e8a | 715 | netdev->vhost_id, name); |
1a64eb93 | 716 | err = vhost_construct_helper(netdev_); |
7d1ced01 | 717 | } |
e5c0f5a4 | 718 | |
7d1ced01 | 719 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
720 | return err; |
721 | } | |
722 | ||
95fb793a | 723 | static int |
724 | netdev_dpdk_construct(struct netdev *netdev) | |
725 | { | |
726 | unsigned int port_no; | |
727 | int err; | |
728 | ||
729 | if (rte_eal_init_ret) { | |
730 | return rte_eal_init_ret; | |
731 | } | |
732 | ||
733 | /* Names always start with "dpdk" */ | |
734 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
735 | if (err) { | |
736 | return err; | |
737 | } | |
738 | ||
739 | ovs_mutex_lock(&dpdk_mutex); | |
58397e6c | 740 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
8a9562d2 PS |
741 | ovs_mutex_unlock(&dpdk_mutex); |
742 | return err; | |
743 | } | |
744 | ||
745 | static void | |
746 | netdev_dpdk_destruct(struct netdev *netdev_) | |
747 | { | |
748 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
749 | ||
750 | ovs_mutex_lock(&dev->mutex); | |
751 | rte_eth_dev_stop(dev->port_id); | |
752 | ovs_mutex_unlock(&dev->mutex); | |
753 | ||
754 | ovs_mutex_lock(&dpdk_mutex); | |
5a034064 | 755 | rte_free(dev->tx_q); |
8a9562d2 PS |
756 | list_remove(&dev->list_node); |
757 | dpdk_mp_put(dev->dpdk_mp); | |
758 | ovs_mutex_unlock(&dpdk_mutex); | |
58397e6c | 759 | } |
8a9562d2 | 760 | |
58397e6c KT |
761 | static void |
762 | netdev_dpdk_vhost_destruct(struct netdev *netdev_) | |
763 | { | |
764 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
765 | ||
766 | /* Can't remove a port while a guest is attached to it. */ | |
767 | if (netdev_dpdk_get_virtio(dev) != NULL) { | |
768 | VLOG_ERR("Can not remove port, vhost device still attached"); | |
769 | return; | |
770 | } | |
771 | ||
e04f7e4f CL |
772 | if (rte_vhost_driver_unregister(dev->vhost_id)) { |
773 | VLOG_ERR("Unable to remove vhost-user socket %s", dev->vhost_id); | |
e5c0f5a4 IM |
774 | } else { |
775 | fatal_signal_remove_file_to_unlink(dev->vhost_id); | |
e04f7e4f CL |
776 | } |
777 | ||
58397e6c KT |
778 | ovs_mutex_lock(&dpdk_mutex); |
779 | list_remove(&dev->list_node); | |
780 | dpdk_mp_put(dev->dpdk_mp); | |
781 | ovs_mutex_unlock(&dpdk_mutex); | |
8a9562d2 PS |
782 | } |
783 | ||
784 | static void | |
785 | netdev_dpdk_dealloc(struct netdev *netdev_) | |
786 | { | |
787 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
788 | ||
789 | rte_free(netdev); | |
790 | } | |
791 | ||
792 | static int | |
a14b8947 | 793 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 794 | { |
a14b8947 | 795 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
796 | |
797 | ovs_mutex_lock(&dev->mutex); | |
798 | ||
a14b8947 IM |
799 | smap_add_format(args, "requested_rx_queues", "%d", netdev->requested_n_rxq); |
800 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); | |
801 | smap_add_format(args, "requested_tx_queues", "%d", netdev->n_txq); | |
a0cb2d66 | 802 | smap_add_format(args, "configured_tx_queues", "%d", dev->real_n_txq); |
8a9562d2 PS |
803 | ovs_mutex_unlock(&dev->mutex); |
804 | ||
805 | return 0; | |
806 | } | |
807 | ||
a14b8947 IM |
808 | static int |
809 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args) | |
810 | { | |
811 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
812 | ||
813 | ovs_mutex_lock(&dev->mutex); | |
814 | netdev->requested_n_rxq = MAX(smap_get_int(args, "n_rxq", | |
815 | netdev->requested_n_rxq), 1); | |
816 | netdev_change_seq_changed(netdev); | |
817 | ovs_mutex_unlock(&dev->mutex); | |
818 | ||
819 | return 0; | |
820 | } | |
821 | ||
7dec44fe AW |
822 | static int |
823 | netdev_dpdk_get_numa_id(const struct netdev *netdev_) | |
824 | { | |
825 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
826 | ||
827 | return netdev->socket_id; | |
828 | } | |
829 | ||
5496878c AW |
830 | /* Sets the number of tx queues and rx queues for the dpdk interface. |
831 | * If the configuration fails, do not try restoring its old configuration | |
832 | * and just returns the error. */ | |
833 | static int | |
834 | netdev_dpdk_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
835 | unsigned int n_rxq) | |
836 | { | |
837 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
838 | int err = 0; | |
4b8df037 | 839 | int old_rxq, old_txq; |
5496878c AW |
840 | |
841 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
842 | return err; | |
843 | } | |
844 | ||
b7ccaf67 | 845 | ovs_mutex_lock(&dpdk_mutex); |
5496878c | 846 | ovs_mutex_lock(&netdev->mutex); |
91968eb0 | 847 | |
5496878c | 848 | rte_eth_dev_stop(netdev->port_id); |
91968eb0 | 849 | |
4b8df037 DDP |
850 | old_txq = netdev->up.n_txq; |
851 | old_rxq = netdev->up.n_rxq; | |
5496878c AW |
852 | netdev->up.n_txq = n_txq; |
853 | netdev->up.n_rxq = n_rxq; | |
58397e6c | 854 | |
91968eb0 | 855 | rte_free(netdev->tx_q); |
5496878c | 856 | err = dpdk_eth_dev_init(netdev); |
a0cb2d66 | 857 | netdev_dpdk_alloc_txq(netdev, netdev->real_n_txq); |
4b8df037 DDP |
858 | if (err) { |
859 | /* If there has been an error, it means that the requested queues | |
860 | * have not been created. Restore the old numbers. */ | |
861 | netdev->up.n_txq = old_txq; | |
862 | netdev->up.n_rxq = old_rxq; | |
863 | } | |
a0cb2d66 DDP |
864 | |
865 | netdev->txq_needs_locking = netdev->real_n_txq != netdev->up.n_txq; | |
91968eb0 | 866 | |
5496878c | 867 | ovs_mutex_unlock(&netdev->mutex); |
b7ccaf67 | 868 | ovs_mutex_unlock(&dpdk_mutex); |
5496878c AW |
869 | |
870 | return err; | |
871 | } | |
872 | ||
58397e6c | 873 | static int |
4573fbd3 | 874 | netdev_dpdk_vhost_cuse_set_multiq(struct netdev *netdev_, unsigned int n_txq, |
a0cb2d66 | 875 | unsigned int n_rxq) |
58397e6c KT |
876 | { |
877 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
878 | int err = 0; | |
879 | ||
880 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
881 | return err; | |
882 | } | |
883 | ||
884 | ovs_mutex_lock(&dpdk_mutex); | |
885 | ovs_mutex_lock(&netdev->mutex); | |
886 | ||
887 | netdev->up.n_txq = n_txq; | |
a0cb2d66 DDP |
888 | netdev->real_n_txq = 1; |
889 | netdev->up.n_rxq = 1; | |
4573fbd3 FL |
890 | netdev->txq_needs_locking = netdev->real_n_txq != netdev->up.n_txq; |
891 | ||
892 | ovs_mutex_unlock(&netdev->mutex); | |
893 | ovs_mutex_unlock(&dpdk_mutex); | |
894 | ||
895 | return err; | |
896 | } | |
897 | ||
898 | static int | |
899 | netdev_dpdk_vhost_set_multiq(struct netdev *netdev_, unsigned int n_txq, | |
900 | unsigned int n_rxq) | |
901 | { | |
902 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
903 | int err = 0; | |
904 | ||
905 | if (netdev->up.n_txq == n_txq && netdev->up.n_rxq == n_rxq) { | |
906 | return err; | |
907 | } | |
908 | ||
909 | ovs_mutex_lock(&dpdk_mutex); | |
910 | ovs_mutex_lock(&netdev->mutex); | |
911 | ||
4573fbd3 FL |
912 | netdev->up.n_txq = n_txq; |
913 | netdev->up.n_rxq = n_rxq; | |
58397e6c KT |
914 | |
915 | ovs_mutex_unlock(&netdev->mutex); | |
916 | ovs_mutex_unlock(&dpdk_mutex); | |
917 | ||
918 | return err; | |
919 | } | |
920 | ||
8a9562d2 PS |
921 | static struct netdev_rxq * |
922 | netdev_dpdk_rxq_alloc(void) | |
923 | { | |
924 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
925 | ||
926 | return &rx->up; | |
927 | } | |
928 | ||
929 | static struct netdev_rxq_dpdk * | |
930 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rx) | |
931 | { | |
932 | return CONTAINER_OF(rx, struct netdev_rxq_dpdk, up); | |
933 | } | |
934 | ||
935 | static int | |
936 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq_) | |
937 | { | |
938 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
939 | struct netdev_dpdk *netdev = netdev_dpdk_cast(rx->up.netdev); | |
940 | ||
941 | ovs_mutex_lock(&netdev->mutex); | |
942 | rx->port_id = netdev->port_id; | |
943 | ovs_mutex_unlock(&netdev->mutex); | |
944 | ||
945 | return 0; | |
946 | } | |
947 | ||
948 | static void | |
949 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq_ OVS_UNUSED) | |
950 | { | |
951 | } | |
952 | ||
953 | static void | |
954 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq_) | |
955 | { | |
956 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
957 | ||
958 | rte_free(rx); | |
959 | } | |
960 | ||
b170db2a RW |
961 | static inline void |
962 | dpdk_queue_flush__(struct netdev_dpdk *dev, int qid) | |
8a9562d2 PS |
963 | { |
964 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1304f1f8 DDP |
965 | uint32_t nb_tx = 0; |
966 | ||
967 | while (nb_tx != txq->count) { | |
968 | uint32_t ret; | |
969 | ||
970 | ret = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts + nb_tx, | |
971 | txq->count - nb_tx); | |
972 | if (!ret) { | |
973 | break; | |
974 | } | |
975 | ||
976 | nb_tx += ret; | |
977 | } | |
8a9562d2 | 978 | |
b170db2a | 979 | if (OVS_UNLIKELY(nb_tx != txq->count)) { |
db73f716 DDP |
980 | /* free buffers, which we couldn't transmit, one at a time (each |
981 | * packet could come from a different mempool) */ | |
982 | int i; | |
983 | ||
984 | for (i = nb_tx; i < txq->count; i++) { | |
985 | rte_pktmbuf_free_seg(txq->burst_pkts[i]); | |
986 | } | |
45d947c4 | 987 | rte_spinlock_lock(&dev->stats_lock); |
1304f1f8 | 988 | dev->stats.tx_dropped += txq->count-nb_tx; |
45d947c4 | 989 | rte_spinlock_unlock(&dev->stats_lock); |
8a9562d2 | 990 | } |
1304f1f8 | 991 | |
8a9562d2 | 992 | txq->count = 0; |
844f2d74 | 993 | txq->tsc = rte_get_timer_cycles(); |
b170db2a RW |
994 | } |
995 | ||
996 | static inline void | |
997 | dpdk_queue_flush(struct netdev_dpdk *dev, int qid) | |
998 | { | |
999 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1000 | ||
1001 | if (txq->count == 0) { | |
1002 | return; | |
1003 | } | |
b170db2a | 1004 | dpdk_queue_flush__(dev, qid); |
8a9562d2 PS |
1005 | } |
1006 | ||
58397e6c KT |
1007 | static bool |
1008 | is_vhost_running(struct virtio_net *dev) | |
1009 | { | |
1010 | return (dev != NULL && (dev->flags & VIRTIO_DEV_RUNNING)); | |
1011 | } | |
1012 | ||
9e3ddd45 TP |
1013 | static inline void |
1014 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
1015 | struct dp_packet **packets, int count) | |
1016 | { | |
1017 | int i; | |
1018 | struct dp_packet *packet; | |
1019 | ||
1020 | stats->rx_packets += count; | |
1021 | for (i = 0; i < count; i++) { | |
1022 | packet = packets[i]; | |
1023 | ||
1024 | if (OVS_UNLIKELY(dp_packet_size(packet) < ETH_HEADER_LEN)) { | |
1025 | /* This only protects the following multicast counting from | |
1026 | * too short packets, but it does not stop the packet from | |
1027 | * further processing. */ | |
1028 | stats->rx_errors++; | |
1029 | stats->rx_length_errors++; | |
1030 | continue; | |
1031 | } | |
1032 | ||
1033 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); | |
1034 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1035 | stats->multicast++; | |
1036 | } | |
1037 | ||
1038 | stats->rx_bytes += dp_packet_size(packet); | |
1039 | } | |
1040 | } | |
1041 | ||
58397e6c KT |
1042 | /* |
1043 | * The receive path for the vhost port is the TX path out from guest. | |
1044 | */ | |
1045 | static int | |
1046 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq_, | |
1047 | struct dp_packet **packets, int *c) | |
1048 | { | |
1049 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
1050 | struct netdev *netdev = rx->up.netdev; | |
1051 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
1052 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
4573fbd3 | 1053 | int qid = rxq_->queue_id; |
58397e6c KT |
1054 | uint16_t nb_rx = 0; |
1055 | ||
1056 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
1057 | return EAGAIN; | |
1058 | } | |
1059 | ||
4573fbd3 FL |
1060 | if (rxq_->queue_id >= vhost_dev->real_n_rxq) { |
1061 | return EOPNOTSUPP; | |
1062 | } | |
1063 | ||
1064 | nb_rx = rte_vhost_dequeue_burst(virtio_dev, qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
58397e6c KT |
1065 | vhost_dev->dpdk_mp->mp, |
1066 | (struct rte_mbuf **)packets, | |
cd159f1a | 1067 | NETDEV_MAX_BURST); |
58397e6c KT |
1068 | if (!nb_rx) { |
1069 | return EAGAIN; | |
1070 | } | |
1071 | ||
45d947c4 | 1072 | rte_spinlock_lock(&vhost_dev->stats_lock); |
9e3ddd45 | 1073 | netdev_dpdk_vhost_update_rx_counters(&vhost_dev->stats, packets, nb_rx); |
45d947c4 DDP |
1074 | rte_spinlock_unlock(&vhost_dev->stats_lock); |
1075 | ||
58397e6c KT |
1076 | *c = (int) nb_rx; |
1077 | return 0; | |
1078 | } | |
1079 | ||
8a9562d2 | 1080 | static int |
e14deea0 | 1081 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq_, struct dp_packet **packets, |
91088554 | 1082 | int *c) |
8a9562d2 PS |
1083 | { |
1084 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_); | |
1085 | struct netdev *netdev = rx->up.netdev; | |
1086 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1087 | int nb_rx; |
8a9562d2 | 1088 | |
5496878c | 1089 | /* There is only one tx queue for this core. Do not flush other |
dc6ba5dc WL |
1090 | * queues. |
1091 | * Do not flush tx queue which is shared among CPUs | |
1092 | * since it is always flushed */ | |
1093 | if (rxq_->queue_id == rte_lcore_id() && | |
1094 | OVS_LIKELY(!dev->txq_needs_locking)) { | |
5496878c AW |
1095 | dpdk_queue_flush(dev, rxq_->queue_id); |
1096 | } | |
8a9562d2 PS |
1097 | |
1098 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq_->queue_id, | |
7d08d53e | 1099 | (struct rte_mbuf **) packets, |
cd159f1a | 1100 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1101 | if (!nb_rx) { |
1102 | return EAGAIN; | |
1103 | } | |
1104 | ||
8a9562d2 PS |
1105 | *c = nb_rx; |
1106 | ||
1107 | return 0; | |
1108 | } | |
1109 | ||
9e3ddd45 TP |
1110 | static inline void |
1111 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1112 | struct dp_packet **packets, | |
1113 | int attempted, | |
1114 | int dropped) | |
1115 | { | |
1116 | int i; | |
1117 | int sent = attempted - dropped; | |
1118 | ||
1119 | stats->tx_packets += sent; | |
1120 | stats->tx_dropped += dropped; | |
1121 | ||
1122 | for (i = 0; i < sent; i++) { | |
1123 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1124 | } | |
1125 | } | |
1126 | ||
58397e6c | 1127 | static void |
4573fbd3 FL |
1128 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1129 | struct dp_packet **pkts, int cnt, | |
1130 | bool may_steal) | |
58397e6c KT |
1131 | { |
1132 | struct netdev_dpdk *vhost_dev = netdev_dpdk_cast(netdev); | |
1133 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(vhost_dev); | |
95e9881f KT |
1134 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1135 | unsigned int total_pkts = cnt; | |
1136 | uint64_t start = 0; | |
58397e6c | 1137 | |
585a5bea IM |
1138 | qid = vhost_dev->tx_q[qid % vhost_dev->real_n_txq].map; |
1139 | ||
1140 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev) || qid == -1)) { | |
45d947c4 | 1141 | rte_spinlock_lock(&vhost_dev->stats_lock); |
1b99bb05 | 1142 | vhost_dev->stats.tx_dropped+= cnt; |
45d947c4 | 1143 | rte_spinlock_unlock(&vhost_dev->stats_lock); |
1b99bb05 | 1144 | goto out; |
58397e6c KT |
1145 | } |
1146 | ||
585a5bea | 1147 | rte_spinlock_lock(&vhost_dev->tx_q[qid].tx_lock); |
58397e6c | 1148 | |
95e9881f | 1149 | do { |
4573fbd3 | 1150 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1151 | unsigned int tx_pkts; |
1152 | ||
4573fbd3 | 1153 | tx_pkts = rte_vhost_enqueue_burst(virtio_dev, vhost_qid, |
95e9881f KT |
1154 | cur_pkts, cnt); |
1155 | if (OVS_LIKELY(tx_pkts)) { | |
1156 | /* Packets have been sent.*/ | |
1157 | cnt -= tx_pkts; | |
1158 | /* Prepare for possible next iteration.*/ | |
1159 | cur_pkts = &cur_pkts[tx_pkts]; | |
1160 | } else { | |
1161 | uint64_t timeout = VHOST_ENQ_RETRY_USECS * rte_get_timer_hz() / 1E6; | |
1162 | unsigned int expired = 0; | |
1163 | ||
1164 | if (!start) { | |
1165 | start = rte_get_timer_cycles(); | |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * Unable to enqueue packets to vhost interface. | |
1170 | * Check available entries before retrying. | |
1171 | */ | |
4573fbd3 | 1172 | while (!rte_vring_available_entries(virtio_dev, vhost_qid)) { |
95e9881f KT |
1173 | if (OVS_UNLIKELY((rte_get_timer_cycles() - start) > timeout)) { |
1174 | expired = 1; | |
1175 | break; | |
1176 | } | |
1177 | } | |
1178 | if (expired) { | |
1179 | /* break out of main loop. */ | |
1180 | break; | |
1181 | } | |
1182 | } | |
1183 | } while (cnt); | |
4573fbd3 | 1184 | |
585a5bea | 1185 | rte_spinlock_unlock(&vhost_dev->tx_q[qid].tx_lock); |
95e9881f | 1186 | |
45d947c4 | 1187 | rte_spinlock_lock(&vhost_dev->stats_lock); |
9e3ddd45 TP |
1188 | netdev_dpdk_vhost_update_tx_counters(&vhost_dev->stats, pkts, total_pkts, |
1189 | cnt); | |
45d947c4 | 1190 | rte_spinlock_unlock(&vhost_dev->stats_lock); |
58397e6c KT |
1191 | |
1192 | out: | |
1193 | if (may_steal) { | |
95e9881f KT |
1194 | int i; |
1195 | ||
1196 | for (i = 0; i < total_pkts; i++) { | |
1b99bb05 MG |
1197 | dp_packet_delete(pkts[i]); |
1198 | } | |
58397e6c KT |
1199 | } |
1200 | } | |
1201 | ||
8a9562d2 | 1202 | inline static void |
f4fd623c DDP |
1203 | dpdk_queue_pkts(struct netdev_dpdk *dev, int qid, |
1204 | struct rte_mbuf **pkts, int cnt) | |
8a9562d2 PS |
1205 | { |
1206 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1207 | uint64_t diff_tsc; | |
8a9562d2 | 1208 | |
f4fd623c DDP |
1209 | int i = 0; |
1210 | ||
f4fd623c DDP |
1211 | while (i < cnt) { |
1212 | int freeslots = MAX_TX_QUEUE_LEN - txq->count; | |
1213 | int tocopy = MIN(freeslots, cnt-i); | |
8a9562d2 | 1214 | |
f4fd623c DDP |
1215 | memcpy(&txq->burst_pkts[txq->count], &pkts[i], |
1216 | tocopy * sizeof (struct rte_mbuf *)); | |
1217 | ||
1218 | txq->count += tocopy; | |
1219 | i += tocopy; | |
1220 | ||
94143fc4 | 1221 | if (txq->count == MAX_TX_QUEUE_LEN || txq->flush_tx) { |
b170db2a | 1222 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 1223 | } |
844f2d74 | 1224 | diff_tsc = rte_get_timer_cycles() - txq->tsc; |
f4fd623c | 1225 | if (diff_tsc >= DRAIN_TSC) { |
b170db2a | 1226 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 1227 | } |
8a9562d2 | 1228 | } |
8a9562d2 PS |
1229 | } |
1230 | ||
1231 | /* Tx function. Transmit packets indefinitely */ | |
1232 | static void | |
58397e6c | 1233 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet **pkts, |
2654cc33 | 1234 | int cnt) |
db73f716 | 1235 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1236 | { |
bce01e3a EJ |
1237 | #if !defined(__CHECKER__) && !defined(_WIN32) |
1238 | const size_t PKT_ARRAY_SIZE = cnt; | |
1239 | #else | |
1240 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1241 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1242 | #endif |
8a9562d2 | 1243 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
bce01e3a | 1244 | struct rte_mbuf *mbufs[PKT_ARRAY_SIZE]; |
175cf4de RW |
1245 | int dropped = 0; |
1246 | int newcnt = 0; | |
1247 | int i; | |
8a9562d2 | 1248 | |
db73f716 DDP |
1249 | /* If we are on a non pmd thread we have to use the mempool mutex, because |
1250 | * every non pmd thread shares the same mempool cache */ | |
1251 | ||
5f17de68 | 1252 | if (!dpdk_thread_is_pmd()) { |
db73f716 DDP |
1253 | ovs_mutex_lock(&nonpmd_mempool_mutex); |
1254 | } | |
1255 | ||
f4fd623c | 1256 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1257 | int size = dp_packet_size(pkts[i]); |
95fb793a | 1258 | |
f98d7864 | 1259 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c DDP |
1260 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1261 | (int)size , dev->max_packet_len); | |
1262 | ||
175cf4de | 1263 | dropped++; |
f4fd623c DDP |
1264 | continue; |
1265 | } | |
8a9562d2 | 1266 | |
f4fd623c | 1267 | mbufs[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1268 | |
f4fd623c | 1269 | if (!mbufs[newcnt]) { |
175cf4de RW |
1270 | dropped += cnt - i; |
1271 | break; | |
f4fd623c DDP |
1272 | } |
1273 | ||
1274 | /* We have to do a copy for now */ | |
b8e57534 | 1275 | memcpy(rte_pktmbuf_mtod(mbufs[newcnt], void *), dp_packet_data(pkts[i]), size); |
f4fd623c DDP |
1276 | |
1277 | rte_pktmbuf_data_len(mbufs[newcnt]) = size; | |
1278 | rte_pktmbuf_pkt_len(mbufs[newcnt]) = size; | |
1279 | ||
1280 | newcnt++; | |
1281 | } | |
8a9562d2 | 1282 | |
f98d7864 | 1283 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1284 | rte_spinlock_lock(&dev->stats_lock); |
175cf4de | 1285 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1286 | rte_spinlock_unlock(&dev->stats_lock); |
175cf4de RW |
1287 | } |
1288 | ||
58397e6c | 1289 | if (dev->type == DPDK_DEV_VHOST) { |
4573fbd3 | 1290 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) mbufs, newcnt, true); |
58397e6c KT |
1291 | } else { |
1292 | dpdk_queue_pkts(dev, qid, mbufs, newcnt); | |
1293 | dpdk_queue_flush(dev, qid); | |
1294 | } | |
db73f716 | 1295 | |
5f17de68 | 1296 | if (!dpdk_thread_is_pmd()) { |
db73f716 DDP |
1297 | ovs_mutex_unlock(&nonpmd_mempool_mutex); |
1298 | } | |
8a9562d2 PS |
1299 | } |
1300 | ||
58397e6c | 1301 | static int |
4573fbd3 | 1302 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, struct dp_packet **pkts, |
58397e6c KT |
1303 | int cnt, bool may_steal) |
1304 | { | |
1305 | if (OVS_UNLIKELY(pkts[0]->source != DPBUF_DPDK)) { | |
1306 | int i; | |
1307 | ||
1308 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); | |
1309 | if (may_steal) { | |
1310 | for (i = 0; i < cnt; i++) { | |
1311 | dp_packet_delete(pkts[i]); | |
1312 | } | |
1313 | } | |
1314 | } else { | |
4573fbd3 | 1315 | __netdev_dpdk_vhost_send(netdev, qid, pkts, cnt, may_steal); |
58397e6c KT |
1316 | } |
1317 | return 0; | |
1318 | } | |
1319 | ||
7251515e DV |
1320 | static inline void |
1321 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
e14deea0 | 1322 | struct dp_packet **pkts, int cnt, bool may_steal) |
8a9562d2 | 1323 | { |
f4fd623c | 1324 | int i; |
8a9562d2 | 1325 | |
a0cb2d66 DDP |
1326 | if (OVS_UNLIKELY(dev->txq_needs_locking)) { |
1327 | qid = qid % dev->real_n_txq; | |
1328 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); | |
1329 | } | |
1330 | ||
7251515e | 1331 | if (OVS_UNLIKELY(!may_steal || |
cf62fa4c | 1332 | pkts[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1333 | struct netdev *netdev = &dev->up; |
1334 | ||
2654cc33 | 1335 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); |
b3cd9f9d PS |
1336 | |
1337 | if (may_steal) { | |
f4fd623c | 1338 | for (i = 0; i < cnt; i++) { |
e14deea0 | 1339 | dp_packet_delete(pkts[i]); |
f4fd623c | 1340 | } |
b3cd9f9d | 1341 | } |
8a9562d2 | 1342 | } else { |
f4fd623c DDP |
1343 | int next_tx_idx = 0; |
1344 | int dropped = 0; | |
8a9562d2 | 1345 | |
f4fd623c | 1346 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1347 | int size = dp_packet_size(pkts[i]); |
1b99bb05 | 1348 | |
f4fd623c DDP |
1349 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
1350 | if (next_tx_idx != i) { | |
1351 | dpdk_queue_pkts(dev, qid, | |
1352 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1353 | i-next_tx_idx); | |
1ebfe1ac | 1354 | } |
f4fd623c | 1355 | |
1ebfe1ac DDP |
1356 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1357 | (int)size , dev->max_packet_len); | |
f4fd623c | 1358 | |
e14deea0 | 1359 | dp_packet_delete(pkts[i]); |
1ebfe1ac | 1360 | dropped++; |
f4fd623c DDP |
1361 | next_tx_idx = i + 1; |
1362 | } | |
1363 | } | |
1364 | if (next_tx_idx != cnt) { | |
1365 | dpdk_queue_pkts(dev, qid, | |
1366 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
1367 | cnt-next_tx_idx); | |
1368 | } | |
8a9562d2 | 1369 | |
f4fd623c | 1370 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1371 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1372 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1373 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1374 | } |
8a9562d2 | 1375 | } |
a0cb2d66 DDP |
1376 | |
1377 | if (OVS_UNLIKELY(dev->txq_needs_locking)) { | |
1378 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); | |
1379 | } | |
7251515e DV |
1380 | } |
1381 | ||
1382 | static int | |
1383 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
e14deea0 | 1384 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1385 | { |
1386 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1387 | |
7251515e DV |
1388 | netdev_dpdk_send__(dev, qid, pkts, cnt, may_steal); |
1389 | return 0; | |
8a9562d2 PS |
1390 | } |
1391 | ||
1392 | static int | |
74ff3298 | 1393 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1394 | { |
1395 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1396 | ||
1397 | ovs_mutex_lock(&dev->mutex); | |
1398 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1399 | dev->hwaddr = mac; |
045c0d1a | 1400 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1401 | } |
1402 | ovs_mutex_unlock(&dev->mutex); | |
1403 | ||
1404 | return 0; | |
1405 | } | |
1406 | ||
1407 | static int | |
74ff3298 | 1408 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1409 | { |
1410 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1411 | ||
1412 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1413 | *mac = dev->hwaddr; |
8a9562d2 PS |
1414 | ovs_mutex_unlock(&dev->mutex); |
1415 | ||
1416 | return 0; | |
1417 | } | |
1418 | ||
1419 | static int | |
1420 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1421 | { | |
1422 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1423 | ||
1424 | ovs_mutex_lock(&dev->mutex); | |
1425 | *mtup = dev->mtu; | |
1426 | ovs_mutex_unlock(&dev->mutex); | |
1427 | ||
1428 | return 0; | |
1429 | } | |
1430 | ||
1431 | static int | |
1432 | netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu) | |
1433 | { | |
1434 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
4be4d22c | 1435 | int old_mtu, err, dpdk_mtu; |
8a9562d2 PS |
1436 | struct dpdk_mp *old_mp; |
1437 | struct dpdk_mp *mp; | |
4be4d22c | 1438 | uint32_t buf_size; |
8a9562d2 PS |
1439 | |
1440 | ovs_mutex_lock(&dpdk_mutex); | |
1441 | ovs_mutex_lock(&dev->mutex); | |
1442 | if (dev->mtu == mtu) { | |
1443 | err = 0; | |
1444 | goto out; | |
1445 | } | |
1446 | ||
4be4d22c MK |
1447 | buf_size = dpdk_buf_size(mtu); |
1448 | dpdk_mtu = FRAME_LEN_TO_MTU(buf_size); | |
1449 | ||
1450 | mp = dpdk_mp_get(dev->socket_id, dpdk_mtu); | |
8a9562d2 PS |
1451 | if (!mp) { |
1452 | err = ENOMEM; | |
1453 | goto out; | |
1454 | } | |
1455 | ||
1456 | rte_eth_dev_stop(dev->port_id); | |
1457 | ||
1458 | old_mtu = dev->mtu; | |
1459 | old_mp = dev->dpdk_mp; | |
1460 | dev->dpdk_mp = mp; | |
1461 | dev->mtu = mtu; | |
4be4d22c | 1462 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
8a9562d2 PS |
1463 | |
1464 | err = dpdk_eth_dev_init(dev); | |
1465 | if (err) { | |
8a9562d2 PS |
1466 | dpdk_mp_put(mp); |
1467 | dev->mtu = old_mtu; | |
1468 | dev->dpdk_mp = old_mp; | |
4be4d22c | 1469 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
8a9562d2 PS |
1470 | dpdk_eth_dev_init(dev); |
1471 | goto out; | |
1472 | } | |
1473 | ||
1474 | dpdk_mp_put(old_mp); | |
045c0d1a | 1475 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1476 | out: |
1477 | ovs_mutex_unlock(&dev->mutex); | |
1478 | ovs_mutex_unlock(&dpdk_mutex); | |
1479 | return err; | |
1480 | } | |
1481 | ||
1482 | static int | |
1483 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier); | |
1484 | ||
58397e6c KT |
1485 | static int |
1486 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1487 | struct netdev_stats *stats) | |
1488 | { | |
1489 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1490 | ||
1491 | ovs_mutex_lock(&dev->mutex); | |
1492 | memset(stats, 0, sizeof(*stats)); | |
1493 | /* Unsupported Stats */ | |
58397e6c KT |
1494 | stats->collisions = UINT64_MAX; |
1495 | stats->rx_crc_errors = UINT64_MAX; | |
1496 | stats->rx_fifo_errors = UINT64_MAX; | |
1497 | stats->rx_frame_errors = UINT64_MAX; | |
58397e6c KT |
1498 | stats->rx_missed_errors = UINT64_MAX; |
1499 | stats->rx_over_errors = UINT64_MAX; | |
1500 | stats->tx_aborted_errors = UINT64_MAX; | |
1501 | stats->tx_carrier_errors = UINT64_MAX; | |
1502 | stats->tx_errors = UINT64_MAX; | |
1503 | stats->tx_fifo_errors = UINT64_MAX; | |
1504 | stats->tx_heartbeat_errors = UINT64_MAX; | |
1505 | stats->tx_window_errors = UINT64_MAX; | |
58397e6c | 1506 | stats->rx_dropped += UINT64_MAX; |
58397e6c | 1507 | |
45d947c4 | 1508 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
1509 | /* Supported Stats */ |
1510 | stats->rx_packets += dev->stats.rx_packets; | |
1511 | stats->tx_packets += dev->stats.tx_packets; | |
1512 | stats->tx_dropped += dev->stats.tx_dropped; | |
9e3ddd45 TP |
1513 | stats->multicast = dev->stats.multicast; |
1514 | stats->rx_bytes = dev->stats.rx_bytes; | |
1515 | stats->tx_bytes = dev->stats.tx_bytes; | |
1516 | stats->rx_errors = dev->stats.rx_errors; | |
1517 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
45d947c4 | 1518 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 1519 | |
58397e6c KT |
1520 | ovs_mutex_unlock(&dev->mutex); |
1521 | ||
1522 | return 0; | |
1523 | } | |
1524 | ||
8a9562d2 PS |
1525 | static int |
1526 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1527 | { | |
1528 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1529 | struct rte_eth_stats rte_stats; | |
1530 | bool gg; | |
1531 | ||
1532 | netdev_dpdk_get_carrier(netdev, &gg); | |
1533 | ovs_mutex_lock(&dev->mutex); | |
1534 | rte_eth_stats_get(dev->port_id, &rte_stats); | |
1535 | ||
2f9dd77f | 1536 | memset(stats, 0, sizeof(*stats)); |
8a9562d2 | 1537 | |
2f9dd77f PS |
1538 | stats->rx_packets = rte_stats.ipackets; |
1539 | stats->tx_packets = rte_stats.opackets; | |
1540 | stats->rx_bytes = rte_stats.ibytes; | |
1541 | stats->tx_bytes = rte_stats.obytes; | |
9e3ddd45 TP |
1542 | /* DPDK counts imissed as errors, but count them here as dropped instead */ |
1543 | stats->rx_errors = rte_stats.ierrors - rte_stats.imissed; | |
2f9dd77f PS |
1544 | stats->tx_errors = rte_stats.oerrors; |
1545 | stats->multicast = rte_stats.imcasts; | |
8a9562d2 | 1546 | |
45d947c4 | 1547 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 1548 | stats->tx_dropped = dev->stats.tx_dropped; |
45d947c4 | 1549 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
1550 | |
1551 | /* These are the available DPDK counters for packets not received due to | |
1552 | * local resource constraints in DPDK and NIC respectively. */ | |
1553 | stats->rx_dropped = rte_stats.rx_nombuf + rte_stats.imissed; | |
1554 | stats->collisions = UINT64_MAX; | |
1555 | ||
02ab4b1a | 1556 | stats->rx_length_errors = UINT64_MAX; |
9e3ddd45 | 1557 | stats->rx_over_errors = UINT64_MAX; |
02ab4b1a | 1558 | stats->rx_crc_errors = UINT64_MAX; |
9e3ddd45 TP |
1559 | stats->rx_frame_errors = UINT64_MAX; |
1560 | stats->rx_fifo_errors = UINT64_MAX; | |
1561 | stats->rx_missed_errors = rte_stats.imissed; | |
1562 | ||
1563 | stats->tx_aborted_errors = UINT64_MAX; | |
1564 | stats->tx_carrier_errors = UINT64_MAX; | |
1565 | stats->tx_fifo_errors = UINT64_MAX; | |
1566 | stats->tx_heartbeat_errors = UINT64_MAX; | |
1567 | stats->tx_window_errors = UINT64_MAX; | |
1568 | ||
8a9562d2 PS |
1569 | ovs_mutex_unlock(&dev->mutex); |
1570 | ||
1571 | return 0; | |
1572 | } | |
1573 | ||
1574 | static int | |
1575 | netdev_dpdk_get_features(const struct netdev *netdev_, | |
1576 | enum netdev_features *current, | |
1577 | enum netdev_features *advertised OVS_UNUSED, | |
1578 | enum netdev_features *supported OVS_UNUSED, | |
1579 | enum netdev_features *peer OVS_UNUSED) | |
1580 | { | |
1581 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1582 | struct rte_eth_link link; | |
1583 | ||
1584 | ovs_mutex_lock(&dev->mutex); | |
1585 | link = dev->link; | |
1586 | ovs_mutex_unlock(&dev->mutex); | |
1587 | ||
1588 | if (link.link_duplex == ETH_LINK_AUTONEG_DUPLEX) { | |
1589 | if (link.link_speed == ETH_LINK_SPEED_AUTONEG) { | |
1590 | *current = NETDEV_F_AUTONEG; | |
1591 | } | |
1592 | } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { | |
1593 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1594 | *current = NETDEV_F_10MB_HD; | |
1595 | } | |
1596 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1597 | *current = NETDEV_F_100MB_HD; | |
1598 | } | |
1599 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1600 | *current = NETDEV_F_1GB_HD; | |
1601 | } | |
1602 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
1603 | if (link.link_speed == ETH_LINK_SPEED_10) { | |
1604 | *current = NETDEV_F_10MB_FD; | |
1605 | } | |
1606 | if (link.link_speed == ETH_LINK_SPEED_100) { | |
1607 | *current = NETDEV_F_100MB_FD; | |
1608 | } | |
1609 | if (link.link_speed == ETH_LINK_SPEED_1000) { | |
1610 | *current = NETDEV_F_1GB_FD; | |
1611 | } | |
1612 | if (link.link_speed == ETH_LINK_SPEED_10000) { | |
1613 | *current = NETDEV_F_10GB_FD; | |
1614 | } | |
1615 | } | |
1616 | ||
1617 | return 0; | |
1618 | } | |
1619 | ||
1620 | static int | |
1621 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
1622 | { | |
1623 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1624 | int ifindex; | |
1625 | ||
1626 | ovs_mutex_lock(&dev->mutex); | |
1627 | ifindex = dev->port_id; | |
1628 | ovs_mutex_unlock(&dev->mutex); | |
1629 | ||
1630 | return ifindex; | |
1631 | } | |
1632 | ||
1633 | static int | |
1634 | netdev_dpdk_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1635 | { | |
1636 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1637 | ||
1638 | ovs_mutex_lock(&dev->mutex); | |
1639 | check_link_status(dev); | |
1640 | *carrier = dev->link.link_status; | |
58397e6c KT |
1641 | |
1642 | ovs_mutex_unlock(&dev->mutex); | |
1643 | ||
1644 | return 0; | |
1645 | } | |
1646 | ||
1647 | static int | |
1648 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev_, bool *carrier) | |
1649 | { | |
1650 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1651 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(dev); | |
1652 | ||
1653 | ovs_mutex_lock(&dev->mutex); | |
1654 | ||
1655 | if (is_vhost_running(virtio_dev)) { | |
1656 | *carrier = 1; | |
1657 | } else { | |
1658 | *carrier = 0; | |
1659 | } | |
1660 | ||
8a9562d2 PS |
1661 | ovs_mutex_unlock(&dev->mutex); |
1662 | ||
1663 | return 0; | |
1664 | } | |
1665 | ||
1666 | static long long int | |
1667 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev_) | |
1668 | { | |
1669 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1670 | long long int carrier_resets; | |
1671 | ||
1672 | ovs_mutex_lock(&dev->mutex); | |
1673 | carrier_resets = dev->link_reset_cnt; | |
1674 | ovs_mutex_unlock(&dev->mutex); | |
1675 | ||
1676 | return carrier_resets; | |
1677 | } | |
1678 | ||
1679 | static int | |
1680 | netdev_dpdk_set_miimon(struct netdev *netdev_ OVS_UNUSED, | |
1681 | long long int interval OVS_UNUSED) | |
1682 | { | |
ee32150e | 1683 | return EOPNOTSUPP; |
8a9562d2 PS |
1684 | } |
1685 | ||
1686 | static int | |
1687 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
1688 | enum netdev_flags off, enum netdev_flags on, | |
95fb793a | 1689 | enum netdev_flags *old_flagsp) OVS_REQUIRES(dev->mutex) |
8a9562d2 PS |
1690 | { |
1691 | int err; | |
1692 | ||
1693 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
1694 | return EINVAL; | |
1695 | } | |
1696 | ||
1697 | *old_flagsp = dev->flags; | |
1698 | dev->flags |= on; | |
1699 | dev->flags &= ~off; | |
1700 | ||
1701 | if (dev->flags == *old_flagsp) { | |
1702 | return 0; | |
1703 | } | |
1704 | ||
58397e6c KT |
1705 | if (dev->type == DPDK_DEV_ETH) { |
1706 | if (dev->flags & NETDEV_UP) { | |
1707 | err = rte_eth_dev_start(dev->port_id); | |
1708 | if (err) | |
1709 | return -err; | |
1710 | } | |
8a9562d2 | 1711 | |
58397e6c KT |
1712 | if (dev->flags & NETDEV_PROMISC) { |
1713 | rte_eth_promiscuous_enable(dev->port_id); | |
1714 | } | |
8a9562d2 | 1715 | |
58397e6c KT |
1716 | if (!(dev->flags & NETDEV_UP)) { |
1717 | rte_eth_dev_stop(dev->port_id); | |
1718 | } | |
8a9562d2 PS |
1719 | } |
1720 | ||
1721 | return 0; | |
1722 | } | |
1723 | ||
1724 | static int | |
1725 | netdev_dpdk_update_flags(struct netdev *netdev_, | |
1726 | enum netdev_flags off, enum netdev_flags on, | |
1727 | enum netdev_flags *old_flagsp) | |
1728 | { | |
1729 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); | |
1730 | int error; | |
1731 | ||
1732 | ovs_mutex_lock(&netdev->mutex); | |
1733 | error = netdev_dpdk_update_flags__(netdev, off, on, old_flagsp); | |
1734 | ovs_mutex_unlock(&netdev->mutex); | |
1735 | ||
1736 | return error; | |
1737 | } | |
1738 | ||
1739 | static int | |
1740 | netdev_dpdk_get_status(const struct netdev *netdev_, struct smap *args) | |
1741 | { | |
1742 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev_); | |
1743 | struct rte_eth_dev_info dev_info; | |
1744 | ||
e0a801c7 | 1745 | if (dev->port_id < 0) |
8a9562d2 PS |
1746 | return ENODEV; |
1747 | ||
1748 | ovs_mutex_lock(&dev->mutex); | |
1749 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
1750 | ovs_mutex_unlock(&dev->mutex); | |
1751 | ||
1752 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1753 | ||
95fb793a | 1754 | smap_add_format(args, "port_no", "%d", dev->port_id); |
8a9562d2 PS |
1755 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); |
1756 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
1757 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 1758 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
1759 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
1760 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
1761 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
1762 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
1763 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
1764 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
1765 | ||
39c2baa9 | 1766 | if (dev_info.pci_dev) { |
1767 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
1768 | dev_info.pci_dev->id.vendor_id); | |
1769 | smap_add_format(args, "pci-device_id", "0x%x", | |
1770 | dev_info.pci_dev->id.device_id); | |
1771 | } | |
8a9562d2 PS |
1772 | |
1773 | return 0; | |
1774 | } | |
1775 | ||
1776 | static void | |
1777 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
1778 | OVS_REQUIRES(dev->mutex) | |
1779 | { | |
1780 | enum netdev_flags old_flags; | |
1781 | ||
1782 | if (admin_state) { | |
1783 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
1784 | } else { | |
1785 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
1786 | } | |
1787 | } | |
1788 | ||
1789 | static void | |
1790 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
1791 | const char *argv[], void *aux OVS_UNUSED) | |
1792 | { | |
1793 | bool up; | |
1794 | ||
1795 | if (!strcasecmp(argv[argc - 1], "up")) { | |
1796 | up = true; | |
1797 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
1798 | up = false; | |
1799 | } else { | |
1800 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
1801 | return; | |
1802 | } | |
1803 | ||
1804 | if (argc > 2) { | |
1805 | struct netdev *netdev = netdev_from_name(argv[1]); | |
1806 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
1807 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
1808 | ||
1809 | ovs_mutex_lock(&dpdk_dev->mutex); | |
1810 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
1811 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
1812 | ||
1813 | netdev_close(netdev); | |
1814 | } else { | |
1815 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
1816 | netdev_close(netdev); | |
1817 | return; | |
1818 | } | |
1819 | } else { | |
1820 | struct netdev_dpdk *netdev; | |
1821 | ||
1822 | ovs_mutex_lock(&dpdk_mutex); | |
1823 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
1824 | ovs_mutex_lock(&netdev->mutex); | |
1825 | netdev_dpdk_set_admin_state__(netdev, up); | |
1826 | ovs_mutex_unlock(&netdev->mutex); | |
1827 | } | |
1828 | ovs_mutex_unlock(&dpdk_mutex); | |
1829 | } | |
1830 | unixctl_command_reply(conn, "OK"); | |
1831 | } | |
1832 | ||
58397e6c KT |
1833 | /* |
1834 | * Set virtqueue flags so that we do not receive interrupts. | |
1835 | */ | |
1836 | static void | |
1837 | set_irq_status(struct virtio_net *dev) | |
1838 | { | |
4573fbd3 FL |
1839 | uint32_t i; |
1840 | uint64_t idx; | |
1841 | ||
1842 | for (i = 0; i < dev->virt_qp_nb; i++) { | |
1843 | idx = i * VIRTIO_QNUM; | |
1844 | rte_vhost_enable_guest_notification(dev, idx + VIRTIO_RXQ, 0); | |
1845 | rte_vhost_enable_guest_notification(dev, idx + VIRTIO_TXQ, 0); | |
1846 | } | |
1847 | } | |
1848 | ||
585a5bea IM |
1849 | /* |
1850 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
1851 | * enabling/disabling of queues and real_n_txq modifications. | |
1852 | */ | |
1853 | static void | |
1854 | netdev_dpdk_remap_txqs(struct netdev_dpdk *netdev) | |
1855 | OVS_REQUIRES(netdev->mutex) | |
1856 | { | |
1857 | int *enabled_queues, n_enabled = 0; | |
1858 | int i, k, total_txqs = netdev->real_n_txq; | |
1859 | ||
1860 | enabled_queues = dpdk_rte_mzalloc(total_txqs * sizeof *enabled_queues); | |
1861 | ||
1862 | for (i = 0; i < total_txqs; i++) { | |
1863 | /* Enabled queues always mapped to themselves. */ | |
1864 | if (netdev->tx_q[i].map == i) { | |
1865 | enabled_queues[n_enabled++] = i; | |
1866 | } | |
1867 | } | |
1868 | ||
1869 | if (n_enabled == 0 && total_txqs != 0) { | |
1870 | enabled_queues[0] = -1; | |
1871 | n_enabled = 1; | |
1872 | } | |
1873 | ||
1874 | k = 0; | |
1875 | for (i = 0; i < total_txqs; i++) { | |
1876 | if (netdev->tx_q[i].map != i) { | |
1877 | netdev->tx_q[i].map = enabled_queues[k]; | |
1878 | k = (k + 1) % n_enabled; | |
1879 | } | |
1880 | } | |
1881 | ||
1882 | VLOG_DBG("TX queue mapping for %s\n", netdev->vhost_id); | |
1883 | for (i = 0; i < total_txqs; i++) { | |
1884 | VLOG_DBG("%2d --> %2d", i, netdev->tx_q[i].map); | |
1885 | } | |
1886 | ||
1887 | rte_free(enabled_queues); | |
1888 | } | |
4573fbd3 FL |
1889 | |
1890 | static int | |
1891 | netdev_dpdk_vhost_set_queues(struct netdev_dpdk *netdev, struct virtio_net *dev) | |
585a5bea | 1892 | OVS_REQUIRES(netdev->mutex) |
4573fbd3 FL |
1893 | { |
1894 | uint32_t qp_num; | |
1895 | ||
1896 | qp_num = dev->virt_qp_nb; | |
1897 | if (qp_num > netdev->up.n_rxq) { | |
1898 | VLOG_ERR("vHost Device '%s' %"PRIu64" can't be added - " | |
1899 | "too many queues %d > %d", dev->ifname, dev->device_fh, | |
1900 | qp_num, netdev->up.n_rxq); | |
1901 | return -1; | |
1902 | } | |
1903 | ||
1904 | netdev->real_n_rxq = qp_num; | |
1905 | netdev->real_n_txq = qp_num; | |
585a5bea IM |
1906 | netdev->txq_needs_locking = true; |
1907 | ||
1908 | netdev_dpdk_remap_txqs(netdev); | |
4573fbd3 FL |
1909 | |
1910 | return 0; | |
58397e6c KT |
1911 | } |
1912 | ||
1913 | /* | |
1914 | * A new virtio-net device is added to a vhost port. | |
1915 | */ | |
1916 | static int | |
1917 | new_device(struct virtio_net *dev) | |
1918 | { | |
1919 | struct netdev_dpdk *netdev; | |
1920 | bool exists = false; | |
1921 | ||
1922 | ovs_mutex_lock(&dpdk_mutex); | |
1923 | /* Add device to the vhost port with the same name as that passed down. */ | |
1924 | LIST_FOR_EACH(netdev, list_node, &dpdk_list) { | |
7d1ced01 | 1925 | if (strncmp(dev->ifname, netdev->vhost_id, IF_NAME_SZ) == 0) { |
58397e6c | 1926 | ovs_mutex_lock(&netdev->mutex); |
4573fbd3 FL |
1927 | if (netdev_dpdk_vhost_set_queues(netdev, dev)) { |
1928 | ovs_mutex_unlock(&netdev->mutex); | |
1929 | ovs_mutex_unlock(&dpdk_mutex); | |
1930 | return -1; | |
1931 | } | |
58397e6c | 1932 | ovsrcu_set(&netdev->virtio_dev, dev); |
58397e6c KT |
1933 | exists = true; |
1934 | dev->flags |= VIRTIO_DEV_RUNNING; | |
1935 | /* Disable notifications. */ | |
1936 | set_irq_status(dev); | |
4573fbd3 | 1937 | ovs_mutex_unlock(&netdev->mutex); |
58397e6c KT |
1938 | break; |
1939 | } | |
1940 | } | |
1941 | ovs_mutex_unlock(&dpdk_mutex); | |
1942 | ||
1943 | if (!exists) { | |
3a8653d7 AZ |
1944 | VLOG_INFO("vHost Device '%s' %"PRIu64" can't be added - name not " |
1945 | "found", dev->ifname, dev->device_fh); | |
58397e6c KT |
1946 | |
1947 | return -1; | |
1948 | } | |
1949 | ||
3a8653d7 AZ |
1950 | VLOG_INFO("vHost Device '%s' %"PRIu64" has been added", dev->ifname, |
1951 | dev->device_fh); | |
58397e6c KT |
1952 | return 0; |
1953 | } | |
1954 | ||
1955 | /* | |
1956 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
1957 | * flag to stop any more packets from being sent or received to/from a VM and | |
1958 | * ensure all currently queued packets have been sent/received before removing | |
1959 | * the device. | |
1960 | */ | |
1961 | static void | |
1962 | destroy_device(volatile struct virtio_net *dev) | |
1963 | { | |
1964 | struct netdev_dpdk *vhost_dev; | |
afee281f | 1965 | bool exists = false; |
58397e6c KT |
1966 | |
1967 | ovs_mutex_lock(&dpdk_mutex); | |
1968 | LIST_FOR_EACH (vhost_dev, list_node, &dpdk_list) { | |
1969 | if (netdev_dpdk_get_virtio(vhost_dev) == dev) { | |
1970 | ||
1971 | ovs_mutex_lock(&vhost_dev->mutex); | |
1972 | dev->flags &= ~VIRTIO_DEV_RUNNING; | |
1973 | ovsrcu_set(&vhost_dev->virtio_dev, NULL); | |
afee281f | 1974 | exists = true; |
58397e6c | 1975 | ovs_mutex_unlock(&vhost_dev->mutex); |
afee281f | 1976 | break; |
58397e6c KT |
1977 | } |
1978 | } | |
afee281f | 1979 | |
58397e6c KT |
1980 | ovs_mutex_unlock(&dpdk_mutex); |
1981 | ||
afee281f KT |
1982 | if (exists == true) { |
1983 | /* | |
1984 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
1985 | * to NULL, before returning. | |
1986 | */ | |
1987 | ovsrcu_synchronize(); | |
1988 | /* | |
1989 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
1990 | * put thread back into quiescent state before returning. | |
1991 | */ | |
1992 | ovsrcu_quiesce_start(); | |
1993 | VLOG_INFO("vHost Device '%s' %"PRIu64" has been removed", dev->ifname, | |
1994 | dev->device_fh); | |
1995 | } else { | |
1996 | VLOG_INFO("vHost Device '%s' %"PRIu64" not found", dev->ifname, | |
1997 | dev->device_fh); | |
1998 | } | |
1999 | ||
58397e6c KT |
2000 | } |
2001 | ||
585a5bea IM |
2002 | static int |
2003 | vring_state_changed(struct virtio_net *dev, uint16_t queue_id, int enable) | |
2004 | { | |
2005 | struct netdev_dpdk *vhost_dev; | |
2006 | bool exists = false; | |
2007 | int qid = queue_id / VIRTIO_QNUM; | |
2008 | ||
2009 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2010 | return 0; | |
2011 | } | |
2012 | ||
2013 | ovs_mutex_lock(&dpdk_mutex); | |
2014 | LIST_FOR_EACH (vhost_dev, list_node, &dpdk_list) { | |
2015 | if (strncmp(dev->ifname, vhost_dev->vhost_id, IF_NAME_SZ) == 0) { | |
2016 | ovs_mutex_lock(&vhost_dev->mutex); | |
2017 | if (enable) { | |
2018 | vhost_dev->tx_q[qid].map = qid; | |
2019 | } else { | |
2020 | vhost_dev->tx_q[qid].map = -1; | |
2021 | } | |
2022 | netdev_dpdk_remap_txqs(vhost_dev); | |
2023 | exists = true; | |
2024 | ovs_mutex_unlock(&vhost_dev->mutex); | |
2025 | break; | |
2026 | } | |
2027 | } | |
2028 | ovs_mutex_unlock(&dpdk_mutex); | |
2029 | ||
2030 | if (exists) { | |
2031 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s' %" | |
2032 | PRIu64" changed to \'%s\'", queue_id, qid, dev->ifname, | |
2033 | dev->device_fh, (enable == 1) ? "enabled" : "disabled"); | |
2034 | } else { | |
2035 | VLOG_INFO("vHost Device '%s' %"PRIu64" not found", dev->ifname, | |
2036 | dev->device_fh); | |
2037 | return -1; | |
2038 | } | |
2039 | ||
2040 | return 0; | |
2041 | } | |
2042 | ||
58397e6c KT |
2043 | struct virtio_net * |
2044 | netdev_dpdk_get_virtio(const struct netdev_dpdk *dev) | |
2045 | { | |
2046 | return ovsrcu_get(struct virtio_net *, &dev->virtio_dev); | |
2047 | } | |
2048 | ||
2049 | /* | |
2050 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
2051 | * configuration has been fully complete. | |
2052 | */ | |
bce01e3a | 2053 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
2054 | { |
2055 | .new_device = new_device, | |
2056 | .destroy_device = destroy_device, | |
585a5bea | 2057 | .vring_state_changed = vring_state_changed |
58397e6c KT |
2058 | }; |
2059 | ||
2060 | static void * | |
7d1ced01 | 2061 | start_vhost_loop(void *dummy OVS_UNUSED) |
58397e6c KT |
2062 | { |
2063 | pthread_detach(pthread_self()); | |
618f44f7 KT |
2064 | /* Put the cuse thread into quiescent state. */ |
2065 | ovsrcu_quiesce_start(); | |
58397e6c KT |
2066 | rte_vhost_driver_session_start(); |
2067 | return NULL; | |
2068 | } | |
2069 | ||
2070 | static int | |
2071 | dpdk_vhost_class_init(void) | |
7d1ced01 CL |
2072 | { |
2073 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
2074 | ovs_thread_create("vhost_thread", start_vhost_loop, NULL); | |
2075 | return 0; | |
2076 | } | |
2077 | ||
2078 | static int | |
2079 | dpdk_vhost_cuse_class_init(void) | |
58397e6c | 2080 | { |
58397e6c KT |
2081 | int err = -1; |
2082 | ||
58397e6c KT |
2083 | |
2084 | /* Register CUSE device to handle IOCTLs. | |
2085 | * Unless otherwise specified on the vswitchd command line, cuse_dev_name | |
2086 | * is set to vhost-net. | |
2087 | */ | |
2088 | err = rte_vhost_driver_register(cuse_dev_name); | |
2089 | ||
2090 | if (err != 0) { | |
2091 | VLOG_ERR("CUSE device setup failure."); | |
2092 | return -1; | |
2093 | } | |
2094 | ||
7d1ced01 CL |
2095 | dpdk_vhost_class_init(); |
2096 | return 0; | |
2097 | } | |
2098 | ||
2099 | static int | |
2100 | dpdk_vhost_user_class_init(void) | |
2101 | { | |
2102 | dpdk_vhost_class_init(); | |
618f44f7 | 2103 | return 0; |
58397e6c KT |
2104 | } |
2105 | ||
033e9df2 DDP |
2106 | static void |
2107 | dpdk_common_init(void) | |
2108 | { | |
2109 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2110 | "[netdev] up|down", 1, 2, | |
2111 | netdev_dpdk_set_admin_state, NULL); | |
2112 | ||
2113 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
2114 | } | |
2115 | ||
95fb793a | 2116 | /* Client Rings */ |
2117 | ||
95fb793a | 2118 | static int |
2119 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
2120 | unsigned int *eth_port_id) | |
2121 | { | |
2122 | struct dpdk_ring *ivshmem; | |
ca7e7bee | 2123 | char ring_name[RTE_RING_NAMESIZE]; |
95fb793a | 2124 | int err; |
2125 | ||
2126 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
2127 | if (ivshmem == NULL) { | |
2128 | return ENOMEM; | |
2129 | } | |
2130 | ||
7251515e | 2131 | /* XXX: Add support for multiquque ring. */ |
ca7e7bee | 2132 | err = snprintf(ring_name, sizeof(ring_name), "%s_tx", dev_name); |
95fb793a | 2133 | if (err < 0) { |
2134 | return -err; | |
2135 | } | |
2136 | ||
8f0a76c9 | 2137 | /* Create single producer tx ring, netdev does explicit locking. */ |
7251515e | 2138 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2139 | RING_F_SP_ENQ); |
95fb793a | 2140 | if (ivshmem->cring_tx == NULL) { |
2141 | rte_free(ivshmem); | |
2142 | return ENOMEM; | |
2143 | } | |
2144 | ||
ca7e7bee | 2145 | err = snprintf(ring_name, sizeof(ring_name), "%s_rx", dev_name); |
95fb793a | 2146 | if (err < 0) { |
2147 | return -err; | |
2148 | } | |
2149 | ||
8f0a76c9 | 2150 | /* Create single consumer rx ring, netdev does explicit locking. */ |
7251515e | 2151 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2152 | RING_F_SC_DEQ); |
95fb793a | 2153 | if (ivshmem->cring_rx == NULL) { |
2154 | rte_free(ivshmem); | |
2155 | return ENOMEM; | |
2156 | } | |
2157 | ||
d7310583 DDP |
2158 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
2159 | &ivshmem->cring_tx, 1, SOCKET0); | |
2160 | ||
95fb793a | 2161 | if (err < 0) { |
2162 | rte_free(ivshmem); | |
2163 | return ENODEV; | |
2164 | } | |
2165 | ||
2166 | ivshmem->user_port_id = port_no; | |
2167 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
2168 | list_push_back(&dpdk_ring_list, &ivshmem->list_node); | |
2169 | ||
2170 | *eth_port_id = ivshmem->eth_port_id; | |
2171 | return 0; | |
2172 | } | |
2173 | ||
2174 | static int | |
2175 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) OVS_REQUIRES(dpdk_mutex) | |
2176 | { | |
2177 | struct dpdk_ring *ivshmem; | |
2178 | unsigned int port_no; | |
2179 | int err = 0; | |
2180 | ||
2181 | /* Names always start with "dpdkr" */ | |
2182 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2183 | if (err) { | |
2184 | return err; | |
2185 | } | |
2186 | ||
2187 | /* look through our list to find the device */ | |
2188 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { | |
2189 | if (ivshmem->user_port_id == port_no) { | |
58397e6c | 2190 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
95fb793a | 2191 | *eth_port_id = ivshmem->eth_port_id; /* really all that is needed */ |
2192 | return 0; | |
2193 | } | |
2194 | } | |
2195 | /* Need to create the device rings */ | |
2196 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2197 | } | |
2198 | ||
7251515e | 2199 | static int |
a0cb2d66 | 2200 | netdev_dpdk_ring_send(struct netdev *netdev_, int qid, |
e14deea0 | 2201 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e | 2202 | { |
a0cb2d66 | 2203 | struct netdev_dpdk *netdev = netdev_dpdk_cast(netdev_); |
1b99bb05 MG |
2204 | unsigned i; |
2205 | ||
2206 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that the | |
2207 | * rss hash field is clear. This is because the same mbuf may be modified by | |
2208 | * the consumer of the ring and return into the datapath without recalculating | |
2209 | * the RSS hash. */ | |
2210 | for (i = 0; i < cnt; i++) { | |
f2f44f5d | 2211 | dp_packet_rss_invalidate(pkts[i]); |
1b99bb05 | 2212 | } |
7251515e | 2213 | |
a0cb2d66 | 2214 | netdev_dpdk_send__(netdev, qid, pkts, cnt, may_steal); |
7251515e DV |
2215 | return 0; |
2216 | } | |
2217 | ||
95fb793a | 2218 | static int |
2219 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2220 | { | |
2221 | unsigned int port_no = 0; | |
2222 | int err = 0; | |
2223 | ||
2224 | if (rte_eal_init_ret) { | |
2225 | return rte_eal_init_ret; | |
2226 | } | |
2227 | ||
2228 | ovs_mutex_lock(&dpdk_mutex); | |
2229 | ||
2230 | err = dpdk_ring_open(netdev->name, &port_no); | |
2231 | if (err) { | |
2232 | goto unlock_dpdk; | |
2233 | } | |
2234 | ||
58397e6c | 2235 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
95fb793a | 2236 | |
2237 | unlock_dpdk: | |
2238 | ovs_mutex_unlock(&dpdk_mutex); | |
2239 | return err; | |
2240 | } | |
2241 | ||
58397e6c KT |
2242 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, MULTIQ, SEND, \ |
2243 | GET_CARRIER, GET_STATS, GET_FEATURES, GET_STATUS, RXQ_RECV) \ | |
95fb793a | 2244 | { \ |
2245 | NAME, \ | |
2246 | INIT, /* init */ \ | |
2247 | NULL, /* netdev_dpdk_run */ \ | |
2248 | NULL, /* netdev_dpdk_wait */ \ | |
2249 | \ | |
2250 | netdev_dpdk_alloc, \ | |
2251 | CONSTRUCT, \ | |
58397e6c | 2252 | DESTRUCT, \ |
95fb793a | 2253 | netdev_dpdk_dealloc, \ |
2254 | netdev_dpdk_get_config, \ | |
a14b8947 | 2255 | netdev_dpdk_set_config, \ |
95fb793a | 2256 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
2257 | NULL, /* build header */ \ |
2258 | NULL, /* push header */ \ | |
2259 | NULL, /* pop header */ \ | |
7dec44fe | 2260 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
5496878c | 2261 | MULTIQ, /* set_multiq */ \ |
95fb793a | 2262 | \ |
7251515e | 2263 | SEND, /* send */ \ |
95fb793a | 2264 | NULL, /* send_wait */ \ |
2265 | \ | |
2266 | netdev_dpdk_set_etheraddr, \ | |
2267 | netdev_dpdk_get_etheraddr, \ | |
2268 | netdev_dpdk_get_mtu, \ | |
2269 | netdev_dpdk_set_mtu, \ | |
2270 | netdev_dpdk_get_ifindex, \ | |
58397e6c | 2271 | GET_CARRIER, \ |
95fb793a | 2272 | netdev_dpdk_get_carrier_resets, \ |
2273 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
2274 | GET_STATS, \ |
2275 | GET_FEATURES, \ | |
95fb793a | 2276 | NULL, /* set_advertisements */ \ |
2277 | \ | |
2278 | NULL, /* set_policing */ \ | |
2279 | NULL, /* get_qos_types */ \ | |
2280 | NULL, /* get_qos_capabilities */ \ | |
2281 | NULL, /* get_qos */ \ | |
2282 | NULL, /* set_qos */ \ | |
2283 | NULL, /* get_queue */ \ | |
2284 | NULL, /* set_queue */ \ | |
2285 | NULL, /* delete_queue */ \ | |
2286 | NULL, /* get_queue_stats */ \ | |
2287 | NULL, /* queue_dump_start */ \ | |
2288 | NULL, /* queue_dump_next */ \ | |
2289 | NULL, /* queue_dump_done */ \ | |
2290 | NULL, /* dump_queue_stats */ \ | |
2291 | \ | |
2292 | NULL, /* get_in4 */ \ | |
2293 | NULL, /* set_in4 */ \ | |
2294 | NULL, /* get_in6 */ \ | |
2295 | NULL, /* add_router */ \ | |
2296 | NULL, /* get_next_hop */ \ | |
58397e6c | 2297 | GET_STATUS, \ |
95fb793a | 2298 | NULL, /* arp_lookup */ \ |
2299 | \ | |
2300 | netdev_dpdk_update_flags, \ | |
2301 | \ | |
2302 | netdev_dpdk_rxq_alloc, \ | |
2303 | netdev_dpdk_rxq_construct, \ | |
2304 | netdev_dpdk_rxq_destruct, \ | |
2305 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 2306 | RXQ_RECV, \ |
95fb793a | 2307 | NULL, /* rx_wait */ \ |
2308 | NULL, /* rxq_drain */ \ | |
2309 | } | |
8a9562d2 | 2310 | |
7d1ced01 CL |
2311 | static int |
2312 | process_vhost_flags(char *flag, char *default_val, int size, | |
2313 | char **argv, char **new_val) | |
2314 | { | |
2315 | int changed = 0; | |
2316 | ||
2317 | /* Depending on which version of vhost is in use, process the vhost-specific | |
2318 | * flag if it is provided on the vswitchd command line, otherwise resort to | |
2319 | * a default value. | |
2320 | * | |
3cdb27d3 | 2321 | * For vhost-user: Process "-vhost_sock_dir" to set the custom location of |
7d1ced01 | 2322 | * the vhost-user socket(s). |
3cdb27d3 | 2323 | * For vhost-cuse: Process "-cuse_dev_name" to set the custom name of the |
7d1ced01 CL |
2324 | * vhost-cuse character device. |
2325 | */ | |
2326 | if (!strcmp(argv[1], flag) && (strlen(argv[2]) <= size)) { | |
2327 | changed = 1; | |
0d4af148 | 2328 | *new_val = xstrdup(argv[2]); |
7d1ced01 CL |
2329 | VLOG_INFO("User-provided %s in use: %s", flag, *new_val); |
2330 | } else { | |
2331 | VLOG_INFO("No %s provided - defaulting to %s", flag, default_val); | |
2332 | *new_val = default_val; | |
2333 | } | |
2334 | ||
2335 | return changed; | |
2336 | } | |
2337 | ||
8a9562d2 PS |
2338 | int |
2339 | dpdk_init(int argc, char **argv) | |
2340 | { | |
2341 | int result; | |
58397e6c KT |
2342 | int base = 0; |
2343 | char *pragram_name = argv[0]; | |
8a9562d2 | 2344 | |
9441caf3 | 2345 | if (argc < 2 || strcmp(argv[1], "--dpdk")) |
8a9562d2 PS |
2346 | return 0; |
2347 | ||
58397e6c | 2348 | /* Remove the --dpdk argument from arg list.*/ |
8a9562d2 PS |
2349 | argc--; |
2350 | argv++; | |
2351 | ||
dfc89a55 AZ |
2352 | /* Reject --user option */ |
2353 | int i; | |
2354 | for (i = 0; i < argc; i++) { | |
2355 | if (!strcmp(argv[i], "--user")) { | |
2356 | VLOG_ERR("Can not mix --dpdk and --user options, aborting."); | |
2357 | } | |
2358 | } | |
2359 | ||
7d1ced01 | 2360 | #ifdef VHOST_CUSE |
0d4af148 | 2361 | if (process_vhost_flags("-cuse_dev_name", xstrdup("vhost-net"), |
7d1ced01 CL |
2362 | PATH_MAX, argv, &cuse_dev_name)) { |
2363 | #else | |
0d4af148 | 2364 | if (process_vhost_flags("-vhost_sock_dir", xstrdup(ovs_rundir()), |
7d1ced01 CL |
2365 | NAME_MAX, argv, &vhost_sock_dir)) { |
2366 | struct stat s; | |
2367 | int err; | |
58397e6c | 2368 | |
7d1ced01 CL |
2369 | err = stat(vhost_sock_dir, &s); |
2370 | if (err) { | |
2371 | VLOG_ERR("vHostUser socket DIR '%s' does not exist.", | |
2372 | vhost_sock_dir); | |
2373 | return err; | |
2374 | } | |
2375 | #endif | |
2376 | /* Remove the vhost flag configuration parameters from the argument | |
58397e6c KT |
2377 | * list, so that the correct elements are passed to the DPDK |
2378 | * initialization function | |
2379 | */ | |
2380 | argc -= 2; | |
7d1ced01 | 2381 | argv += 2; /* Increment by two to bypass the vhost flag arguments */ |
58397e6c | 2382 | base = 2; |
58397e6c KT |
2383 | } |
2384 | ||
2385 | /* Keep the program name argument as this is needed for call to | |
2386 | * rte_eal_init() | |
2387 | */ | |
2388 | argv[0] = pragram_name; | |
2389 | ||
8a9562d2 PS |
2390 | /* Make sure things are initialized ... */ |
2391 | result = rte_eal_init(argc, argv); | |
451450fa | 2392 | if (result < 0) { |
58397e6c | 2393 | ovs_abort(result, "Cannot init EAL"); |
451450fa | 2394 | } |
8a9562d2 | 2395 | |
d7310583 | 2396 | rte_memzone_dump(stdout); |
8a9562d2 PS |
2397 | rte_eal_init_ret = 0; |
2398 | ||
451450fa | 2399 | if (argc > result) { |
9441caf3 | 2400 | argv[result] = argv[0]; |
451450fa | 2401 | } |
9441caf3 | 2402 | |
db73f716 | 2403 | /* We are called from the main thread here */ |
d5c199ea | 2404 | RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID; |
db73f716 | 2405 | |
58397e6c | 2406 | return result + 1 + base; |
8a9562d2 PS |
2407 | } |
2408 | ||
bce01e3a | 2409 | static const struct netdev_class dpdk_class = |
95fb793a | 2410 | NETDEV_DPDK_CLASS( |
2411 | "dpdk", | |
b8e57534 | 2412 | NULL, |
5496878c | 2413 | netdev_dpdk_construct, |
58397e6c | 2414 | netdev_dpdk_destruct, |
7251515e | 2415 | netdev_dpdk_set_multiq, |
58397e6c KT |
2416 | netdev_dpdk_eth_send, |
2417 | netdev_dpdk_get_carrier, | |
2418 | netdev_dpdk_get_stats, | |
2419 | netdev_dpdk_get_features, | |
2420 | netdev_dpdk_get_status, | |
2421 | netdev_dpdk_rxq_recv); | |
95fb793a | 2422 | |
bce01e3a | 2423 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 2424 | NETDEV_DPDK_CLASS( |
2425 | "dpdkr", | |
033e9df2 | 2426 | NULL, |
5496878c | 2427 | netdev_dpdk_ring_construct, |
58397e6c | 2428 | netdev_dpdk_destruct, |
a0cb2d66 | 2429 | netdev_dpdk_set_multiq, |
58397e6c KT |
2430 | netdev_dpdk_ring_send, |
2431 | netdev_dpdk_get_carrier, | |
2432 | netdev_dpdk_get_stats, | |
2433 | netdev_dpdk_get_features, | |
2434 | netdev_dpdk_get_status, | |
2435 | netdev_dpdk_rxq_recv); | |
2436 | ||
bd8baf47 | 2437 | static const struct netdev_class OVS_UNUSED dpdk_vhost_cuse_class = |
58397e6c | 2438 | NETDEV_DPDK_CLASS( |
7d1ced01 CL |
2439 | "dpdkvhostcuse", |
2440 | dpdk_vhost_cuse_class_init, | |
2441 | netdev_dpdk_vhost_cuse_construct, | |
2442 | netdev_dpdk_vhost_destruct, | |
4573fbd3 | 2443 | netdev_dpdk_vhost_cuse_set_multiq, |
7d1ced01 CL |
2444 | netdev_dpdk_vhost_send, |
2445 | netdev_dpdk_vhost_get_carrier, | |
2446 | netdev_dpdk_vhost_get_stats, | |
2447 | NULL, | |
2448 | NULL, | |
2449 | netdev_dpdk_vhost_rxq_recv); | |
2450 | ||
bd8baf47 | 2451 | static const struct netdev_class OVS_UNUSED dpdk_vhost_user_class = |
7d1ced01 CL |
2452 | NETDEV_DPDK_CLASS( |
2453 | "dpdkvhostuser", | |
2454 | dpdk_vhost_user_class_init, | |
2455 | netdev_dpdk_vhost_user_construct, | |
58397e6c KT |
2456 | netdev_dpdk_vhost_destruct, |
2457 | netdev_dpdk_vhost_set_multiq, | |
2458 | netdev_dpdk_vhost_send, | |
2459 | netdev_dpdk_vhost_get_carrier, | |
2460 | netdev_dpdk_vhost_get_stats, | |
2461 | NULL, | |
7251515e | 2462 | NULL, |
58397e6c | 2463 | netdev_dpdk_vhost_rxq_recv); |
95fb793a | 2464 | |
8a9562d2 PS |
2465 | void |
2466 | netdev_dpdk_register(void) | |
2467 | { | |
95fb793a | 2468 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2469 | ||
033e9df2 DDP |
2470 | if (rte_eal_init_ret) { |
2471 | return; | |
2472 | } | |
2473 | ||
95fb793a | 2474 | if (ovsthread_once_start(&once)) { |
033e9df2 | 2475 | dpdk_common_init(); |
95fb793a | 2476 | netdev_register_provider(&dpdk_class); |
2477 | netdev_register_provider(&dpdk_ring_class); | |
7d1ced01 CL |
2478 | #ifdef VHOST_CUSE |
2479 | netdev_register_provider(&dpdk_vhost_cuse_class); | |
2480 | #else | |
2481 | netdev_register_provider(&dpdk_vhost_user_class); | |
2482 | #endif | |
95fb793a | 2483 | ovsthread_once_done(&once); |
2484 | } | |
8a9562d2 | 2485 | } |
8617afff PS |
2486 | |
2487 | int | |
bd5131ba | 2488 | pmd_thread_setaffinity_cpu(unsigned cpu) |
8617afff PS |
2489 | { |
2490 | cpu_set_t cpuset; | |
2491 | int err; | |
2492 | ||
2493 | CPU_ZERO(&cpuset); | |
2494 | CPU_SET(cpu, &cpuset); | |
2495 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset); | |
2496 | if (err) { | |
2497 | VLOG_ERR("Thread affinity error %d",err); | |
2498 | return err; | |
2499 | } | |
abb5943d AW |
2500 | /* NON_PMD_CORE_ID is reserved for use by non pmd threads. */ |
2501 | ovs_assert(cpu != NON_PMD_CORE_ID); | |
65f13b50 | 2502 | RTE_PER_LCORE(_lcore_id) = cpu; |
8617afff PS |
2503 | |
2504 | return 0; | |
2505 | } | |
db73f716 | 2506 | |
db73f716 | 2507 | static bool |
5f17de68 | 2508 | dpdk_thread_is_pmd(void) |
db73f716 | 2509 | { |
abb5943d | 2510 | return rte_lcore_id() != NON_PMD_CORE_ID; |
db73f716 | 2511 | } |