]>
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> | |
01961bbd | 18 | #include "netdev-dpdk.h" |
8a9562d2 | 19 | |
8a9562d2 PS |
20 | #include <string.h> |
21 | #include <signal.h> | |
22 | #include <stdlib.h> | |
8a9562d2 | 23 | #include <errno.h> |
8a9562d2 | 24 | #include <unistd.h> |
01961bbd DDP |
25 | |
26 | #include <rte_config.h> | |
27 | #include <rte_cycles.h> | |
28 | #include <rte_errno.h> | |
29 | #include <rte_eth_ring.h> | |
30 | #include <rte_ethdev.h> | |
31 | #include <rte_malloc.h> | |
32 | #include <rte_mbuf.h> | |
33 | #include <rte_meter.h> | |
01961bbd | 34 | #include <rte_virtio_net.h> |
8a9562d2 | 35 | |
7d1ced01 | 36 | #include "dirs.h" |
e14deea0 | 37 | #include "dp-packet.h" |
01961bbd | 38 | #include "dpdk.h" |
8a9562d2 | 39 | #include "dpif-netdev.h" |
e5c0f5a4 | 40 | #include "fatal-signal.h" |
8a9562d2 PS |
41 | #include "netdev-provider.h" |
42 | #include "netdev-vport.h" | |
43 | #include "odp-util.h" | |
eac84432 | 44 | #include "openvswitch/dynamic-string.h" |
25d436fb BW |
45 | #include "openvswitch/list.h" |
46 | #include "openvswitch/ofp-print.h" | |
47 | #include "openvswitch/vlog.h" | |
94143fc4 | 48 | #include "ovs-numa.h" |
8a9562d2 PS |
49 | #include "ovs-thread.h" |
50 | #include "ovs-rcu.h" | |
51 | #include "packets.h" | |
ee89ea7b | 52 | #include "openvswitch/shash.h" |
0bf765f7 | 53 | #include "smap.h" |
8a9562d2 PS |
54 | #include "sset.h" |
55 | #include "unaligned.h" | |
56 | #include "timeval.h" | |
57 | #include "unixctl.h" | |
8a9562d2 | 58 | |
05b49df6 | 59 | VLOG_DEFINE_THIS_MODULE(netdev_dpdk); |
8a9562d2 PS |
60 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
61 | ||
62 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
63 | ||
64 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
65 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
66 | ||
67 | /* | |
68 | * need to reserve tons of extra space in the mbufs so we can align the | |
69 | * DMA addresses to 4KB. | |
18f777b2 TP |
70 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
71 | * performance for standard Ethernet MTU. | |
8a9562d2 | 72 | */ |
58be5c0e MK |
73 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN \ |
74 | + (2 * VLAN_HEADER_LEN)) | |
4be4d22c MK |
75 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) |
76 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
58be5c0e MK |
77 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len) \ |
78 | - ETHER_HDR_LEN - ETHER_CRC_LEN) | |
79 | #define MBUF_SIZE(mtu) (MTU_TO_MAX_FRAME_LEN(mtu) \ | |
80 | + sizeof(struct dp_packet) \ | |
81 | + RTE_PKTMBUF_HEADROOM) | |
4be4d22c | 82 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 83 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 84 | |
da79ce2b DDP |
85 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
86 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
87 | * enough hugepages) we keep halving the number until the allocation succeeds | |
88 | * or we reach MIN_NB_MBUF */ | |
89 | ||
90 | #define MAX_NB_MBUF (4096 * 64) | |
91 | #define MIN_NB_MBUF (4096 * 4) | |
92 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
93 | ||
94 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
95 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
96 | ||
97 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
98 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
99 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
100 | % MP_CACHE_SZ == 0); | |
101 | ||
d6e3feb5 | 102 | /* |
103 | * DPDK XSTATS Counter names definition | |
104 | */ | |
105 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
106 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
107 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
108 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
109 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
110 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
111 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
112 | ||
113 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
114 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
115 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
116 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
117 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
118 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
119 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
120 | ||
121 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" | |
122 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
123 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
124 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
125 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
126 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
127 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
128 | ||
8a9562d2 PS |
129 | #define SOCKET0 0 |
130 | ||
b685696b CL |
131 | /* Default size of Physical NIC RXQ */ |
132 | #define NIC_PORT_DEFAULT_RXQ_SIZE 2048 | |
133 | /* Default size of Physical NIC TXQ */ | |
134 | #define NIC_PORT_DEFAULT_TXQ_SIZE 2048 | |
135 | /* Maximum size of Physical NIC Queues */ | |
136 | #define NIC_PORT_MAX_Q_SIZE 4096 | |
79f5354c | 137 | |
585a5bea | 138 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
139 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
140 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
141 | * yet mapped to another queue. */ | |
585a5bea | 142 | |
31871ee3 | 143 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 144 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 145 | |
8a9562d2 | 146 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
147 | .rxmode = { |
148 | .mq_mode = ETH_MQ_RX_RSS, | |
149 | .split_hdr_size = 0, | |
150 | .header_split = 0, /* Header Split disabled */ | |
151 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
152 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
153 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
154 | .hw_strip_crc = 0, | |
155 | }, | |
156 | .rx_adv_conf = { | |
157 | .rss_conf = { | |
158 | .rss_key = NULL, | |
543342a4 | 159 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 160 | }, |
a28ddd11 DDP |
161 | }, |
162 | .txmode = { | |
163 | .mq_mode = ETH_MQ_TX_NONE, | |
164 | }, | |
8a9562d2 PS |
165 | }; |
166 | ||
58f7c37b DDP |
167 | enum { DPDK_RING_SIZE = 256 }; |
168 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
169 | enum { DRAIN_TSC = 200000ULL }; |
170 | ||
58397e6c KT |
171 | enum dpdk_dev_type { |
172 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 173 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
174 | }; |
175 | ||
0bf765f7 IS |
176 | /* Quality of Service */ |
177 | ||
178 | /* An instance of a QoS configuration. Always associated with a particular | |
179 | * network device. | |
180 | * | |
181 | * Each QoS implementation subclasses this with whatever additional data it | |
182 | * needs. | |
183 | */ | |
184 | struct qos_conf { | |
185 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 186 | rte_spinlock_t lock; |
0bf765f7 IS |
187 | }; |
188 | ||
189 | /* A particular implementation of dpdk QoS operations. | |
190 | * | |
191 | * The functions below return 0 if successful or a positive errno value on | |
192 | * failure, except where otherwise noted. All of them must be provided, except | |
193 | * where otherwise noted. | |
194 | */ | |
195 | struct dpdk_qos_ops { | |
196 | ||
197 | /* Name of the QoS type */ | |
198 | const char *qos_name; | |
199 | ||
78bd47cf DDP |
200 | /* Called to construct a qos_conf object. The implementation should make |
201 | * the appropriate calls to configure QoS according to 'details'. | |
0bf765f7 IS |
202 | * |
203 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
204 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
205 | * (which is built as ovs-vswitchd.conf.db(8)). | |
206 | * | |
78bd47cf DDP |
207 | * This function must return 0 if and only if it sets '*conf' to an |
208 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
209 | * |
210 | * For all QoS implementations it should always be non-null. | |
211 | */ | |
78bd47cf | 212 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
213 | |
214 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 215 | * 'qos_conf'. |
0bf765f7 IS |
216 | * |
217 | * For all QoS implementations it should always be non-null. | |
218 | */ | |
78bd47cf | 219 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 220 | |
78bd47cf | 221 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
222 | * |
223 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
224 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
225 | * (which is built as ovs-vswitchd.conf.db(8)). | |
226 | */ | |
78bd47cf | 227 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 228 | |
78bd47cf | 229 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
230 | * |
231 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
232 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
233 | * (which is built as ovs-vswitchd.conf.db(8)). | |
234 | * | |
78bd47cf | 235 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 236 | */ |
78bd47cf DDP |
237 | bool (*qos_is_equal)(const struct qos_conf *conf, |
238 | const struct smap *details); | |
0bf765f7 IS |
239 | |
240 | /* Modify an array of rte_mbufs. The modification is specific to | |
241 | * each qos implementation. | |
242 | * | |
243 | * The function should take and array of mbufs and an int representing | |
244 | * the current number of mbufs present in the array. | |
245 | * | |
246 | * After the function has performed a qos modification to the array of | |
247 | * mbufs it returns an int representing the number of mbufs now present in | |
248 | * the array. This value is can then be passed to the port send function | |
249 | * along with the modified array for transmission. | |
250 | * | |
251 | * For all QoS implementations it should always be non-null. | |
252 | */ | |
78bd47cf DDP |
253 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
254 | int pkt_cnt); | |
0bf765f7 IS |
255 | }; |
256 | ||
257 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
258 | static const struct dpdk_qos_ops egress_policer_ops; | |
259 | ||
260 | /* | |
261 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
262 | * operations. | |
263 | */ | |
264 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
265 | &egress_policer_ops, | |
266 | NULL | |
267 | }; | |
268 | ||
c2adb102 IM |
269 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
270 | ||
8a9562d2 | 271 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 272 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 273 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 274 | |
c2adb102 IM |
275 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
276 | = OVS_MUTEX_INITIALIZER; | |
277 | ||
278 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mp_mutex) | |
55951e15 | 279 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 280 | |
8a9562d2 PS |
281 | struct dpdk_mp { |
282 | struct rte_mempool *mp; | |
283 | int mtu; | |
284 | int socket_id; | |
285 | int refcount; | |
c2adb102 | 286 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mp_mutex); |
8a9562d2 PS |
287 | }; |
288 | ||
5a034064 AW |
289 | /* There should be one 'struct dpdk_tx_queue' created for |
290 | * each cpu core. */ | |
8a9562d2 | 291 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
292 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
293 | * from concurrent access. It is used only | |
294 | * if the queue is shared among different | |
324c8374 | 295 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
296 | int map; /* Mapping of configured vhost-user queues |
297 | * to enabled by guest. */ | |
8a9562d2 PS |
298 | }; |
299 | ||
95fb793a | 300 | /* dpdk has no way to remove dpdk ring ethernet devices |
301 | so we have to keep them around once they've been created | |
302 | */ | |
303 | ||
ca6ba700 | 304 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 305 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 306 | |
307 | struct dpdk_ring { | |
308 | /* For the client rings */ | |
309 | struct rte_ring *cring_tx; | |
310 | struct rte_ring *cring_rx; | |
b83a2df1 | 311 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
95fb793a | 312 | int eth_port_id; /* ethernet device port id */ |
ca6ba700 | 313 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 314 | }; |
315 | ||
9509913a IS |
316 | struct ingress_policer { |
317 | struct rte_meter_srtcm_params app_srtcm_params; | |
318 | struct rte_meter_srtcm in_policer; | |
319 | rte_spinlock_t policer_lock; | |
320 | }; | |
321 | ||
8a9562d2 PS |
322 | struct netdev_dpdk { |
323 | struct netdev up; | |
324 | int port_id; | |
325 | int max_packet_len; | |
58397e6c | 326 | enum dpdk_dev_type type; |
8a9562d2 | 327 | |
5a034064 | 328 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
329 | |
330 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
331 | ||
332 | struct dpdk_mp *dpdk_mp; | |
333 | int mtu; | |
334 | int socket_id; | |
335 | int buf_size; | |
8a9562d2 | 336 | struct netdev_stats stats; |
45d947c4 DDP |
337 | /* Protects stats */ |
338 | rte_spinlock_t stats_lock; | |
8a9562d2 | 339 | |
74ff3298 | 340 | struct eth_addr hwaddr; |
8a9562d2 PS |
341 | enum netdev_flags flags; |
342 | ||
343 | struct rte_eth_link link; | |
344 | int link_reset_cnt; | |
345 | ||
0a0f39df CL |
346 | /* virtio identifier for vhost devices */ |
347 | ovsrcu_index vid; | |
348 | ||
349 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
350 | bool vhost_reconfigured; | |
58397e6c | 351 | |
2d24d165 CL |
352 | /* Identifier used to distinguish vhost devices from each other. */ |
353 | char vhost_id[PATH_MAX]; | |
7d1ced01 | 354 | |
8a9562d2 | 355 | /* In dpdk_list. */ |
ca6ba700 | 356 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
0bf765f7 IS |
357 | |
358 | /* QoS configuration and lock for the device */ | |
78bd47cf | 359 | OVSRCU_TYPE(struct qos_conf *) qos_conf; |
0bf765f7 | 360 | |
050c60bf DDP |
361 | /* The following properties cannot be changed when a device is running, |
362 | * so we remember the request and update them next time | |
363 | * netdev_dpdk*_reconfigure() is called */ | |
0072e931 | 364 | int requested_mtu; |
050c60bf DDP |
365 | int requested_n_txq; |
366 | int requested_n_rxq; | |
b685696b CL |
367 | int requested_rxq_size; |
368 | int requested_txq_size; | |
369 | ||
370 | /* Number of rx/tx descriptors for physical devices */ | |
371 | int rxq_size; | |
372 | int txq_size; | |
9509913a | 373 | |
db8f13b0 CL |
374 | /* Socket ID detected when vHost device is brought up */ |
375 | int requested_socket_id; | |
376 | ||
c1ff66ac CL |
377 | /* Denotes whether vHost port is client/server mode */ |
378 | uint64_t vhost_driver_flags; | |
379 | ||
9509913a IS |
380 | /* Ingress Policer */ |
381 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
382 | uint32_t policer_rate; | |
383 | uint32_t policer_burst; | |
9fd39370 SC |
384 | |
385 | /* DPDK-ETH Flow control */ | |
386 | struct rte_eth_fc_conf fc_conf; | |
8a9562d2 PS |
387 | }; |
388 | ||
389 | struct netdev_rxq_dpdk { | |
390 | struct netdev_rxq up; | |
391 | int port_id; | |
392 | }; | |
393 | ||
b3722dca | 394 | static int netdev_dpdk_class_init(void); |
395 | static int netdev_dpdk_vhost_class_init(void); | |
8a9562d2 | 396 | |
0a0f39df | 397 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 398 | |
9509913a IS |
399 | struct ingress_policer * |
400 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
401 | ||
8a9562d2 PS |
402 | static bool |
403 | is_dpdk_class(const struct netdev_class *class) | |
404 | { | |
b3722dca | 405 | return class->init == netdev_dpdk_class_init |
406 | || class->init == netdev_dpdk_vhost_class_init; | |
8a9562d2 PS |
407 | } |
408 | ||
4be4d22c MK |
409 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
410 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
411 | * value, insufficient buffers are allocated to accomodate the packet in its | |
412 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
413 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
414 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
415 | * behaviour, which reduces performance. To prevent this, use a buffer size |
416 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
417 | */ |
418 | static uint32_t | |
419 | dpdk_buf_size(int mtu) | |
420 | { | |
421 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
422 | NETDEV_DPDK_MBUF_ALIGN); | |
423 | } | |
424 | ||
eff23640 DDP |
425 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
426 | * | |
427 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
428 | static void * |
429 | dpdk_rte_mzalloc(size_t sz) | |
430 | { | |
eff23640 | 431 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
432 | } |
433 | ||
434 | void | |
e14deea0 | 435 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 436 | { |
db73f716 | 437 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 438 | |
b00b4a81 | 439 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
440 | } |
441 | ||
b3cd9f9d PS |
442 | static void |
443 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
444 | void *opaque_arg OVS_UNUSED, | |
2391135c | 445 | void *_p, |
b3cd9f9d PS |
446 | unsigned i OVS_UNUSED) |
447 | { | |
2391135c | 448 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 449 | |
2391135c | 450 | rte_pktmbuf_init(mp, opaque_arg, _p, i); |
b3cd9f9d | 451 | |
2391135c | 452 | dp_packet_init_dpdk((struct dp_packet *) pkt, pkt->buf_len); |
b3cd9f9d PS |
453 | } |
454 | ||
8a9562d2 | 455 | static struct dpdk_mp * |
2ae3d542 | 456 | dpdk_mp_create(int socket_id, int mtu) |
8a9562d2 | 457 | { |
2ae3d542 | 458 | struct rte_pktmbuf_pool_private mbp_priv; |
2ae3d542 | 459 | struct dpdk_mp *dmp; |
da79ce2b | 460 | unsigned mp_size; |
0c6f39e5 | 461 | char *mp_name; |
8a9562d2 PS |
462 | |
463 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
eff23640 DDP |
464 | if (!dmp) { |
465 | return NULL; | |
466 | } | |
8a9562d2 PS |
467 | dmp->socket_id = socket_id; |
468 | dmp->mtu = mtu; | |
469 | dmp->refcount = 1; | |
4be4d22c | 470 | mbp_priv.mbuf_data_room_size = MBUF_SIZE(mtu) - sizeof(struct dp_packet); |
58be5c0e MK |
471 | mbp_priv.mbuf_priv_size = sizeof(struct dp_packet) |
472 | - sizeof(struct rte_mbuf); | |
0072e931 | 473 | /* XXX: this is a really rough method of provisioning memory. |
2ae3d542 | 474 | * It's impossible to determine what the exact memory requirements are |
58be5c0e MK |
475 | * when the number of ports and rxqs that utilize a particular mempool can |
476 | * change dynamically at runtime. For now, use this rough heurisitic. | |
0072e931 MK |
477 | */ |
478 | if (mtu >= ETHER_MTU) { | |
479 | mp_size = MAX_NB_MBUF; | |
480 | } else { | |
481 | mp_size = MIN_NB_MBUF; | |
482 | } | |
8a9562d2 | 483 | |
da79ce2b | 484 | do { |
0c6f39e5 DDP |
485 | mp_name = xasprintf("ovs_mp_%d_%d_%u", dmp->mtu, dmp->socket_id, |
486 | mp_size); | |
95fb793a | 487 | |
da79ce2b DDP |
488 | dmp->mp = rte_mempool_create(mp_name, mp_size, MBUF_SIZE(mtu), |
489 | MP_CACHE_SZ, | |
490 | sizeof(struct rte_pktmbuf_pool_private), | |
4be4d22c | 491 | rte_pktmbuf_pool_init, &mbp_priv, |
da79ce2b DDP |
492 | ovs_rte_pktmbuf_init, NULL, |
493 | socket_id, 0); | |
0c6f39e5 DDP |
494 | if (dmp->mp) { |
495 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", | |
496 | mp_name, mp_size); | |
497 | } | |
498 | free(mp_name); | |
499 | if (dmp->mp) { | |
500 | return dmp; | |
501 | } | |
502 | } while (rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
2ae3d542 | 503 | |
2ae3d542 DDP |
504 | rte_free(dmp); |
505 | return NULL; | |
506 | } | |
507 | ||
508 | static struct dpdk_mp * | |
509 | dpdk_mp_get(int socket_id, int mtu) | |
510 | { | |
511 | struct dpdk_mp *dmp; | |
512 | ||
513 | ovs_mutex_lock(&dpdk_mp_mutex); | |
514 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
515 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
516 | dmp->refcount++; | |
517 | goto out; | |
518 | } | |
519 | } | |
520 | ||
521 | dmp = dpdk_mp_create(socket_id, mtu); | |
417e7e66 | 522 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); |
5f88de0d | 523 | |
c2adb102 IM |
524 | out: |
525 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
526 | ||
c2adb102 | 527 | return dmp; |
8a9562d2 PS |
528 | } |
529 | ||
530 | static void | |
c2adb102 | 531 | dpdk_mp_put(struct dpdk_mp *dmp) |
8a9562d2 | 532 | { |
8a9562d2 PS |
533 | if (!dmp) { |
534 | return; | |
535 | } | |
536 | ||
c2adb102 | 537 | ovs_mutex_lock(&dpdk_mp_mutex); |
8d38823b | 538 | ovs_assert(dmp->refcount); |
8a9562d2 | 539 | |
8d38823b MK |
540 | if (!--dmp->refcount) { |
541 | ovs_list_remove(&dmp->list_node); | |
542 | rte_mempool_free(dmp->mp); | |
5f88de0d | 543 | rte_free(dmp); |
8a9562d2 | 544 | } |
c2adb102 | 545 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
546 | } |
547 | ||
0072e931 MK |
548 | /* Tries to allocate new mempool on requested_socket_id with |
549 | * mbuf size corresponding to requested_mtu. | |
550 | * On success new configuration will be applied. | |
551 | * On error, device will be left unchanged. */ | |
552 | static int | |
553 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
554 | OVS_REQUIRES(dev->mutex) |
555 | { | |
556 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
557 | struct dpdk_mp *mp; | |
558 | ||
559 | mp = dpdk_mp_get(dev->requested_socket_id, FRAME_LEN_TO_MTU(buf_size)); | |
560 | if (!mp) { | |
561 | VLOG_ERR("Insufficient memory to create memory pool for netdev " | |
562 | "%s, with MTU %d on socket %d\n", | |
563 | dev->up.name, dev->requested_mtu, dev->requested_socket_id); | |
564 | return ENOMEM; | |
565 | } else { | |
566 | dpdk_mp_put(dev->dpdk_mp); | |
567 | dev->dpdk_mp = mp; | |
568 | dev->mtu = dev->requested_mtu; | |
569 | dev->socket_id = dev->requested_socket_id; | |
570 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
571 | } | |
572 | ||
573 | return 0; | |
574 | } | |
575 | ||
8a9562d2 PS |
576 | static void |
577 | check_link_status(struct netdev_dpdk *dev) | |
578 | { | |
579 | struct rte_eth_link link; | |
580 | ||
581 | rte_eth_link_get_nowait(dev->port_id, &link); | |
582 | ||
583 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 584 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
585 | |
586 | dev->link_reset_cnt++; | |
587 | dev->link = link; | |
588 | if (dev->link.link_status) { | |
589 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
58be5c0e | 590 | dev->port_id, (unsigned) dev->link.link_speed, |
8a9562d2 PS |
591 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? |
592 | ("full-duplex") : ("half-duplex")); | |
593 | } else { | |
594 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
595 | } | |
596 | } | |
597 | } | |
598 | ||
599 | static void * | |
600 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
601 | { | |
602 | struct netdev_dpdk *dev; | |
603 | ||
604 | pthread_detach(pthread_self()); | |
605 | ||
606 | for (;;) { | |
607 | ovs_mutex_lock(&dpdk_mutex); | |
608 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
609 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
610 | if (dev->type == DPDK_DEV_ETH) { |
611 | check_link_status(dev); | |
612 | } | |
8a9562d2 PS |
613 | ovs_mutex_unlock(&dev->mutex); |
614 | } | |
615 | ovs_mutex_unlock(&dpdk_mutex); | |
616 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
617 | } | |
618 | ||
619 | return NULL; | |
620 | } | |
621 | ||
b98d7669 DDP |
622 | static int |
623 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
624 | { | |
625 | int diag = 0; | |
626 | int i; | |
0072e931 | 627 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 628 | |
0072e931 MK |
629 | if (dev->mtu > ETHER_MTU) { |
630 | conf.rxmode.jumbo_frame = 1; | |
631 | conf.rxmode.max_rx_pkt_len = dev->max_packet_len; | |
632 | } else { | |
633 | conf.rxmode.jumbo_frame = 0; | |
634 | conf.rxmode.max_rx_pkt_len = 0; | |
635 | } | |
b98d7669 DDP |
636 | /* A device may report more queues than it makes available (this has |
637 | * been observed for Intel xl710, which reserves some of them for | |
638 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
639 | * available. When this happens we can retry the configuration | |
640 | * and request less queues */ | |
641 | while (n_rxq && n_txq) { | |
642 | if (diag) { | |
643 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
644 | } | |
645 | ||
0072e931 | 646 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 647 | if (diag) { |
0072e931 MK |
648 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
649 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
650 | break; |
651 | } | |
652 | ||
653 | for (i = 0; i < n_txq; i++) { | |
b685696b | 654 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
655 | dev->socket_id, NULL); |
656 | if (diag) { | |
657 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
658 | dev->up.name, i, rte_strerror(-diag)); | |
659 | break; | |
660 | } | |
661 | } | |
662 | ||
663 | if (i != n_txq) { | |
664 | /* Retry with less tx queues */ | |
665 | n_txq = i; | |
666 | continue; | |
667 | } | |
668 | ||
669 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 670 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
b98d7669 DDP |
671 | dev->socket_id, NULL, |
672 | dev->dpdk_mp->mp); | |
673 | if (diag) { | |
674 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
675 | dev->up.name, i, rte_strerror(-diag)); | |
676 | break; | |
677 | } | |
678 | } | |
679 | ||
680 | if (i != n_rxq) { | |
681 | /* Retry with less rx queues */ | |
682 | n_rxq = i; | |
683 | continue; | |
684 | } | |
685 | ||
686 | dev->up.n_rxq = n_rxq; | |
81acebda | 687 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
688 | |
689 | return 0; | |
690 | } | |
691 | ||
692 | return diag; | |
693 | } | |
694 | ||
9fd39370 SC |
695 | static void |
696 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
697 | { | |
698 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
699 | VLOG_WARN("Failed to enable flow control on device %d", dev->port_id); | |
700 | } | |
701 | } | |
b98d7669 | 702 | |
8a9562d2 | 703 | static int |
c2adb102 IM |
704 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
705 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
706 | { |
707 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 708 | struct rte_eth_dev_info info; |
8a9562d2 PS |
709 | struct ether_addr eth_addr; |
710 | int diag; | |
b98d7669 | 711 | int n_rxq, n_txq; |
8a9562d2 | 712 | |
7cd1261d | 713 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
95fb793a | 714 | return ENODEV; |
8a9562d2 PS |
715 | } |
716 | ||
a0cb2d66 | 717 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 718 | |
b98d7669 DDP |
719 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
720 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
721 | ||
722 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 723 | if (diag) { |
b98d7669 DDP |
724 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
725 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 726 | return -diag; |
8a9562d2 PS |
727 | } |
728 | ||
8a9562d2 PS |
729 | diag = rte_eth_dev_start(dev->port_id); |
730 | if (diag) { | |
b98d7669 DDP |
731 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
732 | rte_strerror(-diag)); | |
95fb793a | 733 | return -diag; |
8a9562d2 PS |
734 | } |
735 | ||
736 | rte_eth_promiscuous_enable(dev->port_id); | |
737 | rte_eth_allmulticast_enable(dev->port_id); | |
738 | ||
739 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
740 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
741 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
ca92d173 | 742 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 743 | |
ca92d173 | 744 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
745 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
746 | ||
747 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
748 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
749 | ||
750 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
9fd39370 SC |
751 | |
752 | /* Get the Flow control configuration for DPDK-ETH */ | |
753 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
754 | if (diag) { | |
755 | VLOG_DBG("cannot get flow control parameters on port=%d, err=%d", | |
756 | dev->port_id, diag); | |
757 | } | |
758 | ||
8a9562d2 PS |
759 | return 0; |
760 | } | |
761 | ||
762 | static struct netdev_dpdk * | |
763 | netdev_dpdk_cast(const struct netdev *netdev) | |
764 | { | |
765 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
766 | } | |
767 | ||
768 | static struct netdev * | |
769 | netdev_dpdk_alloc(void) | |
770 | { | |
bab69409 AC |
771 | struct netdev_dpdk *dev; |
772 | ||
65e19e70 DDP |
773 | dev = dpdk_rte_mzalloc(sizeof *dev); |
774 | if (dev) { | |
775 | return &dev->up; | |
bab69409 | 776 | } |
65e19e70 | 777 | |
bab69409 | 778 | return NULL; |
8a9562d2 PS |
779 | } |
780 | ||
eff23640 DDP |
781 | static struct dpdk_tx_queue * |
782 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 783 | { |
eff23640 | 784 | struct dpdk_tx_queue *txqs; |
bd5131ba | 785 | unsigned i; |
5a034064 | 786 | |
eff23640 DDP |
787 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
788 | if (txqs) { | |
789 | for (i = 0; i < n_txqs; i++) { | |
790 | /* Initialize map for vhost devices. */ | |
791 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
792 | rte_spinlock_init(&txqs[i].tx_lock); | |
793 | } | |
5a034064 | 794 | } |
eff23640 DDP |
795 | |
796 | return txqs; | |
5a034064 AW |
797 | } |
798 | ||
8a9562d2 | 799 | static int |
d46285a2 | 800 | netdev_dpdk_init(struct netdev *netdev, unsigned int port_no, |
58397e6c | 801 | enum dpdk_dev_type type) |
5a034064 | 802 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 803 | { |
d46285a2 | 804 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b7a04e0 | 805 | int sid; |
95fb793a | 806 | int err = 0; |
8a9562d2 | 807 | |
d46285a2 DDP |
808 | ovs_mutex_init(&dev->mutex); |
809 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 810 | |
d46285a2 | 811 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 812 | |
1b7a04e0 AW |
813 | /* If the 'sid' is negative, it means that the kernel fails |
814 | * to obtain the pci numa info. In that situation, always | |
815 | * use 'SOCKET0'. */ | |
58397e6c KT |
816 | if (type == DPDK_DEV_ETH) { |
817 | sid = rte_eth_dev_socket_id(port_no); | |
818 | } else { | |
819 | sid = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
820 | } | |
821 | ||
d46285a2 | 822 | dev->socket_id = sid < 0 ? SOCKET0 : sid; |
db8f13b0 | 823 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
824 | dev->port_id = port_no; |
825 | dev->type = type; | |
826 | dev->flags = 0; | |
0072e931 | 827 | dev->requested_mtu = dev->mtu = ETHER_MTU; |
d46285a2 | 828 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
829 | ovsrcu_index_init(&dev->vid, -1); |
830 | dev->vhost_reconfigured = false; | |
8a9562d2 | 831 | |
0072e931 MK |
832 | err = netdev_dpdk_mempool_configure(dev); |
833 | if (err) { | |
95fb793a | 834 | goto unlock; |
8a9562d2 PS |
835 | } |
836 | ||
78bd47cf | 837 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 838 | |
9509913a IS |
839 | ovsrcu_init(&dev->ingress_policer, NULL); |
840 | dev->policer_rate = 0; | |
841 | dev->policer_burst = 0; | |
842 | ||
d46285a2 | 843 | netdev->n_rxq = NR_QUEUE; |
81acebda IM |
844 | netdev->n_txq = NR_QUEUE; |
845 | dev->requested_n_rxq = netdev->n_rxq; | |
846 | dev->requested_n_txq = netdev->n_txq; | |
b685696b CL |
847 | dev->rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; |
848 | dev->txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
849 | dev->requested_rxq_size = dev->rxq_size; | |
850 | dev->requested_txq_size = dev->txq_size; | |
58397e6c | 851 | |
9fd39370 SC |
852 | /* Initialize the flow control to NULL */ |
853 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
58397e6c | 854 | if (type == DPDK_DEV_ETH) { |
d46285a2 | 855 | err = dpdk_eth_dev_init(dev); |
1b99bb05 MG |
856 | if (err) { |
857 | goto unlock; | |
858 | } | |
eff23640 | 859 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
585a5bea | 860 | } else { |
eff23640 | 861 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); |
e543851d ZB |
862 | /* Enable DPDK_DEV_VHOST device and set promiscuous mode flag. */ |
863 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
8a9562d2 | 864 | } |
8a9562d2 | 865 | |
eff23640 DDP |
866 | if (!dev->tx_q) { |
867 | err = ENOMEM; | |
868 | goto unlock; | |
869 | } | |
870 | ||
d46285a2 | 871 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 872 | |
95fb793a | 873 | unlock: |
d46285a2 | 874 | ovs_mutex_unlock(&dev->mutex); |
95fb793a | 875 | return err; |
876 | } | |
877 | ||
b83a2df1 MV |
878 | /* dev_name must be the prefix followed by a positive decimal number. |
879 | * (no leading + or - signs are allowed) */ | |
95fb793a | 880 | static int |
881 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
882 | unsigned int *port_no) | |
883 | { | |
884 | const char *cport; | |
885 | ||
886 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
887 | return ENODEV; | |
888 | } | |
889 | ||
890 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
891 | |
892 | if (str_to_uint(cport, 10, port_no)) { | |
893 | return 0; | |
894 | } else { | |
895 | return ENODEV; | |
896 | } | |
95fb793a | 897 | } |
898 | ||
7d1ced01 | 899 | static int |
53f50d24 | 900 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 901 | { |
d46285a2 DDP |
902 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
903 | const char *name = netdev->name; | |
7d1ced01 | 904 | int err; |
a0cb2d66 | 905 | |
1af27e8a DDP |
906 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
907 | * the file system. '/' or '\' would traverse directories, so they're not | |
908 | * acceptable in 'name'. */ | |
909 | if (strchr(name, '/') || strchr(name, '\\')) { | |
910 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
911 | "A valid name must not include '/' or '\\'", | |
912 | name); | |
913 | return EINVAL; | |
914 | } | |
915 | ||
7d1ced01 CL |
916 | ovs_mutex_lock(&dpdk_mutex); |
917 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 918 | * the socket is to be created, then register the socket. |
7d1ced01 | 919 | */ |
2d24d165 | 920 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 921 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 922 | |
2d24d165 CL |
923 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
924 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
925 | if (err) { |
926 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 927 | dev->vhost_id); |
e5c0f5a4 | 928 | } else { |
2d24d165 CL |
929 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
930 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
931 | dev->vhost_id, name); | |
932 | } | |
933 | err = netdev_dpdk_init(netdev, -1, DPDK_DEV_VHOST); | |
934 | ||
935 | ovs_mutex_unlock(&dpdk_mutex); | |
936 | return err; | |
937 | } | |
938 | ||
939 | static int | |
940 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
941 | { | |
942 | int err; | |
943 | ||
2d24d165 CL |
944 | ovs_mutex_lock(&dpdk_mutex); |
945 | err = netdev_dpdk_init(netdev, -1, DPDK_DEV_VHOST); | |
7d1ced01 | 946 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
947 | return err; |
948 | } | |
949 | ||
95fb793a | 950 | static int |
951 | netdev_dpdk_construct(struct netdev *netdev) | |
952 | { | |
953 | unsigned int port_no; | |
954 | int err; | |
955 | ||
95fb793a | 956 | /* Names always start with "dpdk" */ |
957 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
958 | if (err) { | |
959 | return err; | |
960 | } | |
961 | ||
962 | ovs_mutex_lock(&dpdk_mutex); | |
58397e6c | 963 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
8a9562d2 PS |
964 | ovs_mutex_unlock(&dpdk_mutex); |
965 | return err; | |
966 | } | |
967 | ||
968 | static void | |
d46285a2 | 969 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 970 | { |
d46285a2 | 971 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 972 | |
8d38823b | 973 | ovs_mutex_lock(&dpdk_mutex); |
8a9562d2 | 974 | ovs_mutex_lock(&dev->mutex); |
8d38823b | 975 | |
8a9562d2 | 976 | rte_eth_dev_stop(dev->port_id); |
9509913a IS |
977 | free(ovsrcu_get_protected(struct ingress_policer *, |
978 | &dev->ingress_policer)); | |
8a9562d2 | 979 | |
5a034064 | 980 | rte_free(dev->tx_q); |
417e7e66 | 981 | ovs_list_remove(&dev->list_node); |
8a9562d2 | 982 | dpdk_mp_put(dev->dpdk_mp); |
8d38823b MK |
983 | |
984 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 | 985 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 986 | } |
8a9562d2 | 987 | |
3f891bbe DDP |
988 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
989 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
990 | * deadlock, none of the mutexes must be held while calling this function. */ | |
991 | static int | |
c1ff66ac CL |
992 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
993 | char *vhost_id) | |
3f891bbe DDP |
994 | OVS_EXCLUDED(dpdk_mutex) |
995 | OVS_EXCLUDED(dev->mutex) | |
996 | { | |
c1ff66ac | 997 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
998 | } |
999 | ||
58397e6c | 1000 | static void |
d46285a2 | 1001 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1002 | { |
d46285a2 | 1003 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1004 | char *vhost_id; |
58397e6c | 1005 | |
8d38823b MK |
1006 | ovs_mutex_lock(&dpdk_mutex); |
1007 | ovs_mutex_lock(&dev->mutex); | |
1008 | ||
c62da695 | 1009 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1010 | if (netdev_dpdk_get_vid(dev) >= 0 |
1011 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1012 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1013 | netdev->name); |
58be5c0e MK |
1014 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1015 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1016 | } |
1017 | ||
9509913a IS |
1018 | free(ovsrcu_get_protected(struct ingress_policer *, |
1019 | &dev->ingress_policer)); | |
9509913a | 1020 | |
6019cb63 | 1021 | rte_free(dev->tx_q); |
417e7e66 | 1022 | ovs_list_remove(&dev->list_node); |
58397e6c | 1023 | dpdk_mp_put(dev->dpdk_mp); |
8d38823b | 1024 | |
2d24d165 | 1025 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1026 | |
8d38823b | 1027 | ovs_mutex_unlock(&dev->mutex); |
58397e6c | 1028 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1029 | |
c1ff66ac | 1030 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1031 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1032 | netdev->name, vhost_id); | |
c1ff66ac CL |
1033 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1034 | /* OVS server mode - remove this socket from list for deletion */ | |
1035 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1036 | } |
c1ff66ac | 1037 | free(vhost_id); |
8a9562d2 PS |
1038 | } |
1039 | ||
1040 | static void | |
d46285a2 | 1041 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1042 | { |
d46285a2 | 1043 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1044 | |
d46285a2 | 1045 | rte_free(dev); |
8a9562d2 PS |
1046 | } |
1047 | ||
1048 | static int | |
a14b8947 | 1049 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1050 | { |
a14b8947 | 1051 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1052 | |
1053 | ovs_mutex_lock(&dev->mutex); | |
1054 | ||
050c60bf | 1055 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1056 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1057 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1058 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1059 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1060 | |
1061 | if (dev->type == DPDK_DEV_ETH) { | |
1062 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1063 | dev->requested_rxq_size); | |
1064 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1065 | dev->rxq_size); | |
1066 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1067 | dev->requested_txq_size); | |
1068 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1069 | dev->txq_size); | |
1070 | } | |
8a9562d2 PS |
1071 | ovs_mutex_unlock(&dev->mutex); |
1072 | ||
1073 | return 0; | |
1074 | } | |
1075 | ||
c3d062a7 CL |
1076 | static void |
1077 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1078 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1079 | { |
050c60bf | 1080 | int new_n_rxq; |
a14b8947 | 1081 | |
050c60bf DDP |
1082 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", dev->requested_n_rxq), 1); |
1083 | if (new_n_rxq != dev->requested_n_rxq) { | |
1084 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1085 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1086 | } |
c3d062a7 CL |
1087 | } |
1088 | ||
b685696b CL |
1089 | static void |
1090 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1091 | const char *flag, int default_size, int *new_size) | |
1092 | { | |
1093 | int queue_size = smap_get_int(args, flag, default_size); | |
1094 | ||
1095 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1096 | || !is_pow2(queue_size)) { | |
1097 | queue_size = default_size; | |
1098 | } | |
1099 | ||
1100 | if (queue_size != *new_size) { | |
1101 | *new_size = queue_size; | |
1102 | netdev_request_reconfigure(netdev); | |
1103 | } | |
1104 | } | |
1105 | ||
c3d062a7 CL |
1106 | static int |
1107 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args) | |
1108 | { | |
1109 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
b614c894 IM |
1110 | bool rx_fc_en, tx_fc_en, autoneg; |
1111 | enum rte_eth_fc_mode fc_mode; | |
1112 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1113 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1114 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1115 | }; | |
c3d062a7 CL |
1116 | |
1117 | ovs_mutex_lock(&dev->mutex); | |
1118 | ||
1119 | dpdk_set_rxq_config(dev, args); | |
1120 | ||
b685696b CL |
1121 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1122 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1123 | &dev->requested_rxq_size); | |
1124 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1125 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1126 | &dev->requested_txq_size); | |
1127 | ||
c3d062a7 CL |
1128 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1129 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1130 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1131 | |
b614c894 IM |
1132 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1133 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1134 | dev->fc_conf.mode = fc_mode; | |
1135 | dev->fc_conf.autoneg = autoneg; | |
1136 | dpdk_eth_flow_ctrl_setup(dev); | |
1137 | } | |
9fd39370 | 1138 | |
c3d062a7 CL |
1139 | ovs_mutex_unlock(&dev->mutex); |
1140 | ||
1141 | return 0; | |
1142 | } | |
1143 | ||
1144 | static int | |
1145 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args) | |
1146 | { | |
1147 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1148 | ||
1149 | ovs_mutex_lock(&dev->mutex); | |
1150 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1151 | ovs_mutex_unlock(&dev->mutex); |
1152 | ||
1153 | return 0; | |
1154 | } | |
1155 | ||
c1ff66ac | 1156 | static int |
2d24d165 CL |
1157 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
1158 | const struct smap *args) | |
c1ff66ac CL |
1159 | { |
1160 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1161 | const char *path; | |
1162 | ||
6881885a | 1163 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1164 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1165 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1166 | if (path && strcmp(path, dev->vhost_id)) { |
1167 | strcpy(dev->vhost_id, path); | |
c1ff66ac CL |
1168 | netdev_request_reconfigure(netdev); |
1169 | } | |
1170 | } | |
6881885a | 1171 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1172 | |
1173 | return 0; | |
1174 | } | |
1175 | ||
7dec44fe | 1176 | static int |
d46285a2 | 1177 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1178 | { |
d46285a2 | 1179 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1180 | |
d46285a2 | 1181 | return dev->socket_id; |
7dec44fe AW |
1182 | } |
1183 | ||
050c60bf | 1184 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1185 | static int |
050c60bf | 1186 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1187 | { |
d46285a2 | 1188 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1189 | |
d46285a2 | 1190 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1191 | |
050c60bf DDP |
1192 | if (dev->requested_n_txq == n_txq) { |
1193 | goto out; | |
4573fbd3 FL |
1194 | } |
1195 | ||
050c60bf DDP |
1196 | dev->requested_n_txq = n_txq; |
1197 | netdev_request_reconfigure(netdev); | |
58397e6c | 1198 | |
050c60bf | 1199 | out: |
d46285a2 | 1200 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1201 | return 0; |
58397e6c KT |
1202 | } |
1203 | ||
8a9562d2 PS |
1204 | static struct netdev_rxq * |
1205 | netdev_dpdk_rxq_alloc(void) | |
1206 | { | |
1207 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1208 | ||
eff23640 DDP |
1209 | if (rx) { |
1210 | return &rx->up; | |
1211 | } | |
1212 | ||
1213 | return NULL; | |
8a9562d2 PS |
1214 | } |
1215 | ||
1216 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1217 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1218 | { |
d46285a2 | 1219 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1220 | } |
1221 | ||
1222 | static int | |
d46285a2 | 1223 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1224 | { |
d46285a2 DDP |
1225 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1226 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1227 | |
d46285a2 DDP |
1228 | ovs_mutex_lock(&dev->mutex); |
1229 | rx->port_id = dev->port_id; | |
1230 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1231 | |
1232 | return 0; | |
1233 | } | |
1234 | ||
1235 | static void | |
d46285a2 | 1236 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1237 | { |
1238 | } | |
1239 | ||
1240 | static void | |
d46285a2 | 1241 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1242 | { |
d46285a2 | 1243 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1244 | |
1245 | rte_free(rx); | |
1246 | } | |
1247 | ||
819f13bd DDP |
1248 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1249 | * 'pkts', even in case of failure. | |
1250 | * | |
1251 | * Returns the number of packets that weren't transmitted. */ | |
1252 | static inline int | |
b59cc14e | 1253 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 1254 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 1255 | { |
1304f1f8 DDP |
1256 | uint32_t nb_tx = 0; |
1257 | ||
b59cc14e | 1258 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1259 | uint32_t ret; |
1260 | ||
b59cc14e | 1261 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1262 | if (!ret) { |
1263 | break; | |
1264 | } | |
1265 | ||
1266 | nb_tx += ret; | |
1267 | } | |
8a9562d2 | 1268 | |
b59cc14e | 1269 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 1270 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
1271 | * packet could come from a different mempool) */ |
1272 | int i; | |
1273 | ||
b59cc14e IM |
1274 | for (i = nb_tx; i < cnt; i++) { |
1275 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1276 | } |
8a9562d2 | 1277 | } |
819f13bd DDP |
1278 | |
1279 | return cnt - nb_tx; | |
8a9562d2 PS |
1280 | } |
1281 | ||
f3926f29 IS |
1282 | static inline bool |
1283 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1284 | struct rte_mbuf *pkt, uint64_t time) | |
1285 | { | |
1286 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1287 | ||
1288 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1289 | e_RTE_METER_GREEN; | |
1290 | } | |
1291 | ||
1292 | static int | |
1293 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
1294 | struct rte_mbuf **pkts, int pkt_cnt) | |
1295 | { | |
1296 | int i = 0; | |
1297 | int cnt = 0; | |
1298 | struct rte_mbuf *pkt = NULL; | |
1299 | uint64_t current_time = rte_rdtsc(); | |
1300 | ||
1301 | for (i = 0; i < pkt_cnt; i++) { | |
1302 | pkt = pkts[i]; | |
1303 | /* Handle current packet */ | |
1304 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1305 | if (cnt != i) { | |
1306 | pkts[cnt] = pkt; | |
1307 | } | |
1308 | cnt++; | |
1309 | } else { | |
1310 | rte_pktmbuf_free(pkt); | |
1311 | } | |
1312 | } | |
1313 | ||
1314 | return cnt; | |
1315 | } | |
1316 | ||
9509913a IS |
1317 | static int |
1318 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
1319 | int pkt_cnt) | |
1320 | { | |
1321 | int cnt = 0; | |
1322 | ||
1323 | rte_spinlock_lock(&policer->policer_lock); | |
1324 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, pkt_cnt); | |
1325 | rte_spinlock_unlock(&policer->policer_lock); | |
1326 | ||
1327 | return cnt; | |
1328 | } | |
1329 | ||
58397e6c | 1330 | static bool |
0a0f39df | 1331 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1332 | { |
0a0f39df | 1333 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1334 | } |
1335 | ||
d6e3feb5 | 1336 | static inline void |
1337 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1338 | unsigned int packet_size) | |
1339 | { | |
1340 | /* Hard-coded search for the size bucket. */ | |
1341 | if (packet_size < 256) { | |
1342 | if (packet_size >= 128) { | |
1343 | stats->rx_128_to_255_packets++; | |
1344 | } else if (packet_size <= 64) { | |
1345 | stats->rx_1_to_64_packets++; | |
1346 | } else { | |
1347 | stats->rx_65_to_127_packets++; | |
1348 | } | |
1349 | } else { | |
1350 | if (packet_size >= 1523) { | |
1351 | stats->rx_1523_to_max_packets++; | |
1352 | } else if (packet_size >= 1024) { | |
1353 | stats->rx_1024_to_1522_packets++; | |
1354 | } else if (packet_size < 512) { | |
1355 | stats->rx_256_to_511_packets++; | |
1356 | } else { | |
1357 | stats->rx_512_to_1023_packets++; | |
1358 | } | |
1359 | } | |
1360 | } | |
1361 | ||
9e3ddd45 TP |
1362 | static inline void |
1363 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1364 | struct dp_packet **packets, int count, |
1365 | int dropped) | |
9e3ddd45 TP |
1366 | { |
1367 | int i; | |
d6e3feb5 | 1368 | unsigned int packet_size; |
9e3ddd45 TP |
1369 | struct dp_packet *packet; |
1370 | ||
1371 | stats->rx_packets += count; | |
9509913a | 1372 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1373 | for (i = 0; i < count; i++) { |
1374 | packet = packets[i]; | |
d6e3feb5 | 1375 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1376 | |
d6e3feb5 | 1377 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1378 | /* This only protects the following multicast counting from |
1379 | * too short packets, but it does not stop the packet from | |
1380 | * further processing. */ | |
1381 | stats->rx_errors++; | |
1382 | stats->rx_length_errors++; | |
1383 | continue; | |
1384 | } | |
1385 | ||
d6e3feb5 | 1386 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1387 | ||
9e3ddd45 TP |
1388 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1389 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1390 | stats->multicast++; | |
1391 | } | |
1392 | ||
d6e3feb5 | 1393 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1394 | } |
1395 | } | |
1396 | ||
58397e6c KT |
1397 | /* |
1398 | * The receive path for the vhost port is the TX path out from guest. | |
1399 | */ | |
1400 | static int | |
d46285a2 | 1401 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1402 | struct dp_packet_batch *batch) |
58397e6c | 1403 | { |
d46285a2 | 1404 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
d46285a2 | 1405 | int qid = rxq->queue_id; |
9509913a | 1406 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1407 | uint16_t nb_rx = 0; |
9509913a | 1408 | uint16_t dropped = 0; |
58397e6c | 1409 | |
0a0f39df | 1410 | if (OVS_UNLIKELY(!is_vhost_running(dev) |
e543851d | 1411 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1412 | return EAGAIN; |
1413 | } | |
1414 | ||
0a0f39df CL |
1415 | nb_rx = rte_vhost_dequeue_burst(netdev_dpdk_get_vid(dev), |
1416 | qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
d46285a2 | 1417 | dev->dpdk_mp->mp, |
64839cf4 | 1418 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1419 | NETDEV_MAX_BURST); |
58397e6c KT |
1420 | if (!nb_rx) { |
1421 | return EAGAIN; | |
1422 | } | |
1423 | ||
9509913a IS |
1424 | if (policer) { |
1425 | dropped = nb_rx; | |
64839cf4 WT |
1426 | nb_rx = ingress_policer_run(policer, |
1427 | (struct rte_mbuf **) batch->packets, | |
1428 | nb_rx); | |
9509913a IS |
1429 | dropped -= nb_rx; |
1430 | } | |
1431 | ||
d46285a2 | 1432 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1433 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1434 | nb_rx, dropped); | |
d46285a2 | 1435 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1436 | |
64839cf4 | 1437 | batch->count = (int) nb_rx; |
58397e6c KT |
1438 | return 0; |
1439 | } | |
1440 | ||
8a9562d2 | 1441 | static int |
64839cf4 | 1442 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1443 | { |
d46285a2 DDP |
1444 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1445 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1446 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1447 | int nb_rx; |
9509913a | 1448 | int dropped = 0; |
8a9562d2 | 1449 | |
d46285a2 | 1450 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1451 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1452 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1453 | if (!nb_rx) { |
1454 | return EAGAIN; | |
1455 | } | |
1456 | ||
9509913a IS |
1457 | if (policer) { |
1458 | dropped = nb_rx; | |
64839cf4 | 1459 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 1460 | (struct rte_mbuf **) batch->packets, |
64839cf4 | 1461 | nb_rx); |
9509913a IS |
1462 | dropped -= nb_rx; |
1463 | } | |
1464 | ||
1465 | /* Update stats to reflect dropped packets */ | |
1466 | if (OVS_UNLIKELY(dropped)) { | |
1467 | rte_spinlock_lock(&dev->stats_lock); | |
1468 | dev->stats.rx_dropped += dropped; | |
1469 | rte_spinlock_unlock(&dev->stats_lock); | |
1470 | } | |
1471 | ||
64839cf4 | 1472 | batch->count = nb_rx; |
8a9562d2 PS |
1473 | |
1474 | return 0; | |
1475 | } | |
1476 | ||
0bf765f7 | 1477 | static inline int |
78bd47cf DDP |
1478 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
1479 | int cnt) | |
0bf765f7 | 1480 | { |
78bd47cf | 1481 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 1482 | |
78bd47cf DDP |
1483 | if (qos_conf) { |
1484 | rte_spinlock_lock(&qos_conf->lock); | |
1485 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt); | |
1486 | rte_spinlock_unlock(&qos_conf->lock); | |
0bf765f7 IS |
1487 | } |
1488 | ||
1489 | return cnt; | |
1490 | } | |
1491 | ||
c6ec9d17 IM |
1492 | static int |
1493 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1494 | int pkt_cnt) | |
1495 | { | |
1496 | int i = 0; | |
1497 | int cnt = 0; | |
1498 | struct rte_mbuf *pkt; | |
1499 | ||
1500 | for (i = 0; i < pkt_cnt; i++) { | |
1501 | pkt = pkts[i]; | |
1502 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
1503 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
1504 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
1505 | rte_pktmbuf_free(pkt); | |
1506 | continue; | |
1507 | } | |
1508 | ||
1509 | if (OVS_UNLIKELY(i != cnt)) { | |
1510 | pkts[cnt] = pkt; | |
1511 | } | |
1512 | cnt++; | |
1513 | } | |
1514 | ||
1515 | return cnt; | |
1516 | } | |
1517 | ||
9e3ddd45 TP |
1518 | static inline void |
1519 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1520 | struct dp_packet **packets, | |
1521 | int attempted, | |
1522 | int dropped) | |
1523 | { | |
1524 | int i; | |
1525 | int sent = attempted - dropped; | |
1526 | ||
1527 | stats->tx_packets += sent; | |
1528 | stats->tx_dropped += dropped; | |
1529 | ||
1530 | for (i = 0; i < sent; i++) { | |
1531 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1532 | } | |
1533 | } | |
1534 | ||
58397e6c | 1535 | static void |
4573fbd3 | 1536 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 1537 | struct dp_packet **pkts, int cnt) |
58397e6c | 1538 | { |
d46285a2 | 1539 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1540 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1541 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 1542 | unsigned int dropped = 0; |
dd52de45 | 1543 | int i, retries = 0; |
58397e6c | 1544 | |
81acebda | 1545 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1546 | |
0a0f39df | 1547 | if (OVS_UNLIKELY(!is_vhost_running(dev) || qid < 0 |
e543851d | 1548 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1549 | rte_spinlock_lock(&dev->stats_lock); |
1550 | dev->stats.tx_dropped+= cnt; | |
1551 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1552 | goto out; |
58397e6c KT |
1553 | } |
1554 | ||
d46285a2 | 1555 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1556 | |
c6ec9d17 | 1557 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 1558 | /* Check has QoS has been configured for the netdev */ |
78bd47cf | 1559 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt); |
c6ec9d17 | 1560 | dropped = total_pkts - cnt; |
0bf765f7 | 1561 | |
95e9881f | 1562 | do { |
4573fbd3 | 1563 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1564 | unsigned int tx_pkts; |
1565 | ||
0a0f39df | 1566 | tx_pkts = rte_vhost_enqueue_burst(netdev_dpdk_get_vid(dev), |
c6ec9d17 | 1567 | vhost_qid, cur_pkts, cnt); |
95e9881f KT |
1568 | if (OVS_LIKELY(tx_pkts)) { |
1569 | /* Packets have been sent.*/ | |
1570 | cnt -= tx_pkts; | |
31871ee3 | 1571 | /* Prepare for possible retry.*/ |
95e9881f KT |
1572 | cur_pkts = &cur_pkts[tx_pkts]; |
1573 | } else { | |
31871ee3 KT |
1574 | /* No packets sent - do not retry.*/ |
1575 | break; | |
95e9881f | 1576 | } |
c6ec9d17 | 1577 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1578 | |
d46285a2 | 1579 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1580 | |
d46285a2 | 1581 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 1582 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 1583 | cnt + dropped); |
d46285a2 | 1584 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1585 | |
1586 | out: | |
c6ec9d17 | 1587 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 1588 | dp_packet_delete(pkts[i]); |
58397e6c KT |
1589 | } |
1590 | } | |
1591 | ||
8a9562d2 PS |
1592 | /* Tx function. Transmit packets indefinitely */ |
1593 | static void | |
64839cf4 | 1594 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1595 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1596 | { |
bce01e3a | 1597 | #if !defined(__CHECKER__) && !defined(_WIN32) |
64839cf4 | 1598 | const size_t PKT_ARRAY_SIZE = batch->count; |
bce01e3a EJ |
1599 | #else |
1600 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1601 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1602 | #endif |
8a9562d2 | 1603 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 1604 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
175cf4de RW |
1605 | int dropped = 0; |
1606 | int newcnt = 0; | |
1607 | int i; | |
8a9562d2 | 1608 | |
7d6d1a40 WT |
1609 | dp_packet_batch_apply_cutlen(batch); |
1610 | ||
64839cf4 WT |
1611 | for (i = 0; i < batch->count; i++) { |
1612 | int size = dp_packet_size(batch->packets[i]); | |
95fb793a | 1613 | |
f98d7864 | 1614 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c | 1615 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
58be5c0e | 1616 | (int) size, dev->max_packet_len); |
f4fd623c | 1617 | |
175cf4de | 1618 | dropped++; |
f4fd623c DDP |
1619 | continue; |
1620 | } | |
8a9562d2 | 1621 | |
2391135c | 1622 | pkts[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1623 | |
2391135c | 1624 | if (!pkts[newcnt]) { |
64839cf4 | 1625 | dropped += batch->count - i; |
175cf4de | 1626 | break; |
f4fd623c DDP |
1627 | } |
1628 | ||
1629 | /* We have to do a copy for now */ | |
2391135c | 1630 | memcpy(rte_pktmbuf_mtod(pkts[newcnt], void *), |
64839cf4 | 1631 | dp_packet_data(batch->packets[i]), size); |
f4fd623c | 1632 | |
2391135c MK |
1633 | rte_pktmbuf_data_len(pkts[newcnt]) = size; |
1634 | rte_pktmbuf_pkt_len(pkts[newcnt]) = size; | |
f4fd623c DDP |
1635 | |
1636 | newcnt++; | |
1637 | } | |
8a9562d2 | 1638 | |
58397e6c | 1639 | if (dev->type == DPDK_DEV_VHOST) { |
2391135c | 1640 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, |
dd52de45 | 1641 | newcnt); |
58397e6c | 1642 | } else { |
0bf765f7 IS |
1643 | unsigned int qos_pkts = newcnt; |
1644 | ||
1645 | /* Check if QoS has been configured for this netdev. */ | |
78bd47cf | 1646 | newcnt = netdev_dpdk_qos_run(dev, pkts, newcnt); |
0bf765f7 IS |
1647 | |
1648 | dropped += qos_pkts - newcnt; | |
819f13bd | 1649 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, newcnt); |
58397e6c | 1650 | } |
db73f716 | 1651 | |
0bf765f7 IS |
1652 | if (OVS_UNLIKELY(dropped)) { |
1653 | rte_spinlock_lock(&dev->stats_lock); | |
1654 | dev->stats.tx_dropped += dropped; | |
1655 | rte_spinlock_unlock(&dev->stats_lock); | |
1656 | } | |
8a9562d2 PS |
1657 | } |
1658 | ||
58397e6c | 1659 | static int |
64839cf4 WT |
1660 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1661 | struct dp_packet_batch *batch, | |
324c8374 | 1662 | bool may_steal, bool concurrent_txq OVS_UNUSED) |
58397e6c | 1663 | { |
58397e6c | 1664 | |
dd52de45 | 1665 | if (OVS_UNLIKELY(!may_steal || batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 WT |
1666 | dpdk_do_tx_copy(netdev, qid, batch); |
1667 | dp_packet_delete_batch(batch, may_steal); | |
58397e6c | 1668 | } else { |
64839cf4 | 1669 | dp_packet_batch_apply_cutlen(batch); |
dd52de45 | 1670 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
1671 | } |
1672 | return 0; | |
1673 | } | |
1674 | ||
7251515e DV |
1675 | static inline void |
1676 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
324c8374 IM |
1677 | struct dp_packet_batch *batch, bool may_steal, |
1678 | bool concurrent_txq) | |
8a9562d2 | 1679 | { |
324c8374 | 1680 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 1681 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
1682 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
1683 | } | |
1684 | ||
7251515e | 1685 | if (OVS_UNLIKELY(!may_steal || |
64839cf4 | 1686 | batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1687 | struct netdev *netdev = &dev->up; |
1688 | ||
64839cf4 WT |
1689 | dpdk_do_tx_copy(netdev, qid, batch); |
1690 | dp_packet_delete_batch(batch, may_steal); | |
8a9562d2 | 1691 | } else { |
6b094bf4 | 1692 | int dropped; |
64839cf4 | 1693 | int cnt = batch->count; |
2391135c | 1694 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 1695 | |
7d6d1a40 WT |
1696 | dp_packet_batch_apply_cutlen(batch); |
1697 | ||
2391135c | 1698 | cnt = netdev_dpdk_filter_packet_len(dev, pkts, cnt); |
78bd47cf | 1699 | cnt = netdev_dpdk_qos_run(dev, pkts, cnt); |
6b094bf4 | 1700 | dropped = batch->count - cnt; |
1b99bb05 | 1701 | |
819f13bd | 1702 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, cnt); |
8a9562d2 | 1703 | |
f4fd623c | 1704 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1705 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1706 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1707 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1708 | } |
8a9562d2 | 1709 | } |
a0cb2d66 | 1710 | |
324c8374 | 1711 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
1712 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
1713 | } | |
7251515e DV |
1714 | } |
1715 | ||
1716 | static int | |
1717 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
324c8374 IM |
1718 | struct dp_packet_batch *batch, bool may_steal, |
1719 | bool concurrent_txq) | |
7251515e DV |
1720 | { |
1721 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1722 | |
324c8374 | 1723 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e | 1724 | return 0; |
8a9562d2 PS |
1725 | } |
1726 | ||
1727 | static int | |
74ff3298 | 1728 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1729 | { |
1730 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1731 | ||
1732 | ovs_mutex_lock(&dev->mutex); | |
1733 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1734 | dev->hwaddr = mac; |
045c0d1a | 1735 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1736 | } |
1737 | ovs_mutex_unlock(&dev->mutex); | |
1738 | ||
1739 | return 0; | |
1740 | } | |
1741 | ||
1742 | static int | |
74ff3298 | 1743 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1744 | { |
1745 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1746 | ||
1747 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1748 | *mac = dev->hwaddr; |
8a9562d2 PS |
1749 | ovs_mutex_unlock(&dev->mutex); |
1750 | ||
1751 | return 0; | |
1752 | } | |
1753 | ||
1754 | static int | |
1755 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1756 | { | |
1757 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1758 | ||
1759 | ovs_mutex_lock(&dev->mutex); | |
1760 | *mtup = dev->mtu; | |
1761 | ovs_mutex_unlock(&dev->mutex); | |
1762 | ||
1763 | return 0; | |
1764 | } | |
1765 | ||
0072e931 MK |
1766 | static int |
1767 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
1768 | { | |
1769 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1770 | ||
1771 | if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
1772 | || mtu < ETHER_MIN_MTU) { | |
1773 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
1774 | return EINVAL; | |
1775 | } | |
1776 | ||
1777 | ovs_mutex_lock(&dev->mutex); | |
1778 | if (dev->requested_mtu != mtu) { | |
1779 | dev->requested_mtu = mtu; | |
1780 | netdev_request_reconfigure(netdev); | |
1781 | } | |
1782 | ovs_mutex_unlock(&dev->mutex); | |
1783 | ||
1784 | return 0; | |
1785 | } | |
1786 | ||
8a9562d2 | 1787 | static int |
d46285a2 | 1788 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 1789 | |
58397e6c KT |
1790 | static int |
1791 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1792 | struct netdev_stats *stats) | |
1793 | { | |
1794 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1795 | ||
1796 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 1797 | |
45d947c4 | 1798 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
1799 | /* Supported Stats */ |
1800 | stats->rx_packets += dev->stats.rx_packets; | |
1801 | stats->tx_packets += dev->stats.tx_packets; | |
9509913a | 1802 | stats->rx_dropped = dev->stats.rx_dropped; |
58397e6c | 1803 | stats->tx_dropped += dev->stats.tx_dropped; |
9e3ddd45 TP |
1804 | stats->multicast = dev->stats.multicast; |
1805 | stats->rx_bytes = dev->stats.rx_bytes; | |
1806 | stats->tx_bytes = dev->stats.tx_bytes; | |
1807 | stats->rx_errors = dev->stats.rx_errors; | |
1808 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 1809 | |
1810 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
1811 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
1812 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
1813 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
1814 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
1815 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
1816 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
1817 | ||
45d947c4 | 1818 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 1819 | |
58397e6c KT |
1820 | ovs_mutex_unlock(&dev->mutex); |
1821 | ||
1822 | return 0; | |
1823 | } | |
1824 | ||
d6e3feb5 | 1825 | static void |
1826 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
1827 | const struct rte_eth_xstat *xstats, |
1828 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 1829 | const unsigned int size) |
1830 | { | |
d6e3feb5 | 1831 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 1832 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1833 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1834 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1835 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1836 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1837 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1838 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1839 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1840 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1841 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1842 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1843 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 1844 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1845 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 1846 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1847 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1848 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1849 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1850 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1851 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1852 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1853 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1854 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1855 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1856 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1857 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 1858 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1859 | stats->tx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 1860 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1861 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 1862 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1863 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 1864 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1865 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 1866 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1867 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 1868 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1869 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 1870 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1871 | stats->rx_jabber_errors = xstats[i].value; |
1872 | } | |
1873 | } | |
1874 | } | |
1875 | ||
8a9562d2 PS |
1876 | static int |
1877 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1878 | { | |
1879 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1880 | struct rte_eth_stats rte_stats; | |
1881 | bool gg; | |
1882 | ||
1883 | netdev_dpdk_get_carrier(netdev, &gg); | |
1884 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 1885 | |
0a0f39df CL |
1886 | struct rte_eth_xstat *rte_xstats = NULL; |
1887 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
1888 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 1889 | |
1890 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
1891 | VLOG_ERR("Can't get ETH statistics for port: %i.", dev->port_id); | |
f9256822 | 1892 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 1893 | return EPROTO; |
1894 | } | |
1895 | ||
0a0f39df CL |
1896 | /* Get length of statistics */ |
1897 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
1898 | if (rte_xstats_len < 0) { | |
1899 | VLOG_WARN("Cannot get XSTATS values for port: %i", dev->port_id); | |
1900 | goto out; | |
1901 | } | |
1902 | /* Reserve memory for xstats names and values */ | |
1903 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
1904 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
1905 | ||
1906 | /* Retreive xstats names */ | |
1907 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
1908 | rte_xstats_names, | |
1909 | rte_xstats_len); | |
1910 | if (rte_xstats_new_len != rte_xstats_len) { | |
1911 | VLOG_WARN("Cannot get XSTATS names for port: %i.", dev->port_id); | |
1912 | goto out; | |
1913 | } | |
1914 | /* Retreive xstats values */ | |
1915 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
1916 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
1917 | rte_xstats_len); | |
1918 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
1919 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
1920 | rte_xstats_len); | |
d6e3feb5 | 1921 | } else { |
0a0f39df | 1922 | VLOG_WARN("Cannot get XSTATS values for port: %i.", dev->port_id); |
d6e3feb5 | 1923 | } |
8a9562d2 | 1924 | |
0a0f39df CL |
1925 | out: |
1926 | free(rte_xstats); | |
1927 | free(rte_xstats_names); | |
1928 | ||
2f9dd77f PS |
1929 | stats->rx_packets = rte_stats.ipackets; |
1930 | stats->tx_packets = rte_stats.opackets; | |
1931 | stats->rx_bytes = rte_stats.ibytes; | |
1932 | stats->tx_bytes = rte_stats.obytes; | |
9e3ddd45 TP |
1933 | /* DPDK counts imissed as errors, but count them here as dropped instead */ |
1934 | stats->rx_errors = rte_stats.ierrors - rte_stats.imissed; | |
2f9dd77f | 1935 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 1936 | |
45d947c4 | 1937 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 1938 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 1939 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 1940 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
1941 | |
1942 | /* These are the available DPDK counters for packets not received due to | |
1943 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 1944 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
1945 | stats->rx_missed_errors = rte_stats.imissed; |
1946 | ||
8a9562d2 PS |
1947 | ovs_mutex_unlock(&dev->mutex); |
1948 | ||
1949 | return 0; | |
1950 | } | |
1951 | ||
1952 | static int | |
d46285a2 | 1953 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 1954 | enum netdev_features *current, |
ca3d4f55 BX |
1955 | enum netdev_features *advertised, |
1956 | enum netdev_features *supported, | |
1957 | enum netdev_features *peer) | |
8a9562d2 | 1958 | { |
d46285a2 | 1959 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1960 | struct rte_eth_link link; |
1961 | ||
1962 | ovs_mutex_lock(&dev->mutex); | |
1963 | link = dev->link; | |
1964 | ovs_mutex_unlock(&dev->mutex); | |
1965 | ||
362ca396 | 1966 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
1967 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
1968 | *current = NETDEV_F_10MB_HD; |
1969 | } | |
362ca396 | 1970 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
1971 | *current = NETDEV_F_100MB_HD; |
1972 | } | |
362ca396 | 1973 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
1974 | *current = NETDEV_F_1GB_HD; |
1975 | } | |
1976 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 1977 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
1978 | *current = NETDEV_F_10MB_FD; |
1979 | } | |
362ca396 | 1980 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
1981 | *current = NETDEV_F_100MB_FD; |
1982 | } | |
362ca396 | 1983 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
1984 | *current = NETDEV_F_1GB_FD; |
1985 | } | |
362ca396 | 1986 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
1987 | *current = NETDEV_F_10GB_FD; |
1988 | } | |
1989 | } | |
1990 | ||
362ca396 | 1991 | if (link.link_autoneg) { |
1992 | *current |= NETDEV_F_AUTONEG; | |
1993 | } | |
1994 | ||
ca3d4f55 BX |
1995 | *advertised = *supported = *peer = 0; |
1996 | ||
8a9562d2 PS |
1997 | return 0; |
1998 | } | |
1999 | ||
9509913a IS |
2000 | static struct ingress_policer * |
2001 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2002 | { | |
2003 | struct ingress_policer *policer = NULL; | |
2004 | uint64_t rate_bytes; | |
2005 | uint64_t burst_bytes; | |
2006 | int err = 0; | |
2007 | ||
2008 | policer = xmalloc(sizeof *policer); | |
2009 | rte_spinlock_init(&policer->policer_lock); | |
2010 | ||
2011 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
2012 | rate_bytes = rate * 1000/8; | |
2013 | burst_bytes = burst * 1000/8; | |
2014 | ||
2015 | policer->app_srtcm_params.cir = rate_bytes; | |
2016 | policer->app_srtcm_params.cbs = burst_bytes; | |
2017 | policer->app_srtcm_params.ebs = 0; | |
2018 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2019 | &policer->app_srtcm_params); | |
58be5c0e | 2020 | if (err) { |
9509913a IS |
2021 | VLOG_ERR("Could not create rte meter for ingress policer"); |
2022 | return NULL; | |
2023 | } | |
2024 | ||
2025 | return policer; | |
2026 | } | |
2027 | ||
2028 | static int | |
2029 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2030 | uint32_t policer_burst) | |
2031 | { | |
2032 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2033 | struct ingress_policer *policer; | |
2034 | ||
2035 | /* Force to 0 if no rate specified, | |
2036 | * default to 8000 kbits if burst is 0, | |
2037 | * else stick with user-specified value. | |
2038 | */ | |
2039 | policer_burst = (!policer_rate ? 0 | |
2040 | : !policer_burst ? 8000 | |
2041 | : policer_burst); | |
2042 | ||
2043 | ovs_mutex_lock(&dev->mutex); | |
2044 | ||
2045 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2046 | &dev->ingress_policer); | |
2047 | ||
2048 | if (dev->policer_rate == policer_rate && | |
2049 | dev->policer_burst == policer_burst) { | |
2050 | /* Assume that settings haven't changed since we last set them. */ | |
2051 | ovs_mutex_unlock(&dev->mutex); | |
2052 | return 0; | |
2053 | } | |
2054 | ||
2055 | /* Destroy any existing ingress policer for the device if one exists */ | |
2056 | if (policer) { | |
2057 | ovsrcu_postpone(free, policer); | |
2058 | } | |
2059 | ||
2060 | if (policer_rate != 0) { | |
2061 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2062 | } else { | |
2063 | policer = NULL; | |
2064 | } | |
2065 | ovsrcu_set(&dev->ingress_policer, policer); | |
2066 | dev->policer_rate = policer_rate; | |
2067 | dev->policer_burst = policer_burst; | |
2068 | ovs_mutex_unlock(&dev->mutex); | |
2069 | ||
2070 | return 0; | |
2071 | } | |
2072 | ||
8a9562d2 PS |
2073 | static int |
2074 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2075 | { | |
2076 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2077 | int ifindex; | |
2078 | ||
2079 | ovs_mutex_lock(&dev->mutex); | |
2080 | ifindex = dev->port_id; | |
2081 | ovs_mutex_unlock(&dev->mutex); | |
2082 | ||
2083 | return ifindex; | |
2084 | } | |
2085 | ||
2086 | static int | |
d46285a2 | 2087 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2088 | { |
d46285a2 | 2089 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2090 | |
2091 | ovs_mutex_lock(&dev->mutex); | |
2092 | check_link_status(dev); | |
2093 | *carrier = dev->link.link_status; | |
58397e6c KT |
2094 | |
2095 | ovs_mutex_unlock(&dev->mutex); | |
2096 | ||
2097 | return 0; | |
2098 | } | |
2099 | ||
2100 | static int | |
d46285a2 | 2101 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2102 | { |
d46285a2 | 2103 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2104 | |
2105 | ovs_mutex_lock(&dev->mutex); | |
2106 | ||
0a0f39df | 2107 | if (is_vhost_running(dev)) { |
58397e6c KT |
2108 | *carrier = 1; |
2109 | } else { | |
2110 | *carrier = 0; | |
2111 | } | |
2112 | ||
8a9562d2 PS |
2113 | ovs_mutex_unlock(&dev->mutex); |
2114 | ||
2115 | return 0; | |
2116 | } | |
2117 | ||
2118 | static long long int | |
d46285a2 | 2119 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2120 | { |
d46285a2 | 2121 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2122 | long long int carrier_resets; |
2123 | ||
2124 | ovs_mutex_lock(&dev->mutex); | |
2125 | carrier_resets = dev->link_reset_cnt; | |
2126 | ovs_mutex_unlock(&dev->mutex); | |
2127 | ||
2128 | return carrier_resets; | |
2129 | } | |
2130 | ||
2131 | static int | |
d46285a2 | 2132 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2133 | long long int interval OVS_UNUSED) |
2134 | { | |
ee32150e | 2135 | return EOPNOTSUPP; |
8a9562d2 PS |
2136 | } |
2137 | ||
2138 | static int | |
2139 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2140 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2141 | enum netdev_flags *old_flagsp) |
2142 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
2143 | { |
2144 | int err; | |
2145 | ||
2146 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
2147 | return EINVAL; | |
2148 | } | |
2149 | ||
2150 | *old_flagsp = dev->flags; | |
2151 | dev->flags |= on; | |
2152 | dev->flags &= ~off; | |
2153 | ||
2154 | if (dev->flags == *old_flagsp) { | |
2155 | return 0; | |
2156 | } | |
2157 | ||
58397e6c KT |
2158 | if (dev->type == DPDK_DEV_ETH) { |
2159 | if (dev->flags & NETDEV_UP) { | |
2160 | err = rte_eth_dev_start(dev->port_id); | |
2161 | if (err) | |
2162 | return -err; | |
2163 | } | |
8a9562d2 | 2164 | |
58397e6c KT |
2165 | if (dev->flags & NETDEV_PROMISC) { |
2166 | rte_eth_promiscuous_enable(dev->port_id); | |
2167 | } | |
8a9562d2 | 2168 | |
58397e6c KT |
2169 | if (!(dev->flags & NETDEV_UP)) { |
2170 | rte_eth_dev_stop(dev->port_id); | |
2171 | } | |
314fb5ad | 2172 | |
2173 | netdev_change_seq_changed(&dev->up); | |
e543851d ZB |
2174 | } else { |
2175 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2176 | * running then change netdev's change_seq to trigger link state | |
2177 | * update. */ | |
e543851d ZB |
2178 | |
2179 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2180 | && is_vhost_running(dev)) { |
e543851d ZB |
2181 | netdev_change_seq_changed(&dev->up); |
2182 | ||
2183 | /* Clear statistics if device is getting up. */ | |
2184 | if (NETDEV_UP & on) { | |
2185 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 2186 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
2187 | rte_spinlock_unlock(&dev->stats_lock); |
2188 | } | |
2189 | } | |
8a9562d2 PS |
2190 | } |
2191 | ||
2192 | return 0; | |
2193 | } | |
2194 | ||
2195 | static int | |
d46285a2 | 2196 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2197 | enum netdev_flags off, enum netdev_flags on, |
2198 | enum netdev_flags *old_flagsp) | |
2199 | { | |
d46285a2 | 2200 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2201 | int error; |
2202 | ||
d46285a2 DDP |
2203 | ovs_mutex_lock(&dev->mutex); |
2204 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2205 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2206 | |
2207 | return error; | |
2208 | } | |
2209 | ||
2210 | static int | |
d46285a2 | 2211 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2212 | { |
d46285a2 | 2213 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2214 | struct rte_eth_dev_info dev_info; |
2215 | ||
7cd1261d | 2216 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2217 | return ENODEV; |
7cd1261d | 2218 | } |
8a9562d2 PS |
2219 | |
2220 | ovs_mutex_lock(&dev->mutex); | |
2221 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2222 | ovs_mutex_unlock(&dev->mutex); | |
2223 | ||
95fb793a | 2224 | smap_add_format(args, "port_no", "%d", dev->port_id); |
58be5c0e MK |
2225 | smap_add_format(args, "numa_id", "%d", |
2226 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
2227 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
2228 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2229 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2230 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2231 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2232 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
2233 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
2234 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
2235 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
2236 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2237 | ||
39c2baa9 | 2238 | if (dev_info.pci_dev) { |
2239 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2240 | dev_info.pci_dev->id.vendor_id); | |
2241 | smap_add_format(args, "pci-device_id", "0x%x", | |
2242 | dev_info.pci_dev->id.device_id); | |
2243 | } | |
8a9562d2 PS |
2244 | |
2245 | return 0; | |
2246 | } | |
2247 | ||
2248 | static void | |
2249 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2250 | OVS_REQUIRES(dev->mutex) | |
2251 | { | |
2252 | enum netdev_flags old_flags; | |
2253 | ||
2254 | if (admin_state) { | |
2255 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2256 | } else { | |
2257 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2258 | } | |
2259 | } | |
2260 | ||
2261 | static void | |
2262 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2263 | const char *argv[], void *aux OVS_UNUSED) | |
2264 | { | |
2265 | bool up; | |
2266 | ||
2267 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2268 | up = true; | |
2269 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2270 | up = false; | |
2271 | } else { | |
2272 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2273 | return; | |
2274 | } | |
2275 | ||
2276 | if (argc > 2) { | |
2277 | struct netdev *netdev = netdev_from_name(argv[1]); | |
2278 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
2279 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
2280 | ||
2281 | ovs_mutex_lock(&dpdk_dev->mutex); | |
2282 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
2283 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
2284 | ||
2285 | netdev_close(netdev); | |
2286 | } else { | |
2287 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2288 | netdev_close(netdev); | |
2289 | return; | |
2290 | } | |
2291 | } else { | |
2292 | struct netdev_dpdk *netdev; | |
2293 | ||
2294 | ovs_mutex_lock(&dpdk_mutex); | |
2295 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
2296 | ovs_mutex_lock(&netdev->mutex); | |
2297 | netdev_dpdk_set_admin_state__(netdev, up); | |
2298 | ovs_mutex_unlock(&netdev->mutex); | |
2299 | } | |
2300 | ovs_mutex_unlock(&dpdk_mutex); | |
2301 | } | |
2302 | unixctl_command_reply(conn, "OK"); | |
2303 | } | |
2304 | ||
58397e6c KT |
2305 | /* |
2306 | * Set virtqueue flags so that we do not receive interrupts. | |
2307 | */ | |
2308 | static void | |
0a0f39df | 2309 | set_irq_status(int vid) |
58397e6c | 2310 | { |
4573fbd3 FL |
2311 | uint32_t i; |
2312 | uint64_t idx; | |
2313 | ||
0a0f39df | 2314 | for (i = 0; i < rte_vhost_get_queue_num(vid); i++) { |
4573fbd3 | 2315 | idx = i * VIRTIO_QNUM; |
0a0f39df CL |
2316 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_RXQ, 0); |
2317 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_TXQ, 0); | |
4573fbd3 FL |
2318 | } |
2319 | } | |
2320 | ||
585a5bea IM |
2321 | /* |
2322 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2323 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2324 | */ |
2325 | static void | |
d46285a2 DDP |
2326 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2327 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2328 | { |
2329 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2330 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 2331 | |
eff23640 | 2332 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
2333 | |
2334 | for (i = 0; i < total_txqs; i++) { | |
2335 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2336 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2337 | enabled_queues[n_enabled++] = i; |
2338 | } | |
2339 | } | |
2340 | ||
2341 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2342 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2343 | n_enabled = 1; |
2344 | } | |
2345 | ||
2346 | k = 0; | |
2347 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2348 | if (dev->tx_q[i].map != i) { |
2349 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2350 | k = (k + 1) % n_enabled; |
2351 | } | |
2352 | } | |
2353 | ||
2d24d165 | 2354 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2355 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2356 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2357 | } |
2358 | ||
eff23640 | 2359 | free(enabled_queues); |
585a5bea | 2360 | } |
4573fbd3 | 2361 | |
58397e6c KT |
2362 | /* |
2363 | * A new virtio-net device is added to a vhost port. | |
2364 | */ | |
2365 | static int | |
0a0f39df | 2366 | new_device(int vid) |
58397e6c | 2367 | { |
d46285a2 | 2368 | struct netdev_dpdk *dev; |
58397e6c | 2369 | bool exists = false; |
db8f13b0 | 2370 | int newnode = 0; |
0a0f39df CL |
2371 | char ifname[IF_NAME_SZ]; |
2372 | ||
58be5c0e | 2373 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2374 | |
2375 | ovs_mutex_lock(&dpdk_mutex); | |
2376 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2377 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 2378 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2379 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
0a0f39df | 2380 | uint32_t qp_num = rte_vhost_get_queue_num(vid); |
db8f13b0 CL |
2381 | |
2382 | /* Get NUMA information */ | |
0a0f39df CL |
2383 | newnode = rte_vhost_get_numa_node(vid); |
2384 | if (newnode == -1) { | |
5b9bf9e0 | 2385 | #ifdef VHOST_NUMA |
db8f13b0 | 2386 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2387 | ifname); |
5b9bf9e0 | 2388 | #endif |
db8f13b0 | 2389 | newnode = dev->socket_id; |
db8f13b0 CL |
2390 | } |
2391 | ||
7f5f2bd0 IM |
2392 | if (dev->requested_n_txq != qp_num |
2393 | || dev->requested_n_rxq != qp_num | |
2394 | || dev->requested_socket_id != newnode) { | |
2395 | dev->requested_socket_id = newnode; | |
2396 | dev->requested_n_rxq = qp_num; | |
2397 | dev->requested_n_txq = qp_num; | |
2398 | netdev_request_reconfigure(&dev->up); | |
2399 | } else { | |
2400 | /* Reconfiguration not required. */ | |
2401 | dev->vhost_reconfigured = true; | |
2402 | } | |
81acebda | 2403 | |
0a0f39df | 2404 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2405 | exists = true; |
2406 | ||
58397e6c | 2407 | /* Disable notifications. */ |
0a0f39df | 2408 | set_irq_status(vid); |
e543851d | 2409 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2410 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2411 | break; |
2412 | } | |
c1ff66ac | 2413 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2414 | } |
2415 | ovs_mutex_unlock(&dpdk_mutex); | |
2416 | ||
2417 | if (!exists) { | |
0a0f39df | 2418 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2419 | |
2420 | return -1; | |
2421 | } | |
2422 | ||
0a0f39df CL |
2423 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2424 | ifname, newnode); | |
2425 | ||
58397e6c KT |
2426 | return 0; |
2427 | } | |
2428 | ||
f3ea2ad2 IM |
2429 | /* Clears mapping for all available queues of vhost interface. */ |
2430 | static void | |
2431 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2432 | OVS_REQUIRES(dev->mutex) | |
2433 | { | |
2434 | int i; | |
2435 | ||
81acebda | 2436 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
2437 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
2438 | } | |
2439 | } | |
2440 | ||
58397e6c KT |
2441 | /* |
2442 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2443 | * flag to stop any more packets from being sent or received to/from a VM and | |
2444 | * ensure all currently queued packets have been sent/received before removing | |
2445 | * the device. | |
2446 | */ | |
2447 | static void | |
0a0f39df | 2448 | destroy_device(int vid) |
58397e6c | 2449 | { |
d46285a2 | 2450 | struct netdev_dpdk *dev; |
afee281f | 2451 | bool exists = false; |
0a0f39df CL |
2452 | char ifname[IF_NAME_SZ]; |
2453 | ||
58be5c0e | 2454 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2455 | |
2456 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2457 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2458 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 2459 | |
d46285a2 | 2460 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
2461 | dev->vhost_reconfigured = false; |
2462 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 2463 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 2464 | |
e543851d | 2465 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2466 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 2467 | exists = true; |
afee281f | 2468 | break; |
58397e6c KT |
2469 | } |
2470 | } | |
afee281f | 2471 | |
58397e6c KT |
2472 | ovs_mutex_unlock(&dpdk_mutex); |
2473 | ||
0a0f39df | 2474 | if (exists) { |
afee281f KT |
2475 | /* |
2476 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2477 | * to NULL, before returning. | |
2478 | */ | |
2479 | ovsrcu_synchronize(); | |
2480 | /* | |
2481 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2482 | * put thread back into quiescent state before returning. | |
2483 | */ | |
2484 | ovsrcu_quiesce_start(); | |
0a0f39df | 2485 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 2486 | } else { |
0a0f39df | 2487 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 2488 | } |
58397e6c KT |
2489 | } |
2490 | ||
585a5bea | 2491 | static int |
0a0f39df | 2492 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 2493 | { |
d46285a2 | 2494 | struct netdev_dpdk *dev; |
585a5bea IM |
2495 | bool exists = false; |
2496 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
2497 | char ifname[IF_NAME_SZ]; |
2498 | ||
58be5c0e | 2499 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
2500 | |
2501 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2502 | return 0; | |
2503 | } | |
2504 | ||
2505 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2506 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 2507 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2508 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 2509 | if (enable) { |
d46285a2 | 2510 | dev->tx_q[qid].map = qid; |
585a5bea | 2511 | } else { |
d46285a2 | 2512 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2513 | } |
d46285a2 | 2514 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2515 | exists = true; |
d46285a2 | 2516 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2517 | break; |
2518 | } | |
c1ff66ac | 2519 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2520 | } |
2521 | ovs_mutex_unlock(&dpdk_mutex); | |
2522 | ||
2523 | if (exists) { | |
0a0f39df CL |
2524 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
2525 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 2526 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 2527 | } else { |
0a0f39df | 2528 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
2529 | return -1; |
2530 | } | |
2531 | ||
2532 | return 0; | |
2533 | } | |
2534 | ||
0a0f39df CL |
2535 | int |
2536 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 2537 | { |
0a0f39df | 2538 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
2539 | } |
2540 | ||
9509913a IS |
2541 | struct ingress_policer * |
2542 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
2543 | { | |
2544 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
2545 | } | |
2546 | ||
58397e6c KT |
2547 | /* |
2548 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
2549 | * configuration has been fully complete. | |
2550 | */ | |
bce01e3a | 2551 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
2552 | { |
2553 | .new_device = new_device, | |
2554 | .destroy_device = destroy_device, | |
585a5bea | 2555 | .vring_state_changed = vring_state_changed |
58397e6c KT |
2556 | }; |
2557 | ||
2558 | static void * | |
7d1ced01 | 2559 | start_vhost_loop(void *dummy OVS_UNUSED) |
58397e6c KT |
2560 | { |
2561 | pthread_detach(pthread_self()); | |
41987644 | 2562 | /* Put the vhost thread into quiescent state. */ |
618f44f7 | 2563 | ovsrcu_quiesce_start(); |
58397e6c KT |
2564 | rte_vhost_driver_session_start(); |
2565 | return NULL; | |
2566 | } | |
2567 | ||
2568 | static int | |
ecc1a34e | 2569 | netdev_dpdk_class_init(void) |
7d1ced01 | 2570 | { |
ecc1a34e DDP |
2571 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2572 | ||
2573 | /* This function can be called for different classes. The initialization | |
2574 | * needs to be done only once */ | |
2575 | if (ovsthread_once_start(&once)) { | |
2576 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
2577 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2578 | "[netdev] up|down", 1, 2, | |
2579 | netdev_dpdk_set_admin_state, NULL); | |
2580 | ||
2581 | ovsthread_once_done(&once); | |
2582 | } | |
362ca396 | 2583 | |
7d1ced01 CL |
2584 | return 0; |
2585 | } | |
2586 | ||
ecc1a34e DDP |
2587 | static int |
2588 | netdev_dpdk_vhost_class_init(void) | |
033e9df2 | 2589 | { |
ecc1a34e | 2590 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
033e9df2 | 2591 | |
ecc1a34e DDP |
2592 | /* This function can be called for different classes. The initialization |
2593 | * needs to be done only once */ | |
2594 | if (ovsthread_once_start(&once)) { | |
2595 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
2596 | rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4 | |
2597 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
04de404e CL |
2598 | | 1ULL << VIRTIO_NET_F_CSUM |
2599 | | 1ULL << VIRTIO_RING_F_INDIRECT_DESC); | |
ecc1a34e DDP |
2600 | ovs_thread_create("vhost_thread", start_vhost_loop, NULL); |
2601 | ||
2602 | ovsthread_once_done(&once); | |
2603 | } | |
2604 | ||
2605 | return 0; | |
033e9df2 DDP |
2606 | } |
2607 | ||
95fb793a | 2608 | /* Client Rings */ |
2609 | ||
95fb793a | 2610 | static int |
2611 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
2612 | unsigned int *eth_port_id) | |
2613 | { | |
2614 | struct dpdk_ring *ivshmem; | |
0c6f39e5 | 2615 | char *ring_name; |
95fb793a | 2616 | int err; |
2617 | ||
2618 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
eff23640 | 2619 | if (!ivshmem) { |
95fb793a | 2620 | return ENOMEM; |
2621 | } | |
2622 | ||
7251515e | 2623 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 2624 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 2625 | |
8f0a76c9 | 2626 | /* Create single producer tx ring, netdev does explicit locking. */ |
7251515e | 2627 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2628 | RING_F_SP_ENQ); |
0c6f39e5 | 2629 | free(ring_name); |
95fb793a | 2630 | if (ivshmem->cring_tx == NULL) { |
2631 | rte_free(ivshmem); | |
2632 | return ENOMEM; | |
2633 | } | |
2634 | ||
0c6f39e5 | 2635 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 2636 | |
8f0a76c9 | 2637 | /* Create single consumer rx ring, netdev does explicit locking. */ |
7251515e | 2638 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2639 | RING_F_SC_DEQ); |
0c6f39e5 | 2640 | free(ring_name); |
95fb793a | 2641 | if (ivshmem->cring_rx == NULL) { |
2642 | rte_free(ivshmem); | |
2643 | return ENOMEM; | |
2644 | } | |
2645 | ||
d7310583 DDP |
2646 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
2647 | &ivshmem->cring_tx, 1, SOCKET0); | |
2648 | ||
95fb793a | 2649 | if (err < 0) { |
2650 | rte_free(ivshmem); | |
2651 | return ENODEV; | |
2652 | } | |
2653 | ||
2654 | ivshmem->user_port_id = port_no; | |
2655 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
417e7e66 | 2656 | ovs_list_push_back(&dpdk_ring_list, &ivshmem->list_node); |
95fb793a | 2657 | |
2658 | *eth_port_id = ivshmem->eth_port_id; | |
2659 | return 0; | |
2660 | } | |
2661 | ||
2662 | static int | |
64839cf4 WT |
2663 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) |
2664 | OVS_REQUIRES(dpdk_mutex) | |
95fb793a | 2665 | { |
2666 | struct dpdk_ring *ivshmem; | |
2667 | unsigned int port_no; | |
2668 | int err = 0; | |
2669 | ||
2670 | /* Names always start with "dpdkr" */ | |
2671 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2672 | if (err) { | |
2673 | return err; | |
2674 | } | |
2675 | ||
58be5c0e | 2676 | /* Look through our list to find the device */ |
95fb793a | 2677 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { |
2678 | if (ivshmem->user_port_id == port_no) { | |
58397e6c | 2679 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e MK |
2680 | /* Really all that is needed */ |
2681 | *eth_port_id = ivshmem->eth_port_id; | |
95fb793a | 2682 | return 0; |
2683 | } | |
2684 | } | |
2685 | /* Need to create the device rings */ | |
2686 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2687 | } | |
2688 | ||
7251515e | 2689 | static int |
d46285a2 | 2690 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
324c8374 IM |
2691 | struct dp_packet_batch *batch, bool may_steal, |
2692 | bool concurrent_txq) | |
7251515e | 2693 | { |
d46285a2 | 2694 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b99bb05 MG |
2695 | unsigned i; |
2696 | ||
58be5c0e MK |
2697 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
2698 | * the rss hash field is clear. This is because the same mbuf may be | |
2699 | * modified by the consumer of the ring and return into the datapath | |
2700 | * without recalculating the RSS hash. */ | |
64839cf4 WT |
2701 | for (i = 0; i < batch->count; i++) { |
2702 | dp_packet_rss_invalidate(batch->packets[i]); | |
1b99bb05 | 2703 | } |
7251515e | 2704 | |
324c8374 | 2705 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e DV |
2706 | return 0; |
2707 | } | |
2708 | ||
95fb793a | 2709 | static int |
2710 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2711 | { | |
2712 | unsigned int port_no = 0; | |
2713 | int err = 0; | |
2714 | ||
95fb793a | 2715 | ovs_mutex_lock(&dpdk_mutex); |
2716 | ||
2717 | err = dpdk_ring_open(netdev->name, &port_no); | |
2718 | if (err) { | |
2719 | goto unlock_dpdk; | |
2720 | } | |
2721 | ||
58397e6c | 2722 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
95fb793a | 2723 | |
2724 | unlock_dpdk: | |
2725 | ovs_mutex_unlock(&dpdk_mutex); | |
2726 | return err; | |
2727 | } | |
2728 | ||
0bf765f7 IS |
2729 | /* QoS Functions */ |
2730 | ||
2731 | /* | |
2732 | * Initialize QoS configuration operations. | |
2733 | */ | |
2734 | static void | |
2735 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
2736 | { | |
2737 | conf->ops = ops; | |
78bd47cf | 2738 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
2739 | } |
2740 | ||
2741 | /* | |
2742 | * Search existing QoS operations in qos_ops and compare each set of | |
2743 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
2744 | * else return NULL | |
2745 | */ | |
2746 | static const struct dpdk_qos_ops * | |
2747 | qos_lookup_name(const char *name) | |
2748 | { | |
2749 | const struct dpdk_qos_ops *const *opsp; | |
2750 | ||
2751 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2752 | const struct dpdk_qos_ops *ops = *opsp; | |
2753 | if (!strcmp(name, ops->qos_name)) { | |
2754 | return ops; | |
2755 | } | |
2756 | } | |
2757 | return NULL; | |
2758 | } | |
2759 | ||
0bf765f7 IS |
2760 | static int |
2761 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
2762 | struct sset *types) | |
2763 | { | |
2764 | const struct dpdk_qos_ops *const *opsp; | |
2765 | ||
2766 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2767 | const struct dpdk_qos_ops *ops = *opsp; | |
2768 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
2769 | sset_add(types, ops->qos_name); | |
2770 | } | |
2771 | } | |
2772 | return 0; | |
2773 | } | |
2774 | ||
2775 | static int | |
d46285a2 | 2776 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
2777 | const char **typep, struct smap *details) |
2778 | { | |
d46285a2 | 2779 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 2780 | struct qos_conf *qos_conf; |
0bf765f7 IS |
2781 | int error = 0; |
2782 | ||
d46285a2 | 2783 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
2784 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
2785 | if (qos_conf) { | |
2786 | *typep = qos_conf->ops->qos_name; | |
2787 | error = (qos_conf->ops->qos_get | |
2788 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
2789 | } else { |
2790 | /* No QoS configuration set, return an empty string */ | |
2791 | *typep = ""; | |
0bf765f7 | 2792 | } |
d46285a2 | 2793 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2794 | |
2795 | return error; | |
2796 | } | |
2797 | ||
2798 | static int | |
78bd47cf DDP |
2799 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
2800 | const struct smap *details) | |
0bf765f7 | 2801 | { |
d46285a2 | 2802 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 2803 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 2804 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
2805 | int error = 0; |
2806 | ||
d46285a2 | 2807 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 2808 | |
78bd47cf | 2809 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 2810 | |
78bd47cf DDP |
2811 | new_ops = qos_lookup_name(type); |
2812 | ||
2813 | if (!new_ops || !new_ops->qos_construct) { | |
2814 | new_qos_conf = NULL; | |
2815 | if (type && type[0]) { | |
2816 | error = EOPNOTSUPP; | |
0bf765f7 | 2817 | } |
44975bb0 | 2818 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
2819 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
2820 | new_qos_conf = qos_conf; | |
0bf765f7 | 2821 | } else { |
78bd47cf | 2822 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
2823 | } |
2824 | ||
7ea266e9 | 2825 | if (error) { |
78bd47cf DDP |
2826 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
2827 | type, netdev->name, rte_strerror(error)); | |
2828 | } | |
2829 | ||
2830 | if (new_qos_conf != qos_conf) { | |
2831 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
2832 | if (qos_conf) { | |
2833 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
2834 | } | |
0bf765f7 IS |
2835 | } |
2836 | ||
d46285a2 | 2837 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 2838 | |
0bf765f7 IS |
2839 | return error; |
2840 | } | |
2841 | ||
2842 | /* egress-policer details */ | |
2843 | ||
2844 | struct egress_policer { | |
2845 | struct qos_conf qos_conf; | |
2846 | struct rte_meter_srtcm_params app_srtcm_params; | |
2847 | struct rte_meter_srtcm egress_meter; | |
2848 | }; | |
2849 | ||
78bd47cf DDP |
2850 | static void |
2851 | egress_policer_details_to_param(const struct smap *details, | |
2852 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 2853 | { |
78bd47cf DDP |
2854 | memset(params, 0, sizeof *params); |
2855 | params->cir = smap_get_ullong(details, "cir", 0); | |
2856 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
2857 | params->ebs = 0; | |
0bf765f7 IS |
2858 | } |
2859 | ||
2860 | static int | |
78bd47cf DDP |
2861 | egress_policer_qos_construct(const struct smap *details, |
2862 | struct qos_conf **conf) | |
0bf765f7 | 2863 | { |
0bf765f7 | 2864 | struct egress_policer *policer; |
0bf765f7 IS |
2865 | int err = 0; |
2866 | ||
0bf765f7 IS |
2867 | policer = xmalloc(sizeof *policer); |
2868 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 2869 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
0bf765f7 | 2870 | err = rte_meter_srtcm_config(&policer->egress_meter, |
78bd47cf DDP |
2871 | &policer->app_srtcm_params); |
2872 | if (!err) { | |
2873 | *conf = &policer->qos_conf; | |
2874 | } else { | |
7ea266e9 | 2875 | free(policer); |
78bd47cf | 2876 | *conf = NULL; |
7ea266e9 IS |
2877 | err = -err; |
2878 | } | |
0bf765f7 IS |
2879 | |
2880 | return err; | |
2881 | } | |
2882 | ||
2883 | static void | |
78bd47cf | 2884 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
2885 | { |
2886 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 2887 | qos_conf); |
0bf765f7 IS |
2888 | free(policer); |
2889 | } | |
2890 | ||
2891 | static int | |
78bd47cf | 2892 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 2893 | { |
78bd47cf DDP |
2894 | struct egress_policer *policer = |
2895 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
2896 | ||
2897 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
2898 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 2899 | |
0bf765f7 IS |
2900 | return 0; |
2901 | } | |
2902 | ||
78bd47cf DDP |
2903 | static bool |
2904 | egress_policer_qos_is_equal(const struct qos_conf *conf, const struct smap *details) | |
0bf765f7 | 2905 | { |
78bd47cf DDP |
2906 | struct egress_policer *policer = |
2907 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
2908 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 2909 | |
78bd47cf | 2910 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 2911 | |
78bd47cf | 2912 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
2913 | } |
2914 | ||
0bf765f7 | 2915 | static int |
78bd47cf | 2916 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt) |
0bf765f7 | 2917 | { |
0bf765f7 | 2918 | int cnt = 0; |
78bd47cf DDP |
2919 | struct egress_policer *policer = |
2920 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 2921 | |
f3926f29 | 2922 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, pkt_cnt); |
0bf765f7 IS |
2923 | |
2924 | return cnt; | |
2925 | } | |
2926 | ||
2927 | static const struct dpdk_qos_ops egress_policer_ops = { | |
2928 | "egress-policer", /* qos_name */ | |
2929 | egress_policer_qos_construct, | |
2930 | egress_policer_qos_destruct, | |
2931 | egress_policer_qos_get, | |
78bd47cf | 2932 | egress_policer_qos_is_equal, |
0bf765f7 IS |
2933 | egress_policer_run |
2934 | }; | |
2935 | ||
050c60bf DDP |
2936 | static int |
2937 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
2938 | { | |
2939 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2940 | int err = 0; | |
2941 | ||
050c60bf DDP |
2942 | ovs_mutex_lock(&dev->mutex); |
2943 | ||
2944 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 2945 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b CL |
2946 | && dev->mtu == dev->requested_mtu |
2947 | && dev->rxq_size == dev->requested_rxq_size | |
2948 | && dev->txq_size == dev->requested_txq_size) { | |
050c60bf DDP |
2949 | /* Reconfiguration is unnecessary */ |
2950 | ||
2951 | goto out; | |
2952 | } | |
2953 | ||
2954 | rte_eth_dev_stop(dev->port_id); | |
2955 | ||
0072e931 MK |
2956 | if (dev->mtu != dev->requested_mtu) { |
2957 | netdev_dpdk_mempool_configure(dev); | |
2958 | } | |
2959 | ||
050c60bf DDP |
2960 | netdev->n_txq = dev->requested_n_txq; |
2961 | netdev->n_rxq = dev->requested_n_rxq; | |
2962 | ||
b685696b CL |
2963 | dev->rxq_size = dev->requested_rxq_size; |
2964 | dev->txq_size = dev->requested_txq_size; | |
2965 | ||
050c60bf DDP |
2966 | rte_free(dev->tx_q); |
2967 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
2968 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
2969 | if (!dev->tx_q) { | |
2970 | err = ENOMEM; | |
2971 | } | |
050c60bf | 2972 | |
0072e931 MK |
2973 | netdev_change_seq_changed(netdev); |
2974 | ||
050c60bf | 2975 | out: |
050c60bf | 2976 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
2977 | return err; |
2978 | } | |
2979 | ||
2d24d165 CL |
2980 | static void |
2981 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) | |
2d24d165 | 2982 | OVS_REQUIRES(dev->mutex) |
050c60bf | 2983 | { |
2d24d165 CL |
2984 | dev->up.n_txq = dev->requested_n_txq; |
2985 | dev->up.n_rxq = dev->requested_n_rxq; | |
050c60bf | 2986 | |
81acebda IM |
2987 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
2988 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
2989 | dev->tx_q[0].map = 0; | |
2990 | } | |
2991 | ||
2992 | netdev_dpdk_remap_txqs(dev); | |
2993 | ||
0072e931 MK |
2994 | if (dev->requested_socket_id != dev->socket_id |
2995 | || dev->requested_mtu != dev->mtu) { | |
2996 | if (!netdev_dpdk_mempool_configure(dev)) { | |
2d24d165 | 2997 | netdev_change_seq_changed(&dev->up); |
db8f13b0 CL |
2998 | } |
2999 | } | |
3000 | ||
0a0f39df CL |
3001 | if (netdev_dpdk_get_vid(dev) >= 0) { |
3002 | dev->vhost_reconfigured = true; | |
81acebda | 3003 | } |
2d24d165 CL |
3004 | } |
3005 | ||
3006 | static int | |
3007 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
3008 | { | |
3009 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3010 | ||
2d24d165 | 3011 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 3012 | dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 3013 | ovs_mutex_unlock(&dev->mutex); |
2d24d165 CL |
3014 | return 0; |
3015 | } | |
3016 | ||
3017 | static int | |
3018 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3019 | { | |
3020 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3021 | int err = 0; | |
3022 | ||
2d24d165 CL |
3023 | ovs_mutex_lock(&dev->mutex); |
3024 | ||
3025 | dpdk_vhost_reconfigure_helper(dev); | |
81acebda | 3026 | |
c1ff66ac CL |
3027 | /* Configure vHost client mode if requested and if the following criteria |
3028 | * are met: | |
2d24d165 CL |
3029 | * 1. Device hasn't been registered yet. |
3030 | * 2. A path has been specified. | |
c1ff66ac CL |
3031 | */ |
3032 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 CL |
3033 | && strlen(dev->vhost_id)) { |
3034 | /* Register client-mode device */ | |
3035 | err = rte_vhost_driver_register(dev->vhost_id, | |
3036 | RTE_VHOST_USER_CLIENT); | |
c1ff66ac | 3037 | if (err) { |
2d24d165 CL |
3038 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3039 | dev->vhost_id); | |
c1ff66ac | 3040 | } else { |
2d24d165 CL |
3041 | /* Configuration successful */ |
3042 | dev->vhost_driver_flags |= RTE_VHOST_USER_CLIENT; | |
3043 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " | |
3044 | "using client socket '%s'", | |
3045 | dev->up.name, dev->vhost_id); | |
c1ff66ac CL |
3046 | } |
3047 | } | |
3048 | ||
050c60bf | 3049 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 3050 | |
0072e931 | 3051 | return 0; |
050c60bf DDP |
3052 | } |
3053 | ||
ecc1a34e | 3054 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
81acebda IM |
3055 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
3056 | GET_CARRIER, GET_STATS, \ | |
3057 | GET_FEATURES, GET_STATUS, \ | |
3058 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3059 | { \ |
3060 | NAME, \ | |
118c77b1 | 3061 | true, /* is_pmd */ \ |
ecc1a34e | 3062 | INIT, /* init */ \ |
95fb793a | 3063 | NULL, /* netdev_dpdk_run */ \ |
3064 | NULL, /* netdev_dpdk_wait */ \ | |
3065 | \ | |
3066 | netdev_dpdk_alloc, \ | |
3067 | CONSTRUCT, \ | |
58397e6c | 3068 | DESTRUCT, \ |
95fb793a | 3069 | netdev_dpdk_dealloc, \ |
3070 | netdev_dpdk_get_config, \ | |
81acebda | 3071 | SET_CONFIG, \ |
95fb793a | 3072 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3073 | NULL, /* build header */ \ |
3074 | NULL, /* push header */ \ | |
3075 | NULL, /* pop header */ \ | |
7dec44fe | 3076 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3077 | SET_TX_MULTIQ, \ |
95fb793a | 3078 | \ |
7251515e | 3079 | SEND, /* send */ \ |
95fb793a | 3080 | NULL, /* send_wait */ \ |
3081 | \ | |
3082 | netdev_dpdk_set_etheraddr, \ | |
3083 | netdev_dpdk_get_etheraddr, \ | |
3084 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3085 | netdev_dpdk_set_mtu, \ |
95fb793a | 3086 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3087 | GET_CARRIER, \ |
95fb793a | 3088 | netdev_dpdk_get_carrier_resets, \ |
3089 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
3090 | GET_STATS, \ |
3091 | GET_FEATURES, \ | |
95fb793a | 3092 | NULL, /* set_advertisements */ \ |
3093 | \ | |
9509913a | 3094 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3095 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3096 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3097 | netdev_dpdk_get_qos, \ |
3098 | netdev_dpdk_set_qos, \ | |
95fb793a | 3099 | NULL, /* get_queue */ \ |
3100 | NULL, /* set_queue */ \ | |
3101 | NULL, /* delete_queue */ \ | |
3102 | NULL, /* get_queue_stats */ \ | |
3103 | NULL, /* queue_dump_start */ \ | |
3104 | NULL, /* queue_dump_next */ \ | |
3105 | NULL, /* queue_dump_done */ \ | |
3106 | NULL, /* dump_queue_stats */ \ | |
3107 | \ | |
95fb793a | 3108 | NULL, /* set_in4 */ \ |
a8704b50 | 3109 | NULL, /* get_addr_list */ \ |
95fb793a | 3110 | NULL, /* add_router */ \ |
3111 | NULL, /* get_next_hop */ \ | |
58397e6c | 3112 | GET_STATUS, \ |
95fb793a | 3113 | NULL, /* arp_lookup */ \ |
3114 | \ | |
3115 | netdev_dpdk_update_flags, \ | |
050c60bf | 3116 | RECONFIGURE, \ |
95fb793a | 3117 | \ |
3118 | netdev_dpdk_rxq_alloc, \ | |
3119 | netdev_dpdk_rxq_construct, \ | |
3120 | netdev_dpdk_rxq_destruct, \ | |
3121 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3122 | RXQ_RECV, \ |
95fb793a | 3123 | NULL, /* rx_wait */ \ |
3124 | NULL, /* rxq_drain */ \ | |
3125 | } | |
8a9562d2 | 3126 | |
bce01e3a | 3127 | static const struct netdev_class dpdk_class = |
95fb793a | 3128 | NETDEV_DPDK_CLASS( |
3129 | "dpdk", | |
ecc1a34e | 3130 | netdev_dpdk_class_init, |
5496878c | 3131 | netdev_dpdk_construct, |
58397e6c | 3132 | netdev_dpdk_destruct, |
81acebda IM |
3133 | netdev_dpdk_set_config, |
3134 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3135 | netdev_dpdk_eth_send, |
3136 | netdev_dpdk_get_carrier, | |
3137 | netdev_dpdk_get_stats, | |
3138 | netdev_dpdk_get_features, | |
3139 | netdev_dpdk_get_status, | |
050c60bf | 3140 | netdev_dpdk_reconfigure, |
58397e6c | 3141 | netdev_dpdk_rxq_recv); |
95fb793a | 3142 | |
bce01e3a | 3143 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3144 | NETDEV_DPDK_CLASS( |
3145 | "dpdkr", | |
ecc1a34e | 3146 | netdev_dpdk_class_init, |
5496878c | 3147 | netdev_dpdk_ring_construct, |
58397e6c | 3148 | netdev_dpdk_destruct, |
c3d062a7 | 3149 | netdev_dpdk_ring_set_config, |
81acebda | 3150 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3151 | netdev_dpdk_ring_send, |
3152 | netdev_dpdk_get_carrier, | |
3153 | netdev_dpdk_get_stats, | |
3154 | netdev_dpdk_get_features, | |
3155 | netdev_dpdk_get_status, | |
050c60bf | 3156 | netdev_dpdk_reconfigure, |
58397e6c KT |
3157 | netdev_dpdk_rxq_recv); |
3158 | ||
53f50d24 | 3159 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3160 | NETDEV_DPDK_CLASS( |
3161 | "dpdkvhostuser", | |
ecc1a34e | 3162 | netdev_dpdk_vhost_class_init, |
53f50d24 | 3163 | netdev_dpdk_vhost_construct, |
58397e6c | 3164 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3165 | NULL, |
81acebda | 3166 | NULL, |
58397e6c KT |
3167 | netdev_dpdk_vhost_send, |
3168 | netdev_dpdk_vhost_get_carrier, | |
3169 | netdev_dpdk_vhost_get_stats, | |
3170 | NULL, | |
7251515e | 3171 | NULL, |
53f50d24 | 3172 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3173 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3174 | static const struct netdev_class dpdk_vhost_client_class = |
3175 | NETDEV_DPDK_CLASS( | |
3176 | "dpdkvhostuserclient", | |
ecc1a34e | 3177 | netdev_dpdk_vhost_class_init, |
2d24d165 CL |
3178 | netdev_dpdk_vhost_client_construct, |
3179 | netdev_dpdk_vhost_destruct, | |
3180 | netdev_dpdk_vhost_client_set_config, | |
3181 | NULL, | |
3182 | netdev_dpdk_vhost_send, | |
3183 | netdev_dpdk_vhost_get_carrier, | |
3184 | netdev_dpdk_vhost_get_stats, | |
3185 | NULL, | |
3186 | NULL, | |
3187 | netdev_dpdk_vhost_client_reconfigure, | |
3188 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 3189 | |
8a9562d2 PS |
3190 | void |
3191 | netdev_dpdk_register(void) | |
3192 | { | |
bab69409 AC |
3193 | netdev_register_provider(&dpdk_class); |
3194 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3195 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 3196 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 3197 | } |