]>
Commit | Line | Data |
---|---|---|
8a9562d2 | 1 | /* |
12d0d124 | 2 | * Copyright (c) 2014, 2015, 2016, 2017 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> |
f3e7ec25 MW |
25 | #include <linux/virtio_net.h> |
26 | #include <sys/socket.h> | |
27 | #include <linux/if.h> | |
01961bbd | 28 | |
5e925ccc | 29 | #include <rte_bus_pci.h> |
01961bbd DDP |
30 | #include <rte_config.h> |
31 | #include <rte_cycles.h> | |
32 | #include <rte_errno.h> | |
33 | #include <rte_eth_ring.h> | |
34 | #include <rte_ethdev.h> | |
35 | #include <rte_malloc.h> | |
36 | #include <rte_mbuf.h> | |
37 | #include <rte_meter.h> | |
fc56f5e0 | 38 | #include <rte_pci.h> |
f3e7ec25 | 39 | #include <rte_vhost.h> |
3eb8d4fa | 40 | #include <rte_version.h> |
8a9562d2 | 41 | |
7d1ced01 | 42 | #include "dirs.h" |
e14deea0 | 43 | #include "dp-packet.h" |
01961bbd | 44 | #include "dpdk.h" |
8a9562d2 | 45 | #include "dpif-netdev.h" |
e5c0f5a4 | 46 | #include "fatal-signal.h" |
8a9562d2 PS |
47 | #include "netdev-provider.h" |
48 | #include "netdev-vport.h" | |
49 | #include "odp-util.h" | |
eac84432 | 50 | #include "openvswitch/dynamic-string.h" |
25d436fb BW |
51 | #include "openvswitch/list.h" |
52 | #include "openvswitch/ofp-print.h" | |
53 | #include "openvswitch/vlog.h" | |
94143fc4 | 54 | #include "ovs-numa.h" |
8a9562d2 PS |
55 | #include "ovs-thread.h" |
56 | #include "ovs-rcu.h" | |
57 | #include "packets.h" | |
ee89ea7b | 58 | #include "openvswitch/shash.h" |
0bf765f7 | 59 | #include "smap.h" |
8a9562d2 PS |
60 | #include "sset.h" |
61 | #include "unaligned.h" | |
62 | #include "timeval.h" | |
63 | #include "unixctl.h" | |
8a9562d2 | 64 | |
f3e7ec25 MW |
65 | enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM}; |
66 | ||
05b49df6 | 67 | VLOG_DEFINE_THIS_MODULE(netdev_dpdk); |
8a9562d2 PS |
68 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
69 | ||
70 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
71 | ||
72 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
73 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
74 | ||
75 | /* | |
76 | * need to reserve tons of extra space in the mbufs so we can align the | |
77 | * DMA addresses to 4KB. | |
18f777b2 TP |
78 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
79 | * performance for standard Ethernet MTU. | |
8a9562d2 | 80 | */ |
58be5c0e MK |
81 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN \ |
82 | + (2 * VLAN_HEADER_LEN)) | |
4be4d22c MK |
83 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) |
84 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
58be5c0e MK |
85 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len) \ |
86 | - ETHER_HDR_LEN - ETHER_CRC_LEN) | |
31b88c97 SS |
87 | #define MBUF_SIZE(mtu) ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) \ |
88 | + sizeof(struct dp_packet) \ | |
89 | + RTE_PKTMBUF_HEADROOM), \ | |
90 | RTE_CACHE_LINE_SIZE) | |
4be4d22c | 91 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 92 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 93 | |
bc57ed90 IM |
94 | /* Min number of packets in the mempool. OVS tries to allocate a mempool with |
95 | * roughly estimated number of mbufs: if this fails (because the system doesn't | |
96 | * have enough hugepages) we keep halving the number until the allocation | |
97 | * succeeds or we reach MIN_NB_MBUF */ | |
da79ce2b DDP |
98 | #define MIN_NB_MBUF (4096 * 4) |
99 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
100 | ||
d6e3feb5 | 101 | /* |
102 | * DPDK XSTATS Counter names definition | |
103 | */ | |
104 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
105 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
106 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
107 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
108 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
109 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
110 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
111 | ||
112 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
113 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
114 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
115 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
116 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
117 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
118 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
119 | ||
d57f777f | 120 | #define XSTAT_RX_MULTICAST_PACKETS "rx_multicast_packets" |
d6e3feb5 | 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 | |
bb37956a IM |
143 | #define DPDK_ETH_PORT_ID_INVALID RTE_MAX_ETHPORTS |
144 | ||
5e925ccc MK |
145 | /* DPDK library uses uint16_t for port_id. */ |
146 | typedef uint16_t dpdk_port_t; | |
bb37956a | 147 | |
31871ee3 | 148 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 149 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 150 | |
8a9562d2 | 151 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
152 | .rxmode = { |
153 | .mq_mode = ETH_MQ_RX_RSS, | |
154 | .split_hdr_size = 0, | |
155 | .header_split = 0, /* Header Split disabled */ | |
156 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
157 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
158 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
159 | .hw_strip_crc = 0, | |
160 | }, | |
161 | .rx_adv_conf = { | |
162 | .rss_conf = { | |
163 | .rss_key = NULL, | |
543342a4 | 164 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 165 | }, |
a28ddd11 DDP |
166 | }, |
167 | .txmode = { | |
168 | .mq_mode = ETH_MQ_TX_NONE, | |
169 | }, | |
8a9562d2 PS |
170 | }; |
171 | ||
f3e7ec25 MW |
172 | /* |
173 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
174 | * configuration has been fully completed. | |
175 | */ | |
176 | static int new_device(int vid); | |
177 | static void destroy_device(int vid); | |
178 | static int vring_state_changed(int vid, uint16_t queue_id, int enable); | |
179 | static const struct vhost_device_ops virtio_net_device_ops = | |
180 | { | |
181 | .new_device = new_device, | |
182 | .destroy_device = destroy_device, | |
183 | .vring_state_changed = vring_state_changed, | |
184 | .features_changed = NULL | |
185 | }; | |
186 | ||
58f7c37b DDP |
187 | enum { DPDK_RING_SIZE = 256 }; |
188 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
189 | enum { DRAIN_TSC = 200000ULL }; |
190 | ||
58397e6c KT |
191 | enum dpdk_dev_type { |
192 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 193 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
194 | }; |
195 | ||
0bf765f7 IS |
196 | /* Quality of Service */ |
197 | ||
198 | /* An instance of a QoS configuration. Always associated with a particular | |
199 | * network device. | |
200 | * | |
201 | * Each QoS implementation subclasses this with whatever additional data it | |
202 | * needs. | |
203 | */ | |
204 | struct qos_conf { | |
205 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 206 | rte_spinlock_t lock; |
0bf765f7 IS |
207 | }; |
208 | ||
209 | /* A particular implementation of dpdk QoS operations. | |
210 | * | |
211 | * The functions below return 0 if successful or a positive errno value on | |
212 | * failure, except where otherwise noted. All of them must be provided, except | |
213 | * where otherwise noted. | |
214 | */ | |
215 | struct dpdk_qos_ops { | |
216 | ||
217 | /* Name of the QoS type */ | |
218 | const char *qos_name; | |
219 | ||
78bd47cf DDP |
220 | /* Called to construct a qos_conf object. The implementation should make |
221 | * the appropriate calls to configure QoS according to '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 DDP |
227 | * This function must return 0 if and only if it sets '*conf' to an |
228 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
229 | * |
230 | * For all QoS implementations it should always be non-null. | |
231 | */ | |
78bd47cf | 232 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
233 | |
234 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 235 | * 'qos_conf'. |
0bf765f7 IS |
236 | * |
237 | * For all QoS implementations it should always be non-null. | |
238 | */ | |
78bd47cf | 239 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 240 | |
78bd47cf | 241 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
242 | * |
243 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
244 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
245 | * (which is built as ovs-vswitchd.conf.db(8)). | |
246 | */ | |
78bd47cf | 247 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 248 | |
78bd47cf | 249 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
250 | * |
251 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
252 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
253 | * (which is built as ovs-vswitchd.conf.db(8)). | |
254 | * | |
78bd47cf | 255 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 256 | */ |
78bd47cf DDP |
257 | bool (*qos_is_equal)(const struct qos_conf *conf, |
258 | const struct smap *details); | |
0bf765f7 IS |
259 | |
260 | /* Modify an array of rte_mbufs. The modification is specific to | |
261 | * each qos implementation. | |
262 | * | |
263 | * The function should take and array of mbufs and an int representing | |
264 | * the current number of mbufs present in the array. | |
265 | * | |
266 | * After the function has performed a qos modification to the array of | |
267 | * mbufs it returns an int representing the number of mbufs now present in | |
268 | * the array. This value is can then be passed to the port send function | |
269 | * along with the modified array for transmission. | |
270 | * | |
271 | * For all QoS implementations it should always be non-null. | |
272 | */ | |
78bd47cf | 273 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
3e90f7d7 | 274 | int pkt_cnt, bool may_steal); |
0bf765f7 IS |
275 | }; |
276 | ||
277 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
278 | static const struct dpdk_qos_ops egress_policer_ops; | |
279 | ||
280 | /* | |
281 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
282 | * operations. | |
283 | */ | |
284 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
285 | &egress_policer_ops, | |
286 | NULL | |
287 | }; | |
288 | ||
c2adb102 IM |
289 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
290 | ||
8a9562d2 | 291 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 292 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 293 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 294 | |
c2adb102 IM |
295 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
296 | = OVS_MUTEX_INITIALIZER; | |
297 | ||
5a034064 AW |
298 | /* There should be one 'struct dpdk_tx_queue' created for |
299 | * each cpu core. */ | |
8a9562d2 | 300 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
301 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
302 | * from concurrent access. It is used only | |
303 | * if the queue is shared among different | |
324c8374 | 304 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
305 | int map; /* Mapping of configured vhost-user queues |
306 | * to enabled by guest. */ | |
8a9562d2 PS |
307 | }; |
308 | ||
95fb793a | 309 | /* dpdk has no way to remove dpdk ring ethernet devices |
310 | so we have to keep them around once they've been created | |
311 | */ | |
312 | ||
ca6ba700 | 313 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 314 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 315 | |
316 | struct dpdk_ring { | |
317 | /* For the client rings */ | |
318 | struct rte_ring *cring_tx; | |
319 | struct rte_ring *cring_rx; | |
b83a2df1 | 320 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
bb37956a | 321 | dpdk_port_t eth_port_id; /* ethernet device port id */ |
ca6ba700 | 322 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 323 | }; |
324 | ||
9509913a IS |
325 | struct ingress_policer { |
326 | struct rte_meter_srtcm_params app_srtcm_params; | |
327 | struct rte_meter_srtcm in_policer; | |
328 | rte_spinlock_t policer_lock; | |
329 | }; | |
330 | ||
1a2bb118 SC |
331 | enum dpdk_hw_ol_features { |
332 | NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0, | |
333 | }; | |
334 | ||
b2e72a9c IM |
335 | /* |
336 | * In order to avoid confusion in variables names, following naming convention | |
337 | * should be used, if possible: | |
338 | * | |
339 | * 'struct netdev' : 'netdev' | |
340 | * 'struct netdev_dpdk' : 'dev' | |
341 | * 'struct netdev_rxq' : 'rxq' | |
342 | * 'struct netdev_rxq_dpdk' : 'rx' | |
343 | * | |
344 | * Example: | |
345 | * struct netdev *netdev = netdev_from_name(name); | |
346 | * struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
347 | * | |
348 | * Also, 'netdev' should be used instead of 'dev->up', where 'netdev' was | |
349 | * already defined. | |
350 | */ | |
351 | ||
8a9562d2 | 352 | struct netdev_dpdk { |
23d4d53f BB |
353 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline0, |
354 | dpdk_port_t port_id; | |
355 | ||
356 | /* If true, device was attached by rte_eth_dev_attach(). */ | |
357 | bool attached; | |
358 | struct eth_addr hwaddr; | |
359 | int mtu; | |
360 | int socket_id; | |
361 | int buf_size; | |
362 | int max_packet_len; | |
363 | enum dpdk_dev_type type; | |
364 | enum netdev_flags flags; | |
365 | char *devargs; /* Device arguments for dpdk ports */ | |
366 | struct dpdk_tx_queue *tx_q; | |
367 | struct rte_eth_link link; | |
368 | int link_reset_cnt; | |
369 | /* 4 pad bytes here. */ | |
370 | ); | |
371 | ||
372 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline1, | |
373 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
24e78f93 | 374 | struct rte_mempool *mp; |
23d4d53f BB |
375 | |
376 | /* virtio identifier for vhost devices */ | |
377 | ovsrcu_index vid; | |
378 | ||
379 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
380 | bool vhost_reconfigured; | |
381 | /* 3 pad bytes here. */ | |
382 | ); | |
383 | ||
384 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
385 | /* Identifier used to distinguish vhost devices from each other. */ | |
386 | char vhost_id[PATH_MAX]; | |
387 | ); | |
388 | ||
389 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
390 | struct netdev up; | |
391 | /* In dpdk_list. */ | |
392 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); | |
393 | ||
394 | /* QoS configuration and lock for the device */ | |
395 | OVSRCU_TYPE(struct qos_conf *) qos_conf; | |
396 | ||
397 | /* Ingress Policer */ | |
398 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
399 | uint32_t policer_rate; | |
400 | uint32_t policer_burst; | |
401 | ); | |
402 | ||
403 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
404 | struct netdev_stats stats; | |
405 | /* Protects stats */ | |
406 | rte_spinlock_t stats_lock; | |
407 | /* 44 pad bytes here. */ | |
408 | ); | |
409 | ||
410 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
411 | /* The following properties cannot be changed when a device is running, | |
412 | * so we remember the request and update them next time | |
413 | * netdev_dpdk*_reconfigure() is called */ | |
414 | int requested_mtu; | |
415 | int requested_n_txq; | |
416 | int requested_n_rxq; | |
417 | int requested_rxq_size; | |
418 | int requested_txq_size; | |
419 | ||
420 | /* Number of rx/tx descriptors for physical devices */ | |
421 | int rxq_size; | |
422 | int txq_size; | |
423 | ||
424 | /* Socket ID detected when vHost device is brought up */ | |
425 | int requested_socket_id; | |
426 | ||
427 | /* Denotes whether vHost port is client/server mode */ | |
428 | uint64_t vhost_driver_flags; | |
429 | ||
430 | /* DPDK-ETH Flow control */ | |
431 | struct rte_eth_fc_conf fc_conf; | |
432 | ||
433 | /* DPDK-ETH hardware offload features, | |
434 | * from the enum set 'dpdk_hw_ol_features' */ | |
435 | uint32_t hw_ol_features; | |
436 | ); | |
971f4b39 MW |
437 | |
438 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
439 | /* Names of all XSTATS counters */ | |
440 | struct rte_eth_xstat_name *rte_xstats_names; | |
441 | int rte_xstats_names_size; | |
442 | int rte_xstats_ids_size; | |
443 | uint64_t *rte_xstats_ids; | |
444 | ); | |
8a9562d2 PS |
445 | }; |
446 | ||
447 | struct netdev_rxq_dpdk { | |
448 | struct netdev_rxq up; | |
bb37956a | 449 | dpdk_port_t port_id; |
8a9562d2 PS |
450 | }; |
451 | ||
f3e7ec25 MW |
452 | static void netdev_dpdk_destruct(struct netdev *netdev); |
453 | static void netdev_dpdk_vhost_destruct(struct netdev *netdev); | |
8a9562d2 | 454 | |
0a0f39df | 455 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 456 | |
9509913a IS |
457 | struct ingress_policer * |
458 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
459 | ||
8a9562d2 PS |
460 | static bool |
461 | is_dpdk_class(const struct netdev_class *class) | |
462 | { | |
f3e7ec25 MW |
463 | return class->destruct == netdev_dpdk_destruct |
464 | || class->destruct == netdev_dpdk_vhost_destruct; | |
8a9562d2 PS |
465 | } |
466 | ||
4be4d22c MK |
467 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
468 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
469 | * value, insufficient buffers are allocated to accomodate the packet in its | |
470 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
471 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
472 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
473 | * behaviour, which reduces performance. To prevent this, use a buffer size |
474 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
475 | */ |
476 | static uint32_t | |
477 | dpdk_buf_size(int mtu) | |
478 | { | |
479 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
480 | NETDEV_DPDK_MBUF_ALIGN); | |
481 | } | |
482 | ||
eff23640 DDP |
483 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
484 | * | |
485 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
486 | static void * |
487 | dpdk_rte_mzalloc(size_t sz) | |
488 | { | |
eff23640 | 489 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
490 | } |
491 | ||
492 | void | |
e14deea0 | 493 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 494 | { |
db73f716 | 495 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 496 | |
b00b4a81 | 497 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
498 | } |
499 | ||
b3cd9f9d | 500 | static void |
401b70d6 | 501 | ovs_rte_pktmbuf_init(struct rte_mempool *mp OVS_UNUSED, |
b3cd9f9d | 502 | void *opaque_arg OVS_UNUSED, |
2391135c | 503 | void *_p, |
b3cd9f9d PS |
504 | unsigned i OVS_UNUSED) |
505 | { | |
2391135c | 506 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 507 | |
2391135c | 508 | dp_packet_init_dpdk((struct dp_packet *) pkt, pkt->buf_len); |
b3cd9f9d PS |
509 | } |
510 | ||
24e78f93 IM |
511 | /* Returns a valid pointer when either of the following is true: |
512 | * - a new mempool was just created; | |
513 | * - a matching mempool already exists. */ | |
514 | static struct rte_mempool * | |
515 | dpdk_mp_create(struct netdev_dpdk *dev, int mtu) | |
8a9562d2 | 516 | { |
24e78f93 IM |
517 | char mp_name[RTE_MEMPOOL_NAMESIZE]; |
518 | const char *netdev_name = netdev_get_name(&dev->up); | |
519 | int socket_id = dev->requested_socket_id; | |
520 | uint32_t n_mbufs; | |
521 | uint32_t hash = hash_string(netdev_name, 0); | |
522 | struct rte_mempool *mp = NULL; | |
d555d9bd RW |
523 | |
524 | /* | |
ad9b5b9b | 525 | * XXX: rough estimation of number of mbufs required for this port: |
d555d9bd RW |
526 | * <packets required to fill the device rxqs> |
527 | * + <packets that could be stuck on other ports txqs> | |
528 | * + <packets in the pmd threads> | |
529 | * + <additional memory for corner cases> | |
0072e931 | 530 | */ |
24e78f93 IM |
531 | n_mbufs = dev->requested_n_rxq * dev->requested_rxq_size |
532 | + dev->requested_n_txq * dev->requested_txq_size | |
533 | + MIN(RTE_MAX_LCORE, dev->requested_n_rxq) * NETDEV_MAX_BURST | |
534 | + MIN_NB_MBUF; | |
d555d9bd | 535 | |
24e78f93 | 536 | ovs_mutex_lock(&dpdk_mp_mutex); |
da79ce2b | 537 | do { |
24e78f93 IM |
538 | /* Full DPDK memory pool name must be unique and cannot be |
539 | * longer than RTE_MEMPOOL_NAMESIZE. */ | |
af5b0dad IM |
540 | int ret = snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, |
541 | "ovs%08x%02d%05d%07u", | |
24e78f93 IM |
542 | hash, socket_id, mtu, n_mbufs); |
543 | if (ret < 0 || ret >= RTE_MEMPOOL_NAMESIZE) { | |
544 | VLOG_DBG("snprintf returned %d. " | |
545 | "Failed to generate a mempool name for \"%s\". " | |
546 | "Hash:0x%x, socket_id: %d, mtu:%d, mbufs:%u.", | |
547 | ret, netdev_name, hash, socket_id, mtu, n_mbufs); | |
548 | break; | |
65056fd7 | 549 | } |
95fb793a | 550 | |
f06546a5 FA |
551 | VLOG_DBG("Port %s: Requesting a mempool of %u mbufs " |
552 | "on socket %d for %d Rx and %d Tx queues.", | |
24e78f93 | 553 | netdev_name, n_mbufs, socket_id, |
f06546a5 | 554 | dev->requested_n_rxq, dev->requested_n_txq); |
d555d9bd | 555 | |
24e78f93 IM |
556 | mp = rte_pktmbuf_pool_create(mp_name, n_mbufs, MP_CACHE_SZ, |
557 | sizeof (struct dp_packet) - sizeof (struct rte_mbuf), | |
558 | MBUF_SIZE(mtu) - sizeof(struct dp_packet), socket_id); | |
559 | ||
560 | if (mp) { | |
561 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", | |
562 | mp_name, n_mbufs); | |
837c1761 FA |
563 | /* rte_pktmbuf_pool_create has done some initialization of the |
564 | * rte_mbuf part of each dp_packet. Some OvS specific fields | |
565 | * of the packet still need to be initialized by | |
566 | * ovs_rte_pktmbuf_init. */ | |
24e78f93 | 567 | rte_mempool_obj_iter(mp, ovs_rte_pktmbuf_init, NULL); |
d555d9bd RW |
568 | } else if (rte_errno == EEXIST) { |
569 | /* A mempool with the same name already exists. We just | |
570 | * retrieve its pointer to be returned to the caller. */ | |
24e78f93 | 571 | mp = rte_mempool_lookup(mp_name); |
d555d9bd RW |
572 | /* As the mempool create returned EEXIST we can expect the |
573 | * lookup has returned a valid pointer. If for some reason | |
574 | * that's not the case we keep track of it. */ | |
24e78f93 IM |
575 | VLOG_DBG("A mempool with name \"%s\" already exists at %p.", |
576 | mp_name, mp); | |
d555d9bd RW |
577 | } else { |
578 | VLOG_ERR("Failed mempool \"%s\" create request of %u mbufs", | |
24e78f93 | 579 | mp_name, n_mbufs); |
0c6f39e5 | 580 | } |
24e78f93 | 581 | } while (!mp && rte_errno == ENOMEM && (n_mbufs /= 2) >= MIN_NB_MBUF); |
2ae3d542 | 582 | |
c2adb102 | 583 | ovs_mutex_unlock(&dpdk_mp_mutex); |
24e78f93 | 584 | return mp; |
8a9562d2 PS |
585 | } |
586 | ||
a08a115d | 587 | /* Release an existing mempool. */ |
8a9562d2 | 588 | static void |
24e78f93 | 589 | dpdk_mp_free(struct rte_mempool *mp) |
8a9562d2 | 590 | { |
24e78f93 | 591 | if (!mp) { |
8a9562d2 PS |
592 | return; |
593 | } | |
594 | ||
c2adb102 | 595 | ovs_mutex_lock(&dpdk_mp_mutex); |
24e78f93 IM |
596 | VLOG_DBG("Releasing \"%s\" mempool", mp->name); |
597 | rte_mempool_free(mp); | |
c2adb102 | 598 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
599 | } |
600 | ||
b6b26021 FA |
601 | /* Tries to allocate a new mempool - or re-use an existing one where |
602 | * appropriate - on requested_socket_id with a size determined by | |
603 | * requested_mtu and requested Rx/Tx queues. | |
604 | * On success - or when re-using an existing mempool - the new configuration | |
605 | * will be applied. | |
0072e931 MK |
606 | * On error, device will be left unchanged. */ |
607 | static int | |
608 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
609 | OVS_REQUIRES(dev->mutex) |
610 | { | |
611 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
24e78f93 IM |
612 | struct rte_mempool *mp; |
613 | int ret = 0; | |
0072e931 | 614 | |
24e78f93 | 615 | mp = dpdk_mp_create(dev, FRAME_LEN_TO_MTU(buf_size)); |
0072e931 | 616 | if (!mp) { |
c67e46c0 MK |
617 | VLOG_ERR("Failed to create memory pool for netdev " |
618 | "%s, with MTU %d on socket %d: %s\n", | |
619 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
620 | rte_strerror(rte_errno)); | |
24e78f93 | 621 | ret = rte_errno; |
0072e931 | 622 | } else { |
24e78f93 IM |
623 | /* If a new MTU was requested and its rounded value equals the one |
624 | * that is currently used, then the existing mempool is returned. */ | |
625 | if (dev->mp != mp) { | |
626 | /* A new mempool was created, release the previous one. */ | |
627 | dpdk_mp_free(dev->mp); | |
628 | } else { | |
629 | ret = EEXIST; | |
630 | } | |
631 | dev->mp = mp; | |
0072e931 MK |
632 | dev->mtu = dev->requested_mtu; |
633 | dev->socket_id = dev->requested_socket_id; | |
634 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
635 | } | |
636 | ||
24e78f93 | 637 | return ret; |
0072e931 MK |
638 | } |
639 | ||
8a9562d2 PS |
640 | static void |
641 | check_link_status(struct netdev_dpdk *dev) | |
642 | { | |
643 | struct rte_eth_link link; | |
644 | ||
645 | rte_eth_link_get_nowait(dev->port_id, &link); | |
646 | ||
647 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 648 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
649 | |
650 | dev->link_reset_cnt++; | |
651 | dev->link = link; | |
652 | if (dev->link.link_status) { | |
bb37956a | 653 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Up - speed %u Mbps - %s", |
58be5c0e | 654 | dev->port_id, (unsigned) dev->link.link_speed, |
8a9562d2 PS |
655 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? |
656 | ("full-duplex") : ("half-duplex")); | |
657 | } else { | |
bb37956a | 658 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Down", dev->port_id); |
8a9562d2 PS |
659 | } |
660 | } | |
661 | } | |
662 | ||
663 | static void * | |
664 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
665 | { | |
666 | struct netdev_dpdk *dev; | |
667 | ||
668 | pthread_detach(pthread_self()); | |
669 | ||
670 | for (;;) { | |
671 | ovs_mutex_lock(&dpdk_mutex); | |
672 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
673 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
674 | if (dev->type == DPDK_DEV_ETH) { |
675 | check_link_status(dev); | |
676 | } | |
8a9562d2 PS |
677 | ovs_mutex_unlock(&dev->mutex); |
678 | } | |
679 | ovs_mutex_unlock(&dpdk_mutex); | |
680 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
681 | } | |
682 | ||
683 | return NULL; | |
684 | } | |
685 | ||
b98d7669 DDP |
686 | static int |
687 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
688 | { | |
689 | int diag = 0; | |
690 | int i; | |
0072e931 | 691 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 692 | |
67fe6d63 MK |
693 | /* For some NICs (e.g. Niantic), scatter_rx mode needs to be explicitly |
694 | * enabled. */ | |
0072e931 | 695 | if (dev->mtu > ETHER_MTU) { |
67fe6d63 | 696 | conf.rxmode.enable_scatter = 1; |
0072e931 | 697 | } |
67fe6d63 | 698 | |
1a2bb118 SC |
699 | conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & |
700 | NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
b98d7669 DDP |
701 | /* A device may report more queues than it makes available (this has |
702 | * been observed for Intel xl710, which reserves some of them for | |
703 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
704 | * available. When this happens we can retry the configuration | |
705 | * and request less queues */ | |
706 | while (n_rxq && n_txq) { | |
707 | if (diag) { | |
708 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
709 | } | |
710 | ||
0072e931 | 711 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 712 | if (diag) { |
0072e931 MK |
713 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
714 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
715 | break; |
716 | } | |
717 | ||
67fe6d63 MK |
718 | diag = rte_eth_dev_set_mtu(dev->port_id, dev->mtu); |
719 | if (diag) { | |
720 | VLOG_ERR("Interface %s MTU (%d) setup error: %s", | |
721 | dev->up.name, dev->mtu, rte_strerror(-diag)); | |
722 | break; | |
723 | } | |
724 | ||
b98d7669 | 725 | for (i = 0; i < n_txq; i++) { |
b685696b | 726 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
727 | dev->socket_id, NULL); |
728 | if (diag) { | |
729 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
730 | dev->up.name, i, rte_strerror(-diag)); | |
731 | break; | |
732 | } | |
733 | } | |
734 | ||
735 | if (i != n_txq) { | |
736 | /* Retry with less tx queues */ | |
737 | n_txq = i; | |
738 | continue; | |
739 | } | |
740 | ||
741 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 742 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
24e78f93 | 743 | dev->socket_id, NULL, dev->mp); |
b98d7669 DDP |
744 | if (diag) { |
745 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
746 | dev->up.name, i, rte_strerror(-diag)); | |
747 | break; | |
748 | } | |
749 | } | |
750 | ||
751 | if (i != n_rxq) { | |
752 | /* Retry with less rx queues */ | |
753 | n_rxq = i; | |
754 | continue; | |
755 | } | |
756 | ||
757 | dev->up.n_rxq = n_rxq; | |
81acebda | 758 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
759 | |
760 | return 0; | |
761 | } | |
762 | ||
763 | return diag; | |
764 | } | |
765 | ||
9fd39370 SC |
766 | static void |
767 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
768 | { | |
769 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
bb37956a IM |
770 | VLOG_WARN("Failed to enable flow control on device %"PRIu8, |
771 | dev->port_id); | |
9fd39370 SC |
772 | } |
773 | } | |
b98d7669 | 774 | |
8a9562d2 | 775 | static int |
c2adb102 IM |
776 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
777 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
778 | { |
779 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 780 | struct rte_eth_dev_info info; |
8a9562d2 PS |
781 | struct ether_addr eth_addr; |
782 | int diag; | |
b98d7669 | 783 | int n_rxq, n_txq; |
d4f5282c KT |
784 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | |
785 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
786 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
8a9562d2 | 787 | |
a0cb2d66 | 788 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 789 | |
d4f5282c KT |
790 | if ((info.rx_offload_capa & rx_chksm_offload_capa) != |
791 | rx_chksm_offload_capa) { | |
2cfe866f | 792 | VLOG_WARN("Rx checksum offload is not supported on port %"PRIu8, |
d4f5282c KT |
793 | dev->port_id); |
794 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; | |
795 | } else { | |
796 | dev->hw_ol_features |= NETDEV_RX_CHECKSUM_OFFLOAD; | |
797 | } | |
798 | ||
b98d7669 DDP |
799 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
800 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
801 | ||
802 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 803 | if (diag) { |
b98d7669 DDP |
804 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
805 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 806 | return -diag; |
8a9562d2 PS |
807 | } |
808 | ||
8a9562d2 PS |
809 | diag = rte_eth_dev_start(dev->port_id); |
810 | if (diag) { | |
b98d7669 DDP |
811 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
812 | rte_strerror(-diag)); | |
95fb793a | 813 | return -diag; |
8a9562d2 PS |
814 | } |
815 | ||
816 | rte_eth_promiscuous_enable(dev->port_id); | |
817 | rte_eth_allmulticast_enable(dev->port_id); | |
818 | ||
819 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
820 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
bb37956a | 821 | VLOG_INFO_RL(&rl, "Port %"PRIu8": "ETH_ADDR_FMT, |
ca92d173 | 822 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 823 | |
ca92d173 | 824 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
825 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
826 | ||
24e78f93 | 827 | mbp_priv = rte_mempool_get_priv(dev->mp); |
8a9562d2 PS |
828 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; |
829 | ||
9fd39370 SC |
830 | /* Get the Flow control configuration for DPDK-ETH */ |
831 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
832 | if (diag) { | |
bb37956a | 833 | VLOG_DBG("cannot get flow control parameters on port=%"PRIu8", err=%d", |
9fd39370 SC |
834 | dev->port_id, diag); |
835 | } | |
836 | ||
8a9562d2 PS |
837 | return 0; |
838 | } | |
839 | ||
840 | static struct netdev_dpdk * | |
841 | netdev_dpdk_cast(const struct netdev *netdev) | |
842 | { | |
843 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
844 | } | |
845 | ||
846 | static struct netdev * | |
847 | netdev_dpdk_alloc(void) | |
848 | { | |
bab69409 AC |
849 | struct netdev_dpdk *dev; |
850 | ||
65e19e70 DDP |
851 | dev = dpdk_rte_mzalloc(sizeof *dev); |
852 | if (dev) { | |
853 | return &dev->up; | |
bab69409 | 854 | } |
65e19e70 | 855 | |
bab69409 | 856 | return NULL; |
8a9562d2 PS |
857 | } |
858 | ||
eff23640 DDP |
859 | static struct dpdk_tx_queue * |
860 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 861 | { |
eff23640 | 862 | struct dpdk_tx_queue *txqs; |
bd5131ba | 863 | unsigned i; |
5a034064 | 864 | |
eff23640 DDP |
865 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
866 | if (txqs) { | |
867 | for (i = 0; i < n_txqs; i++) { | |
868 | /* Initialize map for vhost devices. */ | |
869 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
870 | rte_spinlock_init(&txqs[i].tx_lock); | |
871 | } | |
5a034064 | 872 | } |
eff23640 DDP |
873 | |
874 | return txqs; | |
5a034064 AW |
875 | } |
876 | ||
8a9562d2 | 877 | static int |
bb37956a | 878 | common_construct(struct netdev *netdev, dpdk_port_t port_no, |
1ce30dfd | 879 | enum dpdk_dev_type type, int socket_id) |
5a034064 | 880 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 881 | { |
d46285a2 | 882 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 883 | |
d46285a2 | 884 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 885 | |
d46285a2 | 886 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 887 | |
1b7a04e0 AW |
888 | /* If the 'sid' is negative, it means that the kernel fails |
889 | * to obtain the pci numa info. In that situation, always | |
890 | * use 'SOCKET0'. */ | |
1ce30dfd | 891 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 892 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
893 | dev->port_id = port_no; |
894 | dev->type = type; | |
895 | dev->flags = 0; | |
7f381c2e | 896 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 897 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
898 | ovsrcu_index_init(&dev->vid, -1); |
899 | dev->vhost_reconfigured = false; | |
5dcde09c | 900 | dev->attached = false; |
8a9562d2 | 901 | |
78bd47cf | 902 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 903 | |
9509913a IS |
904 | ovsrcu_init(&dev->ingress_policer, NULL); |
905 | dev->policer_rate = 0; | |
906 | dev->policer_burst = 0; | |
907 | ||
7f381c2e DDP |
908 | netdev->n_rxq = 0; |
909 | netdev->n_txq = 0; | |
910 | dev->requested_n_rxq = NR_QUEUE; | |
911 | dev->requested_n_txq = NR_QUEUE; | |
912 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
913 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 914 | |
9fd39370 SC |
915 | /* Initialize the flow control to NULL */ |
916 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
917 | |
918 | /* Initilize the hardware offload flags to 0 */ | |
919 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
920 | |
921 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
922 | ||
d46285a2 | 923 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 924 | |
7f381c2e DDP |
925 | netdev_request_reconfigure(netdev); |
926 | ||
971f4b39 MW |
927 | dev->rte_xstats_names = NULL; |
928 | dev->rte_xstats_names_size = 0; | |
929 | ||
930 | dev->rte_xstats_ids = NULL; | |
931 | dev->rte_xstats_ids_size = 0; | |
932 | ||
1ce30dfd | 933 | return 0; |
95fb793a | 934 | } |
935 | ||
b83a2df1 MV |
936 | /* dev_name must be the prefix followed by a positive decimal number. |
937 | * (no leading + or - signs are allowed) */ | |
95fb793a | 938 | static int |
939 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
940 | unsigned int *port_no) | |
941 | { | |
942 | const char *cport; | |
943 | ||
944 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
945 | return ENODEV; | |
946 | } | |
947 | ||
948 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
949 | |
950 | if (str_to_uint(cport, 10, port_no)) { | |
951 | return 0; | |
952 | } else { | |
953 | return ENODEV; | |
954 | } | |
95fb793a | 955 | } |
956 | ||
1ce30dfd DDP |
957 | static int |
958 | vhost_common_construct(struct netdev *netdev) | |
959 | OVS_REQUIRES(dpdk_mutex) | |
960 | { | |
961 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
962 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
963 | ||
964 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); | |
965 | if (!dev->tx_q) { | |
966 | return ENOMEM; | |
967 | } | |
968 | ||
bb37956a IM |
969 | return common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
970 | DPDK_DEV_VHOST, socket_id); | |
1ce30dfd DDP |
971 | } |
972 | ||
7d1ced01 | 973 | static int |
53f50d24 | 974 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 975 | { |
d46285a2 DDP |
976 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
977 | const char *name = netdev->name; | |
7d1ced01 | 978 | int err; |
a0cb2d66 | 979 | |
1af27e8a DDP |
980 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
981 | * the file system. '/' or '\' would traverse directories, so they're not | |
982 | * acceptable in 'name'. */ | |
983 | if (strchr(name, '/') || strchr(name, '\\')) { | |
984 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
985 | "A valid name must not include '/' or '\\'", | |
986 | name); | |
987 | return EINVAL; | |
988 | } | |
989 | ||
7d1ced01 CL |
990 | ovs_mutex_lock(&dpdk_mutex); |
991 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 992 | * the socket is to be created, then register the socket. |
7d1ced01 | 993 | */ |
2d24d165 | 994 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 995 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 996 | |
2d24d165 CL |
997 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
998 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
999 | if (err) { |
1000 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 1001 | dev->vhost_id); |
f3e7ec25 | 1002 | goto out; |
e5c0f5a4 | 1003 | } else { |
2d24d165 CL |
1004 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
1005 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
1006 | dev->vhost_id, name); | |
1007 | } | |
f3e7ec25 MW |
1008 | |
1009 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
1010 | &virtio_net_device_ops); | |
1011 | if (err) { | |
1012 | VLOG_ERR("rte_vhost_driver_callback_register failed for vhost user " | |
1013 | "port: %s\n", name); | |
1014 | goto out; | |
1015 | } | |
1016 | ||
1017 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
1018 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
1019 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
1020 | | 1ULL << VIRTIO_NET_F_CSUM); | |
1021 | if (err) { | |
1022 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
1023 | "port: %s\n", name); | |
1024 | goto out; | |
1025 | } | |
1026 | ||
1027 | err = rte_vhost_driver_start(dev->vhost_id); | |
1028 | if (err) { | |
1029 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
1030 | "port: %s\n", name); | |
1031 | goto out; | |
1032 | } | |
1033 | ||
1ce30dfd | 1034 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1035 | if (err) { |
1036 | VLOG_ERR("vhost_common_construct failed for vhost user " | |
1037 | "port: %s\n", name); | |
1038 | } | |
2d24d165 | 1039 | |
f3e7ec25 | 1040 | out: |
2d24d165 | 1041 | ovs_mutex_unlock(&dpdk_mutex); |
28ca969e AC |
1042 | VLOG_WARN_ONCE("dpdkvhostuser ports are considered deprecated; " |
1043 | "please migrate to dpdkvhostuserclient ports."); | |
2d24d165 CL |
1044 | return err; |
1045 | } | |
1046 | ||
1047 | static int | |
1048 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
1049 | { | |
1050 | int err; | |
1051 | ||
2d24d165 | 1052 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 1053 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1054 | if (err) { |
1055 | VLOG_ERR("vhost_common_construct failed for vhost user client" | |
1056 | "port: %s\n", netdev->name); | |
1057 | } | |
7d1ced01 | 1058 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
1059 | return err; |
1060 | } | |
1061 | ||
95fb793a | 1062 | static int |
1063 | netdev_dpdk_construct(struct netdev *netdev) | |
1064 | { | |
95fb793a | 1065 | int err; |
1066 | ||
95fb793a | 1067 | ovs_mutex_lock(&dpdk_mutex); |
bb37956a IM |
1068 | err = common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1069 | DPDK_DEV_ETH, SOCKET0); | |
8a9562d2 PS |
1070 | ovs_mutex_unlock(&dpdk_mutex); |
1071 | return err; | |
1072 | } | |
1073 | ||
1ce30dfd DDP |
1074 | static void |
1075 | common_destruct(struct netdev_dpdk *dev) | |
1076 | OVS_REQUIRES(dpdk_mutex) | |
1077 | OVS_EXCLUDED(dev->mutex) | |
1078 | { | |
1079 | rte_free(dev->tx_q); | |
24e78f93 | 1080 | dpdk_mp_free(dev->mp); |
1ce30dfd DDP |
1081 | |
1082 | ovs_list_remove(&dev->list_node); | |
1083 | free(ovsrcu_get_protected(struct ingress_policer *, | |
1084 | &dev->ingress_policer)); | |
1085 | ovs_mutex_destroy(&dev->mutex); | |
1086 | } | |
1087 | ||
8a9562d2 | 1088 | static void |
d46285a2 | 1089 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 1090 | { |
d46285a2 | 1091 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5dcde09c | 1092 | char devname[RTE_ETH_NAME_MAX_LEN]; |
8a9562d2 | 1093 | |
8d38823b | 1094 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1095 | |
8a9562d2 | 1096 | rte_eth_dev_stop(dev->port_id); |
5dcde09c IM |
1097 | |
1098 | if (dev->attached) { | |
1099 | rte_eth_dev_close(dev->port_id); | |
1100 | if (rte_eth_dev_detach(dev->port_id, devname) < 0) { | |
1101 | VLOG_ERR("Device '%s' can not be detached", dev->devargs); | |
1102 | } else { | |
0ee821c2 | 1103 | VLOG_INFO("Device '%s' has been detached", devname); |
5dcde09c IM |
1104 | } |
1105 | } | |
1106 | ||
55e075e6 | 1107 | free(dev->devargs); |
1ce30dfd | 1108 | common_destruct(dev); |
8d38823b | 1109 | |
8a9562d2 | 1110 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1111 | } |
8a9562d2 | 1112 | |
3f891bbe DDP |
1113 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1114 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1115 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1116 | static int | |
c1ff66ac CL |
1117 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1118 | char *vhost_id) | |
3f891bbe DDP |
1119 | OVS_EXCLUDED(dpdk_mutex) |
1120 | OVS_EXCLUDED(dev->mutex) | |
1121 | { | |
c1ff66ac | 1122 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1123 | } |
1124 | ||
58397e6c | 1125 | static void |
d46285a2 | 1126 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1127 | { |
d46285a2 | 1128 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1129 | char *vhost_id; |
58397e6c | 1130 | |
8d38823b | 1131 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1132 | |
c62da695 | 1133 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1134 | if (netdev_dpdk_get_vid(dev) >= 0 |
1135 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1136 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1137 | netdev->name); |
58be5c0e MK |
1138 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1139 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1140 | } |
1141 | ||
2d24d165 | 1142 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1143 | |
1ce30dfd DDP |
1144 | common_destruct(dev); |
1145 | ||
58397e6c | 1146 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1147 | |
569c26da | 1148 | if (!vhost_id[0]) { |
821b8664 IM |
1149 | goto out; |
1150 | } | |
1151 | ||
c1ff66ac | 1152 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1153 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1154 | netdev->name, vhost_id); | |
c1ff66ac CL |
1155 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1156 | /* OVS server mode - remove this socket from list for deletion */ | |
1157 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1158 | } |
821b8664 | 1159 | out: |
c1ff66ac | 1160 | free(vhost_id); |
8a9562d2 PS |
1161 | } |
1162 | ||
1163 | static void | |
d46285a2 | 1164 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1165 | { |
d46285a2 | 1166 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1167 | |
d46285a2 | 1168 | rte_free(dev); |
8a9562d2 PS |
1169 | } |
1170 | ||
971f4b39 MW |
1171 | static void |
1172 | netdev_dpdk_clear_xstats(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
1173 | { | |
1174 | /* If statistics are already allocated, we have to | |
1175 | * reconfigure, as port_id could have been changed. */ | |
1176 | if (dev->rte_xstats_names) { | |
1177 | free(dev->rte_xstats_names); | |
1178 | dev->rte_xstats_names = NULL; | |
1179 | dev->rte_xstats_names_size = 0; | |
1180 | } | |
1181 | if (dev->rte_xstats_ids) { | |
1182 | free(dev->rte_xstats_ids); | |
1183 | dev->rte_xstats_ids = NULL; | |
1184 | dev->rte_xstats_ids_size = 0; | |
1185 | } | |
1186 | } | |
1187 | ||
1188 | static const char* | |
1189 | netdev_dpdk_get_xstat_name(struct netdev_dpdk *dev, uint64_t id) | |
1190 | { | |
1191 | if (id >= dev->rte_xstats_names_size) { | |
1192 | return "UNKNOWN"; | |
1193 | } | |
1194 | return dev->rte_xstats_names[id].name; | |
1195 | } | |
1196 | ||
1197 | static bool | |
1198 | netdev_dpdk_configure_xstats(struct netdev_dpdk *dev) | |
1199 | OVS_REQUIRES(dev->mutex) | |
1200 | { | |
1201 | int rte_xstats_len; | |
1202 | bool ret; | |
1203 | struct rte_eth_xstat *rte_xstats; | |
1204 | uint64_t id; | |
1205 | int xstats_no; | |
1206 | const char *name; | |
1207 | ||
1208 | /* Retrieving all XSTATS names. If something will go wrong | |
1209 | * or amount of counters will be equal 0, rte_xstats_names | |
1210 | * buffer will be marked as NULL, and any further xstats | |
1211 | * query won't be performed (e.g. during netdev_dpdk_get_stats | |
1212 | * execution). */ | |
1213 | ||
1214 | ret = false; | |
1215 | rte_xstats = NULL; | |
1216 | ||
1217 | if (dev->rte_xstats_names == NULL || dev->rte_xstats_ids == NULL) { | |
1218 | dev->rte_xstats_names_size = | |
1219 | rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
1220 | ||
1221 | if (dev->rte_xstats_names_size < 0) { | |
1222 | VLOG_WARN("Cannot get XSTATS for port: %"PRIu8, dev->port_id); | |
1223 | dev->rte_xstats_names_size = 0; | |
1224 | } else { | |
1225 | /* Reserve memory for xstats names and values */ | |
1226 | dev->rte_xstats_names = xcalloc(dev->rte_xstats_names_size, | |
1227 | sizeof *dev->rte_xstats_names); | |
1228 | ||
1229 | if (dev->rte_xstats_names) { | |
1230 | /* Retreive xstats names */ | |
1231 | rte_xstats_len = | |
1232 | rte_eth_xstats_get_names(dev->port_id, | |
1233 | dev->rte_xstats_names, | |
1234 | dev->rte_xstats_names_size); | |
1235 | ||
1236 | if (rte_xstats_len < 0) { | |
1237 | VLOG_WARN("Cannot get XSTATS names for port: %"PRIu8, | |
1238 | dev->port_id); | |
1239 | goto out; | |
1240 | } else if (rte_xstats_len != dev->rte_xstats_names_size) { | |
1241 | VLOG_WARN("XSTATS size doesn't match for port: %"PRIu8, | |
1242 | dev->port_id); | |
1243 | goto out; | |
1244 | } | |
1245 | ||
1246 | dev->rte_xstats_ids = xcalloc(dev->rte_xstats_names_size, | |
1247 | sizeof(uint64_t)); | |
1248 | ||
1249 | /* We have to calculate number of counters */ | |
1250 | rte_xstats = xmalloc(rte_xstats_len * sizeof *rte_xstats); | |
1251 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
1252 | ||
1253 | /* Retreive xstats values */ | |
1254 | if (rte_eth_xstats_get(dev->port_id, rte_xstats, | |
1255 | rte_xstats_len) > 0) { | |
1256 | dev->rte_xstats_ids_size = 0; | |
1257 | xstats_no = 0; | |
1258 | for (uint32_t i = 0; i < rte_xstats_len; i++) { | |
1259 | id = rte_xstats[i].id; | |
1260 | name = netdev_dpdk_get_xstat_name(dev, id); | |
1261 | /* We need to filter out everything except | |
1262 | * dropped, error and management counters */ | |
1263 | if (string_ends_with(name, "_errors") || | |
1264 | strstr(name, "_management_") || | |
1265 | string_ends_with(name, "_dropped")) { | |
1266 | ||
1267 | dev->rte_xstats_ids[xstats_no] = id; | |
1268 | xstats_no++; | |
1269 | } | |
1270 | } | |
1271 | dev->rte_xstats_ids_size = xstats_no; | |
1272 | ret = true; | |
1273 | } else { | |
1274 | VLOG_WARN("Can't get XSTATS IDs for port: %"PRIu8, | |
1275 | dev->port_id); | |
1276 | } | |
1277 | } | |
1278 | } | |
1279 | } else { | |
1280 | /* Already configured */ | |
1281 | ret = true; | |
1282 | } | |
1283 | ||
1284 | out: | |
1285 | if (!ret) { | |
1286 | netdev_dpdk_clear_xstats(dev); | |
1287 | } | |
1288 | return ret; | |
1289 | } | |
1290 | ||
8a9562d2 | 1291 | static int |
a14b8947 | 1292 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1293 | { |
a14b8947 | 1294 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1295 | |
1296 | ovs_mutex_lock(&dev->mutex); | |
1297 | ||
050c60bf | 1298 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1299 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1300 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1301 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1302 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1303 | |
1304 | if (dev->type == DPDK_DEV_ETH) { | |
1305 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1306 | dev->requested_rxq_size); | |
1307 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1308 | dev->rxq_size); | |
1309 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1310 | dev->requested_txq_size); | |
1311 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1312 | dev->txq_size); | |
1a2bb118 SC |
1313 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1314 | smap_add(args, "rx_csum_offload", "true"); | |
8155ab7e KT |
1315 | } else { |
1316 | smap_add(args, "rx_csum_offload", "false"); | |
1a2bb118 | 1317 | } |
451f26fd | 1318 | } |
8a9562d2 PS |
1319 | ovs_mutex_unlock(&dev->mutex); |
1320 | ||
1321 | return 0; | |
1322 | } | |
1323 | ||
55e075e6 | 1324 | static struct netdev_dpdk * |
bb37956a | 1325 | netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id) |
55e075e6 CL |
1326 | OVS_REQUIRES(dpdk_mutex) |
1327 | { | |
1328 | struct netdev_dpdk *dev; | |
1329 | ||
1330 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1331 | if (dev->port_id == port_id) { | |
1332 | return dev; | |
1333 | } | |
1334 | } | |
1335 | ||
1336 | return NULL; | |
1337 | } | |
1338 | ||
bb37956a | 1339 | static dpdk_port_t |
5dcde09c IM |
1340 | netdev_dpdk_process_devargs(struct netdev_dpdk *dev, |
1341 | const char *devargs, char **errp) | |
55e075e6 | 1342 | { |
03d6399e IM |
1343 | /* Get the name up to the first comma. */ |
1344 | char *name = xmemdup0(devargs, strcspn(devargs, ",")); | |
bb37956a | 1345 | dpdk_port_t new_port_id = DPDK_ETH_PORT_ID_INVALID; |
55e075e6 | 1346 | |
255b7bda | 1347 | if (rte_eth_dev_get_port_by_name(name, &new_port_id) |
69876ed7 CL |
1348 | || !rte_eth_dev_is_valid_port(new_port_id)) { |
1349 | /* Device not found in DPDK, attempt to attach it */ | |
1350 | if (!rte_eth_dev_attach(devargs, &new_port_id)) { | |
1351 | /* Attach successful */ | |
5dcde09c | 1352 | dev->attached = true; |
69876ed7 CL |
1353 | VLOG_INFO("Device '%s' attached to DPDK", devargs); |
1354 | } else { | |
1355 | /* Attach unsuccessful */ | |
bb37956a | 1356 | new_port_id = DPDK_ETH_PORT_ID_INVALID; |
47a45d86 KT |
1357 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", |
1358 | devargs); | |
55e075e6 CL |
1359 | } |
1360 | } | |
1361 | ||
03d6399e | 1362 | free(name); |
55e075e6 CL |
1363 | return new_port_id; |
1364 | } | |
1365 | ||
c3d062a7 CL |
1366 | static void |
1367 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1368 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1369 | { |
050c60bf | 1370 | int new_n_rxq; |
a14b8947 | 1371 | |
2a21e757 | 1372 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1373 | if (new_n_rxq != dev->requested_n_rxq) { |
1374 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1375 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1376 | } |
c3d062a7 CL |
1377 | } |
1378 | ||
b685696b CL |
1379 | static void |
1380 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1381 | const char *flag, int default_size, int *new_size) | |
1382 | { | |
1383 | int queue_size = smap_get_int(args, flag, default_size); | |
1384 | ||
1385 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1386 | || !is_pow2(queue_size)) { | |
1387 | queue_size = default_size; | |
1388 | } | |
1389 | ||
1390 | if (queue_size != *new_size) { | |
1391 | *new_size = queue_size; | |
1392 | netdev_request_reconfigure(netdev); | |
1393 | } | |
1394 | } | |
1395 | ||
c3d062a7 | 1396 | static int |
9fff138e DDP |
1397 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1398 | char **errp) | |
c3d062a7 CL |
1399 | { |
1400 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
b614c894 IM |
1401 | bool rx_fc_en, tx_fc_en, autoneg; |
1402 | enum rte_eth_fc_mode fc_mode; | |
1403 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1404 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1405 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1406 | }; | |
55e075e6 CL |
1407 | const char *new_devargs; |
1408 | int err = 0; | |
c3d062a7 | 1409 | |
55e075e6 | 1410 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1411 | ovs_mutex_lock(&dev->mutex); |
1412 | ||
1413 | dpdk_set_rxq_config(dev, args); | |
1414 | ||
b685696b CL |
1415 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1416 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1417 | &dev->requested_rxq_size); | |
1418 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1419 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1420 | &dev->requested_txq_size); | |
1421 | ||
55e075e6 CL |
1422 | new_devargs = smap_get(args, "dpdk-devargs"); |
1423 | ||
1424 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1425 | /* The user requested a new device. If we return error, the caller | |
1426 | * will delete this netdev and try to recreate it. */ | |
1427 | err = EAGAIN; | |
1428 | goto out; | |
1429 | } | |
1430 | ||
1431 | /* dpdk-devargs is required for device configuration */ | |
1432 | if (new_devargs && new_devargs[0]) { | |
1433 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1434 | * is valid */ | |
1435 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1436 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
bb37956a IM |
1437 | dpdk_port_t new_port_id = netdev_dpdk_process_devargs(dev, |
1438 | new_devargs, | |
1439 | errp); | |
55e075e6 CL |
1440 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1441 | err = EINVAL; | |
1442 | } else if (new_port_id == dev->port_id) { | |
1443 | /* Already configured, do not reconfigure again */ | |
1444 | err = 0; | |
1445 | } else { | |
1446 | struct netdev_dpdk *dup_dev; | |
bb37956a | 1447 | |
55e075e6 CL |
1448 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); |
1449 | if (dup_dev) { | |
9fff138e DDP |
1450 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
1451 | "which is already in use by '%s'", | |
1452 | netdev_get_name(netdev), new_devargs, | |
1453 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1454 | err = EADDRINUSE; |
1455 | } else { | |
bd4e172b | 1456 | int sid = rte_eth_dev_socket_id(new_port_id); |
bb37956a | 1457 | |
bd4e172b | 1458 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; |
55e075e6 CL |
1459 | dev->devargs = xstrdup(new_devargs); |
1460 | dev->port_id = new_port_id; | |
1461 | netdev_request_reconfigure(&dev->up); | |
971f4b39 | 1462 | netdev_dpdk_clear_xstats(dev); |
55e075e6 CL |
1463 | err = 0; |
1464 | } | |
1465 | } | |
1466 | } | |
1467 | } else { | |
9fff138e DDP |
1468 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1469 | "The old 'dpdk<port_id>' names are not supported", | |
1470 | netdev_get_name(netdev)); | |
55e075e6 CL |
1471 | err = EINVAL; |
1472 | } | |
1473 | ||
1474 | if (err) { | |
1475 | goto out; | |
1476 | } | |
1477 | ||
c3d062a7 CL |
1478 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1479 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1480 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1481 | |
b614c894 IM |
1482 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1483 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1484 | dev->fc_conf.mode = fc_mode; | |
1485 | dev->fc_conf.autoneg = autoneg; | |
1486 | dpdk_eth_flow_ctrl_setup(dev); | |
1487 | } | |
9fd39370 | 1488 | |
55e075e6 | 1489 | out: |
c3d062a7 | 1490 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1491 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1492 | |
55e075e6 | 1493 | return err; |
c3d062a7 CL |
1494 | } |
1495 | ||
1496 | static int | |
9fff138e DDP |
1497 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1498 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1499 | { |
1500 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1501 | ||
1502 | ovs_mutex_lock(&dev->mutex); | |
1503 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1504 | ovs_mutex_unlock(&dev->mutex); |
1505 | ||
1506 | return 0; | |
1507 | } | |
1508 | ||
c1ff66ac | 1509 | static int |
2d24d165 | 1510 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1511 | const struct smap *args, |
1512 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1513 | { |
1514 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1515 | const char *path; | |
1516 | ||
6881885a | 1517 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1518 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1519 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1520 | if (path && strcmp(path, dev->vhost_id)) { |
1521 | strcpy(dev->vhost_id, path); | |
c1ff66ac CL |
1522 | netdev_request_reconfigure(netdev); |
1523 | } | |
1524 | } | |
6881885a | 1525 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1526 | |
1527 | return 0; | |
1528 | } | |
1529 | ||
7dec44fe | 1530 | static int |
d46285a2 | 1531 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1532 | { |
d46285a2 | 1533 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1534 | |
d46285a2 | 1535 | return dev->socket_id; |
7dec44fe AW |
1536 | } |
1537 | ||
050c60bf | 1538 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1539 | static int |
050c60bf | 1540 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1541 | { |
d46285a2 | 1542 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1543 | |
d46285a2 | 1544 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1545 | |
050c60bf DDP |
1546 | if (dev->requested_n_txq == n_txq) { |
1547 | goto out; | |
4573fbd3 FL |
1548 | } |
1549 | ||
050c60bf DDP |
1550 | dev->requested_n_txq = n_txq; |
1551 | netdev_request_reconfigure(netdev); | |
58397e6c | 1552 | |
050c60bf | 1553 | out: |
d46285a2 | 1554 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1555 | return 0; |
58397e6c KT |
1556 | } |
1557 | ||
8a9562d2 PS |
1558 | static struct netdev_rxq * |
1559 | netdev_dpdk_rxq_alloc(void) | |
1560 | { | |
1561 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1562 | ||
eff23640 DDP |
1563 | if (rx) { |
1564 | return &rx->up; | |
1565 | } | |
1566 | ||
1567 | return NULL; | |
8a9562d2 PS |
1568 | } |
1569 | ||
1570 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1571 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1572 | { |
d46285a2 | 1573 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1574 | } |
1575 | ||
1576 | static int | |
d46285a2 | 1577 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1578 | { |
d46285a2 DDP |
1579 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1580 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1581 | |
d46285a2 DDP |
1582 | ovs_mutex_lock(&dev->mutex); |
1583 | rx->port_id = dev->port_id; | |
1584 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1585 | |
1586 | return 0; | |
1587 | } | |
1588 | ||
1589 | static void | |
d46285a2 | 1590 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1591 | { |
1592 | } | |
1593 | ||
1594 | static void | |
d46285a2 | 1595 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1596 | { |
d46285a2 | 1597 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1598 | |
1599 | rte_free(rx); | |
1600 | } | |
1601 | ||
819f13bd DDP |
1602 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1603 | * 'pkts', even in case of failure. | |
1604 | * | |
1605 | * Returns the number of packets that weren't transmitted. */ | |
1606 | static inline int | |
b59cc14e | 1607 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 1608 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 1609 | { |
1304f1f8 DDP |
1610 | uint32_t nb_tx = 0; |
1611 | ||
b59cc14e | 1612 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1613 | uint32_t ret; |
1614 | ||
b59cc14e | 1615 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1616 | if (!ret) { |
1617 | break; | |
1618 | } | |
1619 | ||
1620 | nb_tx += ret; | |
1621 | } | |
8a9562d2 | 1622 | |
b59cc14e | 1623 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 1624 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
1625 | * packet could come from a different mempool) */ |
1626 | int i; | |
1627 | ||
b59cc14e IM |
1628 | for (i = nb_tx; i < cnt; i++) { |
1629 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1630 | } |
8a9562d2 | 1631 | } |
819f13bd DDP |
1632 | |
1633 | return cnt - nb_tx; | |
8a9562d2 PS |
1634 | } |
1635 | ||
f3926f29 IS |
1636 | static inline bool |
1637 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1638 | struct rte_mbuf *pkt, uint64_t time) | |
1639 | { | |
1640 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1641 | ||
1642 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1643 | e_RTE_METER_GREEN; | |
1644 | } | |
1645 | ||
1646 | static int | |
1647 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
3e90f7d7 GZ |
1648 | struct rte_mbuf **pkts, int pkt_cnt, |
1649 | bool may_steal) | |
f3926f29 IS |
1650 | { |
1651 | int i = 0; | |
1652 | int cnt = 0; | |
1653 | struct rte_mbuf *pkt = NULL; | |
1654 | uint64_t current_time = rte_rdtsc(); | |
1655 | ||
1656 | for (i = 0; i < pkt_cnt; i++) { | |
1657 | pkt = pkts[i]; | |
1658 | /* Handle current packet */ | |
1659 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1660 | if (cnt != i) { | |
1661 | pkts[cnt] = pkt; | |
1662 | } | |
1663 | cnt++; | |
1664 | } else { | |
3e90f7d7 GZ |
1665 | if (may_steal) { |
1666 | rte_pktmbuf_free(pkt); | |
1667 | } | |
f3926f29 IS |
1668 | } |
1669 | } | |
1670 | ||
1671 | return cnt; | |
1672 | } | |
1673 | ||
9509913a IS |
1674 | static int |
1675 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
3e90f7d7 | 1676 | int pkt_cnt, bool may_steal) |
9509913a IS |
1677 | { |
1678 | int cnt = 0; | |
1679 | ||
1680 | rte_spinlock_lock(&policer->policer_lock); | |
3e90f7d7 GZ |
1681 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, |
1682 | pkt_cnt, may_steal); | |
9509913a IS |
1683 | rte_spinlock_unlock(&policer->policer_lock); |
1684 | ||
1685 | return cnt; | |
1686 | } | |
1687 | ||
58397e6c | 1688 | static bool |
0a0f39df | 1689 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1690 | { |
0a0f39df | 1691 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1692 | } |
1693 | ||
d6e3feb5 | 1694 | static inline void |
1695 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1696 | unsigned int packet_size) | |
1697 | { | |
1698 | /* Hard-coded search for the size bucket. */ | |
1699 | if (packet_size < 256) { | |
1700 | if (packet_size >= 128) { | |
1701 | stats->rx_128_to_255_packets++; | |
1702 | } else if (packet_size <= 64) { | |
1703 | stats->rx_1_to_64_packets++; | |
1704 | } else { | |
1705 | stats->rx_65_to_127_packets++; | |
1706 | } | |
1707 | } else { | |
1708 | if (packet_size >= 1523) { | |
1709 | stats->rx_1523_to_max_packets++; | |
1710 | } else if (packet_size >= 1024) { | |
1711 | stats->rx_1024_to_1522_packets++; | |
1712 | } else if (packet_size < 512) { | |
1713 | stats->rx_256_to_511_packets++; | |
1714 | } else { | |
1715 | stats->rx_512_to_1023_packets++; | |
1716 | } | |
1717 | } | |
1718 | } | |
1719 | ||
9e3ddd45 TP |
1720 | static inline void |
1721 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1722 | struct dp_packet **packets, int count, |
1723 | int dropped) | |
9e3ddd45 TP |
1724 | { |
1725 | int i; | |
d6e3feb5 | 1726 | unsigned int packet_size; |
9e3ddd45 TP |
1727 | struct dp_packet *packet; |
1728 | ||
1729 | stats->rx_packets += count; | |
9509913a | 1730 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1731 | for (i = 0; i < count; i++) { |
1732 | packet = packets[i]; | |
d6e3feb5 | 1733 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1734 | |
d6e3feb5 | 1735 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1736 | /* This only protects the following multicast counting from |
1737 | * too short packets, but it does not stop the packet from | |
1738 | * further processing. */ | |
1739 | stats->rx_errors++; | |
1740 | stats->rx_length_errors++; | |
1741 | continue; | |
1742 | } | |
1743 | ||
d6e3feb5 | 1744 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1745 | ||
9e3ddd45 TP |
1746 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1747 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1748 | stats->multicast++; | |
1749 | } | |
1750 | ||
d6e3feb5 | 1751 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1752 | } |
1753 | } | |
1754 | ||
58397e6c KT |
1755 | /* |
1756 | * The receive path for the vhost port is the TX path out from guest. | |
1757 | */ | |
1758 | static int | |
d46285a2 | 1759 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1760 | struct dp_packet_batch *batch) |
58397e6c | 1761 | { |
d46285a2 | 1762 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
9509913a | 1763 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1764 | uint16_t nb_rx = 0; |
9509913a | 1765 | uint16_t dropped = 0; |
daf22bf7 IM |
1766 | int qid = rxq->queue_id; |
1767 | int vid = netdev_dpdk_get_vid(dev); | |
58397e6c | 1768 | |
daf22bf7 | 1769 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured |
e543851d | 1770 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1771 | return EAGAIN; |
1772 | } | |
1773 | ||
daf22bf7 | 1774 | nb_rx = rte_vhost_dequeue_burst(vid, qid * VIRTIO_QNUM + VIRTIO_TXQ, |
24e78f93 | 1775 | dev->mp, |
64839cf4 | 1776 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1777 | NETDEV_MAX_BURST); |
58397e6c KT |
1778 | if (!nb_rx) { |
1779 | return EAGAIN; | |
1780 | } | |
1781 | ||
9509913a IS |
1782 | if (policer) { |
1783 | dropped = nb_rx; | |
64839cf4 WT |
1784 | nb_rx = ingress_policer_run(policer, |
1785 | (struct rte_mbuf **) batch->packets, | |
3e90f7d7 | 1786 | nb_rx, true); |
9509913a IS |
1787 | dropped -= nb_rx; |
1788 | } | |
1789 | ||
d46285a2 | 1790 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1791 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1792 | nb_rx, dropped); | |
d46285a2 | 1793 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1794 | |
75fb9148 ZB |
1795 | batch->count = nb_rx; |
1796 | dp_packet_batch_init_packet_fields(batch); | |
1797 | ||
58397e6c KT |
1798 | return 0; |
1799 | } | |
1800 | ||
8a9562d2 | 1801 | static int |
64839cf4 | 1802 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1803 | { |
d46285a2 DDP |
1804 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1805 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1806 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1807 | int nb_rx; |
9509913a | 1808 | int dropped = 0; |
8a9562d2 | 1809 | |
3b1fb077 DDP |
1810 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1811 | return EAGAIN; | |
1812 | } | |
1813 | ||
d46285a2 | 1814 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1815 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1816 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1817 | if (!nb_rx) { |
1818 | return EAGAIN; | |
1819 | } | |
1820 | ||
9509913a IS |
1821 | if (policer) { |
1822 | dropped = nb_rx; | |
64839cf4 | 1823 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 1824 | (struct rte_mbuf **) batch->packets, |
3e90f7d7 | 1825 | nb_rx, true); |
9509913a IS |
1826 | dropped -= nb_rx; |
1827 | } | |
1828 | ||
1829 | /* Update stats to reflect dropped packets */ | |
1830 | if (OVS_UNLIKELY(dropped)) { | |
1831 | rte_spinlock_lock(&dev->stats_lock); | |
1832 | dev->stats.rx_dropped += dropped; | |
1833 | rte_spinlock_unlock(&dev->stats_lock); | |
1834 | } | |
1835 | ||
64839cf4 | 1836 | batch->count = nb_rx; |
75fb9148 | 1837 | dp_packet_batch_init_packet_fields(batch); |
8a9562d2 PS |
1838 | |
1839 | return 0; | |
1840 | } | |
1841 | ||
0bf765f7 | 1842 | static inline int |
78bd47cf | 1843 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
3e90f7d7 | 1844 | int cnt, bool may_steal) |
0bf765f7 | 1845 | { |
78bd47cf | 1846 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 1847 | |
78bd47cf DDP |
1848 | if (qos_conf) { |
1849 | rte_spinlock_lock(&qos_conf->lock); | |
3e90f7d7 | 1850 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt, may_steal); |
78bd47cf | 1851 | rte_spinlock_unlock(&qos_conf->lock); |
0bf765f7 IS |
1852 | } |
1853 | ||
1854 | return cnt; | |
1855 | } | |
1856 | ||
c6ec9d17 IM |
1857 | static int |
1858 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1859 | int pkt_cnt) | |
1860 | { | |
1861 | int i = 0; | |
1862 | int cnt = 0; | |
1863 | struct rte_mbuf *pkt; | |
1864 | ||
1865 | for (i = 0; i < pkt_cnt; i++) { | |
1866 | pkt = pkts[i]; | |
1867 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
1868 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
1869 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
1870 | rte_pktmbuf_free(pkt); | |
1871 | continue; | |
1872 | } | |
1873 | ||
1874 | if (OVS_UNLIKELY(i != cnt)) { | |
1875 | pkts[cnt] = pkt; | |
1876 | } | |
1877 | cnt++; | |
1878 | } | |
1879 | ||
1880 | return cnt; | |
1881 | } | |
1882 | ||
9e3ddd45 TP |
1883 | static inline void |
1884 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1885 | struct dp_packet **packets, | |
1886 | int attempted, | |
1887 | int dropped) | |
1888 | { | |
1889 | int i; | |
1890 | int sent = attempted - dropped; | |
1891 | ||
1892 | stats->tx_packets += sent; | |
1893 | stats->tx_dropped += dropped; | |
1894 | ||
1895 | for (i = 0; i < sent; i++) { | |
1896 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1897 | } | |
1898 | } | |
1899 | ||
58397e6c | 1900 | static void |
4573fbd3 | 1901 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 1902 | struct dp_packet **pkts, int cnt) |
58397e6c | 1903 | { |
d46285a2 | 1904 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1905 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1906 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 1907 | unsigned int dropped = 0; |
dd52de45 | 1908 | int i, retries = 0; |
daf22bf7 | 1909 | int vid = netdev_dpdk_get_vid(dev); |
58397e6c | 1910 | |
81acebda | 1911 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1912 | |
daf22bf7 | 1913 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured || qid < 0 |
e543851d | 1914 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1915 | rte_spinlock_lock(&dev->stats_lock); |
1916 | dev->stats.tx_dropped+= cnt; | |
1917 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1918 | goto out; |
58397e6c KT |
1919 | } |
1920 | ||
d46285a2 | 1921 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1922 | |
c6ec9d17 | 1923 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 1924 | /* Check has QoS has been configured for the netdev */ |
3e90f7d7 | 1925 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt, true); |
c6ec9d17 | 1926 | dropped = total_pkts - cnt; |
0bf765f7 | 1927 | |
95e9881f | 1928 | do { |
4573fbd3 | 1929 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1930 | unsigned int tx_pkts; |
1931 | ||
daf22bf7 | 1932 | tx_pkts = rte_vhost_enqueue_burst(vid, vhost_qid, cur_pkts, cnt); |
95e9881f KT |
1933 | if (OVS_LIKELY(tx_pkts)) { |
1934 | /* Packets have been sent.*/ | |
1935 | cnt -= tx_pkts; | |
31871ee3 | 1936 | /* Prepare for possible retry.*/ |
95e9881f KT |
1937 | cur_pkts = &cur_pkts[tx_pkts]; |
1938 | } else { | |
31871ee3 KT |
1939 | /* No packets sent - do not retry.*/ |
1940 | break; | |
95e9881f | 1941 | } |
c6ec9d17 | 1942 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1943 | |
d46285a2 | 1944 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1945 | |
d46285a2 | 1946 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 1947 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 1948 | cnt + dropped); |
d46285a2 | 1949 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1950 | |
1951 | out: | |
c6ec9d17 | 1952 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 1953 | dp_packet_delete(pkts[i]); |
58397e6c KT |
1954 | } |
1955 | } | |
1956 | ||
8a9562d2 PS |
1957 | /* Tx function. Transmit packets indefinitely */ |
1958 | static void | |
64839cf4 | 1959 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1960 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1961 | { |
8a14bd7b | 1962 | const size_t batch_cnt = dp_packet_batch_size(batch); |
bce01e3a | 1963 | #if !defined(__CHECKER__) && !defined(_WIN32) |
8a14bd7b | 1964 | const size_t PKT_ARRAY_SIZE = batch_cnt; |
bce01e3a EJ |
1965 | #else |
1966 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1967 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1968 | #endif |
8a9562d2 | 1969 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 1970 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
8a14bd7b | 1971 | uint32_t cnt = batch_cnt; |
3e90f7d7 GZ |
1972 | uint32_t dropped = 0; |
1973 | ||
1974 | if (dev->type != DPDK_DEV_VHOST) { | |
1975 | /* Check if QoS has been configured for this netdev. */ | |
1976 | cnt = netdev_dpdk_qos_run(dev, (struct rte_mbuf **) batch->packets, | |
8a14bd7b BB |
1977 | batch_cnt, false); |
1978 | dropped += batch_cnt - cnt; | |
3e90f7d7 | 1979 | } |
8a9562d2 | 1980 | |
3e90f7d7 GZ |
1981 | uint32_t txcnt = 0; |
1982 | ||
1983 | for (uint32_t i = 0; i < cnt; i++) { | |
8a14bd7b BB |
1984 | struct dp_packet *packet = batch->packets[i]; |
1985 | uint32_t size = dp_packet_size(packet); | |
95fb793a | 1986 | |
f98d7864 | 1987 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
3e90f7d7 GZ |
1988 | VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d", |
1989 | size, dev->max_packet_len); | |
f4fd623c | 1990 | |
175cf4de | 1991 | dropped++; |
f4fd623c DDP |
1992 | continue; |
1993 | } | |
8a9562d2 | 1994 | |
24e78f93 | 1995 | pkts[txcnt] = rte_pktmbuf_alloc(dev->mp); |
8a14bd7b | 1996 | if (OVS_UNLIKELY(!pkts[txcnt])) { |
3e90f7d7 | 1997 | dropped += cnt - i; |
175cf4de | 1998 | break; |
f4fd623c DDP |
1999 | } |
2000 | ||
2001 | /* We have to do a copy for now */ | |
3e90f7d7 | 2002 | memcpy(rte_pktmbuf_mtod(pkts[txcnt], void *), |
8a14bd7b BB |
2003 | dp_packet_data(packet), size); |
2004 | dp_packet_set_size((struct dp_packet *)pkts[txcnt], size); | |
f4fd623c | 2005 | |
3e90f7d7 | 2006 | txcnt++; |
f4fd623c | 2007 | } |
8a9562d2 | 2008 | |
3e90f7d7 GZ |
2009 | if (OVS_LIKELY(txcnt)) { |
2010 | if (dev->type == DPDK_DEV_VHOST) { | |
2011 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, | |
2012 | txcnt); | |
2013 | } else { | |
2014 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, txcnt); | |
2015 | } | |
58397e6c | 2016 | } |
db73f716 | 2017 | |
0bf765f7 IS |
2018 | if (OVS_UNLIKELY(dropped)) { |
2019 | rte_spinlock_lock(&dev->stats_lock); | |
2020 | dev->stats.tx_dropped += dropped; | |
2021 | rte_spinlock_unlock(&dev->stats_lock); | |
2022 | } | |
8a9562d2 PS |
2023 | } |
2024 | ||
58397e6c | 2025 | static int |
64839cf4 WT |
2026 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
2027 | struct dp_packet_batch *batch, | |
b30896c9 | 2028 | bool concurrent_txq OVS_UNUSED) |
58397e6c | 2029 | { |
58397e6c | 2030 | |
b30896c9 | 2031 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 | 2032 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2033 | dp_packet_delete_batch(batch, true); |
58397e6c | 2034 | } else { |
dd52de45 | 2035 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
2036 | } |
2037 | return 0; | |
2038 | } | |
2039 | ||
7251515e DV |
2040 | static inline void |
2041 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
b30896c9 | 2042 | struct dp_packet_batch *batch, |
324c8374 | 2043 | bool concurrent_txq) |
8a9562d2 | 2044 | { |
3b1fb077 | 2045 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
b30896c9 | 2046 | dp_packet_delete_batch(batch, true); |
3b1fb077 DDP |
2047 | return; |
2048 | } | |
2049 | ||
324c8374 | 2050 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 2051 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
2052 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
2053 | } | |
2054 | ||
b30896c9 | 2055 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
2056 | struct netdev *netdev = &dev->up; |
2057 | ||
64839cf4 | 2058 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2059 | dp_packet_delete_batch(batch, true); |
8a9562d2 | 2060 | } else { |
fd57eeba BB |
2061 | int tx_cnt, dropped; |
2062 | int batch_cnt = dp_packet_batch_size(batch); | |
2391135c | 2063 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 2064 | |
fd57eeba BB |
2065 | tx_cnt = netdev_dpdk_filter_packet_len(dev, pkts, batch_cnt); |
2066 | tx_cnt = netdev_dpdk_qos_run(dev, pkts, tx_cnt, true); | |
2067 | dropped = batch_cnt - tx_cnt; | |
1b99bb05 | 2068 | |
fd57eeba | 2069 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, tx_cnt); |
8a9562d2 | 2070 | |
f4fd623c | 2071 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 2072 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 2073 | dev->stats.tx_dropped += dropped; |
45d947c4 | 2074 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 2075 | } |
8a9562d2 | 2076 | } |
a0cb2d66 | 2077 | |
324c8374 | 2078 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
2079 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
2080 | } | |
7251515e DV |
2081 | } |
2082 | ||
2083 | static int | |
2084 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
b30896c9 | 2085 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e DV |
2086 | { |
2087 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 2088 | |
b30896c9 | 2089 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e | 2090 | return 0; |
8a9562d2 PS |
2091 | } |
2092 | ||
2093 | static int | |
74ff3298 | 2094 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
2095 | { |
2096 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2097 | ||
2098 | ovs_mutex_lock(&dev->mutex); | |
2099 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 2100 | dev->hwaddr = mac; |
045c0d1a | 2101 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
2102 | } |
2103 | ovs_mutex_unlock(&dev->mutex); | |
2104 | ||
2105 | return 0; | |
2106 | } | |
2107 | ||
2108 | static int | |
74ff3298 | 2109 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
2110 | { |
2111 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2112 | ||
2113 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 2114 | *mac = dev->hwaddr; |
8a9562d2 PS |
2115 | ovs_mutex_unlock(&dev->mutex); |
2116 | ||
2117 | return 0; | |
2118 | } | |
2119 | ||
2120 | static int | |
2121 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
2122 | { | |
2123 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2124 | ||
2125 | ovs_mutex_lock(&dev->mutex); | |
2126 | *mtup = dev->mtu; | |
2127 | ovs_mutex_unlock(&dev->mutex); | |
2128 | ||
2129 | return 0; | |
2130 | } | |
2131 | ||
0072e931 MK |
2132 | static int |
2133 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
2134 | { | |
2135 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2136 | ||
2137 | if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
2138 | || mtu < ETHER_MIN_MTU) { | |
2139 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
2140 | return EINVAL; | |
2141 | } | |
2142 | ||
2143 | ovs_mutex_lock(&dev->mutex); | |
2144 | if (dev->requested_mtu != mtu) { | |
2145 | dev->requested_mtu = mtu; | |
2146 | netdev_request_reconfigure(netdev); | |
2147 | } | |
2148 | ovs_mutex_unlock(&dev->mutex); | |
2149 | ||
2150 | return 0; | |
2151 | } | |
2152 | ||
8a9562d2 | 2153 | static int |
d46285a2 | 2154 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 2155 | |
58397e6c KT |
2156 | static int |
2157 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
2158 | struct netdev_stats *stats) | |
2159 | { | |
2160 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2161 | ||
2162 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 2163 | |
45d947c4 | 2164 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c | 2165 | /* Supported Stats */ |
50986e78 | 2166 | stats->rx_packets = dev->stats.rx_packets; |
2167 | stats->tx_packets = dev->stats.tx_packets; | |
9509913a | 2168 | stats->rx_dropped = dev->stats.rx_dropped; |
50986e78 | 2169 | stats->tx_dropped = dev->stats.tx_dropped; |
9e3ddd45 TP |
2170 | stats->multicast = dev->stats.multicast; |
2171 | stats->rx_bytes = dev->stats.rx_bytes; | |
2172 | stats->tx_bytes = dev->stats.tx_bytes; | |
2173 | stats->rx_errors = dev->stats.rx_errors; | |
2174 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 2175 | |
2176 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
2177 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
2178 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
2179 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
2180 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
2181 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
2182 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
2183 | ||
45d947c4 | 2184 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 2185 | |
58397e6c KT |
2186 | ovs_mutex_unlock(&dev->mutex); |
2187 | ||
2188 | return 0; | |
2189 | } | |
2190 | ||
d6e3feb5 | 2191 | static void |
2192 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
2193 | const struct rte_eth_xstat *xstats, |
2194 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 2195 | const unsigned int size) |
2196 | { | |
d6e3feb5 | 2197 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 2198 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2199 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2200 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2201 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2202 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2203 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2204 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2205 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2206 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2207 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2208 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2209 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2210 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2211 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 2212 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2213 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2214 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2215 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2216 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2217 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2218 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2219 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2220 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2221 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2222 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2223 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2224 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2225 | stats->tx_1523_to_max_packets = xstats[i].value; |
d57f777f PS |
2226 | } else if (strcmp(XSTAT_RX_MULTICAST_PACKETS, names[i].name) == 0) { |
2227 | stats->multicast = xstats[i].value; | |
0a0f39df | 2228 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2229 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 2230 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2231 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2232 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2233 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2234 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2235 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2236 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2237 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2238 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2239 | stats->rx_jabber_errors = xstats[i].value; |
2240 | } | |
2241 | } | |
2242 | } | |
2243 | ||
8a9562d2 PS |
2244 | static int |
2245 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2246 | { | |
2247 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2248 | struct rte_eth_stats rte_stats; | |
2249 | bool gg; | |
2250 | ||
2251 | netdev_dpdk_get_carrier(netdev, &gg); | |
2252 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2253 | |
0a0f39df CL |
2254 | struct rte_eth_xstat *rte_xstats = NULL; |
2255 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2256 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2257 | |
2258 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
bb37956a | 2259 | VLOG_ERR("Can't get ETH statistics for port: %"PRIu8, dev->port_id); |
f9256822 | 2260 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2261 | return EPROTO; |
2262 | } | |
2263 | ||
0a0f39df CL |
2264 | /* Get length of statistics */ |
2265 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2266 | if (rte_xstats_len < 0) { | |
bb37956a | 2267 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2268 | goto out; |
2269 | } | |
2270 | /* Reserve memory for xstats names and values */ | |
2271 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2272 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2273 | ||
2274 | /* Retreive xstats names */ | |
2275 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2276 | rte_xstats_names, | |
2277 | rte_xstats_len); | |
2278 | if (rte_xstats_new_len != rte_xstats_len) { | |
bb37956a | 2279 | VLOG_WARN("Cannot get XSTATS names for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2280 | goto out; |
2281 | } | |
2282 | /* Retreive xstats values */ | |
2283 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2284 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2285 | rte_xstats_len); | |
2286 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2287 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2288 | rte_xstats_len); | |
d6e3feb5 | 2289 | } else { |
bb37956a | 2290 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
d6e3feb5 | 2291 | } |
8a9562d2 | 2292 | |
0a0f39df CL |
2293 | out: |
2294 | free(rte_xstats); | |
2295 | free(rte_xstats_names); | |
2296 | ||
2f9dd77f PS |
2297 | stats->rx_packets = rte_stats.ipackets; |
2298 | stats->tx_packets = rte_stats.opackets; | |
2299 | stats->rx_bytes = rte_stats.ibytes; | |
2300 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2301 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2302 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2303 | |
45d947c4 | 2304 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2305 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2306 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2307 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2308 | |
2309 | /* These are the available DPDK counters for packets not received due to | |
2310 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2311 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2312 | stats->rx_missed_errors = rte_stats.imissed; |
2313 | ||
8a9562d2 PS |
2314 | ovs_mutex_unlock(&dev->mutex); |
2315 | ||
2316 | return 0; | |
2317 | } | |
2318 | ||
971f4b39 MW |
2319 | static int |
2320 | netdev_dpdk_get_custom_stats(const struct netdev *netdev, | |
2321 | struct netdev_custom_stats *custom_stats) | |
2322 | { | |
2323 | ||
2324 | uint32_t i; | |
2325 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2326 | int rte_xstats_ret; | |
2327 | ||
2328 | ovs_mutex_lock(&dev->mutex); | |
2329 | ||
2330 | if (netdev_dpdk_configure_xstats(dev)) { | |
2331 | uint64_t *values = xcalloc(dev->rte_xstats_ids_size, | |
2332 | sizeof(uint64_t)); | |
2333 | ||
2334 | rte_xstats_ret = | |
2335 | rte_eth_xstats_get_by_id(dev->port_id, dev->rte_xstats_ids, | |
2336 | values, dev->rte_xstats_ids_size); | |
2337 | ||
2338 | if (rte_xstats_ret > 0 && | |
2339 | rte_xstats_ret <= dev->rte_xstats_ids_size) { | |
2340 | ||
2341 | custom_stats->size = rte_xstats_ret; | |
2342 | custom_stats->counters = | |
2343 | (struct netdev_custom_counter *) xcalloc(rte_xstats_ret, | |
2344 | sizeof(struct netdev_custom_counter)); | |
2345 | ||
2346 | for (i = 0; i < rte_xstats_ret; i++) { | |
2347 | ovs_strlcpy(custom_stats->counters[i].name, | |
2348 | netdev_dpdk_get_xstat_name(dev, | |
2349 | dev->rte_xstats_ids[i]), | |
2350 | NETDEV_CUSTOM_STATS_NAME_SIZE); | |
2351 | custom_stats->counters[i].value = values[i]; | |
2352 | } | |
2353 | } else { | |
2354 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, | |
2355 | dev->port_id); | |
2356 | custom_stats->counters = NULL; | |
2357 | custom_stats->size = 0; | |
2358 | /* Let's clear statistics cache, so it will be | |
2359 | * reconfigured */ | |
2360 | netdev_dpdk_clear_xstats(dev); | |
2361 | } | |
2362 | } | |
2363 | ||
2364 | ovs_mutex_unlock(&dev->mutex); | |
2365 | ||
2366 | return 0; | |
2367 | } | |
2368 | ||
8a9562d2 | 2369 | static int |
d46285a2 | 2370 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2371 | enum netdev_features *current, |
ca3d4f55 BX |
2372 | enum netdev_features *advertised, |
2373 | enum netdev_features *supported, | |
2374 | enum netdev_features *peer) | |
8a9562d2 | 2375 | { |
d46285a2 | 2376 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2377 | struct rte_eth_link link; |
2378 | ||
2379 | ovs_mutex_lock(&dev->mutex); | |
2380 | link = dev->link; | |
2381 | ovs_mutex_unlock(&dev->mutex); | |
2382 | ||
362ca396 | 2383 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2384 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
2385 | *current = NETDEV_F_10MB_HD; |
2386 | } | |
362ca396 | 2387 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2388 | *current = NETDEV_F_100MB_HD; |
2389 | } | |
362ca396 | 2390 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2391 | *current = NETDEV_F_1GB_HD; |
2392 | } | |
2393 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 2394 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
2395 | *current = NETDEV_F_10MB_FD; |
2396 | } | |
362ca396 | 2397 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2398 | *current = NETDEV_F_100MB_FD; |
2399 | } | |
362ca396 | 2400 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2401 | *current = NETDEV_F_1GB_FD; |
2402 | } | |
362ca396 | 2403 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
2404 | *current = NETDEV_F_10GB_FD; |
2405 | } | |
2406 | } | |
2407 | ||
362ca396 | 2408 | if (link.link_autoneg) { |
2409 | *current |= NETDEV_F_AUTONEG; | |
2410 | } | |
2411 | ||
ca3d4f55 BX |
2412 | *advertised = *supported = *peer = 0; |
2413 | ||
8a9562d2 PS |
2414 | return 0; |
2415 | } | |
2416 | ||
9509913a IS |
2417 | static struct ingress_policer * |
2418 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2419 | { | |
2420 | struct ingress_policer *policer = NULL; | |
2421 | uint64_t rate_bytes; | |
2422 | uint64_t burst_bytes; | |
2423 | int err = 0; | |
2424 | ||
2425 | policer = xmalloc(sizeof *policer); | |
2426 | rte_spinlock_init(&policer->policer_lock); | |
2427 | ||
2428 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
602c8668 LR |
2429 | rate_bytes = rate * 1000ULL / 8; |
2430 | burst_bytes = burst * 1000ULL / 8; | |
9509913a IS |
2431 | |
2432 | policer->app_srtcm_params.cir = rate_bytes; | |
2433 | policer->app_srtcm_params.cbs = burst_bytes; | |
2434 | policer->app_srtcm_params.ebs = 0; | |
2435 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2436 | &policer->app_srtcm_params); | |
58be5c0e | 2437 | if (err) { |
9509913a | 2438 | VLOG_ERR("Could not create rte meter for ingress policer"); |
4c47ddde | 2439 | free(policer); |
9509913a IS |
2440 | return NULL; |
2441 | } | |
2442 | ||
2443 | return policer; | |
2444 | } | |
2445 | ||
2446 | static int | |
2447 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2448 | uint32_t policer_burst) | |
2449 | { | |
2450 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2451 | struct ingress_policer *policer; | |
2452 | ||
2453 | /* Force to 0 if no rate specified, | |
2454 | * default to 8000 kbits if burst is 0, | |
2455 | * else stick with user-specified value. | |
2456 | */ | |
2457 | policer_burst = (!policer_rate ? 0 | |
2458 | : !policer_burst ? 8000 | |
2459 | : policer_burst); | |
2460 | ||
2461 | ovs_mutex_lock(&dev->mutex); | |
2462 | ||
2463 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2464 | &dev->ingress_policer); | |
2465 | ||
2466 | if (dev->policer_rate == policer_rate && | |
2467 | dev->policer_burst == policer_burst) { | |
2468 | /* Assume that settings haven't changed since we last set them. */ | |
2469 | ovs_mutex_unlock(&dev->mutex); | |
2470 | return 0; | |
2471 | } | |
2472 | ||
2473 | /* Destroy any existing ingress policer for the device if one exists */ | |
2474 | if (policer) { | |
2475 | ovsrcu_postpone(free, policer); | |
2476 | } | |
2477 | ||
2478 | if (policer_rate != 0) { | |
2479 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2480 | } else { | |
2481 | policer = NULL; | |
2482 | } | |
2483 | ovsrcu_set(&dev->ingress_policer, policer); | |
2484 | dev->policer_rate = policer_rate; | |
2485 | dev->policer_burst = policer_burst; | |
2486 | ovs_mutex_unlock(&dev->mutex); | |
2487 | ||
2488 | return 0; | |
2489 | } | |
2490 | ||
8a9562d2 PS |
2491 | static int |
2492 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2493 | { | |
2494 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
2495 | |
2496 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
2497 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
2498 | * postive integer to meet RFC 2863 recommendations. | |
2499 | */ | |
2500 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
2501 | ovs_mutex_unlock(&dev->mutex); |
2502 | ||
2503 | return ifindex; | |
2504 | } | |
2505 | ||
2506 | static int | |
d46285a2 | 2507 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2508 | { |
d46285a2 | 2509 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2510 | |
2511 | ovs_mutex_lock(&dev->mutex); | |
2512 | check_link_status(dev); | |
2513 | *carrier = dev->link.link_status; | |
58397e6c KT |
2514 | |
2515 | ovs_mutex_unlock(&dev->mutex); | |
2516 | ||
2517 | return 0; | |
2518 | } | |
2519 | ||
2520 | static int | |
d46285a2 | 2521 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2522 | { |
d46285a2 | 2523 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2524 | |
2525 | ovs_mutex_lock(&dev->mutex); | |
2526 | ||
0a0f39df | 2527 | if (is_vhost_running(dev)) { |
58397e6c KT |
2528 | *carrier = 1; |
2529 | } else { | |
2530 | *carrier = 0; | |
2531 | } | |
2532 | ||
8a9562d2 PS |
2533 | ovs_mutex_unlock(&dev->mutex); |
2534 | ||
2535 | return 0; | |
2536 | } | |
2537 | ||
2538 | static long long int | |
d46285a2 | 2539 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2540 | { |
d46285a2 | 2541 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2542 | long long int carrier_resets; |
2543 | ||
2544 | ovs_mutex_lock(&dev->mutex); | |
2545 | carrier_resets = dev->link_reset_cnt; | |
2546 | ovs_mutex_unlock(&dev->mutex); | |
2547 | ||
2548 | return carrier_resets; | |
2549 | } | |
2550 | ||
2551 | static int | |
d46285a2 | 2552 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2553 | long long int interval OVS_UNUSED) |
2554 | { | |
ee32150e | 2555 | return EOPNOTSUPP; |
8a9562d2 PS |
2556 | } |
2557 | ||
2558 | static int | |
2559 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2560 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2561 | enum netdev_flags *old_flagsp) |
2562 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 2563 | { |
8a9562d2 PS |
2564 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
2565 | return EINVAL; | |
2566 | } | |
2567 | ||
2568 | *old_flagsp = dev->flags; | |
2569 | dev->flags |= on; | |
2570 | dev->flags &= ~off; | |
2571 | ||
2572 | if (dev->flags == *old_flagsp) { | |
2573 | return 0; | |
2574 | } | |
2575 | ||
58397e6c | 2576 | if (dev->type == DPDK_DEV_ETH) { |
58397e6c KT |
2577 | if (dev->flags & NETDEV_PROMISC) { |
2578 | rte_eth_promiscuous_enable(dev->port_id); | |
2579 | } | |
8a9562d2 | 2580 | |
314fb5ad | 2581 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
2582 | } else { |
2583 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2584 | * running then change netdev's change_seq to trigger link state | |
2585 | * update. */ | |
e543851d ZB |
2586 | |
2587 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2588 | && is_vhost_running(dev)) { |
e543851d ZB |
2589 | netdev_change_seq_changed(&dev->up); |
2590 | ||
2591 | /* Clear statistics if device is getting up. */ | |
2592 | if (NETDEV_UP & on) { | |
2593 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 2594 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
2595 | rte_spinlock_unlock(&dev->stats_lock); |
2596 | } | |
2597 | } | |
8a9562d2 PS |
2598 | } |
2599 | ||
2600 | return 0; | |
2601 | } | |
2602 | ||
2603 | static int | |
d46285a2 | 2604 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2605 | enum netdev_flags off, enum netdev_flags on, |
2606 | enum netdev_flags *old_flagsp) | |
2607 | { | |
d46285a2 | 2608 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2609 | int error; |
2610 | ||
d46285a2 DDP |
2611 | ovs_mutex_lock(&dev->mutex); |
2612 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2613 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2614 | |
2615 | return error; | |
2616 | } | |
2617 | ||
b2e8b12f FL |
2618 | static int |
2619 | netdev_dpdk_vhost_user_get_status(const struct netdev *netdev, | |
2620 | struct smap *args) | |
2621 | { | |
2622 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2623 | ||
2624 | ovs_mutex_lock(&dev->mutex); | |
2625 | ||
2626 | bool client_mode = dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT; | |
2627 | smap_add_format(args, "mode", "%s", client_mode ? "client" : "server"); | |
2628 | ||
2629 | int vid = netdev_dpdk_get_vid(dev); | |
2630 | if (vid < 0) { | |
2631 | smap_add_format(args, "status", "disconnected"); | |
2632 | ovs_mutex_unlock(&dev->mutex); | |
2633 | return 0; | |
2634 | } else { | |
2635 | smap_add_format(args, "status", "connected"); | |
2636 | } | |
2637 | ||
2638 | char socket_name[PATH_MAX]; | |
2639 | if (!rte_vhost_get_ifname(vid, socket_name, PATH_MAX)) { | |
2640 | smap_add_format(args, "socket", "%s", socket_name); | |
2641 | } | |
2642 | ||
2643 | uint64_t features; | |
2644 | if (!rte_vhost_get_negotiated_features(vid, &features)) { | |
2645 | smap_add_format(args, "features", "0x%016"PRIx64, features); | |
2646 | } | |
2647 | ||
2648 | uint16_t mtu; | |
2649 | if (!rte_vhost_get_mtu(vid, &mtu)) { | |
2650 | smap_add_format(args, "mtu", "%d", mtu); | |
2651 | } | |
2652 | ||
2653 | int numa = rte_vhost_get_numa_node(vid); | |
2654 | if (numa >= 0) { | |
2655 | smap_add_format(args, "numa", "%d", numa); | |
2656 | } | |
2657 | ||
2658 | uint16_t vring_num = rte_vhost_get_vring_num(vid); | |
2659 | if (vring_num) { | |
2660 | smap_add_format(args, "num_of_vrings", "%d", vring_num); | |
2661 | } | |
2662 | ||
2663 | for (int i = 0; i < vring_num; i++) { | |
2664 | struct rte_vhost_vring vring; | |
2665 | char vhost_vring[16]; | |
2666 | ||
2667 | rte_vhost_get_vhost_vring(vid, i, &vring); | |
2668 | snprintf(vhost_vring, 16, "vring_%d_size", i); | |
2669 | smap_add_format(args, vhost_vring, "%d", vring.size); | |
2670 | } | |
2671 | ||
2672 | ovs_mutex_unlock(&dev->mutex); | |
2673 | return 0; | |
2674 | } | |
2675 | ||
8a9562d2 | 2676 | static int |
d46285a2 | 2677 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2678 | { |
d46285a2 | 2679 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2680 | struct rte_eth_dev_info dev_info; |
2681 | ||
7cd1261d | 2682 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2683 | return ENODEV; |
7cd1261d | 2684 | } |
8a9562d2 PS |
2685 | |
2686 | ovs_mutex_lock(&dev->mutex); | |
2687 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2688 | ovs_mutex_unlock(&dev->mutex); | |
2689 | ||
95fb793a | 2690 | smap_add_format(args, "port_no", "%d", dev->port_id); |
58be5c0e MK |
2691 | smap_add_format(args, "numa_id", "%d", |
2692 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
2693 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
2694 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2695 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2696 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2697 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2698 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
2699 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
2700 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
2701 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
2702 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2703 | ||
3eb8d4fa MW |
2704 | /* Querying the DPDK library for iftype may be done in future, pending |
2705 | * support; cf. RFC 3635 Section 3.2.4. */ | |
2706 | enum { IF_TYPE_ETHERNETCSMACD = 6 }; | |
2707 | ||
2708 | smap_add_format(args, "if_type", "%"PRIu32, IF_TYPE_ETHERNETCSMACD); | |
2709 | smap_add_format(args, "if_descr", "%s %s", rte_version(), | |
2710 | dev_info.driver_name); | |
2711 | ||
39c2baa9 | 2712 | if (dev_info.pci_dev) { |
2713 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2714 | dev_info.pci_dev->id.vendor_id); | |
2715 | smap_add_format(args, "pci-device_id", "0x%x", | |
2716 | dev_info.pci_dev->id.device_id); | |
2717 | } | |
8a9562d2 PS |
2718 | |
2719 | return 0; | |
2720 | } | |
2721 | ||
2722 | static void | |
2723 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2724 | OVS_REQUIRES(dev->mutex) | |
2725 | { | |
2726 | enum netdev_flags old_flags; | |
2727 | ||
2728 | if (admin_state) { | |
2729 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2730 | } else { | |
2731 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2732 | } | |
2733 | } | |
2734 | ||
2735 | static void | |
2736 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2737 | const char *argv[], void *aux OVS_UNUSED) | |
2738 | { | |
2739 | bool up; | |
2740 | ||
2741 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2742 | up = true; | |
2743 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2744 | up = false; | |
2745 | } else { | |
2746 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2747 | return; | |
2748 | } | |
2749 | ||
2750 | if (argc > 2) { | |
2751 | struct netdev *netdev = netdev_from_name(argv[1]); | |
3d0d5ab1 | 2752 | |
8a9562d2 | 2753 | if (netdev && is_dpdk_class(netdev->netdev_class)) { |
3d0d5ab1 | 2754 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 2755 | |
3d0d5ab1 IM |
2756 | ovs_mutex_lock(&dev->mutex); |
2757 | netdev_dpdk_set_admin_state__(dev, up); | |
2758 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2759 | |
2760 | netdev_close(netdev); | |
2761 | } else { | |
2762 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2763 | netdev_close(netdev); | |
2764 | return; | |
2765 | } | |
2766 | } else { | |
3d0d5ab1 | 2767 | struct netdev_dpdk *dev; |
8a9562d2 PS |
2768 | |
2769 | ovs_mutex_lock(&dpdk_mutex); | |
3d0d5ab1 IM |
2770 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
2771 | ovs_mutex_lock(&dev->mutex); | |
2772 | netdev_dpdk_set_admin_state__(dev, up); | |
2773 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2774 | } |
2775 | ovs_mutex_unlock(&dpdk_mutex); | |
2776 | } | |
2777 | unixctl_command_reply(conn, "OK"); | |
2778 | } | |
2779 | ||
0ee821c2 DB |
2780 | static void |
2781 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2782 | const char *argv[], void *aux OVS_UNUSED) | |
2783 | { | |
2784 | int ret; | |
2785 | char *response; | |
7ee94cba | 2786 | dpdk_port_t port_id; |
0ee821c2 DB |
2787 | char devname[RTE_ETH_NAME_MAX_LEN]; |
2788 | struct netdev_dpdk *dev; | |
2789 | ||
2790 | ovs_mutex_lock(&dpdk_mutex); | |
2791 | ||
255b7bda | 2792 | if (rte_eth_dev_get_port_by_name(argv[1], &port_id)) { |
0ee821c2 DB |
2793 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); |
2794 | goto error; | |
2795 | } | |
2796 | ||
2797 | dev = netdev_dpdk_lookup_by_port_id(port_id); | |
2798 | if (dev) { | |
2799 | response = xasprintf("Device '%s' is being used by interface '%s'. " | |
2800 | "Remove it before detaching", | |
2801 | argv[1], netdev_get_name(&dev->up)); | |
2802 | goto error; | |
2803 | } | |
2804 | ||
2805 | rte_eth_dev_close(port_id); | |
2806 | ||
2807 | ret = rte_eth_dev_detach(port_id, devname); | |
2808 | if (ret < 0) { | |
2809 | response = xasprintf("Device '%s' can not be detached", argv[1]); | |
2810 | goto error; | |
2811 | } | |
2812 | ||
2813 | response = xasprintf("Device '%s' has been detached", argv[1]); | |
2814 | ||
2815 | ovs_mutex_unlock(&dpdk_mutex); | |
2816 | unixctl_command_reply(conn, response); | |
2817 | free(response); | |
2818 | return; | |
2819 | ||
2820 | error: | |
2821 | ovs_mutex_unlock(&dpdk_mutex); | |
2822 | unixctl_command_reply_error(conn, response); | |
2823 | free(response); | |
2824 | } | |
2825 | ||
be481733 IM |
2826 | static void |
2827 | netdev_dpdk_get_mempool_info(struct unixctl_conn *conn, | |
2828 | int argc, const char *argv[], | |
2829 | void *aux OVS_UNUSED) | |
2830 | { | |
2831 | size_t size; | |
2832 | FILE *stream; | |
2833 | char *response = NULL; | |
2834 | struct netdev *netdev = NULL; | |
2835 | ||
2836 | if (argc == 2) { | |
2837 | netdev = netdev_from_name(argv[1]); | |
2838 | if (!netdev || !is_dpdk_class(netdev->netdev_class)) { | |
2839 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2840 | goto out; | |
2841 | } | |
2842 | } | |
2843 | ||
2844 | stream = open_memstream(&response, &size); | |
2845 | if (!stream) { | |
2846 | response = xasprintf("Unable to open memstream: %s.", | |
2847 | ovs_strerror(errno)); | |
2848 | unixctl_command_reply_error(conn, response); | |
2849 | goto out; | |
2850 | } | |
2851 | ||
2852 | if (netdev) { | |
2853 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2854 | ||
2855 | ovs_mutex_lock(&dev->mutex); | |
2856 | ovs_mutex_lock(&dpdk_mp_mutex); | |
2857 | ||
2858 | rte_mempool_dump(stream, dev->mp); | |
2859 | ||
2860 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
2861 | ovs_mutex_unlock(&dev->mutex); | |
2862 | } else { | |
2863 | ovs_mutex_lock(&dpdk_mp_mutex); | |
2864 | rte_mempool_list_dump(stream); | |
2865 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
2866 | } | |
2867 | ||
2868 | fclose(stream); | |
2869 | ||
2870 | unixctl_command_reply(conn, response); | |
2871 | out: | |
2872 | free(response); | |
2873 | netdev_close(netdev); | |
2874 | } | |
2875 | ||
58397e6c KT |
2876 | /* |
2877 | * Set virtqueue flags so that we do not receive interrupts. | |
2878 | */ | |
2879 | static void | |
0a0f39df | 2880 | set_irq_status(int vid) |
58397e6c | 2881 | { |
4573fbd3 | 2882 | uint32_t i; |
4573fbd3 | 2883 | |
f3e7ec25 MW |
2884 | for (i = 0; i < rte_vhost_get_vring_num(vid); i++) { |
2885 | rte_vhost_enable_guest_notification(vid, i, 0); | |
4573fbd3 FL |
2886 | } |
2887 | } | |
2888 | ||
585a5bea IM |
2889 | /* |
2890 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2891 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2892 | */ |
2893 | static void | |
d46285a2 DDP |
2894 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2895 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2896 | { |
2897 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2898 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 2899 | |
eff23640 | 2900 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
2901 | |
2902 | for (i = 0; i < total_txqs; i++) { | |
2903 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2904 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2905 | enabled_queues[n_enabled++] = i; |
2906 | } | |
2907 | } | |
2908 | ||
2909 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2910 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2911 | n_enabled = 1; |
2912 | } | |
2913 | ||
2914 | k = 0; | |
2915 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2916 | if (dev->tx_q[i].map != i) { |
2917 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2918 | k = (k + 1) % n_enabled; |
2919 | } | |
2920 | } | |
2921 | ||
2d24d165 | 2922 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2923 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2924 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2925 | } |
2926 | ||
eff23640 | 2927 | free(enabled_queues); |
585a5bea | 2928 | } |
4573fbd3 | 2929 | |
58397e6c KT |
2930 | /* |
2931 | * A new virtio-net device is added to a vhost port. | |
2932 | */ | |
2933 | static int | |
0a0f39df | 2934 | new_device(int vid) |
58397e6c | 2935 | { |
d46285a2 | 2936 | struct netdev_dpdk *dev; |
58397e6c | 2937 | bool exists = false; |
db8f13b0 | 2938 | int newnode = 0; |
0a0f39df CL |
2939 | char ifname[IF_NAME_SZ]; |
2940 | ||
58be5c0e | 2941 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2942 | |
2943 | ovs_mutex_lock(&dpdk_mutex); | |
2944 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2945 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 2946 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2947 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
f3e7ec25 | 2948 | uint32_t qp_num = rte_vhost_get_vring_num(vid)/VIRTIO_QNUM; |
db8f13b0 CL |
2949 | |
2950 | /* Get NUMA information */ | |
0a0f39df CL |
2951 | newnode = rte_vhost_get_numa_node(vid); |
2952 | if (newnode == -1) { | |
5b9bf9e0 | 2953 | #ifdef VHOST_NUMA |
db8f13b0 | 2954 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2955 | ifname); |
5b9bf9e0 | 2956 | #endif |
db8f13b0 | 2957 | newnode = dev->socket_id; |
db8f13b0 CL |
2958 | } |
2959 | ||
7f5f2bd0 IM |
2960 | if (dev->requested_n_txq != qp_num |
2961 | || dev->requested_n_rxq != qp_num | |
2962 | || dev->requested_socket_id != newnode) { | |
2963 | dev->requested_socket_id = newnode; | |
2964 | dev->requested_n_rxq = qp_num; | |
2965 | dev->requested_n_txq = qp_num; | |
2966 | netdev_request_reconfigure(&dev->up); | |
2967 | } else { | |
2968 | /* Reconfiguration not required. */ | |
2969 | dev->vhost_reconfigured = true; | |
2970 | } | |
81acebda | 2971 | |
0a0f39df | 2972 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2973 | exists = true; |
2974 | ||
58397e6c | 2975 | /* Disable notifications. */ |
0a0f39df | 2976 | set_irq_status(vid); |
e543851d | 2977 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2978 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2979 | break; |
2980 | } | |
c1ff66ac | 2981 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2982 | } |
2983 | ovs_mutex_unlock(&dpdk_mutex); | |
2984 | ||
2985 | if (!exists) { | |
0a0f39df | 2986 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2987 | |
2988 | return -1; | |
2989 | } | |
2990 | ||
0a0f39df CL |
2991 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2992 | ifname, newnode); | |
2993 | ||
58397e6c KT |
2994 | return 0; |
2995 | } | |
2996 | ||
f3ea2ad2 IM |
2997 | /* Clears mapping for all available queues of vhost interface. */ |
2998 | static void | |
2999 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
3000 | OVS_REQUIRES(dev->mutex) | |
3001 | { | |
3002 | int i; | |
3003 | ||
81acebda | 3004 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
3005 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
3006 | } | |
3007 | } | |
3008 | ||
58397e6c KT |
3009 | /* |
3010 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
3011 | * flag to stop any more packets from being sent or received to/from a VM and | |
3012 | * ensure all currently queued packets have been sent/received before removing | |
3013 | * the device. | |
3014 | */ | |
3015 | static void | |
0a0f39df | 3016 | destroy_device(int vid) |
58397e6c | 3017 | { |
d46285a2 | 3018 | struct netdev_dpdk *dev; |
afee281f | 3019 | bool exists = false; |
0a0f39df CL |
3020 | char ifname[IF_NAME_SZ]; |
3021 | ||
58be5c0e | 3022 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
3023 | |
3024 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 3025 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 3026 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 3027 | |
d46285a2 | 3028 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
3029 | dev->vhost_reconfigured = false; |
3030 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 3031 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 3032 | |
e543851d | 3033 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 3034 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 3035 | exists = true; |
afee281f | 3036 | break; |
58397e6c KT |
3037 | } |
3038 | } | |
afee281f | 3039 | |
58397e6c KT |
3040 | ovs_mutex_unlock(&dpdk_mutex); |
3041 | ||
0a0f39df | 3042 | if (exists) { |
afee281f KT |
3043 | /* |
3044 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
3045 | * to NULL, before returning. | |
3046 | */ | |
3047 | ovsrcu_synchronize(); | |
3048 | /* | |
3049 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
3050 | * put thread back into quiescent state before returning. | |
3051 | */ | |
3052 | ovsrcu_quiesce_start(); | |
0a0f39df | 3053 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 3054 | } else { |
0a0f39df | 3055 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 3056 | } |
58397e6c KT |
3057 | } |
3058 | ||
585a5bea | 3059 | static int |
0a0f39df | 3060 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 3061 | { |
d46285a2 | 3062 | struct netdev_dpdk *dev; |
585a5bea IM |
3063 | bool exists = false; |
3064 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
3065 | char ifname[IF_NAME_SZ]; |
3066 | ||
58be5c0e | 3067 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
3068 | |
3069 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
3070 | return 0; | |
3071 | } | |
3072 | ||
3073 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 3074 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 3075 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 3076 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 3077 | if (enable) { |
d46285a2 | 3078 | dev->tx_q[qid].map = qid; |
585a5bea | 3079 | } else { |
d46285a2 | 3080 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 3081 | } |
d46285a2 | 3082 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 3083 | exists = true; |
d46285a2 | 3084 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3085 | break; |
3086 | } | |
c1ff66ac | 3087 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3088 | } |
3089 | ovs_mutex_unlock(&dpdk_mutex); | |
3090 | ||
3091 | if (exists) { | |
0a0f39df CL |
3092 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
3093 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 3094 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 3095 | } else { |
0a0f39df | 3096 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
3097 | return -1; |
3098 | } | |
3099 | ||
3100 | return 0; | |
3101 | } | |
3102 | ||
0a0f39df CL |
3103 | int |
3104 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 3105 | { |
0a0f39df | 3106 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
3107 | } |
3108 | ||
9509913a IS |
3109 | struct ingress_policer * |
3110 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
3111 | { | |
3112 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
3113 | } | |
3114 | ||
58397e6c | 3115 | static int |
ecc1a34e | 3116 | netdev_dpdk_class_init(void) |
7d1ced01 | 3117 | { |
ecc1a34e DDP |
3118 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
3119 | ||
3120 | /* This function can be called for different classes. The initialization | |
3121 | * needs to be done only once */ | |
3122 | if (ovsthread_once_start(&once)) { | |
3123 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
3124 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
3125 | "[netdev] up|down", 1, 2, | |
3126 | netdev_dpdk_set_admin_state, NULL); | |
3127 | ||
0ee821c2 DB |
3128 | unixctl_command_register("netdev-dpdk/detach", |
3129 | "pci address of device", 1, 1, | |
3130 | netdev_dpdk_detach, NULL); | |
3131 | ||
be481733 IM |
3132 | unixctl_command_register("netdev-dpdk/get-mempool-info", |
3133 | "[netdev]", 0, 1, | |
3134 | netdev_dpdk_get_mempool_info, NULL); | |
3135 | ||
ecc1a34e DDP |
3136 | ovsthread_once_done(&once); |
3137 | } | |
362ca396 | 3138 | |
7d1ced01 CL |
3139 | return 0; |
3140 | } | |
3141 | ||
033e9df2 | 3142 | |
95fb793a | 3143 | /* Client Rings */ |
3144 | ||
95fb793a | 3145 | static int |
3146 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
bb37956a | 3147 | dpdk_port_t *eth_port_id) |
95fb793a | 3148 | { |
48fffdee | 3149 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 3150 | char *ring_name; |
b8374d0d | 3151 | int port_id; |
95fb793a | 3152 | |
48fffdee KT |
3153 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
3154 | if (!ring_pair) { | |
95fb793a | 3155 | return ENOMEM; |
3156 | } | |
3157 | ||
7251515e | 3158 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 3159 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 3160 | |
8f0a76c9 | 3161 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 3162 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3163 | RING_F_SP_ENQ); |
0c6f39e5 | 3164 | free(ring_name); |
48fffdee KT |
3165 | if (ring_pair->cring_tx == NULL) { |
3166 | rte_free(ring_pair); | |
95fb793a | 3167 | return ENOMEM; |
3168 | } | |
3169 | ||
0c6f39e5 | 3170 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 3171 | |
8f0a76c9 | 3172 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 3173 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3174 | RING_F_SC_DEQ); |
0c6f39e5 | 3175 | free(ring_name); |
48fffdee KT |
3176 | if (ring_pair->cring_rx == NULL) { |
3177 | rte_free(ring_pair); | |
95fb793a | 3178 | return ENOMEM; |
3179 | } | |
3180 | ||
b8374d0d MV |
3181 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
3182 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 3183 | |
b8374d0d | 3184 | if (port_id < 0) { |
48fffdee | 3185 | rte_free(ring_pair); |
95fb793a | 3186 | return ENODEV; |
3187 | } | |
3188 | ||
48fffdee | 3189 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
3190 | ring_pair->eth_port_id = port_id; |
3191 | *eth_port_id = port_id; | |
3192 | ||
48fffdee | 3193 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 3194 | |
95fb793a | 3195 | return 0; |
3196 | } | |
3197 | ||
3198 | static int | |
bb37956a | 3199 | dpdk_ring_open(const char dev_name[], dpdk_port_t *eth_port_id) |
64839cf4 | 3200 | OVS_REQUIRES(dpdk_mutex) |
95fb793a | 3201 | { |
48fffdee | 3202 | struct dpdk_ring *ring_pair; |
95fb793a | 3203 | unsigned int port_no; |
3204 | int err = 0; | |
3205 | ||
3206 | /* Names always start with "dpdkr" */ | |
3207 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
3208 | if (err) { | |
3209 | return err; | |
3210 | } | |
3211 | ||
58be5c0e | 3212 | /* Look through our list to find the device */ |
48fffdee KT |
3213 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
3214 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 3215 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 3216 | /* Really all that is needed */ |
48fffdee | 3217 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 3218 | return 0; |
3219 | } | |
3220 | } | |
3221 | /* Need to create the device rings */ | |
3222 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
3223 | } | |
3224 | ||
7251515e | 3225 | static int |
d46285a2 | 3226 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
b30896c9 | 3227 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e | 3228 | { |
d46285a2 | 3229 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a543eb0 | 3230 | struct dp_packet *packet; |
1b99bb05 | 3231 | |
58be5c0e MK |
3232 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
3233 | * the rss hash field is clear. This is because the same mbuf may be | |
3234 | * modified by the consumer of the ring and return into the datapath | |
3235 | * without recalculating the RSS hash. */ | |
8a543eb0 BB |
3236 | DP_PACKET_BATCH_FOR_EACH (packet, batch) { |
3237 | dp_packet_mbuf_rss_flag_reset(packet); | |
1b99bb05 | 3238 | } |
7251515e | 3239 | |
b30896c9 | 3240 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e DV |
3241 | return 0; |
3242 | } | |
3243 | ||
95fb793a | 3244 | static int |
3245 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
3246 | { | |
bb37956a | 3247 | dpdk_port_t port_no = 0; |
95fb793a | 3248 | int err = 0; |
3249 | ||
95fb793a | 3250 | ovs_mutex_lock(&dpdk_mutex); |
3251 | ||
3252 | err = dpdk_ring_open(netdev->name, &port_no); | |
3253 | if (err) { | |
3254 | goto unlock_dpdk; | |
3255 | } | |
3256 | ||
1ce30dfd DDP |
3257 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
3258 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 3259 | unlock_dpdk: |
3260 | ovs_mutex_unlock(&dpdk_mutex); | |
3261 | return err; | |
3262 | } | |
3263 | ||
0bf765f7 IS |
3264 | /* QoS Functions */ |
3265 | ||
3266 | /* | |
3267 | * Initialize QoS configuration operations. | |
3268 | */ | |
3269 | static void | |
3270 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
3271 | { | |
3272 | conf->ops = ops; | |
78bd47cf | 3273 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
3274 | } |
3275 | ||
3276 | /* | |
3277 | * Search existing QoS operations in qos_ops and compare each set of | |
3278 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
3279 | * else return NULL | |
3280 | */ | |
3281 | static const struct dpdk_qos_ops * | |
3282 | qos_lookup_name(const char *name) | |
3283 | { | |
3284 | const struct dpdk_qos_ops *const *opsp; | |
3285 | ||
3286 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3287 | const struct dpdk_qos_ops *ops = *opsp; | |
3288 | if (!strcmp(name, ops->qos_name)) { | |
3289 | return ops; | |
3290 | } | |
3291 | } | |
3292 | return NULL; | |
3293 | } | |
3294 | ||
0bf765f7 IS |
3295 | static int |
3296 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
3297 | struct sset *types) | |
3298 | { | |
3299 | const struct dpdk_qos_ops *const *opsp; | |
3300 | ||
3301 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3302 | const struct dpdk_qos_ops *ops = *opsp; | |
3303 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
3304 | sset_add(types, ops->qos_name); | |
3305 | } | |
3306 | } | |
3307 | return 0; | |
3308 | } | |
3309 | ||
3310 | static int | |
d46285a2 | 3311 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
3312 | const char **typep, struct smap *details) |
3313 | { | |
d46285a2 | 3314 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 3315 | struct qos_conf *qos_conf; |
0bf765f7 IS |
3316 | int error = 0; |
3317 | ||
d46285a2 | 3318 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
3319 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
3320 | if (qos_conf) { | |
3321 | *typep = qos_conf->ops->qos_name; | |
3322 | error = (qos_conf->ops->qos_get | |
3323 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
3324 | } else { |
3325 | /* No QoS configuration set, return an empty string */ | |
3326 | *typep = ""; | |
0bf765f7 | 3327 | } |
d46285a2 | 3328 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
3329 | |
3330 | return error; | |
3331 | } | |
3332 | ||
3333 | static int | |
78bd47cf DDP |
3334 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
3335 | const struct smap *details) | |
0bf765f7 | 3336 | { |
d46285a2 | 3337 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 3338 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 3339 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
3340 | int error = 0; |
3341 | ||
d46285a2 | 3342 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 3343 | |
78bd47cf | 3344 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 3345 | |
78bd47cf DDP |
3346 | new_ops = qos_lookup_name(type); |
3347 | ||
3348 | if (!new_ops || !new_ops->qos_construct) { | |
3349 | new_qos_conf = NULL; | |
3350 | if (type && type[0]) { | |
3351 | error = EOPNOTSUPP; | |
0bf765f7 | 3352 | } |
44975bb0 | 3353 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
3354 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
3355 | new_qos_conf = qos_conf; | |
0bf765f7 | 3356 | } else { |
78bd47cf | 3357 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
3358 | } |
3359 | ||
7ea266e9 | 3360 | if (error) { |
78bd47cf DDP |
3361 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
3362 | type, netdev->name, rte_strerror(error)); | |
3363 | } | |
3364 | ||
3365 | if (new_qos_conf != qos_conf) { | |
3366 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
3367 | if (qos_conf) { | |
3368 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
3369 | } | |
0bf765f7 IS |
3370 | } |
3371 | ||
d46285a2 | 3372 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 3373 | |
0bf765f7 IS |
3374 | return error; |
3375 | } | |
3376 | ||
3377 | /* egress-policer details */ | |
3378 | ||
3379 | struct egress_policer { | |
3380 | struct qos_conf qos_conf; | |
3381 | struct rte_meter_srtcm_params app_srtcm_params; | |
3382 | struct rte_meter_srtcm egress_meter; | |
3383 | }; | |
3384 | ||
78bd47cf DDP |
3385 | static void |
3386 | egress_policer_details_to_param(const struct smap *details, | |
3387 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 3388 | { |
78bd47cf DDP |
3389 | memset(params, 0, sizeof *params); |
3390 | params->cir = smap_get_ullong(details, "cir", 0); | |
3391 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
3392 | params->ebs = 0; | |
0bf765f7 IS |
3393 | } |
3394 | ||
3395 | static int | |
78bd47cf DDP |
3396 | egress_policer_qos_construct(const struct smap *details, |
3397 | struct qos_conf **conf) | |
0bf765f7 | 3398 | { |
0bf765f7 | 3399 | struct egress_policer *policer; |
0bf765f7 IS |
3400 | int err = 0; |
3401 | ||
0bf765f7 IS |
3402 | policer = xmalloc(sizeof *policer); |
3403 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 3404 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
0bf765f7 | 3405 | err = rte_meter_srtcm_config(&policer->egress_meter, |
78bd47cf DDP |
3406 | &policer->app_srtcm_params); |
3407 | if (!err) { | |
3408 | *conf = &policer->qos_conf; | |
3409 | } else { | |
7ea266e9 | 3410 | free(policer); |
78bd47cf | 3411 | *conf = NULL; |
7ea266e9 IS |
3412 | err = -err; |
3413 | } | |
0bf765f7 IS |
3414 | |
3415 | return err; | |
3416 | } | |
3417 | ||
3418 | static void | |
78bd47cf | 3419 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
3420 | { |
3421 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 3422 | qos_conf); |
0bf765f7 IS |
3423 | free(policer); |
3424 | } | |
3425 | ||
3426 | static int | |
78bd47cf | 3427 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 3428 | { |
78bd47cf DDP |
3429 | struct egress_policer *policer = |
3430 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3431 | ||
3432 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
3433 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 3434 | |
0bf765f7 IS |
3435 | return 0; |
3436 | } | |
3437 | ||
78bd47cf | 3438 | static bool |
47a45d86 KT |
3439 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
3440 | const struct smap *details) | |
0bf765f7 | 3441 | { |
78bd47cf DDP |
3442 | struct egress_policer *policer = |
3443 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3444 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 3445 | |
78bd47cf | 3446 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 3447 | |
78bd47cf | 3448 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
3449 | } |
3450 | ||
0bf765f7 | 3451 | static int |
3e90f7d7 GZ |
3452 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt, |
3453 | bool may_steal) | |
0bf765f7 | 3454 | { |
0bf765f7 | 3455 | int cnt = 0; |
78bd47cf DDP |
3456 | struct egress_policer *policer = |
3457 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 3458 | |
3e90f7d7 GZ |
3459 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, |
3460 | pkt_cnt, may_steal); | |
0bf765f7 IS |
3461 | |
3462 | return cnt; | |
3463 | } | |
3464 | ||
3465 | static const struct dpdk_qos_ops egress_policer_ops = { | |
3466 | "egress-policer", /* qos_name */ | |
3467 | egress_policer_qos_construct, | |
3468 | egress_policer_qos_destruct, | |
3469 | egress_policer_qos_get, | |
78bd47cf | 3470 | egress_policer_qos_is_equal, |
0bf765f7 IS |
3471 | egress_policer_run |
3472 | }; | |
3473 | ||
050c60bf DDP |
3474 | static int |
3475 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
3476 | { | |
3477 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3478 | int err = 0; | |
3479 | ||
050c60bf DDP |
3480 | ovs_mutex_lock(&dev->mutex); |
3481 | ||
3482 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 3483 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b CL |
3484 | && dev->mtu == dev->requested_mtu |
3485 | && dev->rxq_size == dev->requested_rxq_size | |
bd4e172b | 3486 | && dev->txq_size == dev->requested_txq_size |
3487 | && dev->socket_id == dev->requested_socket_id) { | |
050c60bf DDP |
3488 | /* Reconfiguration is unnecessary */ |
3489 | ||
3490 | goto out; | |
3491 | } | |
3492 | ||
3493 | rte_eth_dev_stop(dev->port_id); | |
3494 | ||
d555d9bd | 3495 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 | 3496 | if (err && err != EEXIST) { |
d555d9bd | 3497 | goto out; |
0072e931 MK |
3498 | } |
3499 | ||
050c60bf DDP |
3500 | netdev->n_txq = dev->requested_n_txq; |
3501 | netdev->n_rxq = dev->requested_n_rxq; | |
3502 | ||
b685696b CL |
3503 | dev->rxq_size = dev->requested_rxq_size; |
3504 | dev->txq_size = dev->requested_txq_size; | |
3505 | ||
050c60bf DDP |
3506 | rte_free(dev->tx_q); |
3507 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
3508 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
3509 | if (!dev->tx_q) { | |
3510 | err = ENOMEM; | |
3511 | } | |
050c60bf | 3512 | |
0072e931 MK |
3513 | netdev_change_seq_changed(netdev); |
3514 | ||
050c60bf | 3515 | out: |
050c60bf | 3516 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
3517 | return err; |
3518 | } | |
3519 | ||
7f381c2e | 3520 | static int |
2d24d165 | 3521 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 3522 | OVS_REQUIRES(dev->mutex) |
050c60bf | 3523 | { |
2d24d165 CL |
3524 | dev->up.n_txq = dev->requested_n_txq; |
3525 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 3526 | int err; |
050c60bf | 3527 | |
81acebda IM |
3528 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
3529 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
3530 | dev->tx_q[0].map = 0; | |
3531 | } | |
3532 | ||
3533 | netdev_dpdk_remap_txqs(dev); | |
3534 | ||
d555d9bd | 3535 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 FA |
3536 | if (!err) { |
3537 | /* A new mempool was created. */ | |
d555d9bd | 3538 | netdev_change_seq_changed(&dev->up); |
b6b26021 FA |
3539 | } else if (err != EEXIST){ |
3540 | return err; | |
db8f13b0 | 3541 | } |
0a0f39df | 3542 | if (netdev_dpdk_get_vid(dev) >= 0) { |
894af647 | 3543 | if (dev->vhost_reconfigured == false) { |
3544 | dev->vhost_reconfigured = true; | |
3545 | /* Carrier status may need updating. */ | |
3546 | netdev_change_seq_changed(&dev->up); | |
3547 | } | |
81acebda | 3548 | } |
7f381c2e DDP |
3549 | |
3550 | return 0; | |
2d24d165 CL |
3551 | } |
3552 | ||
3553 | static int | |
3554 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
3555 | { | |
3556 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3557 | int err; |
2d24d165 | 3558 | |
2d24d165 | 3559 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 3560 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 3561 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
3562 | |
3563 | return err; | |
2d24d165 CL |
3564 | } |
3565 | ||
3566 | static int | |
3567 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3568 | { | |
3569 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3570 | int err; |
a14d1cc8 | 3571 | uint64_t vhost_flags = 0; |
2d24d165 | 3572 | |
2d24d165 CL |
3573 | ovs_mutex_lock(&dev->mutex); |
3574 | ||
c1ff66ac CL |
3575 | /* Configure vHost client mode if requested and if the following criteria |
3576 | * are met: | |
2d24d165 CL |
3577 | * 1. Device hasn't been registered yet. |
3578 | * 2. A path has been specified. | |
c1ff66ac CL |
3579 | */ |
3580 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 | 3581 | && strlen(dev->vhost_id)) { |
a14d1cc8 MK |
3582 | /* Register client-mode device. */ |
3583 | vhost_flags |= RTE_VHOST_USER_CLIENT; | |
3584 | ||
3585 | /* Enable IOMMU support, if explicitly requested. */ | |
3586 | if (dpdk_vhost_iommu_enabled()) { | |
3587 | vhost_flags |= RTE_VHOST_USER_IOMMU_SUPPORT; | |
3588 | } | |
3589 | err = rte_vhost_driver_register(dev->vhost_id, vhost_flags); | |
c1ff66ac | 3590 | if (err) { |
2d24d165 CL |
3591 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3592 | dev->vhost_id); | |
7f381c2e | 3593 | goto unlock; |
c1ff66ac | 3594 | } else { |
2d24d165 | 3595 | /* Configuration successful */ |
a14d1cc8 | 3596 | dev->vhost_driver_flags |= vhost_flags; |
2d24d165 CL |
3597 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " |
3598 | "using client socket '%s'", | |
3599 | dev->up.name, dev->vhost_id); | |
c1ff66ac | 3600 | } |
f3e7ec25 MW |
3601 | |
3602 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
3603 | &virtio_net_device_ops); | |
3604 | if (err) { | |
3605 | VLOG_ERR("rte_vhost_driver_callback_register failed for " | |
3606 | "vhost user client port: %s\n", dev->up.name); | |
3607 | goto unlock; | |
3608 | } | |
3609 | ||
3610 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
3611 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
3612 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
3613 | | 1ULL << VIRTIO_NET_F_CSUM); | |
3614 | if (err) { | |
3615 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
3616 | "client port: %s\n", dev->up.name); | |
3617 | goto unlock; | |
3618 | } | |
3619 | ||
3620 | err = rte_vhost_driver_start(dev->vhost_id); | |
3621 | if (err) { | |
3622 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
3623 | "client port: %s\n", dev->up.name); | |
3624 | goto unlock; | |
3625 | } | |
c1ff66ac CL |
3626 | } |
3627 | ||
7f381c2e DDP |
3628 | err = dpdk_vhost_reconfigure_helper(dev); |
3629 | ||
3630 | unlock: | |
050c60bf | 3631 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 3632 | |
7f381c2e | 3633 | return err; |
050c60bf DDP |
3634 | } |
3635 | ||
ecc1a34e | 3636 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
81acebda | 3637 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
971f4b39 MW |
3638 | GET_CARRIER, GET_STATS, \ |
3639 | GET_CUSTOM_STATS, \ | |
81acebda IM |
3640 | GET_FEATURES, GET_STATUS, \ |
3641 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3642 | { \ |
3643 | NAME, \ | |
118c77b1 | 3644 | true, /* is_pmd */ \ |
ecc1a34e | 3645 | INIT, /* init */ \ |
95fb793a | 3646 | NULL, /* netdev_dpdk_run */ \ |
3647 | NULL, /* netdev_dpdk_wait */ \ | |
3648 | \ | |
3649 | netdev_dpdk_alloc, \ | |
3650 | CONSTRUCT, \ | |
58397e6c | 3651 | DESTRUCT, \ |
95fb793a | 3652 | netdev_dpdk_dealloc, \ |
3653 | netdev_dpdk_get_config, \ | |
81acebda | 3654 | SET_CONFIG, \ |
95fb793a | 3655 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3656 | NULL, /* build header */ \ |
3657 | NULL, /* push header */ \ | |
3658 | NULL, /* pop header */ \ | |
7dec44fe | 3659 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3660 | SET_TX_MULTIQ, \ |
95fb793a | 3661 | \ |
7251515e | 3662 | SEND, /* send */ \ |
95fb793a | 3663 | NULL, /* send_wait */ \ |
3664 | \ | |
3665 | netdev_dpdk_set_etheraddr, \ | |
3666 | netdev_dpdk_get_etheraddr, \ | |
3667 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3668 | netdev_dpdk_set_mtu, \ |
95fb793a | 3669 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3670 | GET_CARRIER, \ |
95fb793a | 3671 | netdev_dpdk_get_carrier_resets, \ |
3672 | netdev_dpdk_set_miimon, \ | |
58397e6c | 3673 | GET_STATS, \ |
971f4b39 | 3674 | GET_CUSTOM_STATS, \ |
58397e6c | 3675 | GET_FEATURES, \ |
95fb793a | 3676 | NULL, /* set_advertisements */ \ |
875ab130 | 3677 | NULL, /* get_pt_mode */ \ |
95fb793a | 3678 | \ |
9509913a | 3679 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3680 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3681 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3682 | netdev_dpdk_get_qos, \ |
3683 | netdev_dpdk_set_qos, \ | |
95fb793a | 3684 | NULL, /* get_queue */ \ |
3685 | NULL, /* set_queue */ \ | |
3686 | NULL, /* delete_queue */ \ | |
3687 | NULL, /* get_queue_stats */ \ | |
3688 | NULL, /* queue_dump_start */ \ | |
3689 | NULL, /* queue_dump_next */ \ | |
3690 | NULL, /* queue_dump_done */ \ | |
3691 | NULL, /* dump_queue_stats */ \ | |
3692 | \ | |
95fb793a | 3693 | NULL, /* set_in4 */ \ |
a8704b50 | 3694 | NULL, /* get_addr_list */ \ |
95fb793a | 3695 | NULL, /* add_router */ \ |
3696 | NULL, /* get_next_hop */ \ | |
58397e6c | 3697 | GET_STATUS, \ |
95fb793a | 3698 | NULL, /* arp_lookup */ \ |
3699 | \ | |
3700 | netdev_dpdk_update_flags, \ | |
050c60bf | 3701 | RECONFIGURE, \ |
95fb793a | 3702 | \ |
3703 | netdev_dpdk_rxq_alloc, \ | |
3704 | netdev_dpdk_rxq_construct, \ | |
3705 | netdev_dpdk_rxq_destruct, \ | |
3706 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3707 | RXQ_RECV, \ |
95fb793a | 3708 | NULL, /* rx_wait */ \ |
3709 | NULL, /* rxq_drain */ \ | |
18ebd48c | 3710 | NO_OFFLOAD_API \ |
95fb793a | 3711 | } |
8a9562d2 | 3712 | |
bce01e3a | 3713 | static const struct netdev_class dpdk_class = |
95fb793a | 3714 | NETDEV_DPDK_CLASS( |
3715 | "dpdk", | |
ecc1a34e | 3716 | netdev_dpdk_class_init, |
5496878c | 3717 | netdev_dpdk_construct, |
58397e6c | 3718 | netdev_dpdk_destruct, |
81acebda IM |
3719 | netdev_dpdk_set_config, |
3720 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3721 | netdev_dpdk_eth_send, |
3722 | netdev_dpdk_get_carrier, | |
3723 | netdev_dpdk_get_stats, | |
971f4b39 | 3724 | netdev_dpdk_get_custom_stats, |
58397e6c KT |
3725 | netdev_dpdk_get_features, |
3726 | netdev_dpdk_get_status, | |
050c60bf | 3727 | netdev_dpdk_reconfigure, |
58397e6c | 3728 | netdev_dpdk_rxq_recv); |
95fb793a | 3729 | |
bce01e3a | 3730 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3731 | NETDEV_DPDK_CLASS( |
3732 | "dpdkr", | |
ecc1a34e | 3733 | netdev_dpdk_class_init, |
5496878c | 3734 | netdev_dpdk_ring_construct, |
58397e6c | 3735 | netdev_dpdk_destruct, |
c3d062a7 | 3736 | netdev_dpdk_ring_set_config, |
81acebda | 3737 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3738 | netdev_dpdk_ring_send, |
3739 | netdev_dpdk_get_carrier, | |
3740 | netdev_dpdk_get_stats, | |
971f4b39 | 3741 | netdev_dpdk_get_custom_stats, |
58397e6c KT |
3742 | netdev_dpdk_get_features, |
3743 | netdev_dpdk_get_status, | |
050c60bf | 3744 | netdev_dpdk_reconfigure, |
58397e6c KT |
3745 | netdev_dpdk_rxq_recv); |
3746 | ||
53f50d24 | 3747 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3748 | NETDEV_DPDK_CLASS( |
3749 | "dpdkvhostuser", | |
f3e7ec25 | 3750 | NULL, |
53f50d24 | 3751 | netdev_dpdk_vhost_construct, |
58397e6c | 3752 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3753 | NULL, |
81acebda | 3754 | NULL, |
58397e6c KT |
3755 | netdev_dpdk_vhost_send, |
3756 | netdev_dpdk_vhost_get_carrier, | |
3757 | netdev_dpdk_vhost_get_stats, | |
3758 | NULL, | |
7251515e | 3759 | NULL, |
b2e8b12f | 3760 | netdev_dpdk_vhost_user_get_status, |
53f50d24 | 3761 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3762 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3763 | static const struct netdev_class dpdk_vhost_client_class = |
3764 | NETDEV_DPDK_CLASS( | |
3765 | "dpdkvhostuserclient", | |
f3e7ec25 | 3766 | NULL, |
2d24d165 CL |
3767 | netdev_dpdk_vhost_client_construct, |
3768 | netdev_dpdk_vhost_destruct, | |
3769 | netdev_dpdk_vhost_client_set_config, | |
3770 | NULL, | |
3771 | netdev_dpdk_vhost_send, | |
3772 | netdev_dpdk_vhost_get_carrier, | |
3773 | netdev_dpdk_vhost_get_stats, | |
3774 | NULL, | |
3775 | NULL, | |
b2e8b12f | 3776 | netdev_dpdk_vhost_user_get_status, |
2d24d165 CL |
3777 | netdev_dpdk_vhost_client_reconfigure, |
3778 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 3779 | |
8a9562d2 PS |
3780 | void |
3781 | netdev_dpdk_register(void) | |
3782 | { | |
bab69409 AC |
3783 | netdev_register_provider(&dpdk_class); |
3784 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3785 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 3786 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 3787 | } |