]>
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 | |
6ebc4b09 | 20 | #include <errno.h> |
8a9562d2 PS |
21 | #include <signal.h> |
22 | #include <stdlib.h> | |
6ebc4b09 | 23 | #include <string.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> | |
6ebc4b09 | 35 | #include <rte_flow.h> |
01961bbd DDP |
36 | #include <rte_malloc.h> |
37 | #include <rte_mbuf.h> | |
38 | #include <rte_meter.h> | |
fc56f5e0 | 39 | #include <rte_pci.h> |
3eb8d4fa | 40 | #include <rte_version.h> |
6ebc4b09 | 41 | #include <rte_vhost.h> |
8a9562d2 | 42 | |
e8a2b5bf | 43 | #include "cmap.h" |
7d1ced01 | 44 | #include "dirs.h" |
e14deea0 | 45 | #include "dp-packet.h" |
01961bbd | 46 | #include "dpdk.h" |
8a9562d2 | 47 | #include "dpif-netdev.h" |
e5c0f5a4 | 48 | #include "fatal-signal.h" |
8a9562d2 PS |
49 | #include "netdev-provider.h" |
50 | #include "netdev-vport.h" | |
51 | #include "odp-util.h" | |
eac84432 | 52 | #include "openvswitch/dynamic-string.h" |
25d436fb | 53 | #include "openvswitch/list.h" |
6ebc4b09 | 54 | #include "openvswitch/match.h" |
25d436fb | 55 | #include "openvswitch/ofp-print.h" |
6ebc4b09 | 56 | #include "openvswitch/shash.h" |
25d436fb | 57 | #include "openvswitch/vlog.h" |
94143fc4 | 58 | #include "ovs-numa.h" |
8a9562d2 | 59 | #include "ovs-rcu.h" |
6ebc4b09 | 60 | #include "ovs-thread.h" |
8a9562d2 | 61 | #include "packets.h" |
0bf765f7 | 62 | #include "smap.h" |
8a9562d2 | 63 | #include "sset.h" |
8a9562d2 | 64 | #include "timeval.h" |
6ebc4b09 | 65 | #include "unaligned.h" |
8a9562d2 | 66 | #include "unixctl.h" |
6ebc4b09 IM |
67 | #include "util.h" |
68 | #include "uuid.h" | |
8a9562d2 | 69 | |
f3e7ec25 MW |
70 | enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM}; |
71 | ||
05b49df6 | 72 | VLOG_DEFINE_THIS_MODULE(netdev_dpdk); |
8a9562d2 PS |
73 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
74 | ||
75 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
76 | ||
77 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
78 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
79 | ||
80 | /* | |
81 | * need to reserve tons of extra space in the mbufs so we can align the | |
82 | * DMA addresses to 4KB. | |
18f777b2 TP |
83 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
84 | * performance for standard Ethernet MTU. | |
8a9562d2 | 85 | */ |
58be5c0e MK |
86 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN \ |
87 | + (2 * VLAN_HEADER_LEN)) | |
4be4d22c MK |
88 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) |
89 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
58be5c0e MK |
90 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len) \ |
91 | - ETHER_HDR_LEN - ETHER_CRC_LEN) | |
4be4d22c | 92 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 93 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 94 | |
43307ad0 IS |
95 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
96 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
97 | * enough hugepages) we keep halving the number until the allocation succeeds | |
98 | * or we reach MIN_NB_MBUF */ | |
99 | ||
100 | #define MAX_NB_MBUF (4096 * 64) | |
da79ce2b DDP |
101 | #define MIN_NB_MBUF (4096 * 4) |
102 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
103 | ||
43307ad0 IS |
104 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ |
105 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF / MIN_NB_MBUF) | |
106 | == 0); | |
107 | ||
108 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
109 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
110 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF / MIN_NB_MBUF)) | |
111 | % MP_CACHE_SZ == 0); | |
112 | ||
d6e3feb5 | 113 | /* |
114 | * DPDK XSTATS Counter names definition | |
115 | */ | |
116 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
117 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
118 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
119 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
120 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
121 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
122 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
123 | ||
124 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
125 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
126 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
127 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
128 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
129 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
130 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
131 | ||
d57f777f | 132 | #define XSTAT_RX_MULTICAST_PACKETS "rx_multicast_packets" |
d6e3feb5 | 133 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" |
134 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
135 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
136 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
137 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
138 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
139 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
140 | ||
8a9562d2 PS |
141 | #define SOCKET0 0 |
142 | ||
b685696b CL |
143 | /* Default size of Physical NIC RXQ */ |
144 | #define NIC_PORT_DEFAULT_RXQ_SIZE 2048 | |
145 | /* Default size of Physical NIC TXQ */ | |
146 | #define NIC_PORT_DEFAULT_TXQ_SIZE 2048 | |
147 | /* Maximum size of Physical NIC Queues */ | |
148 | #define NIC_PORT_MAX_Q_SIZE 4096 | |
79f5354c | 149 | |
585a5bea | 150 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
151 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
152 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
153 | * yet mapped to another queue. */ | |
585a5bea | 154 | |
bb37956a IM |
155 | #define DPDK_ETH_PORT_ID_INVALID RTE_MAX_ETHPORTS |
156 | ||
5e925ccc MK |
157 | /* DPDK library uses uint16_t for port_id. */ |
158 | typedef uint16_t dpdk_port_t; | |
fa9f4eeb | 159 | #define DPDK_PORT_ID_FMT "%"PRIu16 |
bb37956a | 160 | |
31871ee3 | 161 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 162 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 163 | |
8a9562d2 | 164 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
165 | .rxmode = { |
166 | .mq_mode = ETH_MQ_RX_RSS, | |
167 | .split_hdr_size = 0, | |
03f3f9c0 | 168 | .offloads = 0, |
a28ddd11 DDP |
169 | }, |
170 | .rx_adv_conf = { | |
171 | .rss_conf = { | |
172 | .rss_key = NULL, | |
543342a4 | 173 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 174 | }, |
a28ddd11 DDP |
175 | }, |
176 | .txmode = { | |
177 | .mq_mode = ETH_MQ_TX_NONE, | |
178 | }, | |
8a9562d2 PS |
179 | }; |
180 | ||
e8a2b5bf FC |
181 | /* |
182 | * A mapping from ufid to dpdk rte_flow. | |
183 | */ | |
184 | static struct cmap ufid_to_rte_flow = CMAP_INITIALIZER; | |
185 | ||
186 | struct ufid_to_rte_flow_data { | |
187 | struct cmap_node node; | |
188 | ovs_u128 ufid; | |
189 | struct rte_flow *rte_flow; | |
190 | }; | |
191 | ||
f3e7ec25 MW |
192 | /* |
193 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
194 | * configuration has been fully completed. | |
195 | */ | |
196 | static int new_device(int vid); | |
197 | static void destroy_device(int vid); | |
198 | static int vring_state_changed(int vid, uint16_t queue_id, int enable); | |
199 | static const struct vhost_device_ops virtio_net_device_ops = | |
200 | { | |
201 | .new_device = new_device, | |
202 | .destroy_device = destroy_device, | |
203 | .vring_state_changed = vring_state_changed, | |
204 | .features_changed = NULL | |
205 | }; | |
206 | ||
58f7c37b DDP |
207 | enum { DPDK_RING_SIZE = 256 }; |
208 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
209 | enum { DRAIN_TSC = 200000ULL }; |
210 | ||
58397e6c KT |
211 | enum dpdk_dev_type { |
212 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 213 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
214 | }; |
215 | ||
0bf765f7 IS |
216 | /* Quality of Service */ |
217 | ||
218 | /* An instance of a QoS configuration. Always associated with a particular | |
219 | * network device. | |
220 | * | |
221 | * Each QoS implementation subclasses this with whatever additional data it | |
222 | * needs. | |
223 | */ | |
224 | struct qos_conf { | |
225 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 226 | rte_spinlock_t lock; |
0bf765f7 IS |
227 | }; |
228 | ||
229 | /* A particular implementation of dpdk QoS operations. | |
230 | * | |
231 | * The functions below return 0 if successful or a positive errno value on | |
232 | * failure, except where otherwise noted. All of them must be provided, except | |
233 | * where otherwise noted. | |
234 | */ | |
235 | struct dpdk_qos_ops { | |
236 | ||
237 | /* Name of the QoS type */ | |
238 | const char *qos_name; | |
239 | ||
78bd47cf DDP |
240 | /* Called to construct a qos_conf object. The implementation should make |
241 | * the appropriate calls to configure QoS according to '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 DDP |
247 | * This function must return 0 if and only if it sets '*conf' to an |
248 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
249 | * |
250 | * For all QoS implementations it should always be non-null. | |
251 | */ | |
78bd47cf | 252 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
253 | |
254 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 255 | * 'qos_conf'. |
0bf765f7 IS |
256 | * |
257 | * For all QoS implementations it should always be non-null. | |
258 | */ | |
78bd47cf | 259 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 260 | |
78bd47cf | 261 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
262 | * |
263 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
264 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
265 | * (which is built as ovs-vswitchd.conf.db(8)). | |
266 | */ | |
78bd47cf | 267 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 268 | |
78bd47cf | 269 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
270 | * |
271 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
272 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
273 | * (which is built as ovs-vswitchd.conf.db(8)). | |
274 | * | |
78bd47cf | 275 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 276 | */ |
78bd47cf DDP |
277 | bool (*qos_is_equal)(const struct qos_conf *conf, |
278 | const struct smap *details); | |
0bf765f7 IS |
279 | |
280 | /* Modify an array of rte_mbufs. The modification is specific to | |
281 | * each qos implementation. | |
282 | * | |
283 | * The function should take and array of mbufs and an int representing | |
284 | * the current number of mbufs present in the array. | |
285 | * | |
286 | * After the function has performed a qos modification to the array of | |
287 | * mbufs it returns an int representing the number of mbufs now present in | |
288 | * the array. This value is can then be passed to the port send function | |
289 | * along with the modified array for transmission. | |
290 | * | |
291 | * For all QoS implementations it should always be non-null. | |
292 | */ | |
78bd47cf | 293 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
7d7ded7a | 294 | int pkt_cnt, bool should_steal); |
0bf765f7 IS |
295 | }; |
296 | ||
297 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
298 | static const struct dpdk_qos_ops egress_policer_ops; | |
299 | ||
300 | /* | |
301 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
302 | * operations. | |
303 | */ | |
304 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
305 | &egress_policer_ops, | |
306 | NULL | |
307 | }; | |
308 | ||
c2adb102 IM |
309 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
310 | ||
8a9562d2 | 311 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 312 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 313 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 314 | |
c2adb102 IM |
315 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
316 | = OVS_MUTEX_INITIALIZER; | |
317 | ||
91fccdad | 318 | /* Contains all 'struct dpdk_mp's. */ |
43307ad0 IS |
319 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mp_mutex) |
320 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); | |
91fccdad | 321 | |
91fccdad KT |
322 | struct dpdk_mp { |
323 | struct rte_mempool *mp; | |
43307ad0 IS |
324 | int mtu; |
325 | int socket_id; | |
326 | int refcount; | |
91fccdad KT |
327 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mp_mutex); |
328 | }; | |
329 | ||
5a034064 AW |
330 | /* There should be one 'struct dpdk_tx_queue' created for |
331 | * each cpu core. */ | |
8a9562d2 | 332 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
333 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
334 | * from concurrent access. It is used only | |
335 | * if the queue is shared among different | |
324c8374 | 336 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
337 | int map; /* Mapping of configured vhost-user queues |
338 | * to enabled by guest. */ | |
8a9562d2 PS |
339 | }; |
340 | ||
95fb793a | 341 | /* dpdk has no way to remove dpdk ring ethernet devices |
342 | so we have to keep them around once they've been created | |
343 | */ | |
344 | ||
ca6ba700 | 345 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 346 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 347 | |
348 | struct dpdk_ring { | |
349 | /* For the client rings */ | |
350 | struct rte_ring *cring_tx; | |
351 | struct rte_ring *cring_rx; | |
b83a2df1 | 352 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
bb37956a | 353 | dpdk_port_t eth_port_id; /* ethernet device port id */ |
ca6ba700 | 354 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 355 | }; |
356 | ||
9509913a IS |
357 | struct ingress_policer { |
358 | struct rte_meter_srtcm_params app_srtcm_params; | |
359 | struct rte_meter_srtcm in_policer; | |
03f3f9c0 | 360 | struct rte_meter_srtcm_profile in_prof; |
9509913a IS |
361 | rte_spinlock_t policer_lock; |
362 | }; | |
363 | ||
1a2bb118 SC |
364 | enum dpdk_hw_ol_features { |
365 | NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0, | |
e10ca8b9 | 366 | NETDEV_RX_HW_CRC_STRIP = 1 << 1, |
03f3f9c0 | 367 | NETDEV_RX_HW_SCATTER = 1 << 2 |
1a2bb118 SC |
368 | }; |
369 | ||
b2e72a9c IM |
370 | /* |
371 | * In order to avoid confusion in variables names, following naming convention | |
372 | * should be used, if possible: | |
373 | * | |
374 | * 'struct netdev' : 'netdev' | |
375 | * 'struct netdev_dpdk' : 'dev' | |
376 | * 'struct netdev_rxq' : 'rxq' | |
377 | * 'struct netdev_rxq_dpdk' : 'rx' | |
378 | * | |
379 | * Example: | |
380 | * struct netdev *netdev = netdev_from_name(name); | |
381 | * struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
382 | * | |
383 | * Also, 'netdev' should be used instead of 'dev->up', where 'netdev' was | |
384 | * already defined. | |
385 | */ | |
386 | ||
8a9562d2 | 387 | struct netdev_dpdk { |
23d4d53f BB |
388 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline0, |
389 | dpdk_port_t port_id; | |
390 | ||
391 | /* If true, device was attached by rte_eth_dev_attach(). */ | |
392 | bool attached; | |
606f6650 EC |
393 | /* If true, rte_eth_dev_start() was successfully called */ |
394 | bool started; | |
23d4d53f BB |
395 | struct eth_addr hwaddr; |
396 | int mtu; | |
397 | int socket_id; | |
398 | int buf_size; | |
399 | int max_packet_len; | |
400 | enum dpdk_dev_type type; | |
401 | enum netdev_flags flags; | |
eaa43581 | 402 | int link_reset_cnt; |
23d4d53f BB |
403 | char *devargs; /* Device arguments for dpdk ports */ |
404 | struct dpdk_tx_queue *tx_q; | |
405 | struct rte_eth_link link; | |
23d4d53f BB |
406 | ); |
407 | ||
408 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline1, | |
409 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
43307ad0 | 410 | struct dpdk_mp *dpdk_mp; |
23d4d53f BB |
411 | |
412 | /* virtio identifier for vhost devices */ | |
413 | ovsrcu_index vid; | |
414 | ||
415 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
416 | bool vhost_reconfigured; | |
417 | /* 3 pad bytes here. */ | |
418 | ); | |
419 | ||
420 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
421 | /* Identifier used to distinguish vhost devices from each other. */ | |
422 | char vhost_id[PATH_MAX]; | |
423 | ); | |
424 | ||
425 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
426 | struct netdev up; | |
427 | /* In dpdk_list. */ | |
428 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); | |
429 | ||
430 | /* QoS configuration and lock for the device */ | |
431 | OVSRCU_TYPE(struct qos_conf *) qos_conf; | |
432 | ||
433 | /* Ingress Policer */ | |
434 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
435 | uint32_t policer_rate; | |
436 | uint32_t policer_burst; | |
437 | ); | |
438 | ||
439 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
440 | struct netdev_stats stats; | |
441 | /* Protects stats */ | |
442 | rte_spinlock_t stats_lock; | |
443 | /* 44 pad bytes here. */ | |
444 | ); | |
445 | ||
446 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
447 | /* The following properties cannot be changed when a device is running, | |
448 | * so we remember the request and update them next time | |
449 | * netdev_dpdk*_reconfigure() is called */ | |
450 | int requested_mtu; | |
451 | int requested_n_txq; | |
452 | int requested_n_rxq; | |
453 | int requested_rxq_size; | |
454 | int requested_txq_size; | |
455 | ||
456 | /* Number of rx/tx descriptors for physical devices */ | |
457 | int rxq_size; | |
458 | int txq_size; | |
459 | ||
460 | /* Socket ID detected when vHost device is brought up */ | |
461 | int requested_socket_id; | |
462 | ||
463 | /* Denotes whether vHost port is client/server mode */ | |
464 | uint64_t vhost_driver_flags; | |
465 | ||
466 | /* DPDK-ETH Flow control */ | |
467 | struct rte_eth_fc_conf fc_conf; | |
468 | ||
469 | /* DPDK-ETH hardware offload features, | |
470 | * from the enum set 'dpdk_hw_ol_features' */ | |
471 | uint32_t hw_ol_features; | |
f8b64a61 RM |
472 | |
473 | /* Properties for link state change detection mode. | |
474 | * If lsc_interrupt_mode is set to false, poll mode is used, | |
475 | * otherwise interrupt mode is used. */ | |
476 | bool requested_lsc_interrupt_mode; | |
477 | bool lsc_interrupt_mode; | |
23d4d53f | 478 | ); |
971f4b39 MW |
479 | |
480 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
481 | /* Names of all XSTATS counters */ | |
482 | struct rte_eth_xstat_name *rte_xstats_names; | |
483 | int rte_xstats_names_size; | |
484 | int rte_xstats_ids_size; | |
485 | uint64_t *rte_xstats_ids; | |
486 | ); | |
8a9562d2 PS |
487 | }; |
488 | ||
489 | struct netdev_rxq_dpdk { | |
490 | struct netdev_rxq up; | |
bb37956a | 491 | dpdk_port_t port_id; |
8a9562d2 PS |
492 | }; |
493 | ||
f3e7ec25 MW |
494 | static void netdev_dpdk_destruct(struct netdev *netdev); |
495 | static void netdev_dpdk_vhost_destruct(struct netdev *netdev); | |
8a9562d2 | 496 | |
ac1a9bb9 IM |
497 | static void netdev_dpdk_clear_xstats(struct netdev_dpdk *dev); |
498 | ||
0a0f39df | 499 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 500 | |
9509913a IS |
501 | struct ingress_policer * |
502 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
503 | ||
8a9562d2 PS |
504 | static bool |
505 | is_dpdk_class(const struct netdev_class *class) | |
506 | { | |
f3e7ec25 MW |
507 | return class->destruct == netdev_dpdk_destruct |
508 | || class->destruct == netdev_dpdk_vhost_destruct; | |
8a9562d2 PS |
509 | } |
510 | ||
4be4d22c MK |
511 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
512 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
513 | * value, insufficient buffers are allocated to accomodate the packet in its | |
514 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
515 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
516 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
517 | * behaviour, which reduces performance. To prevent this, use a buffer size |
518 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
519 | */ |
520 | static uint32_t | |
521 | dpdk_buf_size(int mtu) | |
522 | { | |
a32bab26 TL |
523 | return ROUND_UP(MTU_TO_MAX_FRAME_LEN(mtu), NETDEV_DPDK_MBUF_ALIGN) |
524 | + RTE_PKTMBUF_HEADROOM; | |
4be4d22c MK |
525 | } |
526 | ||
eff23640 DDP |
527 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
528 | * | |
529 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
530 | static void * |
531 | dpdk_rte_mzalloc(size_t sz) | |
532 | { | |
eff23640 | 533 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
534 | } |
535 | ||
536 | void | |
e14deea0 | 537 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 538 | { |
db73f716 | 539 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 540 | |
b00b4a81 | 541 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
542 | } |
543 | ||
b3cd9f9d | 544 | static void |
401b70d6 | 545 | ovs_rte_pktmbuf_init(struct rte_mempool *mp OVS_UNUSED, |
b3cd9f9d | 546 | void *opaque_arg OVS_UNUSED, |
2391135c | 547 | void *_p, |
b3cd9f9d PS |
548 | unsigned i OVS_UNUSED) |
549 | { | |
2391135c | 550 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 551 | |
3aaa6201 | 552 | dp_packet_init_dpdk((struct dp_packet *) pkt); |
b3cd9f9d PS |
553 | } |
554 | ||
91fccdad KT |
555 | static int |
556 | dpdk_mp_full(const struct rte_mempool *mp) OVS_REQUIRES(dpdk_mp_mutex) | |
557 | { | |
1f84a2d5 KT |
558 | /* At this point we want to know if all the mbufs are back |
559 | * in the mempool. rte_mempool_full() is not atomic but it's | |
560 | * the best available and as we are no longer requesting mbufs | |
561 | * from the mempool, it means mbufs will not move from | |
562 | * 'mempool ring' --> 'mempool cache'. In rte_mempool_full() | |
563 | * the ring is counted before caches, so we won't get false | |
564 | * positives in this use case and we handle false negatives. | |
565 | * | |
566 | * If future implementations of rte_mempool_full() were to change | |
567 | * it could be possible for a false positive. Even that would | |
568 | * likely be ok, as there are additional checks during mempool | |
569 | * freeing but it would make things racey. | |
91fccdad | 570 | */ |
1f84a2d5 | 571 | return rte_mempool_full(mp); |
91fccdad KT |
572 | } |
573 | ||
574 | /* Free unused mempools. */ | |
575 | static void | |
43307ad0 | 576 | dpdk_mp_sweep(void) OVS_REQUIRES(dpdk_mp_mutex) |
91fccdad KT |
577 | { |
578 | struct dpdk_mp *dmp, *next; | |
579 | ||
43307ad0 IS |
580 | LIST_FOR_EACH_SAFE (dmp, next, list_node, &dpdk_mp_list) { |
581 | if (!dmp->refcount && dpdk_mp_full(dmp->mp)) { | |
91fccdad KT |
582 | VLOG_DBG("Freeing mempool \"%s\"", dmp->mp->name); |
583 | ovs_list_remove(&dmp->list_node); | |
584 | rte_mempool_free(dmp->mp); | |
585 | rte_free(dmp); | |
586 | } | |
587 | } | |
91fccdad KT |
588 | } |
589 | ||
43307ad0 IS |
590 | /* Calculating the required number of mbufs differs depending on the |
591 | * mempool model being used. Check if per port memory is in use before | |
592 | * calculating. | |
593 | */ | |
594 | static uint32_t | |
595 | dpdk_calculate_mbufs(struct netdev_dpdk *dev, int mtu, bool per_port_mp) | |
91fccdad | 596 | { |
43307ad0 | 597 | uint32_t n_mbufs; |
91fccdad | 598 | |
43307ad0 IS |
599 | if (!per_port_mp) { |
600 | /* Shared memory are being used. | |
601 | * XXX: this is a really rough method of provisioning memory. | |
602 | * It's impossible to determine what the exact memory requirements are | |
603 | * when the number of ports and rxqs that utilize a particular mempool | |
604 | * can change dynamically at runtime. For now, use this rough | |
605 | * heurisitic. | |
606 | */ | |
607 | if (mtu >= ETHER_MTU) { | |
608 | n_mbufs = MAX_NB_MBUF; | |
609 | } else { | |
610 | n_mbufs = MIN_NB_MBUF; | |
91fccdad | 611 | } |
43307ad0 IS |
612 | } else { |
613 | /* Per port memory is being used. | |
614 | * XXX: rough estimation of number of mbufs required for this port: | |
615 | * <packets required to fill the device rxqs> | |
616 | * + <packets that could be stuck on other ports txqs> | |
617 | * + <packets in the pmd threads> | |
618 | * + <additional memory for corner cases> | |
619 | */ | |
620 | n_mbufs = dev->requested_n_rxq * dev->requested_rxq_size | |
621 | + dev->requested_n_txq * dev->requested_txq_size | |
622 | + MIN(RTE_MAX_LCORE, dev->requested_n_rxq) * NETDEV_MAX_BURST | |
623 | + MIN_NB_MBUF; | |
91fccdad | 624 | } |
43307ad0 IS |
625 | |
626 | return n_mbufs; | |
91fccdad KT |
627 | } |
628 | ||
43307ad0 IS |
629 | static struct dpdk_mp * |
630 | dpdk_mp_create(struct netdev_dpdk *dev, int mtu, bool per_port_mp) | |
8a9562d2 | 631 | { |
24e78f93 IM |
632 | char mp_name[RTE_MEMPOOL_NAMESIZE]; |
633 | const char *netdev_name = netdev_get_name(&dev->up); | |
634 | int socket_id = dev->requested_socket_id; | |
dfaf00e8 MK |
635 | uint32_t n_mbufs = 0; |
636 | uint32_t mbuf_size = 0; | |
637 | uint32_t aligned_mbuf_size = 0; | |
638 | uint32_t mbuf_priv_data_len = 0; | |
639 | uint32_t pkt_size = 0; | |
24e78f93 | 640 | uint32_t hash = hash_string(netdev_name, 0); |
43307ad0 IS |
641 | struct dpdk_mp *dmp = NULL; |
642 | int ret; | |
643 | ||
644 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
645 | if (!dmp) { | |
646 | return NULL; | |
647 | } | |
648 | dmp->socket_id = socket_id; | |
649 | dmp->mtu = mtu; | |
650 | dmp->refcount = 1; | |
651 | ||
dfaf00e8 MK |
652 | /* Get the size of each mbuf, based on the MTU */ |
653 | mbuf_size = MTU_TO_FRAME_LEN(mtu); | |
654 | ||
43307ad0 | 655 | n_mbufs = dpdk_calculate_mbufs(dev, mtu, per_port_mp); |
d555d9bd | 656 | |
da79ce2b | 657 | do { |
24e78f93 | 658 | /* Full DPDK memory pool name must be unique and cannot be |
43307ad0 IS |
659 | * longer than RTE_MEMPOOL_NAMESIZE. Note that for the shared |
660 | * mempool case this can result in one device using a mempool | |
661 | * which references a different device in it's name. However as | |
662 | * mempool names are hashed, the device name will not be readable | |
663 | * so this is not an issue for tasks such as debugging. | |
664 | */ | |
665 | ret = snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, | |
dfaf00e8 MK |
666 | "ovs%08x%02d%05d%07u", |
667 | hash, socket_id, mtu, n_mbufs); | |
24e78f93 IM |
668 | if (ret < 0 || ret >= RTE_MEMPOOL_NAMESIZE) { |
669 | VLOG_DBG("snprintf returned %d. " | |
670 | "Failed to generate a mempool name for \"%s\". " | |
671 | "Hash:0x%x, socket_id: %d, mtu:%d, mbufs:%u.", | |
672 | ret, netdev_name, hash, socket_id, mtu, n_mbufs); | |
673 | break; | |
65056fd7 | 674 | } |
95fb793a | 675 | |
dfaf00e8 MK |
676 | VLOG_DBG("Port %s: Requesting a mempool of %u mbufs of size %u " |
677 | "on socket %d for %d Rx and %d Tx queues, " | |
678 | "cache line size of %u", | |
679 | netdev_name, n_mbufs, mbuf_size, socket_id, | |
680 | dev->requested_n_rxq, dev->requested_n_txq, | |
681 | RTE_CACHE_LINE_SIZE); | |
682 | ||
a32bab26 TL |
683 | /* The size of the mbuf's private area (i.e. area that holds OvS' |
684 | * dp_packet data)*/ | |
dfaf00e8 MK |
685 | mbuf_priv_data_len = sizeof(struct dp_packet) - |
686 | sizeof(struct rte_mbuf); | |
687 | /* The size of the entire dp_packet. */ | |
688 | pkt_size = sizeof(struct dp_packet) + mbuf_size; | |
689 | /* mbuf size, rounded up to cacheline size. */ | |
690 | aligned_mbuf_size = ROUND_UP(pkt_size, RTE_CACHE_LINE_SIZE); | |
691 | /* If there is a size discrepancy, add padding to mbuf_priv_data_len. | |
692 | * This maintains mbuf size cache alignment, while also honoring RX | |
693 | * buffer alignment in the data portion of the mbuf. If this adjustment | |
694 | * is not made, there is a possiblity later on that for an element of | |
695 | * the mempool, buf, buf->data_len < (buf->buf_len - buf->data_off). | |
696 | * This is problematic in the case of multi-segment mbufs, particularly | |
697 | * when an mbuf segment needs to be resized (when [push|popp]ing a VLAN | |
698 | * header, for example. | |
699 | */ | |
700 | mbuf_priv_data_len += (aligned_mbuf_size - pkt_size); | |
701 | ||
702 | dmp->mp = rte_pktmbuf_pool_create(mp_name, n_mbufs, MP_CACHE_SZ, | |
703 | mbuf_priv_data_len, | |
704 | mbuf_size, | |
43307ad0 | 705 | socket_id); |
24e78f93 | 706 | |
43307ad0 | 707 | if (dmp->mp) { |
24e78f93 IM |
708 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", |
709 | mp_name, n_mbufs); | |
837c1761 | 710 | /* rte_pktmbuf_pool_create has done some initialization of the |
43307ad0 IS |
711 | * rte_mbuf part of each dp_packet, while ovs_rte_pktmbuf_init |
712 | * initializes some OVS specific fields of dp_packet. | |
713 | */ | |
714 | rte_mempool_obj_iter(dmp->mp, ovs_rte_pktmbuf_init, NULL); | |
715 | return dmp; | |
d555d9bd RW |
716 | } else if (rte_errno == EEXIST) { |
717 | /* A mempool with the same name already exists. We just | |
718 | * retrieve its pointer to be returned to the caller. */ | |
43307ad0 | 719 | dmp->mp = rte_mempool_lookup(mp_name); |
d555d9bd RW |
720 | /* As the mempool create returned EEXIST we can expect the |
721 | * lookup has returned a valid pointer. If for some reason | |
722 | * that's not the case we keep track of it. */ | |
24e78f93 | 723 | VLOG_DBG("A mempool with name \"%s\" already exists at %p.", |
43307ad0 IS |
724 | mp_name, dmp->mp); |
725 | return dmp; | |
d555d9bd | 726 | } else { |
43307ad0 IS |
727 | VLOG_DBG("Failed to create mempool \"%s\" with a request of " |
728 | "%u mbufs, retrying with %u mbufs", | |
729 | mp_name, n_mbufs, n_mbufs / 2); | |
0c6f39e5 | 730 | } |
43307ad0 | 731 | } while (!dmp->mp && rte_errno == ENOMEM && (n_mbufs /= 2) >= MIN_NB_MBUF); |
2ae3d542 | 732 | |
43307ad0 IS |
733 | VLOG_ERR("Failed to create mempool \"%s\" with a request of %u mbufs", |
734 | mp_name, n_mbufs); | |
735 | ||
736 | rte_free(dmp); | |
737 | return NULL; | |
8a9562d2 PS |
738 | } |
739 | ||
43307ad0 IS |
740 | static struct dpdk_mp * |
741 | dpdk_mp_get(struct netdev_dpdk *dev, int mtu, bool per_port_mp) | |
8a9562d2 | 742 | { |
43307ad0 IS |
743 | struct dpdk_mp *dmp, *next; |
744 | bool reuse = false; | |
8a9562d2 | 745 | |
c2adb102 | 746 | ovs_mutex_lock(&dpdk_mp_mutex); |
43307ad0 IS |
747 | /* Check if shared memory is being used, if so check existing mempools |
748 | * to see if reuse is possible. */ | |
749 | if (!per_port_mp) { | |
750 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
751 | if (dmp->socket_id == dev->requested_socket_id | |
752 | && dmp->mtu == mtu) { | |
753 | VLOG_DBG("Reusing mempool \"%s\"", dmp->mp->name); | |
754 | dmp->refcount++; | |
755 | reuse = true; | |
756 | break; | |
757 | } | |
758 | } | |
759 | } | |
760 | /* Sweep mempools after reuse or before create. */ | |
761 | dpdk_mp_sweep(); | |
91fccdad | 762 | |
43307ad0 IS |
763 | if (!reuse) { |
764 | dmp = dpdk_mp_create(dev, mtu, per_port_mp); | |
91fccdad | 765 | if (dmp) { |
43307ad0 IS |
766 | /* Shared memory will hit the reuse case above so will not |
767 | * request a mempool that already exists but we need to check | |
768 | * for the EEXIST case for per port memory case. Compare the | |
769 | * mempool returned by dmp to each entry in dpdk_mp_list. If a | |
770 | * match is found, free dmp as a new entry is not required, set | |
771 | * dmp to point to the existing entry and increment the refcount | |
772 | * to avoid being freed at a later stage. | |
773 | */ | |
774 | if (per_port_mp && rte_errno == EEXIST) { | |
775 | LIST_FOR_EACH (next, list_node, &dpdk_mp_list) { | |
776 | if (dmp->mp == next->mp) { | |
777 | rte_free(dmp); | |
778 | dmp = next; | |
779 | dmp->refcount++; | |
780 | } | |
781 | } | |
782 | } else { | |
783 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); | |
784 | } | |
91fccdad KT |
785 | } |
786 | } | |
43307ad0 IS |
787 | |
788 | ||
789 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
790 | ||
791 | return dmp; | |
792 | } | |
793 | ||
794 | /* Decrement reference to a mempool. */ | |
795 | static void | |
796 | dpdk_mp_put(struct dpdk_mp *dmp) | |
797 | { | |
798 | if (!dmp) { | |
799 | return; | |
800 | } | |
801 | ||
802 | ovs_mutex_lock(&dpdk_mp_mutex); | |
803 | ovs_assert(dmp->refcount); | |
804 | dmp->refcount--; | |
c2adb102 | 805 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
806 | } |
807 | ||
43307ad0 IS |
808 | /* Depending on the memory model being used this function tries to |
809 | * identify and reuse an existing mempool or tries to allocate a new | |
810 | * mempool on requested_socket_id with mbuf size corresponding to the | |
811 | * requested_mtu. On success, a new configuration will be applied. | |
0072e931 MK |
812 | * On error, device will be left unchanged. */ |
813 | static int | |
814 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
815 | OVS_REQUIRES(dev->mutex) |
816 | { | |
817 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
43307ad0 | 818 | struct dpdk_mp *dmp; |
24e78f93 | 819 | int ret = 0; |
43307ad0 | 820 | bool per_port_mp = dpdk_per_port_memory(); |
0072e931 | 821 | |
43307ad0 IS |
822 | /* With shared memory we do not need to configure a mempool if the MTU |
823 | * and socket ID have not changed, the previous configuration is still | |
824 | * valid so return 0 */ | |
825 | if (!per_port_mp && dev->mtu == dev->requested_mtu | |
826 | && dev->socket_id == dev->requested_socket_id) { | |
827 | return ret; | |
828 | } | |
91fccdad | 829 | |
43307ad0 IS |
830 | dmp = dpdk_mp_get(dev, FRAME_LEN_TO_MTU(buf_size), per_port_mp); |
831 | if (!dmp) { | |
c67e46c0 MK |
832 | VLOG_ERR("Failed to create memory pool for netdev " |
833 | "%s, with MTU %d on socket %d: %s\n", | |
834 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
835 | rte_strerror(rte_errno)); | |
24e78f93 | 836 | ret = rte_errno; |
0072e931 | 837 | } else { |
43307ad0 IS |
838 | /* Check for any pre-existing dpdk_mp for the device before accessing |
839 | * the associated mempool. | |
840 | */ | |
841 | if (dev->dpdk_mp != NULL) { | |
842 | /* A new MTU was requested, decrement the reference count for the | |
843 | * devices current dpdk_mp. This is required even if a pointer to | |
844 | * same dpdk_mp is returned by dpdk_mp_get. The refcount for dmp | |
845 | * has already been incremented by dpdk_mp_get at this stage so it | |
846 | * must be decremented to keep an accurate refcount for the | |
847 | * dpdk_mp. | |
848 | */ | |
849 | dpdk_mp_put(dev->dpdk_mp); | |
850 | } | |
851 | dev->dpdk_mp = dmp; | |
0072e931 MK |
852 | dev->mtu = dev->requested_mtu; |
853 | dev->socket_id = dev->requested_socket_id; | |
854 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
855 | } | |
856 | ||
24e78f93 | 857 | return ret; |
0072e931 MK |
858 | } |
859 | ||
8a9562d2 PS |
860 | static void |
861 | check_link_status(struct netdev_dpdk *dev) | |
862 | { | |
863 | struct rte_eth_link link; | |
864 | ||
865 | rte_eth_link_get_nowait(dev->port_id, &link); | |
866 | ||
867 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 868 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
869 | |
870 | dev->link_reset_cnt++; | |
871 | dev->link = link; | |
872 | if (dev->link.link_status) { | |
fa9f4eeb IM |
873 | VLOG_DBG_RL(&rl, |
874 | "Port "DPDK_PORT_ID_FMT" Link Up - speed %u Mbps - %s", | |
58be5c0e | 875 | dev->port_id, (unsigned) dev->link.link_speed, |
fa9f4eeb IM |
876 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) |
877 | ? "full-duplex" : "half-duplex"); | |
8a9562d2 | 878 | } else { |
fa9f4eeb IM |
879 | VLOG_DBG_RL(&rl, "Port "DPDK_PORT_ID_FMT" Link Down", |
880 | dev->port_id); | |
8a9562d2 PS |
881 | } |
882 | } | |
883 | } | |
884 | ||
885 | static void * | |
886 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
887 | { | |
888 | struct netdev_dpdk *dev; | |
889 | ||
890 | pthread_detach(pthread_self()); | |
891 | ||
892 | for (;;) { | |
893 | ovs_mutex_lock(&dpdk_mutex); | |
894 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
895 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
896 | if (dev->type == DPDK_DEV_ETH) { |
897 | check_link_status(dev); | |
898 | } | |
8a9562d2 PS |
899 | ovs_mutex_unlock(&dev->mutex); |
900 | } | |
901 | ovs_mutex_unlock(&dpdk_mutex); | |
902 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
903 | } | |
904 | ||
905 | return NULL; | |
906 | } | |
907 | ||
b98d7669 | 908 | static int |
f8b64a61 | 909 | dpdk_eth_dev_port_config(struct netdev_dpdk *dev, int n_rxq, int n_txq) |
b98d7669 DDP |
910 | { |
911 | int diag = 0; | |
912 | int i; | |
0072e931 | 913 | struct rte_eth_conf conf = port_conf; |
65a87968 | 914 | struct rte_eth_dev_info info; |
4dd16ca0 | 915 | uint16_t conf_mtu; |
65a87968 | 916 | |
03f3f9c0 OM |
917 | rte_eth_dev_info_get(dev->port_id, &info); |
918 | ||
65a87968 | 919 | /* As of DPDK 17.11.1 a few PMDs require to explicitly enable |
03f3f9c0 OM |
920 | * scatter to support jumbo RX. |
921 | * Setting scatter for the device is done after checking for | |
922 | * scatter support in the device capabilites. */ | |
0072e931 | 923 | if (dev->mtu > ETHER_MTU) { |
03f3f9c0 OM |
924 | if (dev->hw_ol_features & NETDEV_RX_HW_SCATTER) { |
925 | conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER; | |
65a87968 | 926 | } |
0072e931 | 927 | } |
67fe6d63 | 928 | |
f8b64a61 | 929 | conf.intr_conf.lsc = dev->lsc_interrupt_mode; |
e10ca8b9 | 930 | |
03f3f9c0 OM |
931 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
932 | conf.rxmode.offloads |= DEV_RX_OFFLOAD_CHECKSUM; | |
933 | } | |
934 | ||
935 | if (!(dev->hw_ol_features & NETDEV_RX_HW_CRC_STRIP) | |
936 | && info.rx_offload_capa & DEV_RX_OFFLOAD_KEEP_CRC) { | |
937 | conf.rxmode.offloads |= DEV_RX_OFFLOAD_KEEP_CRC; | |
e10ca8b9 MW |
938 | } |
939 | ||
03f3f9c0 OM |
940 | /* Limit configured rss hash functions to only those supported |
941 | * by the eth device. */ | |
942 | conf.rx_adv_conf.rss_conf.rss_hf &= info.flow_type_rss_offloads; | |
943 | ||
b98d7669 DDP |
944 | /* A device may report more queues than it makes available (this has |
945 | * been observed for Intel xl710, which reserves some of them for | |
946 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
947 | * available. When this happens we can retry the configuration | |
948 | * and request less queues */ | |
949 | while (n_rxq && n_txq) { | |
950 | if (diag) { | |
951 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
952 | } | |
953 | ||
0072e931 | 954 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 955 | if (diag) { |
0072e931 MK |
956 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
957 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
958 | break; |
959 | } | |
960 | ||
67fe6d63 MK |
961 | diag = rte_eth_dev_set_mtu(dev->port_id, dev->mtu); |
962 | if (diag) { | |
4dd16ca0 IS |
963 | /* A device may not support rte_eth_dev_set_mtu, in this case |
964 | * flag a warning to the user and include the devices configured | |
965 | * MTU value that will be used instead. */ | |
966 | if (-ENOTSUP == diag) { | |
967 | rte_eth_dev_get_mtu(dev->port_id, &conf_mtu); | |
968 | VLOG_WARN("Interface %s does not support MTU configuration, " | |
969 | "max packet size supported is %"PRIu16".", | |
970 | dev->up.name, conf_mtu); | |
971 | } else { | |
972 | VLOG_ERR("Interface %s MTU (%d) setup error: %s", | |
973 | dev->up.name, dev->mtu, rte_strerror(-diag)); | |
974 | break; | |
975 | } | |
67fe6d63 MK |
976 | } |
977 | ||
b98d7669 | 978 | for (i = 0; i < n_txq; i++) { |
b685696b | 979 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
980 | dev->socket_id, NULL); |
981 | if (diag) { | |
1dfebee9 | 982 | VLOG_INFO("Interface %s unable to setup txq(%d): %s", |
b98d7669 DDP |
983 | dev->up.name, i, rte_strerror(-diag)); |
984 | break; | |
985 | } | |
986 | } | |
987 | ||
988 | if (i != n_txq) { | |
989 | /* Retry with less tx queues */ | |
990 | n_txq = i; | |
991 | continue; | |
992 | } | |
993 | ||
994 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 995 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
43307ad0 IS |
996 | dev->socket_id, NULL, |
997 | dev->dpdk_mp->mp); | |
b98d7669 | 998 | if (diag) { |
1dfebee9 | 999 | VLOG_INFO("Interface %s unable to setup rxq(%d): %s", |
b98d7669 DDP |
1000 | dev->up.name, i, rte_strerror(-diag)); |
1001 | break; | |
1002 | } | |
1003 | } | |
1004 | ||
1005 | if (i != n_rxq) { | |
1006 | /* Retry with less rx queues */ | |
1007 | n_rxq = i; | |
1008 | continue; | |
1009 | } | |
1010 | ||
1011 | dev->up.n_rxq = n_rxq; | |
81acebda | 1012 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
1013 | |
1014 | return 0; | |
1015 | } | |
1016 | ||
1017 | return diag; | |
1018 | } | |
1019 | ||
9fd39370 SC |
1020 | static void |
1021 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
1022 | { | |
1023 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
fa9f4eeb | 1024 | VLOG_WARN("Failed to enable flow control on device "DPDK_PORT_ID_FMT, |
bb37956a | 1025 | dev->port_id); |
9fd39370 SC |
1026 | } |
1027 | } | |
b98d7669 | 1028 | |
8a9562d2 | 1029 | static int |
c2adb102 IM |
1030 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
1031 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
1032 | { |
1033 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 1034 | struct rte_eth_dev_info info; |
8a9562d2 PS |
1035 | struct ether_addr eth_addr; |
1036 | int diag; | |
b98d7669 | 1037 | int n_rxq, n_txq; |
d4f5282c KT |
1038 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | |
1039 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
1040 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
8a9562d2 | 1041 | |
a0cb2d66 | 1042 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 1043 | |
e10ca8b9 MW |
1044 | if (strstr(info.driver_name, "vf") != NULL) { |
1045 | VLOG_INFO("Virtual function detected, HW_CRC_STRIP will be enabled"); | |
1046 | dev->hw_ol_features |= NETDEV_RX_HW_CRC_STRIP; | |
1047 | } else { | |
1048 | dev->hw_ol_features &= ~NETDEV_RX_HW_CRC_STRIP; | |
1049 | } | |
1050 | ||
d4f5282c KT |
1051 | if ((info.rx_offload_capa & rx_chksm_offload_capa) != |
1052 | rx_chksm_offload_capa) { | |
fa9f4eeb IM |
1053 | VLOG_WARN("Rx checksum offload is not supported on port " |
1054 | DPDK_PORT_ID_FMT, dev->port_id); | |
d4f5282c KT |
1055 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; |
1056 | } else { | |
1057 | dev->hw_ol_features |= NETDEV_RX_CHECKSUM_OFFLOAD; | |
1058 | } | |
1059 | ||
03f3f9c0 OM |
1060 | if (info.rx_offload_capa & DEV_RX_OFFLOAD_SCATTER) { |
1061 | dev->hw_ol_features |= NETDEV_RX_HW_SCATTER; | |
1062 | } else { | |
1063 | /* Do not warn on lack of scatter support */ | |
1064 | dev->hw_ol_features &= ~NETDEV_RX_HW_SCATTER; | |
1065 | } | |
1066 | ||
b98d7669 DDP |
1067 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
1068 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
1069 | ||
f8b64a61 | 1070 | diag = dpdk_eth_dev_port_config(dev, n_rxq, n_txq); |
8a9562d2 | 1071 | if (diag) { |
f8b64a61 RM |
1072 | VLOG_ERR("Interface %s(rxq:%d txq:%d lsc interrupt mode:%s) " |
1073 | "configure error: %s", | |
1074 | dev->up.name, n_rxq, n_txq, | |
1075 | dev->lsc_interrupt_mode ? "true" : "false", | |
1076 | rte_strerror(-diag)); | |
95fb793a | 1077 | return -diag; |
8a9562d2 PS |
1078 | } |
1079 | ||
8a9562d2 PS |
1080 | diag = rte_eth_dev_start(dev->port_id); |
1081 | if (diag) { | |
b98d7669 DDP |
1082 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
1083 | rte_strerror(-diag)); | |
95fb793a | 1084 | return -diag; |
8a9562d2 | 1085 | } |
606f6650 | 1086 | dev->started = true; |
8a9562d2 PS |
1087 | |
1088 | rte_eth_promiscuous_enable(dev->port_id); | |
1089 | rte_eth_allmulticast_enable(dev->port_id); | |
1090 | ||
1091 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
1092 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
fa9f4eeb IM |
1093 | VLOG_INFO_RL(&rl, "Port "DPDK_PORT_ID_FMT": "ETH_ADDR_FMT, |
1094 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); | |
8a9562d2 | 1095 | |
ca92d173 | 1096 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
1097 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
1098 | ||
43307ad0 | 1099 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); |
8a9562d2 | 1100 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; |
8a9562d2 PS |
1101 | return 0; |
1102 | } | |
1103 | ||
1104 | static struct netdev_dpdk * | |
1105 | netdev_dpdk_cast(const struct netdev *netdev) | |
1106 | { | |
1107 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
1108 | } | |
1109 | ||
1110 | static struct netdev * | |
1111 | netdev_dpdk_alloc(void) | |
1112 | { | |
bab69409 AC |
1113 | struct netdev_dpdk *dev; |
1114 | ||
65e19e70 DDP |
1115 | dev = dpdk_rte_mzalloc(sizeof *dev); |
1116 | if (dev) { | |
1117 | return &dev->up; | |
bab69409 | 1118 | } |
65e19e70 | 1119 | |
bab69409 | 1120 | return NULL; |
8a9562d2 PS |
1121 | } |
1122 | ||
eff23640 DDP |
1123 | static struct dpdk_tx_queue * |
1124 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 1125 | { |
eff23640 | 1126 | struct dpdk_tx_queue *txqs; |
bd5131ba | 1127 | unsigned i; |
5a034064 | 1128 | |
eff23640 DDP |
1129 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
1130 | if (txqs) { | |
1131 | for (i = 0; i < n_txqs; i++) { | |
1132 | /* Initialize map for vhost devices. */ | |
1133 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
1134 | rte_spinlock_init(&txqs[i].tx_lock); | |
1135 | } | |
5a034064 | 1136 | } |
eff23640 DDP |
1137 | |
1138 | return txqs; | |
5a034064 AW |
1139 | } |
1140 | ||
8a9562d2 | 1141 | static int |
bb37956a | 1142 | common_construct(struct netdev *netdev, dpdk_port_t port_no, |
1ce30dfd | 1143 | enum dpdk_dev_type type, int socket_id) |
5a034064 | 1144 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 1145 | { |
d46285a2 | 1146 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1147 | |
d46285a2 | 1148 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 1149 | |
d46285a2 | 1150 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 1151 | |
1b7a04e0 AW |
1152 | /* If the 'sid' is negative, it means that the kernel fails |
1153 | * to obtain the pci numa info. In that situation, always | |
1154 | * use 'SOCKET0'. */ | |
1ce30dfd | 1155 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 1156 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
1157 | dev->port_id = port_no; |
1158 | dev->type = type; | |
1159 | dev->flags = 0; | |
7f381c2e | 1160 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 1161 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
f8b64a61 | 1162 | dev->requested_lsc_interrupt_mode = 0; |
0a0f39df CL |
1163 | ovsrcu_index_init(&dev->vid, -1); |
1164 | dev->vhost_reconfigured = false; | |
5dcde09c | 1165 | dev->attached = false; |
8a9562d2 | 1166 | |
78bd47cf | 1167 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 1168 | |
9509913a IS |
1169 | ovsrcu_init(&dev->ingress_policer, NULL); |
1170 | dev->policer_rate = 0; | |
1171 | dev->policer_burst = 0; | |
1172 | ||
7f381c2e DDP |
1173 | netdev->n_rxq = 0; |
1174 | netdev->n_txq = 0; | |
1175 | dev->requested_n_rxq = NR_QUEUE; | |
1176 | dev->requested_n_txq = NR_QUEUE; | |
1177 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
1178 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 1179 | |
9fd39370 SC |
1180 | /* Initialize the flow control to NULL */ |
1181 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
1182 | |
1183 | /* Initilize the hardware offload flags to 0 */ | |
1184 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
1185 | |
1186 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
1187 | ||
d46285a2 | 1188 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 1189 | |
7f381c2e DDP |
1190 | netdev_request_reconfigure(netdev); |
1191 | ||
971f4b39 MW |
1192 | dev->rte_xstats_names = NULL; |
1193 | dev->rte_xstats_names_size = 0; | |
1194 | ||
1195 | dev->rte_xstats_ids = NULL; | |
1196 | dev->rte_xstats_ids_size = 0; | |
1197 | ||
1ce30dfd | 1198 | return 0; |
95fb793a | 1199 | } |
1200 | ||
b83a2df1 MV |
1201 | /* dev_name must be the prefix followed by a positive decimal number. |
1202 | * (no leading + or - signs are allowed) */ | |
95fb793a | 1203 | static int |
1204 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
1205 | unsigned int *port_no) | |
1206 | { | |
1207 | const char *cport; | |
1208 | ||
1209 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
1210 | return ENODEV; | |
1211 | } | |
1212 | ||
1213 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
1214 | |
1215 | if (str_to_uint(cport, 10, port_no)) { | |
1216 | return 0; | |
1217 | } else { | |
1218 | return ENODEV; | |
1219 | } | |
95fb793a | 1220 | } |
1221 | ||
40e940e4 OM |
1222 | /* Get the number of OVS interfaces which have the same DPDK |
1223 | * rte device (e.g. same pci bus address). | |
1224 | * FIXME: avoid direct access to DPDK internal array rte_eth_devices. | |
1225 | */ | |
1226 | static int | |
1227 | netdev_dpdk_get_num_ports(struct rte_device *device) | |
1228 | OVS_REQUIRES(dpdk_mutex) | |
1229 | { | |
1230 | struct netdev_dpdk *dev; | |
1231 | int count = 0; | |
1232 | ||
1233 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1234 | if (rte_eth_devices[dev->port_id].device == device | |
1235 | && rte_eth_devices[dev->port_id].state != RTE_ETH_DEV_UNUSED) { | |
1236 | count++; | |
1237 | } | |
1238 | } | |
1239 | return count; | |
1240 | } | |
1241 | ||
1ce30dfd DDP |
1242 | static int |
1243 | vhost_common_construct(struct netdev *netdev) | |
1244 | OVS_REQUIRES(dpdk_mutex) | |
1245 | { | |
1246 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
1247 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1248 | ||
1249 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); | |
1250 | if (!dev->tx_q) { | |
1251 | return ENOMEM; | |
1252 | } | |
1253 | ||
bb37956a IM |
1254 | return common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1255 | DPDK_DEV_VHOST, socket_id); | |
1ce30dfd DDP |
1256 | } |
1257 | ||
7d1ced01 | 1258 | static int |
53f50d24 | 1259 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 1260 | { |
d46285a2 DDP |
1261 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1262 | const char *name = netdev->name; | |
7d1ced01 | 1263 | int err; |
a0cb2d66 | 1264 | |
1af27e8a DDP |
1265 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
1266 | * the file system. '/' or '\' would traverse directories, so they're not | |
1267 | * acceptable in 'name'. */ | |
1268 | if (strchr(name, '/') || strchr(name, '\\')) { | |
1269 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
1270 | "A valid name must not include '/' or '\\'", | |
1271 | name); | |
1272 | return EINVAL; | |
1273 | } | |
1274 | ||
7d1ced01 CL |
1275 | ovs_mutex_lock(&dpdk_mutex); |
1276 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 1277 | * the socket is to be created, then register the socket. |
7d1ced01 | 1278 | */ |
2d24d165 | 1279 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 1280 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 1281 | |
2d24d165 CL |
1282 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
1283 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
1284 | if (err) { |
1285 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 1286 | dev->vhost_id); |
f3e7ec25 | 1287 | goto out; |
e5c0f5a4 | 1288 | } else { |
2d24d165 CL |
1289 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
1290 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
1291 | dev->vhost_id, name); | |
1292 | } | |
f3e7ec25 MW |
1293 | |
1294 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
1295 | &virtio_net_device_ops); | |
1296 | if (err) { | |
1297 | VLOG_ERR("rte_vhost_driver_callback_register failed for vhost user " | |
1298 | "port: %s\n", name); | |
1299 | goto out; | |
1300 | } | |
1301 | ||
1302 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
1303 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
1304 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
1305 | | 1ULL << VIRTIO_NET_F_CSUM); | |
1306 | if (err) { | |
1307 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
1308 | "port: %s\n", name); | |
1309 | goto out; | |
1310 | } | |
1311 | ||
1312 | err = rte_vhost_driver_start(dev->vhost_id); | |
1313 | if (err) { | |
1314 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
1315 | "port: %s\n", name); | |
1316 | goto out; | |
1317 | } | |
1318 | ||
1ce30dfd | 1319 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1320 | if (err) { |
1321 | VLOG_ERR("vhost_common_construct failed for vhost user " | |
1322 | "port: %s\n", name); | |
1323 | } | |
2d24d165 | 1324 | |
f3e7ec25 | 1325 | out: |
2d24d165 | 1326 | ovs_mutex_unlock(&dpdk_mutex); |
28ca969e AC |
1327 | VLOG_WARN_ONCE("dpdkvhostuser ports are considered deprecated; " |
1328 | "please migrate to dpdkvhostuserclient ports."); | |
2d24d165 CL |
1329 | return err; |
1330 | } | |
1331 | ||
1332 | static int | |
1333 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
1334 | { | |
1335 | int err; | |
1336 | ||
2d24d165 | 1337 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 1338 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1339 | if (err) { |
1340 | VLOG_ERR("vhost_common_construct failed for vhost user client" | |
1341 | "port: %s\n", netdev->name); | |
1342 | } | |
7d1ced01 | 1343 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
1344 | return err; |
1345 | } | |
1346 | ||
95fb793a | 1347 | static int |
1348 | netdev_dpdk_construct(struct netdev *netdev) | |
1349 | { | |
95fb793a | 1350 | int err; |
1351 | ||
95fb793a | 1352 | ovs_mutex_lock(&dpdk_mutex); |
bb37956a IM |
1353 | err = common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1354 | DPDK_DEV_ETH, SOCKET0); | |
8a9562d2 PS |
1355 | ovs_mutex_unlock(&dpdk_mutex); |
1356 | return err; | |
1357 | } | |
1358 | ||
1ce30dfd DDP |
1359 | static void |
1360 | common_destruct(struct netdev_dpdk *dev) | |
1361 | OVS_REQUIRES(dpdk_mutex) | |
1362 | OVS_EXCLUDED(dev->mutex) | |
1363 | { | |
1364 | rte_free(dev->tx_q); | |
43307ad0 | 1365 | dpdk_mp_put(dev->dpdk_mp); |
1ce30dfd DDP |
1366 | |
1367 | ovs_list_remove(&dev->list_node); | |
1368 | free(ovsrcu_get_protected(struct ingress_policer *, | |
1369 | &dev->ingress_policer)); | |
1370 | ovs_mutex_destroy(&dev->mutex); | |
1371 | } | |
1372 | ||
8a9562d2 | 1373 | static void |
d46285a2 | 1374 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 1375 | { |
d46285a2 | 1376 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
40e940e4 OM |
1377 | struct rte_device *rte_dev; |
1378 | struct rte_eth_dev *eth_dev; | |
1379 | bool remove_on_close; | |
8a9562d2 | 1380 | |
8d38823b | 1381 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1382 | |
8a9562d2 | 1383 | rte_eth_dev_stop(dev->port_id); |
606f6650 | 1384 | dev->started = false; |
5dcde09c IM |
1385 | |
1386 | if (dev->attached) { | |
40e940e4 OM |
1387 | /* Retrieve eth device data before closing it. |
1388 | * FIXME: avoid direct access to DPDK internal array rte_eth_devices. | |
1389 | */ | |
1390 | eth_dev = &rte_eth_devices[dev->port_id]; | |
1391 | remove_on_close = | |
1392 | eth_dev->data && | |
1393 | (eth_dev->data->dev_flags & RTE_ETH_DEV_CLOSE_REMOVE); | |
1394 | rte_dev = eth_dev->device; | |
1395 | ||
1396 | /* Remove the eth device. */ | |
5dcde09c | 1397 | rte_eth_dev_close(dev->port_id); |
40e940e4 OM |
1398 | |
1399 | /* Remove this rte device and all its eth devices if flag | |
1400 | * RTE_ETH_DEV_CLOSE_REMOVE is not supported (which means representors | |
1401 | * are not supported), or if all the eth devices belonging to the rte | |
1402 | * device are closed. | |
1403 | */ | |
1404 | if (!remove_on_close || !netdev_dpdk_get_num_ports(rte_dev)) { | |
1405 | if (rte_dev_remove(rte_dev) < 0) { | |
1406 | VLOG_ERR("Device '%s' can not be detached", dev->devargs); | |
1407 | } else { | |
1408 | /* Device was closed and detached. */ | |
1409 | VLOG_INFO("Device '%s' has been removed and detached", | |
1410 | dev->devargs); | |
1411 | } | |
5dcde09c | 1412 | } else { |
40e940e4 OM |
1413 | /* Device was only closed. rte_dev_remove() was not called. */ |
1414 | VLOG_INFO("Device '%s' has been removed", dev->devargs); | |
5dcde09c IM |
1415 | } |
1416 | } | |
1417 | ||
ac1a9bb9 | 1418 | netdev_dpdk_clear_xstats(dev); |
55e075e6 | 1419 | free(dev->devargs); |
1ce30dfd | 1420 | common_destruct(dev); |
8d38823b | 1421 | |
8a9562d2 | 1422 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1423 | } |
8a9562d2 | 1424 | |
3f891bbe DDP |
1425 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1426 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1427 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1428 | static int | |
c1ff66ac CL |
1429 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1430 | char *vhost_id) | |
3f891bbe DDP |
1431 | OVS_EXCLUDED(dpdk_mutex) |
1432 | OVS_EXCLUDED(dev->mutex) | |
1433 | { | |
c1ff66ac | 1434 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1435 | } |
1436 | ||
58397e6c | 1437 | static void |
d46285a2 | 1438 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1439 | { |
d46285a2 | 1440 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1441 | char *vhost_id; |
58397e6c | 1442 | |
8d38823b | 1443 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1444 | |
c62da695 | 1445 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1446 | if (netdev_dpdk_get_vid(dev) >= 0 |
1447 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1448 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1449 | netdev->name); |
58be5c0e MK |
1450 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1451 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1452 | } |
1453 | ||
2d24d165 | 1454 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1455 | |
1ce30dfd DDP |
1456 | common_destruct(dev); |
1457 | ||
58397e6c | 1458 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1459 | |
569c26da | 1460 | if (!vhost_id[0]) { |
821b8664 IM |
1461 | goto out; |
1462 | } | |
1463 | ||
c1ff66ac | 1464 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1465 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1466 | netdev->name, vhost_id); | |
c1ff66ac CL |
1467 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1468 | /* OVS server mode - remove this socket from list for deletion */ | |
1469 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1470 | } |
821b8664 | 1471 | out: |
c1ff66ac | 1472 | free(vhost_id); |
8a9562d2 PS |
1473 | } |
1474 | ||
1475 | static void | |
d46285a2 | 1476 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1477 | { |
d46285a2 | 1478 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1479 | |
d46285a2 | 1480 | rte_free(dev); |
8a9562d2 PS |
1481 | } |
1482 | ||
971f4b39 | 1483 | static void |
ac1a9bb9 | 1484 | netdev_dpdk_clear_xstats(struct netdev_dpdk *dev) |
971f4b39 MW |
1485 | { |
1486 | /* If statistics are already allocated, we have to | |
1487 | * reconfigure, as port_id could have been changed. */ | |
1488 | if (dev->rte_xstats_names) { | |
1489 | free(dev->rte_xstats_names); | |
1490 | dev->rte_xstats_names = NULL; | |
1491 | dev->rte_xstats_names_size = 0; | |
1492 | } | |
1493 | if (dev->rte_xstats_ids) { | |
1494 | free(dev->rte_xstats_ids); | |
1495 | dev->rte_xstats_ids = NULL; | |
1496 | dev->rte_xstats_ids_size = 0; | |
1497 | } | |
1498 | } | |
1499 | ||
1500 | static const char* | |
1501 | netdev_dpdk_get_xstat_name(struct netdev_dpdk *dev, uint64_t id) | |
1502 | { | |
1503 | if (id >= dev->rte_xstats_names_size) { | |
1504 | return "UNKNOWN"; | |
1505 | } | |
1506 | return dev->rte_xstats_names[id].name; | |
1507 | } | |
1508 | ||
1509 | static bool | |
1510 | netdev_dpdk_configure_xstats(struct netdev_dpdk *dev) | |
1511 | OVS_REQUIRES(dev->mutex) | |
1512 | { | |
1513 | int rte_xstats_len; | |
1514 | bool ret; | |
1515 | struct rte_eth_xstat *rte_xstats; | |
1516 | uint64_t id; | |
1517 | int xstats_no; | |
1518 | const char *name; | |
1519 | ||
1520 | /* Retrieving all XSTATS names. If something will go wrong | |
1521 | * or amount of counters will be equal 0, rte_xstats_names | |
1522 | * buffer will be marked as NULL, and any further xstats | |
1523 | * query won't be performed (e.g. during netdev_dpdk_get_stats | |
1524 | * execution). */ | |
1525 | ||
1526 | ret = false; | |
1527 | rte_xstats = NULL; | |
1528 | ||
1529 | if (dev->rte_xstats_names == NULL || dev->rte_xstats_ids == NULL) { | |
1530 | dev->rte_xstats_names_size = | |
1531 | rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
1532 | ||
1533 | if (dev->rte_xstats_names_size < 0) { | |
fa9f4eeb IM |
1534 | VLOG_WARN("Cannot get XSTATS for port: "DPDK_PORT_ID_FMT, |
1535 | dev->port_id); | |
971f4b39 MW |
1536 | dev->rte_xstats_names_size = 0; |
1537 | } else { | |
1538 | /* Reserve memory for xstats names and values */ | |
1539 | dev->rte_xstats_names = xcalloc(dev->rte_xstats_names_size, | |
1540 | sizeof *dev->rte_xstats_names); | |
1541 | ||
1542 | if (dev->rte_xstats_names) { | |
1543 | /* Retreive xstats names */ | |
1544 | rte_xstats_len = | |
1545 | rte_eth_xstats_get_names(dev->port_id, | |
1546 | dev->rte_xstats_names, | |
1547 | dev->rte_xstats_names_size); | |
1548 | ||
1549 | if (rte_xstats_len < 0) { | |
fa9f4eeb IM |
1550 | VLOG_WARN("Cannot get XSTATS names for port: " |
1551 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 MW |
1552 | goto out; |
1553 | } else if (rte_xstats_len != dev->rte_xstats_names_size) { | |
fa9f4eeb IM |
1554 | VLOG_WARN("XSTATS size doesn't match for port: " |
1555 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 MW |
1556 | goto out; |
1557 | } | |
1558 | ||
1559 | dev->rte_xstats_ids = xcalloc(dev->rte_xstats_names_size, | |
1560 | sizeof(uint64_t)); | |
1561 | ||
1562 | /* We have to calculate number of counters */ | |
1563 | rte_xstats = xmalloc(rte_xstats_len * sizeof *rte_xstats); | |
1564 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
1565 | ||
1566 | /* Retreive xstats values */ | |
1567 | if (rte_eth_xstats_get(dev->port_id, rte_xstats, | |
1568 | rte_xstats_len) > 0) { | |
1569 | dev->rte_xstats_ids_size = 0; | |
1570 | xstats_no = 0; | |
1571 | for (uint32_t i = 0; i < rte_xstats_len; i++) { | |
1572 | id = rte_xstats[i].id; | |
1573 | name = netdev_dpdk_get_xstat_name(dev, id); | |
1574 | /* We need to filter out everything except | |
1575 | * dropped, error and management counters */ | |
1576 | if (string_ends_with(name, "_errors") || | |
1577 | strstr(name, "_management_") || | |
1578 | string_ends_with(name, "_dropped")) { | |
1579 | ||
1580 | dev->rte_xstats_ids[xstats_no] = id; | |
1581 | xstats_no++; | |
1582 | } | |
1583 | } | |
1584 | dev->rte_xstats_ids_size = xstats_no; | |
1585 | ret = true; | |
1586 | } else { | |
fa9f4eeb IM |
1587 | VLOG_WARN("Can't get XSTATS IDs for port: " |
1588 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 | 1589 | } |
34eb0863 IM |
1590 | |
1591 | free(rte_xstats); | |
971f4b39 MW |
1592 | } |
1593 | } | |
1594 | } else { | |
1595 | /* Already configured */ | |
1596 | ret = true; | |
1597 | } | |
1598 | ||
1599 | out: | |
1600 | if (!ret) { | |
1601 | netdev_dpdk_clear_xstats(dev); | |
1602 | } | |
1603 | return ret; | |
1604 | } | |
1605 | ||
8a9562d2 | 1606 | static int |
a14b8947 | 1607 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1608 | { |
a14b8947 | 1609 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1610 | |
1611 | ovs_mutex_lock(&dev->mutex); | |
1612 | ||
050c60bf | 1613 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1614 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1615 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1616 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1617 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1618 | |
1619 | if (dev->type == DPDK_DEV_ETH) { | |
1620 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1621 | dev->requested_rxq_size); | |
1622 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1623 | dev->rxq_size); | |
1624 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1625 | dev->requested_txq_size); | |
1626 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1627 | dev->txq_size); | |
1a2bb118 SC |
1628 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1629 | smap_add(args, "rx_csum_offload", "true"); | |
8155ab7e KT |
1630 | } else { |
1631 | smap_add(args, "rx_csum_offload", "false"); | |
1a2bb118 | 1632 | } |
f8b64a61 RM |
1633 | smap_add(args, "lsc_interrupt_mode", |
1634 | dev->lsc_interrupt_mode ? "true" : "false"); | |
451f26fd | 1635 | } |
8a9562d2 PS |
1636 | ovs_mutex_unlock(&dev->mutex); |
1637 | ||
1638 | return 0; | |
1639 | } | |
1640 | ||
55e075e6 | 1641 | static struct netdev_dpdk * |
bb37956a | 1642 | netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id) |
55e075e6 CL |
1643 | OVS_REQUIRES(dpdk_mutex) |
1644 | { | |
1645 | struct netdev_dpdk *dev; | |
1646 | ||
1647 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1648 | if (dev->port_id == port_id) { | |
1649 | return dev; | |
1650 | } | |
1651 | } | |
1652 | ||
1653 | return NULL; | |
1654 | } | |
1655 | ||
5e758818 YL |
1656 | static dpdk_port_t |
1657 | netdev_dpdk_get_port_by_mac(const char *mac_str) | |
1658 | { | |
1659 | dpdk_port_t port_id; | |
1660 | struct eth_addr mac, port_mac; | |
1661 | ||
1662 | if (!eth_addr_from_string(mac_str, &mac)) { | |
1663 | VLOG_ERR("invalid mac: %s", mac_str); | |
1664 | return DPDK_ETH_PORT_ID_INVALID; | |
1665 | } | |
1666 | ||
1667 | RTE_ETH_FOREACH_DEV (port_id) { | |
1668 | struct ether_addr ea; | |
1669 | ||
1670 | rte_eth_macaddr_get(port_id, &ea); | |
1671 | memcpy(port_mac.ea, ea.addr_bytes, ETH_ADDR_LEN); | |
1672 | if (eth_addr_equals(mac, port_mac)) { | |
1673 | return port_id; | |
1674 | } | |
1675 | } | |
1676 | ||
1677 | return DPDK_ETH_PORT_ID_INVALID; | |
1678 | } | |
1679 | ||
40e940e4 OM |
1680 | /* Return the first DPDK port id matching the devargs pattern. */ |
1681 | static dpdk_port_t netdev_dpdk_get_port_by_devargs(const char *devargs) | |
1682 | OVS_REQUIRES(dpdk_mutex) | |
1683 | { | |
1684 | dpdk_port_t port_id; | |
1685 | struct rte_dev_iterator iterator; | |
1686 | ||
1687 | RTE_ETH_FOREACH_MATCHING_DEV (port_id, devargs, &iterator) { | |
1688 | /* If a break is done - must call rte_eth_iterator_cleanup. */ | |
1689 | rte_eth_iterator_cleanup(&iterator); | |
1690 | break; | |
1691 | } | |
1692 | ||
1693 | return port_id; | |
1694 | } | |
1695 | ||
5e758818 | 1696 | /* |
40e940e4 OM |
1697 | * Normally, a PCI id (optionally followed by a representor number) |
1698 | * is enough for identifying a specific DPDK port. | |
5e758818 YL |
1699 | * However, for some NICs having multiple ports sharing the same PCI |
1700 | * id, using PCI id won't work then. | |
1701 | * | |
1702 | * To fix that, here one more method is introduced: "class=eth,mac=$MAC". | |
1703 | * | |
1704 | * Note that the compatibility is fully kept: user can still use the | |
1705 | * PCI id for adding ports (when it's enough for them). | |
1706 | */ | |
bb37956a | 1707 | static dpdk_port_t |
5dcde09c IM |
1708 | netdev_dpdk_process_devargs(struct netdev_dpdk *dev, |
1709 | const char *devargs, char **errp) | |
40e940e4 | 1710 | OVS_REQUIRES(dpdk_mutex) |
55e075e6 | 1711 | { |
40e940e4 | 1712 | dpdk_port_t new_port_id; |
55e075e6 | 1713 | |
5e758818 YL |
1714 | if (strncmp(devargs, "class=eth,mac=", 14) == 0) { |
1715 | new_port_id = netdev_dpdk_get_port_by_mac(&devargs[14]); | |
1716 | } else { | |
40e940e4 OM |
1717 | new_port_id = netdev_dpdk_get_port_by_devargs(devargs); |
1718 | if (!rte_eth_dev_is_valid_port(new_port_id)) { | |
5e758818 | 1719 | /* Device not found in DPDK, attempt to attach it */ |
40e940e4 | 1720 | if (rte_dev_probe(devargs)) { |
5e758818 | 1721 | new_port_id = DPDK_ETH_PORT_ID_INVALID; |
40e940e4 OM |
1722 | } else { |
1723 | new_port_id = netdev_dpdk_get_port_by_devargs(devargs); | |
1724 | if (rte_eth_dev_is_valid_port(new_port_id)) { | |
1725 | /* Attach successful */ | |
1726 | dev->attached = true; | |
1727 | VLOG_INFO("Device '%s' attached to DPDK", devargs); | |
1728 | } else { | |
1729 | /* Attach unsuccessful */ | |
1730 | new_port_id = DPDK_ETH_PORT_ID_INVALID; | |
1731 | } | |
5e758818 | 1732 | } |
55e075e6 | 1733 | } |
5e758818 YL |
1734 | } |
1735 | ||
1736 | if (new_port_id == DPDK_ETH_PORT_ID_INVALID) { | |
1737 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", devargs); | |
55e075e6 CL |
1738 | } |
1739 | ||
1740 | return new_port_id; | |
1741 | } | |
1742 | ||
c3d062a7 CL |
1743 | static void |
1744 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1745 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1746 | { |
050c60bf | 1747 | int new_n_rxq; |
a14b8947 | 1748 | |
2a21e757 | 1749 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1750 | if (new_n_rxq != dev->requested_n_rxq) { |
1751 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1752 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1753 | } |
c3d062a7 CL |
1754 | } |
1755 | ||
b685696b CL |
1756 | static void |
1757 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1758 | const char *flag, int default_size, int *new_size) | |
1759 | { | |
1760 | int queue_size = smap_get_int(args, flag, default_size); | |
1761 | ||
1762 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1763 | || !is_pow2(queue_size)) { | |
1764 | queue_size = default_size; | |
1765 | } | |
1766 | ||
1767 | if (queue_size != *new_size) { | |
1768 | *new_size = queue_size; | |
1769 | netdev_request_reconfigure(netdev); | |
1770 | } | |
1771 | } | |
1772 | ||
c3d062a7 | 1773 | static int |
9fff138e DDP |
1774 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1775 | char **errp) | |
c3d062a7 CL |
1776 | { |
1777 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
f8b64a61 | 1778 | bool rx_fc_en, tx_fc_en, autoneg, lsc_interrupt_mode; |
b614c894 IM |
1779 | enum rte_eth_fc_mode fc_mode; |
1780 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1781 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1782 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1783 | }; | |
55e075e6 CL |
1784 | const char *new_devargs; |
1785 | int err = 0; | |
c3d062a7 | 1786 | |
55e075e6 | 1787 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1788 | ovs_mutex_lock(&dev->mutex); |
1789 | ||
1790 | dpdk_set_rxq_config(dev, args); | |
1791 | ||
b685696b CL |
1792 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1793 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1794 | &dev->requested_rxq_size); | |
1795 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1796 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1797 | &dev->requested_txq_size); | |
1798 | ||
55e075e6 CL |
1799 | new_devargs = smap_get(args, "dpdk-devargs"); |
1800 | ||
1801 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1802 | /* The user requested a new device. If we return error, the caller | |
1803 | * will delete this netdev and try to recreate it. */ | |
1804 | err = EAGAIN; | |
1805 | goto out; | |
1806 | } | |
1807 | ||
1808 | /* dpdk-devargs is required for device configuration */ | |
1809 | if (new_devargs && new_devargs[0]) { | |
1810 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1811 | * is valid */ | |
1812 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1813 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
bb37956a IM |
1814 | dpdk_port_t new_port_id = netdev_dpdk_process_devargs(dev, |
1815 | new_devargs, | |
1816 | errp); | |
55e075e6 CL |
1817 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1818 | err = EINVAL; | |
1819 | } else if (new_port_id == dev->port_id) { | |
1820 | /* Already configured, do not reconfigure again */ | |
1821 | err = 0; | |
1822 | } else { | |
1823 | struct netdev_dpdk *dup_dev; | |
bb37956a | 1824 | |
55e075e6 CL |
1825 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); |
1826 | if (dup_dev) { | |
9fff138e | 1827 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
40e940e4 | 1828 | "which is already in use by '%s'", |
9fff138e DDP |
1829 | netdev_get_name(netdev), new_devargs, |
1830 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1831 | err = EADDRINUSE; |
1832 | } else { | |
bd4e172b | 1833 | int sid = rte_eth_dev_socket_id(new_port_id); |
bb37956a | 1834 | |
bd4e172b | 1835 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; |
55e075e6 CL |
1836 | dev->devargs = xstrdup(new_devargs); |
1837 | dev->port_id = new_port_id; | |
1838 | netdev_request_reconfigure(&dev->up); | |
971f4b39 | 1839 | netdev_dpdk_clear_xstats(dev); |
55e075e6 CL |
1840 | err = 0; |
1841 | } | |
1842 | } | |
1843 | } | |
1844 | } else { | |
9fff138e DDP |
1845 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1846 | "The old 'dpdk<port_id>' names are not supported", | |
1847 | netdev_get_name(netdev)); | |
55e075e6 CL |
1848 | err = EINVAL; |
1849 | } | |
1850 | ||
1851 | if (err) { | |
1852 | goto out; | |
1853 | } | |
1854 | ||
f8b64a61 RM |
1855 | lsc_interrupt_mode = smap_get_bool(args, "dpdk-lsc-interrupt", false); |
1856 | if (dev->requested_lsc_interrupt_mode != lsc_interrupt_mode) { | |
1857 | dev->requested_lsc_interrupt_mode = lsc_interrupt_mode; | |
1858 | netdev_request_reconfigure(netdev); | |
1859 | } | |
1860 | ||
c3d062a7 CL |
1861 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1862 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1863 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1864 | |
b614c894 IM |
1865 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1866 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1867 | dev->fc_conf.mode = fc_mode; | |
1868 | dev->fc_conf.autoneg = autoneg; | |
7e1de65e SC |
1869 | /* Get the Flow control configuration for DPDK-ETH */ |
1870 | err = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
1871 | if (err) { | |
1872 | VLOG_WARN("Cannot get flow control parameters on port " | |
1873 | DPDK_PORT_ID_FMT", err=%d", dev->port_id, err); | |
1874 | } | |
b614c894 IM |
1875 | dpdk_eth_flow_ctrl_setup(dev); |
1876 | } | |
9fd39370 | 1877 | |
55e075e6 | 1878 | out: |
c3d062a7 | 1879 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1880 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1881 | |
55e075e6 | 1882 | return err; |
c3d062a7 CL |
1883 | } |
1884 | ||
1885 | static int | |
9fff138e DDP |
1886 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1887 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1888 | { |
1889 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1890 | ||
1891 | ovs_mutex_lock(&dev->mutex); | |
1892 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1893 | ovs_mutex_unlock(&dev->mutex); |
1894 | ||
1895 | return 0; | |
1896 | } | |
1897 | ||
c1ff66ac | 1898 | static int |
2d24d165 | 1899 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1900 | const struct smap *args, |
1901 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1902 | { |
1903 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1904 | const char *path; | |
1905 | ||
6881885a | 1906 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1907 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1908 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1909 | if (path && strcmp(path, dev->vhost_id)) { |
1910 | strcpy(dev->vhost_id, path); | |
10087cba CL |
1911 | /* check zero copy configuration */ |
1912 | if (smap_get_bool(args, "dq-zero-copy", false)) { | |
1913 | dev->vhost_driver_flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
1914 | } else { | |
1915 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
1916 | } | |
c1ff66ac CL |
1917 | netdev_request_reconfigure(netdev); |
1918 | } | |
1919 | } | |
6881885a | 1920 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1921 | |
1922 | return 0; | |
1923 | } | |
1924 | ||
7dec44fe | 1925 | static int |
d46285a2 | 1926 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1927 | { |
d46285a2 | 1928 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1929 | |
d46285a2 | 1930 | return dev->socket_id; |
7dec44fe AW |
1931 | } |
1932 | ||
050c60bf | 1933 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1934 | static int |
050c60bf | 1935 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1936 | { |
d46285a2 | 1937 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1938 | |
d46285a2 | 1939 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1940 | |
050c60bf DDP |
1941 | if (dev->requested_n_txq == n_txq) { |
1942 | goto out; | |
4573fbd3 FL |
1943 | } |
1944 | ||
050c60bf DDP |
1945 | dev->requested_n_txq = n_txq; |
1946 | netdev_request_reconfigure(netdev); | |
58397e6c | 1947 | |
050c60bf | 1948 | out: |
d46285a2 | 1949 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1950 | return 0; |
58397e6c KT |
1951 | } |
1952 | ||
8a9562d2 PS |
1953 | static struct netdev_rxq * |
1954 | netdev_dpdk_rxq_alloc(void) | |
1955 | { | |
1956 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1957 | ||
eff23640 DDP |
1958 | if (rx) { |
1959 | return &rx->up; | |
1960 | } | |
1961 | ||
1962 | return NULL; | |
8a9562d2 PS |
1963 | } |
1964 | ||
1965 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1966 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1967 | { |
d46285a2 | 1968 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1969 | } |
1970 | ||
1971 | static int | |
d46285a2 | 1972 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1973 | { |
d46285a2 DDP |
1974 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1975 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1976 | |
d46285a2 DDP |
1977 | ovs_mutex_lock(&dev->mutex); |
1978 | rx->port_id = dev->port_id; | |
1979 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1980 | |
1981 | return 0; | |
1982 | } | |
1983 | ||
1984 | static void | |
d46285a2 | 1985 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1986 | { |
1987 | } | |
1988 | ||
1989 | static void | |
d46285a2 | 1990 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1991 | { |
d46285a2 | 1992 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1993 | |
1994 | rte_free(rx); | |
1995 | } | |
1996 | ||
819f13bd DDP |
1997 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1998 | * 'pkts', even in case of failure. | |
1999 | * | |
2000 | * Returns the number of packets that weren't transmitted. */ | |
2001 | static inline int | |
b59cc14e | 2002 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 2003 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 2004 | { |
1304f1f8 DDP |
2005 | uint32_t nb_tx = 0; |
2006 | ||
b59cc14e | 2007 | while (nb_tx != cnt) { |
1304f1f8 DDP |
2008 | uint32_t ret; |
2009 | ||
b59cc14e | 2010 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
2011 | if (!ret) { |
2012 | break; | |
2013 | } | |
2014 | ||
2015 | nb_tx += ret; | |
2016 | } | |
8a9562d2 | 2017 | |
b59cc14e | 2018 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 2019 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
2020 | * packet could come from a different mempool) */ |
2021 | int i; | |
2022 | ||
b59cc14e IM |
2023 | for (i = nb_tx; i < cnt; i++) { |
2024 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 2025 | } |
8a9562d2 | 2026 | } |
819f13bd DDP |
2027 | |
2028 | return cnt - nb_tx; | |
8a9562d2 PS |
2029 | } |
2030 | ||
f3926f29 IS |
2031 | static inline bool |
2032 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
03f3f9c0 | 2033 | struct rte_meter_srtcm_profile *profile, |
f3926f29 IS |
2034 | struct rte_mbuf *pkt, uint64_t time) |
2035 | { | |
2036 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
2037 | ||
03f3f9c0 OM |
2038 | return rte_meter_srtcm_color_blind_check(meter, profile, time, pkt_len) == |
2039 | e_RTE_METER_GREEN; | |
f3926f29 IS |
2040 | } |
2041 | ||
2042 | static int | |
2043 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
03f3f9c0 | 2044 | struct rte_meter_srtcm_profile *profile, |
3e90f7d7 | 2045 | struct rte_mbuf **pkts, int pkt_cnt, |
7d7ded7a | 2046 | bool should_steal) |
f3926f29 IS |
2047 | { |
2048 | int i = 0; | |
2049 | int cnt = 0; | |
2050 | struct rte_mbuf *pkt = NULL; | |
2051 | uint64_t current_time = rte_rdtsc(); | |
2052 | ||
2053 | for (i = 0; i < pkt_cnt; i++) { | |
2054 | pkt = pkts[i]; | |
2055 | /* Handle current packet */ | |
03f3f9c0 OM |
2056 | if (netdev_dpdk_policer_pkt_handle(meter, profile, |
2057 | pkt, current_time)) { | |
f3926f29 IS |
2058 | if (cnt != i) { |
2059 | pkts[cnt] = pkt; | |
2060 | } | |
2061 | cnt++; | |
2062 | } else { | |
7d7ded7a | 2063 | if (should_steal) { |
3e90f7d7 GZ |
2064 | rte_pktmbuf_free(pkt); |
2065 | } | |
f3926f29 IS |
2066 | } |
2067 | } | |
2068 | ||
2069 | return cnt; | |
2070 | } | |
2071 | ||
9509913a IS |
2072 | static int |
2073 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
7d7ded7a | 2074 | int pkt_cnt, bool should_steal) |
9509913a IS |
2075 | { |
2076 | int cnt = 0; | |
2077 | ||
2078 | rte_spinlock_lock(&policer->policer_lock); | |
03f3f9c0 OM |
2079 | cnt = netdev_dpdk_policer_run(&policer->in_policer, &policer->in_prof, |
2080 | pkts, pkt_cnt, should_steal); | |
9509913a IS |
2081 | rte_spinlock_unlock(&policer->policer_lock); |
2082 | ||
2083 | return cnt; | |
2084 | } | |
2085 | ||
58397e6c | 2086 | static bool |
0a0f39df | 2087 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 2088 | { |
0a0f39df | 2089 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
2090 | } |
2091 | ||
d6e3feb5 | 2092 | static inline void |
2093 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
2094 | unsigned int packet_size) | |
2095 | { | |
2096 | /* Hard-coded search for the size bucket. */ | |
2097 | if (packet_size < 256) { | |
2098 | if (packet_size >= 128) { | |
2099 | stats->rx_128_to_255_packets++; | |
2100 | } else if (packet_size <= 64) { | |
2101 | stats->rx_1_to_64_packets++; | |
2102 | } else { | |
2103 | stats->rx_65_to_127_packets++; | |
2104 | } | |
2105 | } else { | |
2106 | if (packet_size >= 1523) { | |
2107 | stats->rx_1523_to_max_packets++; | |
2108 | } else if (packet_size >= 1024) { | |
2109 | stats->rx_1024_to_1522_packets++; | |
2110 | } else if (packet_size < 512) { | |
2111 | stats->rx_256_to_511_packets++; | |
2112 | } else { | |
2113 | stats->rx_512_to_1023_packets++; | |
2114 | } | |
2115 | } | |
2116 | } | |
2117 | ||
9e3ddd45 TP |
2118 | static inline void |
2119 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
2120 | struct dp_packet **packets, int count, |
2121 | int dropped) | |
9e3ddd45 TP |
2122 | { |
2123 | int i; | |
d6e3feb5 | 2124 | unsigned int packet_size; |
9e3ddd45 TP |
2125 | struct dp_packet *packet; |
2126 | ||
2127 | stats->rx_packets += count; | |
9509913a | 2128 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
2129 | for (i = 0; i < count; i++) { |
2130 | packet = packets[i]; | |
d6e3feb5 | 2131 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 2132 | |
d6e3feb5 | 2133 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
2134 | /* This only protects the following multicast counting from |
2135 | * too short packets, but it does not stop the packet from | |
2136 | * further processing. */ | |
2137 | stats->rx_errors++; | |
2138 | stats->rx_length_errors++; | |
2139 | continue; | |
2140 | } | |
2141 | ||
d6e3feb5 | 2142 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
2143 | ||
9e3ddd45 TP |
2144 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
2145 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
2146 | stats->multicast++; | |
2147 | } | |
2148 | ||
d6e3feb5 | 2149 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
2150 | } |
2151 | } | |
2152 | ||
58397e6c KT |
2153 | /* |
2154 | * The receive path for the vhost port is the TX path out from guest. | |
2155 | */ | |
2156 | static int | |
d46285a2 | 2157 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
8492adc2 | 2158 | struct dp_packet_batch *batch, int *qfill) |
58397e6c | 2159 | { |
d46285a2 | 2160 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
9509913a | 2161 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 2162 | uint16_t nb_rx = 0; |
9509913a | 2163 | uint16_t dropped = 0; |
8492adc2 | 2164 | int qid = rxq->queue_id * VIRTIO_QNUM + VIRTIO_TXQ; |
daf22bf7 | 2165 | int vid = netdev_dpdk_get_vid(dev); |
58397e6c | 2166 | |
daf22bf7 | 2167 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured |
e543851d | 2168 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
2169 | return EAGAIN; |
2170 | } | |
2171 | ||
43307ad0 | 2172 | nb_rx = rte_vhost_dequeue_burst(vid, qid, dev->dpdk_mp->mp, |
64839cf4 | 2173 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 2174 | NETDEV_MAX_BURST); |
58397e6c KT |
2175 | if (!nb_rx) { |
2176 | return EAGAIN; | |
2177 | } | |
2178 | ||
8492adc2 JS |
2179 | if (qfill) { |
2180 | if (nb_rx == NETDEV_MAX_BURST) { | |
2181 | /* The DPDK API returns a uint32_t which often has invalid bits in | |
2182 | * the upper 16-bits. Need to restrict the value to uint16_t. */ | |
2183 | *qfill = rte_vhost_rx_queue_count(vid, qid) & UINT16_MAX; | |
2184 | } else { | |
2185 | *qfill = 0; | |
2186 | } | |
2187 | } | |
2188 | ||
9509913a IS |
2189 | if (policer) { |
2190 | dropped = nb_rx; | |
64839cf4 WT |
2191 | nb_rx = ingress_policer_run(policer, |
2192 | (struct rte_mbuf **) batch->packets, | |
3e90f7d7 | 2193 | nb_rx, true); |
9509913a IS |
2194 | dropped -= nb_rx; |
2195 | } | |
2196 | ||
d46285a2 | 2197 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
2198 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
2199 | nb_rx, dropped); | |
d46285a2 | 2200 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 2201 | |
75fb9148 ZB |
2202 | batch->count = nb_rx; |
2203 | dp_packet_batch_init_packet_fields(batch); | |
2204 | ||
58397e6c KT |
2205 | return 0; |
2206 | } | |
2207 | ||
8a9562d2 | 2208 | static int |
8492adc2 JS |
2209 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch, |
2210 | int *qfill) | |
8a9562d2 | 2211 | { |
d46285a2 DDP |
2212 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
2213 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 2214 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 2215 | int nb_rx; |
9509913a | 2216 | int dropped = 0; |
8a9562d2 | 2217 | |
3b1fb077 DDP |
2218 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
2219 | return EAGAIN; | |
2220 | } | |
2221 | ||
d46285a2 | 2222 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 2223 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 2224 | NETDEV_MAX_BURST); |
8a9562d2 PS |
2225 | if (!nb_rx) { |
2226 | return EAGAIN; | |
2227 | } | |
2228 | ||
9509913a IS |
2229 | if (policer) { |
2230 | dropped = nb_rx; | |
64839cf4 | 2231 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 2232 | (struct rte_mbuf **) batch->packets, |
3e90f7d7 | 2233 | nb_rx, true); |
9509913a IS |
2234 | dropped -= nb_rx; |
2235 | } | |
2236 | ||
2237 | /* Update stats to reflect dropped packets */ | |
2238 | if (OVS_UNLIKELY(dropped)) { | |
2239 | rte_spinlock_lock(&dev->stats_lock); | |
2240 | dev->stats.rx_dropped += dropped; | |
2241 | rte_spinlock_unlock(&dev->stats_lock); | |
2242 | } | |
2243 | ||
64839cf4 | 2244 | batch->count = nb_rx; |
75fb9148 | 2245 | dp_packet_batch_init_packet_fields(batch); |
8a9562d2 | 2246 | |
8492adc2 JS |
2247 | if (qfill) { |
2248 | if (nb_rx == NETDEV_MAX_BURST) { | |
2249 | *qfill = rte_eth_rx_queue_count(rx->port_id, rxq->queue_id); | |
2250 | } else { | |
2251 | *qfill = 0; | |
2252 | } | |
2253 | } | |
2254 | ||
8a9562d2 PS |
2255 | return 0; |
2256 | } | |
2257 | ||
0bf765f7 | 2258 | static inline int |
78bd47cf | 2259 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
7d7ded7a | 2260 | int cnt, bool should_steal) |
0bf765f7 | 2261 | { |
78bd47cf | 2262 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 2263 | |
78bd47cf DDP |
2264 | if (qos_conf) { |
2265 | rte_spinlock_lock(&qos_conf->lock); | |
7d7ded7a | 2266 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt, should_steal); |
78bd47cf | 2267 | rte_spinlock_unlock(&qos_conf->lock); |
0bf765f7 IS |
2268 | } |
2269 | ||
2270 | return cnt; | |
2271 | } | |
2272 | ||
c6ec9d17 IM |
2273 | static int |
2274 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
2275 | int pkt_cnt) | |
2276 | { | |
2277 | int i = 0; | |
2278 | int cnt = 0; | |
2279 | struct rte_mbuf *pkt; | |
2280 | ||
2281 | for (i = 0; i < pkt_cnt; i++) { | |
2282 | pkt = pkts[i]; | |
2283 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
2284 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
2285 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
2286 | rte_pktmbuf_free(pkt); | |
2287 | continue; | |
2288 | } | |
2289 | ||
2290 | if (OVS_UNLIKELY(i != cnt)) { | |
2291 | pkts[cnt] = pkt; | |
2292 | } | |
2293 | cnt++; | |
2294 | } | |
2295 | ||
2296 | return cnt; | |
2297 | } | |
2298 | ||
9e3ddd45 TP |
2299 | static inline void |
2300 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
2301 | struct dp_packet **packets, | |
2302 | int attempted, | |
2303 | int dropped) | |
2304 | { | |
2305 | int i; | |
2306 | int sent = attempted - dropped; | |
2307 | ||
2308 | stats->tx_packets += sent; | |
2309 | stats->tx_dropped += dropped; | |
2310 | ||
2311 | for (i = 0; i < sent; i++) { | |
2312 | stats->tx_bytes += dp_packet_size(packets[i]); | |
2313 | } | |
2314 | } | |
2315 | ||
58397e6c | 2316 | static void |
4573fbd3 | 2317 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 2318 | struct dp_packet **pkts, int cnt) |
58397e6c | 2319 | { |
d46285a2 | 2320 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
2321 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
2322 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 2323 | unsigned int dropped = 0; |
dd52de45 | 2324 | int i, retries = 0; |
daf22bf7 | 2325 | int vid = netdev_dpdk_get_vid(dev); |
58397e6c | 2326 | |
81acebda | 2327 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 2328 | |
daf22bf7 | 2329 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured || qid < 0 |
e543851d | 2330 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
2331 | rte_spinlock_lock(&dev->stats_lock); |
2332 | dev->stats.tx_dropped+= cnt; | |
2333 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 2334 | goto out; |
58397e6c KT |
2335 | } |
2336 | ||
d46285a2 | 2337 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 2338 | |
c6ec9d17 | 2339 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 2340 | /* Check has QoS has been configured for the netdev */ |
3e90f7d7 | 2341 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt, true); |
c6ec9d17 | 2342 | dropped = total_pkts - cnt; |
0bf765f7 | 2343 | |
95e9881f | 2344 | do { |
4573fbd3 | 2345 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
2346 | unsigned int tx_pkts; |
2347 | ||
daf22bf7 | 2348 | tx_pkts = rte_vhost_enqueue_burst(vid, vhost_qid, cur_pkts, cnt); |
95e9881f KT |
2349 | if (OVS_LIKELY(tx_pkts)) { |
2350 | /* Packets have been sent.*/ | |
2351 | cnt -= tx_pkts; | |
31871ee3 | 2352 | /* Prepare for possible retry.*/ |
95e9881f KT |
2353 | cur_pkts = &cur_pkts[tx_pkts]; |
2354 | } else { | |
31871ee3 KT |
2355 | /* No packets sent - do not retry.*/ |
2356 | break; | |
95e9881f | 2357 | } |
c6ec9d17 | 2358 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 2359 | |
d46285a2 | 2360 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 2361 | |
d46285a2 | 2362 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 2363 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 2364 | cnt + dropped); |
d46285a2 | 2365 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
2366 | |
2367 | out: | |
c6ec9d17 | 2368 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 2369 | dp_packet_delete(pkts[i]); |
58397e6c KT |
2370 | } |
2371 | } | |
2372 | ||
8a9562d2 PS |
2373 | /* Tx function. Transmit packets indefinitely */ |
2374 | static void | |
64839cf4 | 2375 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 2376 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 2377 | { |
8a14bd7b | 2378 | const size_t batch_cnt = dp_packet_batch_size(batch); |
bce01e3a | 2379 | #if !defined(__CHECKER__) && !defined(_WIN32) |
8a14bd7b | 2380 | const size_t PKT_ARRAY_SIZE = batch_cnt; |
bce01e3a EJ |
2381 | #else |
2382 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 2383 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 2384 | #endif |
8a9562d2 | 2385 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 2386 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
8a14bd7b | 2387 | uint32_t cnt = batch_cnt; |
3e90f7d7 GZ |
2388 | uint32_t dropped = 0; |
2389 | ||
2390 | if (dev->type != DPDK_DEV_VHOST) { | |
2391 | /* Check if QoS has been configured for this netdev. */ | |
2392 | cnt = netdev_dpdk_qos_run(dev, (struct rte_mbuf **) batch->packets, | |
8a14bd7b BB |
2393 | batch_cnt, false); |
2394 | dropped += batch_cnt - cnt; | |
3e90f7d7 | 2395 | } |
8a9562d2 | 2396 | |
3e90f7d7 GZ |
2397 | uint32_t txcnt = 0; |
2398 | ||
2399 | for (uint32_t i = 0; i < cnt; i++) { | |
8a14bd7b BB |
2400 | struct dp_packet *packet = batch->packets[i]; |
2401 | uint32_t size = dp_packet_size(packet); | |
95fb793a | 2402 | |
f98d7864 | 2403 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
3e90f7d7 GZ |
2404 | VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d", |
2405 | size, dev->max_packet_len); | |
f4fd623c | 2406 | |
175cf4de | 2407 | dropped++; |
f4fd623c DDP |
2408 | continue; |
2409 | } | |
8a9562d2 | 2410 | |
43307ad0 | 2411 | pkts[txcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a14bd7b | 2412 | if (OVS_UNLIKELY(!pkts[txcnt])) { |
3e90f7d7 | 2413 | dropped += cnt - i; |
175cf4de | 2414 | break; |
f4fd623c DDP |
2415 | } |
2416 | ||
2417 | /* We have to do a copy for now */ | |
3e90f7d7 | 2418 | memcpy(rte_pktmbuf_mtod(pkts[txcnt], void *), |
8a14bd7b BB |
2419 | dp_packet_data(packet), size); |
2420 | dp_packet_set_size((struct dp_packet *)pkts[txcnt], size); | |
f4fd623c | 2421 | |
3e90f7d7 | 2422 | txcnt++; |
f4fd623c | 2423 | } |
8a9562d2 | 2424 | |
3e90f7d7 GZ |
2425 | if (OVS_LIKELY(txcnt)) { |
2426 | if (dev->type == DPDK_DEV_VHOST) { | |
2427 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, | |
2428 | txcnt); | |
2429 | } else { | |
2430 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, txcnt); | |
2431 | } | |
58397e6c | 2432 | } |
db73f716 | 2433 | |
0bf765f7 IS |
2434 | if (OVS_UNLIKELY(dropped)) { |
2435 | rte_spinlock_lock(&dev->stats_lock); | |
2436 | dev->stats.tx_dropped += dropped; | |
2437 | rte_spinlock_unlock(&dev->stats_lock); | |
2438 | } | |
8a9562d2 PS |
2439 | } |
2440 | ||
58397e6c | 2441 | static int |
64839cf4 WT |
2442 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
2443 | struct dp_packet_batch *batch, | |
b30896c9 | 2444 | bool concurrent_txq OVS_UNUSED) |
58397e6c | 2445 | { |
58397e6c | 2446 | |
b30896c9 | 2447 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 | 2448 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2449 | dp_packet_delete_batch(batch, true); |
58397e6c | 2450 | } else { |
dd52de45 | 2451 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
2452 | } |
2453 | return 0; | |
2454 | } | |
2455 | ||
7251515e DV |
2456 | static inline void |
2457 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
b30896c9 | 2458 | struct dp_packet_batch *batch, |
324c8374 | 2459 | bool concurrent_txq) |
8a9562d2 | 2460 | { |
3b1fb077 | 2461 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
b30896c9 | 2462 | dp_packet_delete_batch(batch, true); |
3b1fb077 DDP |
2463 | return; |
2464 | } | |
2465 | ||
324c8374 | 2466 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 2467 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
2468 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
2469 | } | |
2470 | ||
b30896c9 | 2471 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
2472 | struct netdev *netdev = &dev->up; |
2473 | ||
64839cf4 | 2474 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2475 | dp_packet_delete_batch(batch, true); |
8a9562d2 | 2476 | } else { |
fd57eeba BB |
2477 | int tx_cnt, dropped; |
2478 | int batch_cnt = dp_packet_batch_size(batch); | |
2391135c | 2479 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 2480 | |
fd57eeba BB |
2481 | tx_cnt = netdev_dpdk_filter_packet_len(dev, pkts, batch_cnt); |
2482 | tx_cnt = netdev_dpdk_qos_run(dev, pkts, tx_cnt, true); | |
2483 | dropped = batch_cnt - tx_cnt; | |
1b99bb05 | 2484 | |
fd57eeba | 2485 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, tx_cnt); |
8a9562d2 | 2486 | |
f4fd623c | 2487 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 2488 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 2489 | dev->stats.tx_dropped += dropped; |
45d947c4 | 2490 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 2491 | } |
8a9562d2 | 2492 | } |
a0cb2d66 | 2493 | |
324c8374 | 2494 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
2495 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
2496 | } | |
7251515e DV |
2497 | } |
2498 | ||
2499 | static int | |
2500 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
b30896c9 | 2501 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e DV |
2502 | { |
2503 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 2504 | |
b30896c9 | 2505 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e | 2506 | return 0; |
8a9562d2 PS |
2507 | } |
2508 | ||
2509 | static int | |
74ff3298 | 2510 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
2511 | { |
2512 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2513 | ||
2514 | ovs_mutex_lock(&dev->mutex); | |
2515 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 2516 | dev->hwaddr = mac; |
045c0d1a | 2517 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
2518 | } |
2519 | ovs_mutex_unlock(&dev->mutex); | |
2520 | ||
2521 | return 0; | |
2522 | } | |
2523 | ||
2524 | static int | |
74ff3298 | 2525 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
2526 | { |
2527 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2528 | ||
2529 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 2530 | *mac = dev->hwaddr; |
8a9562d2 PS |
2531 | ovs_mutex_unlock(&dev->mutex); |
2532 | ||
2533 | return 0; | |
2534 | } | |
2535 | ||
2536 | static int | |
2537 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
2538 | { | |
2539 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2540 | ||
2541 | ovs_mutex_lock(&dev->mutex); | |
2542 | *mtup = dev->mtu; | |
2543 | ovs_mutex_unlock(&dev->mutex); | |
2544 | ||
2545 | return 0; | |
2546 | } | |
2547 | ||
0072e931 MK |
2548 | static int |
2549 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
2550 | { | |
2551 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2552 | ||
f6f50552 IS |
2553 | /* XXX: Ensure that the overall frame length of the requested MTU does not |
2554 | * surpass the NETDEV_DPDK_MAX_PKT_LEN. DPDK device drivers differ in how | |
2555 | * the L2 frame length is calculated for a given MTU when | |
2556 | * rte_eth_dev_set_mtu(mtu) is called e.g. i40e driver includes 2 x vlan | |
2557 | * headers, the em driver includes 1 x vlan header, the ixgbe driver does | |
2558 | * not include vlan headers. As such we should use | |
2559 | * MTU_TO_MAX_FRAME_LEN(mtu) which includes an additional 2 x vlan headers | |
2560 | * (8 bytes) for comparison. This avoids a failure later with | |
2561 | * rte_eth_dev_set_mtu(). This approach should be used until DPDK provides | |
2562 | * a method to retrieve the upper bound MTU for a given device. | |
2563 | */ | |
2564 | if (MTU_TO_MAX_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
0072e931 MK |
2565 | || mtu < ETHER_MIN_MTU) { |
2566 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
2567 | return EINVAL; | |
2568 | } | |
2569 | ||
2570 | ovs_mutex_lock(&dev->mutex); | |
2571 | if (dev->requested_mtu != mtu) { | |
2572 | dev->requested_mtu = mtu; | |
2573 | netdev_request_reconfigure(netdev); | |
2574 | } | |
2575 | ovs_mutex_unlock(&dev->mutex); | |
2576 | ||
2577 | return 0; | |
2578 | } | |
2579 | ||
8a9562d2 | 2580 | static int |
d46285a2 | 2581 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 2582 | |
58397e6c KT |
2583 | static int |
2584 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
2585 | struct netdev_stats *stats) | |
2586 | { | |
2587 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2588 | ||
2589 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 2590 | |
45d947c4 | 2591 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c | 2592 | /* Supported Stats */ |
50986e78 | 2593 | stats->rx_packets = dev->stats.rx_packets; |
2594 | stats->tx_packets = dev->stats.tx_packets; | |
9509913a | 2595 | stats->rx_dropped = dev->stats.rx_dropped; |
50986e78 | 2596 | stats->tx_dropped = dev->stats.tx_dropped; |
9e3ddd45 TP |
2597 | stats->multicast = dev->stats.multicast; |
2598 | stats->rx_bytes = dev->stats.rx_bytes; | |
2599 | stats->tx_bytes = dev->stats.tx_bytes; | |
2600 | stats->rx_errors = dev->stats.rx_errors; | |
2601 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 2602 | |
2603 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
2604 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
2605 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
2606 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
2607 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
2608 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
2609 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
2610 | ||
45d947c4 | 2611 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 2612 | |
58397e6c KT |
2613 | ovs_mutex_unlock(&dev->mutex); |
2614 | ||
2615 | return 0; | |
2616 | } | |
2617 | ||
d6e3feb5 | 2618 | static void |
2619 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
2620 | const struct rte_eth_xstat *xstats, |
2621 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 2622 | const unsigned int size) |
2623 | { | |
d6e3feb5 | 2624 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 2625 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2626 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2627 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2628 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2629 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2630 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2631 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2632 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2633 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2634 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2635 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2636 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2637 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2638 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 2639 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2640 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2641 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2642 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2643 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2644 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2645 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2646 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2647 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2648 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2649 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2650 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2651 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2652 | stats->tx_1523_to_max_packets = xstats[i].value; |
d57f777f PS |
2653 | } else if (strcmp(XSTAT_RX_MULTICAST_PACKETS, names[i].name) == 0) { |
2654 | stats->multicast = xstats[i].value; | |
0a0f39df | 2655 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2656 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 2657 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2658 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2659 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2660 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2661 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2662 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2663 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2664 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2665 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2666 | stats->rx_jabber_errors = xstats[i].value; |
2667 | } | |
2668 | } | |
2669 | } | |
2670 | ||
8a9562d2 PS |
2671 | static int |
2672 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2673 | { | |
2674 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2675 | struct rte_eth_stats rte_stats; | |
2676 | bool gg; | |
2677 | ||
2678 | netdev_dpdk_get_carrier(netdev, &gg); | |
2679 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2680 | |
0a0f39df CL |
2681 | struct rte_eth_xstat *rte_xstats = NULL; |
2682 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2683 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2684 | |
2685 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
fa9f4eeb IM |
2686 | VLOG_ERR("Can't get ETH statistics for port: "DPDK_PORT_ID_FMT, |
2687 | dev->port_id); | |
f9256822 | 2688 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2689 | return EPROTO; |
2690 | } | |
2691 | ||
0a0f39df CL |
2692 | /* Get length of statistics */ |
2693 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2694 | if (rte_xstats_len < 0) { | |
fa9f4eeb IM |
2695 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
2696 | dev->port_id); | |
0a0f39df CL |
2697 | goto out; |
2698 | } | |
2699 | /* Reserve memory for xstats names and values */ | |
2700 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2701 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2702 | ||
2703 | /* Retreive xstats names */ | |
2704 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2705 | rte_xstats_names, | |
2706 | rte_xstats_len); | |
2707 | if (rte_xstats_new_len != rte_xstats_len) { | |
fa9f4eeb IM |
2708 | VLOG_WARN("Cannot get XSTATS names for port: "DPDK_PORT_ID_FMT, |
2709 | dev->port_id); | |
0a0f39df CL |
2710 | goto out; |
2711 | } | |
2712 | /* Retreive xstats values */ | |
2713 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2714 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2715 | rte_xstats_len); | |
2716 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2717 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2718 | rte_xstats_len); | |
d6e3feb5 | 2719 | } else { |
fa9f4eeb IM |
2720 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
2721 | dev->port_id); | |
d6e3feb5 | 2722 | } |
8a9562d2 | 2723 | |
0a0f39df CL |
2724 | out: |
2725 | free(rte_xstats); | |
2726 | free(rte_xstats_names); | |
2727 | ||
2f9dd77f PS |
2728 | stats->rx_packets = rte_stats.ipackets; |
2729 | stats->tx_packets = rte_stats.opackets; | |
2730 | stats->rx_bytes = rte_stats.ibytes; | |
2731 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2732 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2733 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2734 | |
45d947c4 | 2735 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2736 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2737 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2738 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2739 | |
2740 | /* These are the available DPDK counters for packets not received due to | |
2741 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2742 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2743 | stats->rx_missed_errors = rte_stats.imissed; |
2744 | ||
8a9562d2 PS |
2745 | ovs_mutex_unlock(&dev->mutex); |
2746 | ||
2747 | return 0; | |
2748 | } | |
2749 | ||
971f4b39 MW |
2750 | static int |
2751 | netdev_dpdk_get_custom_stats(const struct netdev *netdev, | |
2752 | struct netdev_custom_stats *custom_stats) | |
2753 | { | |
2754 | ||
2755 | uint32_t i; | |
2756 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2757 | int rte_xstats_ret; | |
2758 | ||
2759 | ovs_mutex_lock(&dev->mutex); | |
2760 | ||
2761 | if (netdev_dpdk_configure_xstats(dev)) { | |
2762 | uint64_t *values = xcalloc(dev->rte_xstats_ids_size, | |
2763 | sizeof(uint64_t)); | |
2764 | ||
2765 | rte_xstats_ret = | |
2766 | rte_eth_xstats_get_by_id(dev->port_id, dev->rte_xstats_ids, | |
2767 | values, dev->rte_xstats_ids_size); | |
2768 | ||
2769 | if (rte_xstats_ret > 0 && | |
2770 | rte_xstats_ret <= dev->rte_xstats_ids_size) { | |
2771 | ||
2772 | custom_stats->size = rte_xstats_ret; | |
2773 | custom_stats->counters = | |
2774 | (struct netdev_custom_counter *) xcalloc(rte_xstats_ret, | |
2775 | sizeof(struct netdev_custom_counter)); | |
2776 | ||
2777 | for (i = 0; i < rte_xstats_ret; i++) { | |
2778 | ovs_strlcpy(custom_stats->counters[i].name, | |
2779 | netdev_dpdk_get_xstat_name(dev, | |
2780 | dev->rte_xstats_ids[i]), | |
2781 | NETDEV_CUSTOM_STATS_NAME_SIZE); | |
2782 | custom_stats->counters[i].value = values[i]; | |
2783 | } | |
2784 | } else { | |
fa9f4eeb | 2785 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
971f4b39 MW |
2786 | dev->port_id); |
2787 | custom_stats->counters = NULL; | |
2788 | custom_stats->size = 0; | |
2789 | /* Let's clear statistics cache, so it will be | |
2790 | * reconfigured */ | |
2791 | netdev_dpdk_clear_xstats(dev); | |
2792 | } | |
526259f2 IM |
2793 | |
2794 | free(values); | |
971f4b39 MW |
2795 | } |
2796 | ||
2797 | ovs_mutex_unlock(&dev->mutex); | |
2798 | ||
2799 | return 0; | |
2800 | } | |
2801 | ||
8a9562d2 | 2802 | static int |
d46285a2 | 2803 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2804 | enum netdev_features *current, |
ca3d4f55 BX |
2805 | enum netdev_features *advertised, |
2806 | enum netdev_features *supported, | |
2807 | enum netdev_features *peer) | |
8a9562d2 | 2808 | { |
d46285a2 | 2809 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 2810 | struct rte_eth_link link; |
dfcb5b8a | 2811 | uint32_t feature = 0; |
8a9562d2 PS |
2812 | |
2813 | ovs_mutex_lock(&dev->mutex); | |
2814 | link = dev->link; | |
2815 | ovs_mutex_unlock(&dev->mutex); | |
2816 | ||
dfcb5b8a IS |
2817 | /* Match against OpenFlow defined link speed values. */ |
2818 | if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
2819 | switch (link.link_speed) { | |
2820 | case ETH_SPEED_NUM_10M: | |
2821 | feature |= NETDEV_F_10MB_FD; | |
2822 | break; | |
2823 | case ETH_SPEED_NUM_100M: | |
2824 | feature |= NETDEV_F_100MB_FD; | |
2825 | break; | |
2826 | case ETH_SPEED_NUM_1G: | |
2827 | feature |= NETDEV_F_1GB_FD; | |
2828 | break; | |
2829 | case ETH_SPEED_NUM_10G: | |
2830 | feature |= NETDEV_F_10GB_FD; | |
2831 | break; | |
2832 | case ETH_SPEED_NUM_40G: | |
2833 | feature |= NETDEV_F_40GB_FD; | |
2834 | break; | |
2835 | case ETH_SPEED_NUM_100G: | |
2836 | feature |= NETDEV_F_100GB_FD; | |
2837 | break; | |
2838 | default: | |
2839 | feature |= NETDEV_F_OTHER; | |
8a9562d2 | 2840 | } |
dfcb5b8a IS |
2841 | } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2842 | switch (link.link_speed) { | |
2843 | case ETH_SPEED_NUM_10M: | |
2844 | feature |= NETDEV_F_10MB_HD; | |
2845 | break; | |
2846 | case ETH_SPEED_NUM_100M: | |
2847 | feature |= NETDEV_F_100MB_HD; | |
2848 | break; | |
2849 | case ETH_SPEED_NUM_1G: | |
2850 | feature |= NETDEV_F_1GB_HD; | |
2851 | break; | |
2852 | default: | |
2853 | feature |= NETDEV_F_OTHER; | |
74cd69a4 | 2854 | } |
8a9562d2 PS |
2855 | } |
2856 | ||
362ca396 | 2857 | if (link.link_autoneg) { |
dfcb5b8a | 2858 | feature |= NETDEV_F_AUTONEG; |
362ca396 | 2859 | } |
2860 | ||
dfcb5b8a | 2861 | *current = feature; |
ca3d4f55 BX |
2862 | *advertised = *supported = *peer = 0; |
2863 | ||
8a9562d2 PS |
2864 | return 0; |
2865 | } | |
2866 | ||
9509913a IS |
2867 | static struct ingress_policer * |
2868 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2869 | { | |
2870 | struct ingress_policer *policer = NULL; | |
2871 | uint64_t rate_bytes; | |
2872 | uint64_t burst_bytes; | |
2873 | int err = 0; | |
2874 | ||
2875 | policer = xmalloc(sizeof *policer); | |
2876 | rte_spinlock_init(&policer->policer_lock); | |
2877 | ||
2878 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
602c8668 LR |
2879 | rate_bytes = rate * 1000ULL / 8; |
2880 | burst_bytes = burst * 1000ULL / 8; | |
9509913a IS |
2881 | |
2882 | policer->app_srtcm_params.cir = rate_bytes; | |
2883 | policer->app_srtcm_params.cbs = burst_bytes; | |
2884 | policer->app_srtcm_params.ebs = 0; | |
03f3f9c0 OM |
2885 | err = rte_meter_srtcm_profile_config(&policer->in_prof, |
2886 | &policer->app_srtcm_params); | |
2887 | if (!err) { | |
2888 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2889 | &policer->in_prof); | |
2890 | } | |
58be5c0e | 2891 | if (err) { |
9509913a | 2892 | VLOG_ERR("Could not create rte meter for ingress policer"); |
4c47ddde | 2893 | free(policer); |
9509913a IS |
2894 | return NULL; |
2895 | } | |
2896 | ||
2897 | return policer; | |
2898 | } | |
2899 | ||
2900 | static int | |
2901 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2902 | uint32_t policer_burst) | |
2903 | { | |
2904 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2905 | struct ingress_policer *policer; | |
2906 | ||
2907 | /* Force to 0 if no rate specified, | |
2908 | * default to 8000 kbits if burst is 0, | |
2909 | * else stick with user-specified value. | |
2910 | */ | |
2911 | policer_burst = (!policer_rate ? 0 | |
2912 | : !policer_burst ? 8000 | |
2913 | : policer_burst); | |
2914 | ||
2915 | ovs_mutex_lock(&dev->mutex); | |
2916 | ||
2917 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2918 | &dev->ingress_policer); | |
2919 | ||
2920 | if (dev->policer_rate == policer_rate && | |
2921 | dev->policer_burst == policer_burst) { | |
2922 | /* Assume that settings haven't changed since we last set them. */ | |
2923 | ovs_mutex_unlock(&dev->mutex); | |
2924 | return 0; | |
2925 | } | |
2926 | ||
2927 | /* Destroy any existing ingress policer for the device if one exists */ | |
2928 | if (policer) { | |
2929 | ovsrcu_postpone(free, policer); | |
2930 | } | |
2931 | ||
2932 | if (policer_rate != 0) { | |
2933 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2934 | } else { | |
2935 | policer = NULL; | |
2936 | } | |
2937 | ovsrcu_set(&dev->ingress_policer, policer); | |
2938 | dev->policer_rate = policer_rate; | |
2939 | dev->policer_burst = policer_burst; | |
2940 | ovs_mutex_unlock(&dev->mutex); | |
2941 | ||
2942 | return 0; | |
2943 | } | |
2944 | ||
8a9562d2 PS |
2945 | static int |
2946 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2947 | { | |
2948 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
2949 | |
2950 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
2951 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
2952 | * postive integer to meet RFC 2863 recommendations. | |
2953 | */ | |
2954 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
2955 | ovs_mutex_unlock(&dev->mutex); |
2956 | ||
2957 | return ifindex; | |
2958 | } | |
2959 | ||
2960 | static int | |
d46285a2 | 2961 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2962 | { |
d46285a2 | 2963 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2964 | |
2965 | ovs_mutex_lock(&dev->mutex); | |
2966 | check_link_status(dev); | |
2967 | *carrier = dev->link.link_status; | |
58397e6c KT |
2968 | |
2969 | ovs_mutex_unlock(&dev->mutex); | |
2970 | ||
2971 | return 0; | |
2972 | } | |
2973 | ||
2974 | static int | |
d46285a2 | 2975 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2976 | { |
d46285a2 | 2977 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2978 | |
2979 | ovs_mutex_lock(&dev->mutex); | |
2980 | ||
0a0f39df | 2981 | if (is_vhost_running(dev)) { |
58397e6c KT |
2982 | *carrier = 1; |
2983 | } else { | |
2984 | *carrier = 0; | |
2985 | } | |
2986 | ||
8a9562d2 PS |
2987 | ovs_mutex_unlock(&dev->mutex); |
2988 | ||
2989 | return 0; | |
2990 | } | |
2991 | ||
2992 | static long long int | |
d46285a2 | 2993 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2994 | { |
d46285a2 | 2995 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2996 | long long int carrier_resets; |
2997 | ||
2998 | ovs_mutex_lock(&dev->mutex); | |
2999 | carrier_resets = dev->link_reset_cnt; | |
3000 | ovs_mutex_unlock(&dev->mutex); | |
3001 | ||
3002 | return carrier_resets; | |
3003 | } | |
3004 | ||
3005 | static int | |
d46285a2 | 3006 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
3007 | long long int interval OVS_UNUSED) |
3008 | { | |
ee32150e | 3009 | return EOPNOTSUPP; |
8a9562d2 PS |
3010 | } |
3011 | ||
3012 | static int | |
3013 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
3014 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
3015 | enum netdev_flags *old_flagsp) |
3016 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 3017 | { |
8a9562d2 PS |
3018 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
3019 | return EINVAL; | |
3020 | } | |
3021 | ||
3022 | *old_flagsp = dev->flags; | |
3023 | dev->flags |= on; | |
3024 | dev->flags &= ~off; | |
3025 | ||
3026 | if (dev->flags == *old_flagsp) { | |
3027 | return 0; | |
3028 | } | |
3029 | ||
58397e6c | 3030 | if (dev->type == DPDK_DEV_ETH) { |
2d37de73 EC |
3031 | |
3032 | if ((dev->flags ^ *old_flagsp) & NETDEV_UP) { | |
3033 | int err; | |
3034 | ||
3035 | if (dev->flags & NETDEV_UP) { | |
3036 | err = rte_eth_dev_set_link_up(dev->port_id); | |
3037 | } else { | |
3038 | err = rte_eth_dev_set_link_down(dev->port_id); | |
3039 | } | |
3040 | if (err == -ENOTSUP) { | |
3041 | VLOG_INFO("Interface %s does not support link state " | |
3042 | "configuration", netdev_get_name(&dev->up)); | |
3043 | } else if (err < 0) { | |
3044 | VLOG_ERR("Interface %s link change error: %s", | |
3045 | netdev_get_name(&dev->up), rte_strerror(-err)); | |
3046 | dev->flags = *old_flagsp; | |
3047 | return -err; | |
3048 | } | |
3049 | } | |
3050 | ||
58397e6c KT |
3051 | if (dev->flags & NETDEV_PROMISC) { |
3052 | rte_eth_promiscuous_enable(dev->port_id); | |
3053 | } | |
8a9562d2 | 3054 | |
314fb5ad | 3055 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
3056 | } else { |
3057 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
3058 | * running then change netdev's change_seq to trigger link state | |
3059 | * update. */ | |
e543851d ZB |
3060 | |
3061 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 3062 | && is_vhost_running(dev)) { |
e543851d ZB |
3063 | netdev_change_seq_changed(&dev->up); |
3064 | ||
3065 | /* Clear statistics if device is getting up. */ | |
3066 | if (NETDEV_UP & on) { | |
3067 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 3068 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
3069 | rte_spinlock_unlock(&dev->stats_lock); |
3070 | } | |
3071 | } | |
8a9562d2 PS |
3072 | } |
3073 | ||
3074 | return 0; | |
3075 | } | |
3076 | ||
3077 | static int | |
d46285a2 | 3078 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
3079 | enum netdev_flags off, enum netdev_flags on, |
3080 | enum netdev_flags *old_flagsp) | |
3081 | { | |
d46285a2 | 3082 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
3083 | int error; |
3084 | ||
d46285a2 DDP |
3085 | ovs_mutex_lock(&dev->mutex); |
3086 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
3087 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
3088 | |
3089 | return error; | |
3090 | } | |
3091 | ||
b2e8b12f FL |
3092 | static int |
3093 | netdev_dpdk_vhost_user_get_status(const struct netdev *netdev, | |
3094 | struct smap *args) | |
3095 | { | |
3096 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3097 | ||
3098 | ovs_mutex_lock(&dev->mutex); | |
3099 | ||
3100 | bool client_mode = dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT; | |
3101 | smap_add_format(args, "mode", "%s", client_mode ? "client" : "server"); | |
3102 | ||
3103 | int vid = netdev_dpdk_get_vid(dev); | |
3104 | if (vid < 0) { | |
3105 | smap_add_format(args, "status", "disconnected"); | |
3106 | ovs_mutex_unlock(&dev->mutex); | |
3107 | return 0; | |
3108 | } else { | |
3109 | smap_add_format(args, "status", "connected"); | |
3110 | } | |
3111 | ||
3112 | char socket_name[PATH_MAX]; | |
3113 | if (!rte_vhost_get_ifname(vid, socket_name, PATH_MAX)) { | |
3114 | smap_add_format(args, "socket", "%s", socket_name); | |
3115 | } | |
3116 | ||
3117 | uint64_t features; | |
3118 | if (!rte_vhost_get_negotiated_features(vid, &features)) { | |
3119 | smap_add_format(args, "features", "0x%016"PRIx64, features); | |
3120 | } | |
3121 | ||
3122 | uint16_t mtu; | |
3123 | if (!rte_vhost_get_mtu(vid, &mtu)) { | |
3124 | smap_add_format(args, "mtu", "%d", mtu); | |
3125 | } | |
3126 | ||
3127 | int numa = rte_vhost_get_numa_node(vid); | |
3128 | if (numa >= 0) { | |
3129 | smap_add_format(args, "numa", "%d", numa); | |
3130 | } | |
3131 | ||
3132 | uint16_t vring_num = rte_vhost_get_vring_num(vid); | |
3133 | if (vring_num) { | |
3134 | smap_add_format(args, "num_of_vrings", "%d", vring_num); | |
3135 | } | |
3136 | ||
3137 | for (int i = 0; i < vring_num; i++) { | |
3138 | struct rte_vhost_vring vring; | |
b2e8b12f FL |
3139 | |
3140 | rte_vhost_get_vhost_vring(vid, i, &vring); | |
b9a3183d AC |
3141 | smap_add_nocopy(args, xasprintf("vring_%d_size", i), |
3142 | xasprintf("%d", vring.size)); | |
b2e8b12f FL |
3143 | } |
3144 | ||
3145 | ovs_mutex_unlock(&dev->mutex); | |
3146 | return 0; | |
3147 | } | |
3148 | ||
31154f95 IS |
3149 | /* |
3150 | * Convert a given uint32_t link speed defined in DPDK to a string | |
3151 | * equivalent. | |
3152 | */ | |
3153 | static const char * | |
3154 | netdev_dpdk_link_speed_to_str__(uint32_t link_speed) | |
3155 | { | |
3156 | switch (link_speed) { | |
3157 | case ETH_SPEED_NUM_10M: return "10Mbps"; | |
3158 | case ETH_SPEED_NUM_100M: return "100Mbps"; | |
3159 | case ETH_SPEED_NUM_1G: return "1Gbps"; | |
3160 | case ETH_SPEED_NUM_2_5G: return "2.5Gbps"; | |
3161 | case ETH_SPEED_NUM_5G: return "5Gbps"; | |
3162 | case ETH_SPEED_NUM_10G: return "10Gbps"; | |
3163 | case ETH_SPEED_NUM_20G: return "20Gbps"; | |
3164 | case ETH_SPEED_NUM_25G: return "25Gbps"; | |
3165 | case ETH_SPEED_NUM_40G: return "40Gbps"; | |
3166 | case ETH_SPEED_NUM_50G: return "50Gbps"; | |
3167 | case ETH_SPEED_NUM_56G: return "56Gbps"; | |
3168 | case ETH_SPEED_NUM_100G: return "100Gbps"; | |
3169 | default: return "Not Defined"; | |
3170 | } | |
3171 | } | |
3172 | ||
8a9562d2 | 3173 | static int |
d46285a2 | 3174 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 3175 | { |
d46285a2 | 3176 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 3177 | struct rte_eth_dev_info dev_info; |
31154f95 | 3178 | uint32_t link_speed; |
8a9562d2 | 3179 | |
7cd1261d | 3180 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 3181 | return ENODEV; |
7cd1261d | 3182 | } |
8a9562d2 | 3183 | |
03f3f9c0 | 3184 | ovs_mutex_lock(&dpdk_mutex); |
8a9562d2 PS |
3185 | ovs_mutex_lock(&dev->mutex); |
3186 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
31154f95 | 3187 | link_speed = dev->link.link_speed; |
8a9562d2 | 3188 | ovs_mutex_unlock(&dev->mutex); |
03f3f9c0 OM |
3189 | const struct rte_bus *bus; |
3190 | const struct rte_pci_device *pci_dev; | |
3191 | uint16_t vendor_id = PCI_ANY_ID; | |
3192 | uint16_t device_id = PCI_ANY_ID; | |
3193 | bus = rte_bus_find_by_device(dev_info.device); | |
3194 | if (bus && !strcmp(bus->name, "pci")) { | |
3195 | pci_dev = RTE_DEV_TO_PCI(dev_info.device); | |
3196 | if (pci_dev) { | |
3197 | vendor_id = pci_dev->id.vendor_id; | |
3198 | device_id = pci_dev->id.device_id; | |
3199 | } | |
3200 | } | |
3201 | ovs_mutex_unlock(&dpdk_mutex); | |
8a9562d2 | 3202 | |
fa9f4eeb | 3203 | smap_add_format(args, "port_no", DPDK_PORT_ID_FMT, dev->port_id); |
58be5c0e MK |
3204 | smap_add_format(args, "numa_id", "%d", |
3205 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
3206 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
3207 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 3208 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
3209 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
3210 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
3211 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
3212 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
3213 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
3214 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
3215 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
3216 | ||
3eb8d4fa MW |
3217 | /* Querying the DPDK library for iftype may be done in future, pending |
3218 | * support; cf. RFC 3635 Section 3.2.4. */ | |
3219 | enum { IF_TYPE_ETHERNETCSMACD = 6 }; | |
3220 | ||
3221 | smap_add_format(args, "if_type", "%"PRIu32, IF_TYPE_ETHERNETCSMACD); | |
3222 | smap_add_format(args, "if_descr", "%s %s", rte_version(), | |
3223 | dev_info.driver_name); | |
03f3f9c0 OM |
3224 | smap_add_format(args, "pci-vendor_id", "0x%x", vendor_id); |
3225 | smap_add_format(args, "pci-device_id", "0x%x", device_id); | |
8a9562d2 | 3226 | |
31154f95 IS |
3227 | /* Not all link speeds are defined in the OpenFlow specs e.g. 25 Gbps. |
3228 | * In that case the speed will not be reported as part of the usual | |
3229 | * call to get_features(). Get the link speed of the device and add it | |
3230 | * to the device status in an easy to read string format. | |
3231 | */ | |
3232 | smap_add(args, "link_speed", | |
3233 | netdev_dpdk_link_speed_to_str__(link_speed)); | |
3234 | ||
8a9562d2 PS |
3235 | return 0; |
3236 | } | |
3237 | ||
3238 | static void | |
3239 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
3240 | OVS_REQUIRES(dev->mutex) | |
3241 | { | |
3242 | enum netdev_flags old_flags; | |
3243 | ||
3244 | if (admin_state) { | |
3245 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
3246 | } else { | |
3247 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
3248 | } | |
3249 | } | |
3250 | ||
3251 | static void | |
3252 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
3253 | const char *argv[], void *aux OVS_UNUSED) | |
3254 | { | |
3255 | bool up; | |
3256 | ||
3257 | if (!strcasecmp(argv[argc - 1], "up")) { | |
3258 | up = true; | |
3259 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
3260 | up = false; | |
3261 | } else { | |
3262 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
3263 | return; | |
3264 | } | |
3265 | ||
3266 | if (argc > 2) { | |
3267 | struct netdev *netdev = netdev_from_name(argv[1]); | |
3d0d5ab1 | 3268 | |
8a9562d2 | 3269 | if (netdev && is_dpdk_class(netdev->netdev_class)) { |
3d0d5ab1 | 3270 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 3271 | |
3d0d5ab1 IM |
3272 | ovs_mutex_lock(&dev->mutex); |
3273 | netdev_dpdk_set_admin_state__(dev, up); | |
3274 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
3275 | |
3276 | netdev_close(netdev); | |
3277 | } else { | |
3278 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
3279 | netdev_close(netdev); | |
3280 | return; | |
3281 | } | |
3282 | } else { | |
3d0d5ab1 | 3283 | struct netdev_dpdk *dev; |
8a9562d2 PS |
3284 | |
3285 | ovs_mutex_lock(&dpdk_mutex); | |
3d0d5ab1 IM |
3286 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
3287 | ovs_mutex_lock(&dev->mutex); | |
3288 | netdev_dpdk_set_admin_state__(dev, up); | |
3289 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
3290 | } |
3291 | ovs_mutex_unlock(&dpdk_mutex); | |
3292 | } | |
3293 | unixctl_command_reply(conn, "OK"); | |
3294 | } | |
3295 | ||
0ee821c2 DB |
3296 | static void |
3297 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
3298 | const char *argv[], void *aux OVS_UNUSED) | |
3299 | { | |
0ee821c2 | 3300 | char *response; |
7ee94cba | 3301 | dpdk_port_t port_id; |
0ee821c2 | 3302 | struct netdev_dpdk *dev; |
40e940e4 OM |
3303 | struct rte_device *rte_dev; |
3304 | struct ds used_interfaces = DS_EMPTY_INITIALIZER; | |
3305 | bool used = false; | |
0ee821c2 DB |
3306 | |
3307 | ovs_mutex_lock(&dpdk_mutex); | |
3308 | ||
40e940e4 OM |
3309 | port_id = netdev_dpdk_get_port_by_devargs(argv[1]); |
3310 | if (!rte_eth_dev_is_valid_port(port_id)) { | |
0ee821c2 DB |
3311 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); |
3312 | goto error; | |
3313 | } | |
3314 | ||
40e940e4 OM |
3315 | rte_dev = rte_eth_devices[port_id].device; |
3316 | ds_put_format(&used_interfaces, | |
3317 | "Device '%s' is being used by the following interfaces:", | |
3318 | argv[1]); | |
3319 | ||
3320 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
3321 | /* FIXME: avoid direct access to DPDK array rte_eth_devices. */ | |
3322 | if (rte_eth_devices[dev->port_id].device == rte_dev | |
3323 | && rte_eth_devices[dev->port_id].state != RTE_ETH_DEV_UNUSED) { | |
3324 | used = true; | |
3325 | ds_put_format(&used_interfaces, " %s", | |
3326 | netdev_get_name(&dev->up)); | |
3327 | } | |
3328 | } | |
3329 | ||
3330 | if (used) { | |
3331 | ds_put_cstr(&used_interfaces, ". Remove them before detaching."); | |
3332 | response = ds_steal_cstr(&used_interfaces); | |
3333 | ds_destroy(&used_interfaces); | |
0ee821c2 DB |
3334 | goto error; |
3335 | } | |
40e940e4 | 3336 | ds_destroy(&used_interfaces); |
0ee821c2 DB |
3337 | |
3338 | rte_eth_dev_close(port_id); | |
40e940e4 | 3339 | if (rte_dev_remove(rte_dev) < 0) { |
0ee821c2 DB |
3340 | response = xasprintf("Device '%s' can not be detached", argv[1]); |
3341 | goto error; | |
3342 | } | |
3343 | ||
40e940e4 OM |
3344 | response = xasprintf("All devices shared with device '%s' " |
3345 | "have been detached", argv[1]); | |
0ee821c2 DB |
3346 | |
3347 | ovs_mutex_unlock(&dpdk_mutex); | |
3348 | unixctl_command_reply(conn, response); | |
3349 | free(response); | |
3350 | return; | |
3351 | ||
3352 | error: | |
3353 | ovs_mutex_unlock(&dpdk_mutex); | |
3354 | unixctl_command_reply_error(conn, response); | |
3355 | free(response); | |
3356 | } | |
3357 | ||
be481733 IM |
3358 | static void |
3359 | netdev_dpdk_get_mempool_info(struct unixctl_conn *conn, | |
3360 | int argc, const char *argv[], | |
3361 | void *aux OVS_UNUSED) | |
3362 | { | |
3363 | size_t size; | |
3364 | FILE *stream; | |
3365 | char *response = NULL; | |
3366 | struct netdev *netdev = NULL; | |
3367 | ||
3368 | if (argc == 2) { | |
3369 | netdev = netdev_from_name(argv[1]); | |
3370 | if (!netdev || !is_dpdk_class(netdev->netdev_class)) { | |
3371 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
3372 | goto out; | |
3373 | } | |
3374 | } | |
3375 | ||
3376 | stream = open_memstream(&response, &size); | |
3377 | if (!stream) { | |
3378 | response = xasprintf("Unable to open memstream: %s.", | |
3379 | ovs_strerror(errno)); | |
3380 | unixctl_command_reply_error(conn, response); | |
3381 | goto out; | |
3382 | } | |
3383 | ||
3384 | if (netdev) { | |
3385 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3386 | ||
3387 | ovs_mutex_lock(&dev->mutex); | |
3388 | ovs_mutex_lock(&dpdk_mp_mutex); | |
3389 | ||
43307ad0 | 3390 | rte_mempool_dump(stream, dev->dpdk_mp->mp); |
be481733 IM |
3391 | |
3392 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
3393 | ovs_mutex_unlock(&dev->mutex); | |
3394 | } else { | |
3395 | ovs_mutex_lock(&dpdk_mp_mutex); | |
3396 | rte_mempool_list_dump(stream); | |
3397 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
3398 | } | |
3399 | ||
3400 | fclose(stream); | |
3401 | ||
3402 | unixctl_command_reply(conn, response); | |
3403 | out: | |
3404 | free(response); | |
3405 | netdev_close(netdev); | |
3406 | } | |
3407 | ||
58397e6c KT |
3408 | /* |
3409 | * Set virtqueue flags so that we do not receive interrupts. | |
3410 | */ | |
3411 | static void | |
0a0f39df | 3412 | set_irq_status(int vid) |
58397e6c | 3413 | { |
4573fbd3 | 3414 | uint32_t i; |
4573fbd3 | 3415 | |
f3e7ec25 MW |
3416 | for (i = 0; i < rte_vhost_get_vring_num(vid); i++) { |
3417 | rte_vhost_enable_guest_notification(vid, i, 0); | |
4573fbd3 FL |
3418 | } |
3419 | } | |
3420 | ||
585a5bea IM |
3421 | /* |
3422 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 3423 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
3424 | */ |
3425 | static void | |
d46285a2 DDP |
3426 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
3427 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
3428 | { |
3429 | int *enabled_queues, n_enabled = 0; | |
81acebda | 3430 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 3431 | |
eff23640 | 3432 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
3433 | |
3434 | for (i = 0; i < total_txqs; i++) { | |
3435 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 3436 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
3437 | enabled_queues[n_enabled++] = i; |
3438 | } | |
3439 | } | |
3440 | ||
3441 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 3442 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
3443 | n_enabled = 1; |
3444 | } | |
3445 | ||
3446 | k = 0; | |
3447 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
3448 | if (dev->tx_q[i].map != i) { |
3449 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
3450 | k = (k + 1) % n_enabled; |
3451 | } | |
3452 | } | |
3453 | ||
2d24d165 | 3454 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 3455 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 3456 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
3457 | } |
3458 | ||
eff23640 | 3459 | free(enabled_queues); |
585a5bea | 3460 | } |
4573fbd3 | 3461 | |
58397e6c KT |
3462 | /* |
3463 | * A new virtio-net device is added to a vhost port. | |
3464 | */ | |
3465 | static int | |
0a0f39df | 3466 | new_device(int vid) |
58397e6c | 3467 | { |
d46285a2 | 3468 | struct netdev_dpdk *dev; |
58397e6c | 3469 | bool exists = false; |
db8f13b0 | 3470 | int newnode = 0; |
0a0f39df CL |
3471 | char ifname[IF_NAME_SZ]; |
3472 | ||
58be5c0e | 3473 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
3474 | |
3475 | ovs_mutex_lock(&dpdk_mutex); | |
3476 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 3477 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 3478 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 3479 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
f3e7ec25 | 3480 | uint32_t qp_num = rte_vhost_get_vring_num(vid)/VIRTIO_QNUM; |
db8f13b0 CL |
3481 | |
3482 | /* Get NUMA information */ | |
0a0f39df CL |
3483 | newnode = rte_vhost_get_numa_node(vid); |
3484 | if (newnode == -1) { | |
5b9bf9e0 | 3485 | #ifdef VHOST_NUMA |
db8f13b0 | 3486 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 3487 | ifname); |
5b9bf9e0 | 3488 | #endif |
db8f13b0 | 3489 | newnode = dev->socket_id; |
db8f13b0 CL |
3490 | } |
3491 | ||
7f5f2bd0 IM |
3492 | if (dev->requested_n_txq != qp_num |
3493 | || dev->requested_n_rxq != qp_num | |
3494 | || dev->requested_socket_id != newnode) { | |
3495 | dev->requested_socket_id = newnode; | |
3496 | dev->requested_n_rxq = qp_num; | |
3497 | dev->requested_n_txq = qp_num; | |
3498 | netdev_request_reconfigure(&dev->up); | |
3499 | } else { | |
3500 | /* Reconfiguration not required. */ | |
3501 | dev->vhost_reconfigured = true; | |
3502 | } | |
81acebda | 3503 | |
0a0f39df | 3504 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
3505 | exists = true; |
3506 | ||
58397e6c | 3507 | /* Disable notifications. */ |
0a0f39df | 3508 | set_irq_status(vid); |
e543851d | 3509 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 3510 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
3511 | break; |
3512 | } | |
c1ff66ac | 3513 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
3514 | } |
3515 | ovs_mutex_unlock(&dpdk_mutex); | |
3516 | ||
3517 | if (!exists) { | |
0a0f39df | 3518 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
3519 | |
3520 | return -1; | |
3521 | } | |
3522 | ||
0a0f39df CL |
3523 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
3524 | ifname, newnode); | |
3525 | ||
58397e6c KT |
3526 | return 0; |
3527 | } | |
3528 | ||
f3ea2ad2 IM |
3529 | /* Clears mapping for all available queues of vhost interface. */ |
3530 | static void | |
3531 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
3532 | OVS_REQUIRES(dev->mutex) | |
3533 | { | |
3534 | int i; | |
3535 | ||
81acebda | 3536 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
3537 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
3538 | } | |
3539 | } | |
3540 | ||
58397e6c KT |
3541 | /* |
3542 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
3543 | * flag to stop any more packets from being sent or received to/from a VM and | |
3544 | * ensure all currently queued packets have been sent/received before removing | |
3545 | * the device. | |
3546 | */ | |
3547 | static void | |
0a0f39df | 3548 | destroy_device(int vid) |
58397e6c | 3549 | { |
d46285a2 | 3550 | struct netdev_dpdk *dev; |
afee281f | 3551 | bool exists = false; |
0a0f39df CL |
3552 | char ifname[IF_NAME_SZ]; |
3553 | ||
58be5c0e | 3554 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
3555 | |
3556 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 3557 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 3558 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 3559 | |
d46285a2 | 3560 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
3561 | dev->vhost_reconfigured = false; |
3562 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 3563 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 3564 | |
e543851d | 3565 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 3566 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 3567 | exists = true; |
afee281f | 3568 | break; |
58397e6c KT |
3569 | } |
3570 | } | |
afee281f | 3571 | |
58397e6c KT |
3572 | ovs_mutex_unlock(&dpdk_mutex); |
3573 | ||
0a0f39df | 3574 | if (exists) { |
afee281f KT |
3575 | /* |
3576 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
3577 | * to NULL, before returning. | |
3578 | */ | |
3579 | ovsrcu_synchronize(); | |
3580 | /* | |
3581 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
3582 | * put thread back into quiescent state before returning. | |
3583 | */ | |
3584 | ovsrcu_quiesce_start(); | |
0a0f39df | 3585 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 3586 | } else { |
0a0f39df | 3587 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 3588 | } |
58397e6c KT |
3589 | } |
3590 | ||
585a5bea | 3591 | static int |
0a0f39df | 3592 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 3593 | { |
d46285a2 | 3594 | struct netdev_dpdk *dev; |
585a5bea IM |
3595 | bool exists = false; |
3596 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
3597 | char ifname[IF_NAME_SZ]; |
3598 | ||
58be5c0e | 3599 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
3600 | |
3601 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
3602 | return 0; | |
3603 | } | |
3604 | ||
3605 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 3606 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 3607 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 3608 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 3609 | if (enable) { |
d46285a2 | 3610 | dev->tx_q[qid].map = qid; |
585a5bea | 3611 | } else { |
d46285a2 | 3612 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 3613 | } |
d46285a2 | 3614 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 3615 | exists = true; |
d46285a2 | 3616 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3617 | break; |
3618 | } | |
c1ff66ac | 3619 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3620 | } |
3621 | ovs_mutex_unlock(&dpdk_mutex); | |
3622 | ||
3623 | if (exists) { | |
0a0f39df CL |
3624 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
3625 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 3626 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 3627 | } else { |
0a0f39df | 3628 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
3629 | return -1; |
3630 | } | |
3631 | ||
3632 | return 0; | |
3633 | } | |
3634 | ||
8492adc2 JS |
3635 | /* |
3636 | * Retrieve the DPDK virtio device ID (vid) associated with a vhostuser | |
3637 | * or vhostuserclient netdev. | |
3638 | * | |
3639 | * Returns a value greater or equal to zero for a valid vid or '-1' if | |
3640 | * there is no valid vid associated. A vid of '-1' must not be used in | |
3641 | * rte_vhost_ APi calls. | |
3642 | * | |
3643 | * Once obtained and validated, a vid can be used by a PMD for multiple | |
3644 | * subsequent rte_vhost API calls until the PMD quiesces. A PMD should | |
3645 | * not fetch the vid again for each of a series of API calls. | |
3646 | */ | |
3647 | ||
0a0f39df CL |
3648 | int |
3649 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 3650 | { |
0a0f39df | 3651 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
3652 | } |
3653 | ||
9509913a IS |
3654 | struct ingress_policer * |
3655 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
3656 | { | |
3657 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
3658 | } | |
3659 | ||
58397e6c | 3660 | static int |
ecc1a34e | 3661 | netdev_dpdk_class_init(void) |
7d1ced01 | 3662 | { |
ecc1a34e DDP |
3663 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
3664 | ||
3665 | /* This function can be called for different classes. The initialization | |
3666 | * needs to be done only once */ | |
3667 | if (ovsthread_once_start(&once)) { | |
3668 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
3669 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
3670 | "[netdev] up|down", 1, 2, | |
3671 | netdev_dpdk_set_admin_state, NULL); | |
3672 | ||
0ee821c2 DB |
3673 | unixctl_command_register("netdev-dpdk/detach", |
3674 | "pci address of device", 1, 1, | |
3675 | netdev_dpdk_detach, NULL); | |
3676 | ||
be481733 IM |
3677 | unixctl_command_register("netdev-dpdk/get-mempool-info", |
3678 | "[netdev]", 0, 1, | |
3679 | netdev_dpdk_get_mempool_info, NULL); | |
3680 | ||
ecc1a34e DDP |
3681 | ovsthread_once_done(&once); |
3682 | } | |
362ca396 | 3683 | |
7d1ced01 CL |
3684 | return 0; |
3685 | } | |
3686 | ||
033e9df2 | 3687 | |
95fb793a | 3688 | /* Client Rings */ |
3689 | ||
95fb793a | 3690 | static int |
3691 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
bb37956a | 3692 | dpdk_port_t *eth_port_id) |
95fb793a | 3693 | { |
48fffdee | 3694 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 3695 | char *ring_name; |
b8374d0d | 3696 | int port_id; |
95fb793a | 3697 | |
48fffdee KT |
3698 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
3699 | if (!ring_pair) { | |
95fb793a | 3700 | return ENOMEM; |
3701 | } | |
3702 | ||
7251515e | 3703 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 3704 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 3705 | |
8f0a76c9 | 3706 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 3707 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3708 | RING_F_SP_ENQ); |
0c6f39e5 | 3709 | free(ring_name); |
48fffdee KT |
3710 | if (ring_pair->cring_tx == NULL) { |
3711 | rte_free(ring_pair); | |
95fb793a | 3712 | return ENOMEM; |
3713 | } | |
3714 | ||
0c6f39e5 | 3715 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 3716 | |
8f0a76c9 | 3717 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 3718 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3719 | RING_F_SC_DEQ); |
0c6f39e5 | 3720 | free(ring_name); |
48fffdee KT |
3721 | if (ring_pair->cring_rx == NULL) { |
3722 | rte_free(ring_pair); | |
95fb793a | 3723 | return ENOMEM; |
3724 | } | |
3725 | ||
b8374d0d MV |
3726 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
3727 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 3728 | |
b8374d0d | 3729 | if (port_id < 0) { |
48fffdee | 3730 | rte_free(ring_pair); |
95fb793a | 3731 | return ENODEV; |
3732 | } | |
3733 | ||
48fffdee | 3734 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
3735 | ring_pair->eth_port_id = port_id; |
3736 | *eth_port_id = port_id; | |
3737 | ||
48fffdee | 3738 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 3739 | |
95fb793a | 3740 | return 0; |
3741 | } | |
3742 | ||
3743 | static int | |
bb37956a | 3744 | dpdk_ring_open(const char dev_name[], dpdk_port_t *eth_port_id) |
64839cf4 | 3745 | OVS_REQUIRES(dpdk_mutex) |
95fb793a | 3746 | { |
48fffdee | 3747 | struct dpdk_ring *ring_pair; |
95fb793a | 3748 | unsigned int port_no; |
3749 | int err = 0; | |
3750 | ||
3751 | /* Names always start with "dpdkr" */ | |
3752 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
3753 | if (err) { | |
3754 | return err; | |
3755 | } | |
3756 | ||
58be5c0e | 3757 | /* Look through our list to find the device */ |
48fffdee KT |
3758 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
3759 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 3760 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 3761 | /* Really all that is needed */ |
48fffdee | 3762 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 3763 | return 0; |
3764 | } | |
3765 | } | |
3766 | /* Need to create the device rings */ | |
3767 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
3768 | } | |
3769 | ||
7251515e | 3770 | static int |
d46285a2 | 3771 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
b30896c9 | 3772 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e | 3773 | { |
d46285a2 | 3774 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a543eb0 | 3775 | struct dp_packet *packet; |
1b99bb05 | 3776 | |
58be5c0e | 3777 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
a47e2db2 | 3778 | * the offload fields are clear. This is because the same mbuf may be |
58be5c0e | 3779 | * modified by the consumer of the ring and return into the datapath |
a47e2db2 | 3780 | * without recalculating the RSS hash or revalidating the checksums. */ |
e883448e | 3781 | DP_PACKET_BATCH_FOR_EACH (i, packet, batch) { |
a47e2db2 | 3782 | dp_packet_reset_offload(packet); |
1b99bb05 | 3783 | } |
7251515e | 3784 | |
b30896c9 | 3785 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e DV |
3786 | return 0; |
3787 | } | |
3788 | ||
95fb793a | 3789 | static int |
3790 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
3791 | { | |
bb37956a | 3792 | dpdk_port_t port_no = 0; |
95fb793a | 3793 | int err = 0; |
3794 | ||
95fb793a | 3795 | ovs_mutex_lock(&dpdk_mutex); |
3796 | ||
3797 | err = dpdk_ring_open(netdev->name, &port_no); | |
3798 | if (err) { | |
3799 | goto unlock_dpdk; | |
3800 | } | |
3801 | ||
1ce30dfd DDP |
3802 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
3803 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 3804 | unlock_dpdk: |
3805 | ovs_mutex_unlock(&dpdk_mutex); | |
3806 | return err; | |
3807 | } | |
3808 | ||
0bf765f7 IS |
3809 | /* QoS Functions */ |
3810 | ||
3811 | /* | |
3812 | * Initialize QoS configuration operations. | |
3813 | */ | |
3814 | static void | |
3815 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
3816 | { | |
3817 | conf->ops = ops; | |
78bd47cf | 3818 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
3819 | } |
3820 | ||
3821 | /* | |
3822 | * Search existing QoS operations in qos_ops and compare each set of | |
3823 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
3824 | * else return NULL | |
3825 | */ | |
3826 | static const struct dpdk_qos_ops * | |
3827 | qos_lookup_name(const char *name) | |
3828 | { | |
3829 | const struct dpdk_qos_ops *const *opsp; | |
3830 | ||
3831 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3832 | const struct dpdk_qos_ops *ops = *opsp; | |
3833 | if (!strcmp(name, ops->qos_name)) { | |
3834 | return ops; | |
3835 | } | |
3836 | } | |
3837 | return NULL; | |
3838 | } | |
3839 | ||
0bf765f7 IS |
3840 | static int |
3841 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
3842 | struct sset *types) | |
3843 | { | |
3844 | const struct dpdk_qos_ops *const *opsp; | |
3845 | ||
3846 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3847 | const struct dpdk_qos_ops *ops = *opsp; | |
3848 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
3849 | sset_add(types, ops->qos_name); | |
3850 | } | |
3851 | } | |
3852 | return 0; | |
3853 | } | |
3854 | ||
3855 | static int | |
d46285a2 | 3856 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
3857 | const char **typep, struct smap *details) |
3858 | { | |
d46285a2 | 3859 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 3860 | struct qos_conf *qos_conf; |
0bf765f7 IS |
3861 | int error = 0; |
3862 | ||
d46285a2 | 3863 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
3864 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
3865 | if (qos_conf) { | |
3866 | *typep = qos_conf->ops->qos_name; | |
3867 | error = (qos_conf->ops->qos_get | |
3868 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
3869 | } else { |
3870 | /* No QoS configuration set, return an empty string */ | |
3871 | *typep = ""; | |
0bf765f7 | 3872 | } |
d46285a2 | 3873 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
3874 | |
3875 | return error; | |
3876 | } | |
3877 | ||
3878 | static int | |
78bd47cf DDP |
3879 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
3880 | const struct smap *details) | |
0bf765f7 | 3881 | { |
d46285a2 | 3882 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 3883 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 3884 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
3885 | int error = 0; |
3886 | ||
d46285a2 | 3887 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 3888 | |
78bd47cf | 3889 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 3890 | |
78bd47cf DDP |
3891 | new_ops = qos_lookup_name(type); |
3892 | ||
3893 | if (!new_ops || !new_ops->qos_construct) { | |
3894 | new_qos_conf = NULL; | |
3895 | if (type && type[0]) { | |
3896 | error = EOPNOTSUPP; | |
0bf765f7 | 3897 | } |
44975bb0 | 3898 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
3899 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
3900 | new_qos_conf = qos_conf; | |
0bf765f7 | 3901 | } else { |
78bd47cf | 3902 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
3903 | } |
3904 | ||
7ea266e9 | 3905 | if (error) { |
78bd47cf DDP |
3906 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
3907 | type, netdev->name, rte_strerror(error)); | |
3908 | } | |
3909 | ||
3910 | if (new_qos_conf != qos_conf) { | |
3911 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
3912 | if (qos_conf) { | |
3913 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
3914 | } | |
0bf765f7 IS |
3915 | } |
3916 | ||
d46285a2 | 3917 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 3918 | |
0bf765f7 IS |
3919 | return error; |
3920 | } | |
3921 | ||
3922 | /* egress-policer details */ | |
3923 | ||
3924 | struct egress_policer { | |
3925 | struct qos_conf qos_conf; | |
3926 | struct rte_meter_srtcm_params app_srtcm_params; | |
3927 | struct rte_meter_srtcm egress_meter; | |
03f3f9c0 | 3928 | struct rte_meter_srtcm_profile egress_prof; |
0bf765f7 IS |
3929 | }; |
3930 | ||
78bd47cf DDP |
3931 | static void |
3932 | egress_policer_details_to_param(const struct smap *details, | |
3933 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 3934 | { |
78bd47cf DDP |
3935 | memset(params, 0, sizeof *params); |
3936 | params->cir = smap_get_ullong(details, "cir", 0); | |
3937 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
3938 | params->ebs = 0; | |
0bf765f7 IS |
3939 | } |
3940 | ||
3941 | static int | |
78bd47cf DDP |
3942 | egress_policer_qos_construct(const struct smap *details, |
3943 | struct qos_conf **conf) | |
0bf765f7 | 3944 | { |
0bf765f7 | 3945 | struct egress_policer *policer; |
0bf765f7 IS |
3946 | int err = 0; |
3947 | ||
0bf765f7 IS |
3948 | policer = xmalloc(sizeof *policer); |
3949 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 3950 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
03f3f9c0 OM |
3951 | err = rte_meter_srtcm_profile_config(&policer->egress_prof, |
3952 | &policer->app_srtcm_params); | |
3953 | if (!err) { | |
3954 | err = rte_meter_srtcm_config(&policer->egress_meter, | |
3955 | &policer->egress_prof); | |
3956 | } | |
3957 | ||
78bd47cf DDP |
3958 | if (!err) { |
3959 | *conf = &policer->qos_conf; | |
3960 | } else { | |
03f3f9c0 | 3961 | VLOG_ERR("Could not create rte meter for egress policer"); |
7ea266e9 | 3962 | free(policer); |
78bd47cf | 3963 | *conf = NULL; |
7ea266e9 IS |
3964 | err = -err; |
3965 | } | |
0bf765f7 IS |
3966 | |
3967 | return err; | |
3968 | } | |
3969 | ||
3970 | static void | |
78bd47cf | 3971 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
3972 | { |
3973 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 3974 | qos_conf); |
0bf765f7 IS |
3975 | free(policer); |
3976 | } | |
3977 | ||
3978 | static int | |
78bd47cf | 3979 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 3980 | { |
78bd47cf DDP |
3981 | struct egress_policer *policer = |
3982 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3983 | ||
3984 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
3985 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 3986 | |
0bf765f7 IS |
3987 | return 0; |
3988 | } | |
3989 | ||
78bd47cf | 3990 | static bool |
47a45d86 KT |
3991 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
3992 | const struct smap *details) | |
0bf765f7 | 3993 | { |
78bd47cf DDP |
3994 | struct egress_policer *policer = |
3995 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3996 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 3997 | |
78bd47cf | 3998 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 3999 | |
78bd47cf | 4000 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
4001 | } |
4002 | ||
0bf765f7 | 4003 | static int |
3e90f7d7 | 4004 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt, |
7d7ded7a | 4005 | bool should_steal) |
0bf765f7 | 4006 | { |
0bf765f7 | 4007 | int cnt = 0; |
78bd47cf DDP |
4008 | struct egress_policer *policer = |
4009 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 4010 | |
03f3f9c0 OM |
4011 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, |
4012 | &policer->egress_prof, pkts, | |
7d7ded7a | 4013 | pkt_cnt, should_steal); |
0bf765f7 IS |
4014 | |
4015 | return cnt; | |
4016 | } | |
4017 | ||
4018 | static const struct dpdk_qos_ops egress_policer_ops = { | |
4019 | "egress-policer", /* qos_name */ | |
4020 | egress_policer_qos_construct, | |
4021 | egress_policer_qos_destruct, | |
4022 | egress_policer_qos_get, | |
78bd47cf | 4023 | egress_policer_qos_is_equal, |
0bf765f7 IS |
4024 | egress_policer_run |
4025 | }; | |
4026 | ||
050c60bf DDP |
4027 | static int |
4028 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
4029 | { | |
4030 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
4031 | int err = 0; | |
4032 | ||
050c60bf DDP |
4033 | ovs_mutex_lock(&dev->mutex); |
4034 | ||
4035 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 4036 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b | 4037 | && dev->mtu == dev->requested_mtu |
f8b64a61 | 4038 | && dev->lsc_interrupt_mode == dev->requested_lsc_interrupt_mode |
b685696b | 4039 | && dev->rxq_size == dev->requested_rxq_size |
bd4e172b | 4040 | && dev->txq_size == dev->requested_txq_size |
606f6650 EC |
4041 | && dev->socket_id == dev->requested_socket_id |
4042 | && dev->started) { | |
050c60bf DDP |
4043 | /* Reconfiguration is unnecessary */ |
4044 | ||
4045 | goto out; | |
4046 | } | |
4047 | ||
4048 | rte_eth_dev_stop(dev->port_id); | |
606f6650 | 4049 | dev->started = false; |
050c60bf | 4050 | |
d555d9bd | 4051 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 | 4052 | if (err && err != EEXIST) { |
d555d9bd | 4053 | goto out; |
0072e931 MK |
4054 | } |
4055 | ||
f8b64a61 RM |
4056 | dev->lsc_interrupt_mode = dev->requested_lsc_interrupt_mode; |
4057 | ||
050c60bf DDP |
4058 | netdev->n_txq = dev->requested_n_txq; |
4059 | netdev->n_rxq = dev->requested_n_rxq; | |
4060 | ||
b685696b CL |
4061 | dev->rxq_size = dev->requested_rxq_size; |
4062 | dev->txq_size = dev->requested_txq_size; | |
4063 | ||
050c60bf DDP |
4064 | rte_free(dev->tx_q); |
4065 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
4066 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
4067 | if (!dev->tx_q) { | |
4068 | err = ENOMEM; | |
4069 | } | |
050c60bf | 4070 | |
0072e931 MK |
4071 | netdev_change_seq_changed(netdev); |
4072 | ||
050c60bf | 4073 | out: |
050c60bf | 4074 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
4075 | return err; |
4076 | } | |
4077 | ||
7f381c2e | 4078 | static int |
2d24d165 | 4079 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 4080 | OVS_REQUIRES(dev->mutex) |
050c60bf | 4081 | { |
2d24d165 CL |
4082 | dev->up.n_txq = dev->requested_n_txq; |
4083 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 4084 | int err; |
050c60bf | 4085 | |
81acebda IM |
4086 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
4087 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
4088 | dev->tx_q[0].map = 0; | |
4089 | } | |
4090 | ||
4091 | netdev_dpdk_remap_txqs(dev); | |
4092 | ||
d555d9bd | 4093 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 | 4094 | if (!err) { |
43307ad0 | 4095 | /* A new mempool was created or re-used. */ |
d555d9bd | 4096 | netdev_change_seq_changed(&dev->up); |
03f3f9c0 | 4097 | } else if (err != EEXIST) { |
b6b26021 | 4098 | return err; |
db8f13b0 | 4099 | } |
0a0f39df | 4100 | if (netdev_dpdk_get_vid(dev) >= 0) { |
894af647 | 4101 | if (dev->vhost_reconfigured == false) { |
4102 | dev->vhost_reconfigured = true; | |
4103 | /* Carrier status may need updating. */ | |
4104 | netdev_change_seq_changed(&dev->up); | |
4105 | } | |
81acebda | 4106 | } |
7f381c2e DDP |
4107 | |
4108 | return 0; | |
2d24d165 CL |
4109 | } |
4110 | ||
4111 | static int | |
4112 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
4113 | { | |
4114 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 4115 | int err; |
2d24d165 | 4116 | |
2d24d165 | 4117 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 4118 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 4119 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
4120 | |
4121 | return err; | |
2d24d165 CL |
4122 | } |
4123 | ||
4124 | static int | |
4125 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
4126 | { | |
4127 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 4128 | int err; |
a14d1cc8 | 4129 | uint64_t vhost_flags = 0; |
10087cba | 4130 | bool zc_enabled; |
2d24d165 | 4131 | |
2d24d165 CL |
4132 | ovs_mutex_lock(&dev->mutex); |
4133 | ||
c1ff66ac CL |
4134 | /* Configure vHost client mode if requested and if the following criteria |
4135 | * are met: | |
2d24d165 CL |
4136 | * 1. Device hasn't been registered yet. |
4137 | * 2. A path has been specified. | |
c1ff66ac CL |
4138 | */ |
4139 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 | 4140 | && strlen(dev->vhost_id)) { |
a14d1cc8 MK |
4141 | /* Register client-mode device. */ |
4142 | vhost_flags |= RTE_VHOST_USER_CLIENT; | |
4143 | ||
4144 | /* Enable IOMMU support, if explicitly requested. */ | |
4145 | if (dpdk_vhost_iommu_enabled()) { | |
4146 | vhost_flags |= RTE_VHOST_USER_IOMMU_SUPPORT; | |
4147 | } | |
10087cba CL |
4148 | |
4149 | zc_enabled = dev->vhost_driver_flags | |
4150 | & RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
4151 | /* Enable zero copy flag, if requested */ | |
4152 | if (zc_enabled) { | |
4153 | vhost_flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
4154 | } | |
4155 | ||
a14d1cc8 | 4156 | err = rte_vhost_driver_register(dev->vhost_id, vhost_flags); |
c1ff66ac | 4157 | if (err) { |
2d24d165 CL |
4158 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
4159 | dev->vhost_id); | |
7f381c2e | 4160 | goto unlock; |
c1ff66ac | 4161 | } else { |
2d24d165 | 4162 | /* Configuration successful */ |
a14d1cc8 | 4163 | dev->vhost_driver_flags |= vhost_flags; |
2d24d165 CL |
4164 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " |
4165 | "using client socket '%s'", | |
4166 | dev->up.name, dev->vhost_id); | |
10087cba CL |
4167 | if (zc_enabled) { |
4168 | VLOG_INFO("Zero copy enabled for vHost port %s", dev->up.name); | |
4169 | } | |
c1ff66ac | 4170 | } |
f3e7ec25 MW |
4171 | |
4172 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
4173 | &virtio_net_device_ops); | |
4174 | if (err) { | |
4175 | VLOG_ERR("rte_vhost_driver_callback_register failed for " | |
4176 | "vhost user client port: %s\n", dev->up.name); | |
4177 | goto unlock; | |
4178 | } | |
4179 | ||
4180 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
4181 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
4182 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
4183 | | 1ULL << VIRTIO_NET_F_CSUM); | |
4184 | if (err) { | |
4185 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
4186 | "client port: %s\n", dev->up.name); | |
4187 | goto unlock; | |
4188 | } | |
4189 | ||
4190 | err = rte_vhost_driver_start(dev->vhost_id); | |
4191 | if (err) { | |
4192 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
4193 | "client port: %s\n", dev->up.name); | |
4194 | goto unlock; | |
4195 | } | |
c1ff66ac CL |
4196 | } |
4197 | ||
7f381c2e DDP |
4198 | err = dpdk_vhost_reconfigure_helper(dev); |
4199 | ||
4200 | unlock: | |
050c60bf | 4201 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 4202 | |
7f381c2e | 4203 | return err; |
050c60bf DDP |
4204 | } |
4205 | ||
e8a2b5bf FC |
4206 | |
4207 | /* Find rte_flow with @ufid */ | |
4208 | static struct rte_flow * | |
4209 | ufid_to_rte_flow_find(const ovs_u128 *ufid) { | |
4210 | size_t hash = hash_bytes(ufid, sizeof(*ufid), 0); | |
4211 | struct ufid_to_rte_flow_data *data; | |
4212 | ||
4213 | CMAP_FOR_EACH_WITH_HASH (data, node, hash, &ufid_to_rte_flow) { | |
4214 | if (ovs_u128_equals(*ufid, data->ufid)) { | |
4215 | return data->rte_flow; | |
4216 | } | |
4217 | } | |
4218 | ||
4219 | return NULL; | |
4220 | } | |
4221 | ||
4222 | static inline void | |
4223 | ufid_to_rte_flow_associate(const ovs_u128 *ufid, | |
4224 | struct rte_flow *rte_flow) { | |
4225 | size_t hash = hash_bytes(ufid, sizeof(*ufid), 0); | |
4226 | struct ufid_to_rte_flow_data *data = xzalloc(sizeof(*data)); | |
4227 | ||
4228 | /* | |
4229 | * We should not simply overwrite an existing rte flow. | |
4230 | * We should have deleted it first before re-adding it. | |
4231 | * Thus, if following assert triggers, something is wrong: | |
4232 | * the rte_flow is not destroyed. | |
4233 | */ | |
4234 | ovs_assert(ufid_to_rte_flow_find(ufid) == NULL); | |
4235 | ||
4236 | data->ufid = *ufid; | |
4237 | data->rte_flow = rte_flow; | |
4238 | ||
4239 | cmap_insert(&ufid_to_rte_flow, | |
4240 | CONST_CAST(struct cmap_node *, &data->node), hash); | |
4241 | } | |
4242 | ||
4243 | static inline void | |
4244 | ufid_to_rte_flow_disassociate(const ovs_u128 *ufid) { | |
4245 | size_t hash = hash_bytes(ufid, sizeof(*ufid), 0); | |
4246 | struct ufid_to_rte_flow_data *data; | |
4247 | ||
4248 | CMAP_FOR_EACH_WITH_HASH (data, node, hash, &ufid_to_rte_flow) { | |
4249 | if (ovs_u128_equals(*ufid, data->ufid)) { | |
4250 | cmap_remove(&ufid_to_rte_flow, | |
4251 | CONST_CAST(struct cmap_node *, &data->node), hash); | |
5752eae4 | 4252 | ovsrcu_postpone(free, data); |
e8a2b5bf FC |
4253 | return; |
4254 | } | |
4255 | } | |
4256 | ||
4257 | VLOG_WARN("ufid "UUID_FMT" is not associated with an rte flow\n", | |
4258 | UUID_ARGS((struct uuid *)ufid)); | |
4259 | } | |
4260 | ||
4261 | /* | |
4262 | * To avoid individual xrealloc calls for each new element, a 'curent_max' | |
4263 | * is used to keep track of current allocated number of elements. Starts | |
4264 | * by 8 and doubles on each xrealloc call | |
4265 | */ | |
4266 | struct flow_patterns { | |
4267 | struct rte_flow_item *items; | |
4268 | int cnt; | |
4269 | int current_max; | |
4270 | }; | |
4271 | ||
4272 | struct flow_actions { | |
4273 | struct rte_flow_action *actions; | |
4274 | int cnt; | |
4275 | int current_max; | |
4276 | }; | |
4277 | ||
daf90186 YL |
4278 | static void |
4279 | dump_flow_pattern(struct rte_flow_item *item) | |
4280 | { | |
94740736 IM |
4281 | struct ds s; |
4282 | ||
4283 | if (!VLOG_IS_DBG_ENABLED() || item->type == RTE_FLOW_ITEM_TYPE_END) { | |
4284 | return; | |
4285 | } | |
4286 | ||
4287 | ds_init(&s); | |
4288 | ||
daf90186 YL |
4289 | if (item->type == RTE_FLOW_ITEM_TYPE_ETH) { |
4290 | const struct rte_flow_item_eth *eth_spec = item->spec; | |
4291 | const struct rte_flow_item_eth *eth_mask = item->mask; | |
4292 | ||
94740736 | 4293 | ds_put_cstr(&s, "rte flow eth pattern:\n"); |
daf90186 | 4294 | if (eth_spec) { |
94740736 IM |
4295 | ds_put_format(&s, |
4296 | " Spec: src="ETH_ADDR_FMT", dst="ETH_ADDR_FMT", " | |
daf90186 | 4297 | "type=0x%04" PRIx16"\n", |
773c3cb4 BP |
4298 | ETH_ADDR_BYTES_ARGS(eth_spec->src.addr_bytes), |
4299 | ETH_ADDR_BYTES_ARGS(eth_spec->dst.addr_bytes), | |
daf90186 YL |
4300 | ntohs(eth_spec->type)); |
4301 | } else { | |
94740736 | 4302 | ds_put_cstr(&s, " Spec = null\n"); |
daf90186 YL |
4303 | } |
4304 | if (eth_mask) { | |
94740736 IM |
4305 | ds_put_format(&s, |
4306 | " Mask: src="ETH_ADDR_FMT", dst="ETH_ADDR_FMT", " | |
daf90186 | 4307 | "type=0x%04"PRIx16"\n", |
773c3cb4 BP |
4308 | ETH_ADDR_BYTES_ARGS(eth_mask->src.addr_bytes), |
4309 | ETH_ADDR_BYTES_ARGS(eth_mask->dst.addr_bytes), | |
daf90186 YL |
4310 | eth_mask->type); |
4311 | } else { | |
94740736 | 4312 | ds_put_cstr(&s, " Mask = null\n"); |
daf90186 YL |
4313 | } |
4314 | } | |
4315 | ||
4316 | if (item->type == RTE_FLOW_ITEM_TYPE_VLAN) { | |
4317 | const struct rte_flow_item_vlan *vlan_spec = item->spec; | |
4318 | const struct rte_flow_item_vlan *vlan_mask = item->mask; | |
4319 | ||
94740736 | 4320 | ds_put_cstr(&s, "rte flow vlan pattern:\n"); |
daf90186 | 4321 | if (vlan_spec) { |
94740736 | 4322 | ds_put_format(&s, |
03f3f9c0 OM |
4323 | " Spec: inner_type=0x%"PRIx16", tci=0x%"PRIx16"\n", |
4324 | ntohs(vlan_spec->inner_type), ntohs(vlan_spec->tci)); | |
daf90186 | 4325 | } else { |
94740736 | 4326 | ds_put_cstr(&s, " Spec = null\n"); |
daf90186 YL |
4327 | } |
4328 | ||
4329 | if (vlan_mask) { | |
94740736 | 4330 | ds_put_format(&s, |
03f3f9c0 OM |
4331 | " Mask: inner_type=0x%"PRIx16", tci=0x%"PRIx16"\n", |
4332 | ntohs(vlan_mask->inner_type), ntohs(vlan_mask->tci)); | |
daf90186 | 4333 | } else { |
94740736 | 4334 | ds_put_cstr(&s, " Mask = null\n"); |
daf90186 YL |
4335 | } |
4336 | } | |
4337 | ||
4338 | if (item->type == RTE_FLOW_ITEM_TYPE_IPV4) { | |
4339 | const struct rte_flow_item_ipv4 *ipv4_spec = item->spec; | |
4340 | const struct rte_flow_item_ipv4 *ipv4_mask = item->mask; | |
4341 | ||
94740736 | 4342 | ds_put_cstr(&s, "rte flow ipv4 pattern:\n"); |
daf90186 | 4343 | if (ipv4_spec) { |
94740736 IM |
4344 | ds_put_format(&s, |
4345 | " Spec: tos=0x%"PRIx8", ttl=%"PRIx8", proto=0x%"PRIx8 | |
daf90186 YL |
4346 | ", src="IP_FMT", dst="IP_FMT"\n", |
4347 | ipv4_spec->hdr.type_of_service, | |
4348 | ipv4_spec->hdr.time_to_live, | |
4349 | ipv4_spec->hdr.next_proto_id, | |
4350 | IP_ARGS(ipv4_spec->hdr.src_addr), | |
4351 | IP_ARGS(ipv4_spec->hdr.dst_addr)); | |
4352 | } else { | |
94740736 | 4353 | ds_put_cstr(&s, " Spec = null\n"); |
daf90186 YL |
4354 | } |
4355 | if (ipv4_mask) { | |
94740736 IM |
4356 | ds_put_format(&s, |
4357 | " Mask: tos=0x%"PRIx8", ttl=%"PRIx8", proto=0x%"PRIx8 | |
daf90186 YL |
4358 | ", src="IP_FMT", dst="IP_FMT"\n", |
4359 | ipv4_mask->hdr.type_of_service, | |
4360 | ipv4_mask->hdr.time_to_live, | |
4361 | ipv4_mask->hdr.next_proto_id, | |
4362 | IP_ARGS(ipv4_mask->hdr.src_addr), | |
4363 | IP_ARGS(ipv4_mask->hdr.dst_addr)); | |
4364 | } else { | |
94740736 | 4365 | ds_put_cstr(&s, " Mask = null\n"); |
daf90186 YL |
4366 | } |
4367 | } | |
4368 | ||
4369 | if (item->type == RTE_FLOW_ITEM_TYPE_UDP) { | |
4370 | const struct rte_flow_item_udp *udp_spec = item->spec; | |
4371 | const struct rte_flow_item_udp *udp_mask = item->mask; | |
4372 | ||
94740736 | 4373 | ds_put_cstr(&s, "rte flow udp pattern:\n"); |
daf90186 | 4374 | if (udp_spec) { |
94740736 IM |
4375 | ds_put_format(&s, |
4376 | " Spec: src_port=%"PRIu16", dst_port=%"PRIu16"\n", | |
daf90186 YL |
4377 | ntohs(udp_spec->hdr.src_port), |
4378 | ntohs(udp_spec->hdr.dst_port)); | |
4379 | } else { | |
94740736 | 4380 | ds_put_cstr(&s, " Spec = null\n"); |
daf90186 YL |
4381 | } |
4382 | if (udp_mask) { | |
94740736 IM |
4383 | ds_put_format(&s, |
4384 | " Mask: src_port=0x%"PRIx16", dst_port=0x%"PRIx16"\n", | |
daf90186 YL |
4385 | udp_mask->hdr.src_port, |
4386 | udp_mask->hdr.dst_port); | |
4387 | } else { | |
94740736 | 4388 | ds_put_cstr(&s, " Mask = null\n"); |
daf90186 YL |
4389 | } |
4390 | } | |
4391 | ||
4392 | if (item->type == RTE_FLOW_ITEM_TYPE_SCTP) { | |
4393 | const struct rte_flow_item_sctp *sctp_spec = item->spec; | |
4394 | const struct rte_flow_item_sctp *sctp_mask = item->mask; | |
4395 | ||
94740736 | 4396 | ds_put_cstr(&s, "rte flow sctp pattern:\n"); |
daf90186 | 4397 | if (sctp_spec) { |
94740736 IM |
4398 | ds_put_format(&s, |
4399 | " Spec: src_port=%"PRIu16", dst_port=%"PRIu16"\n", | |
daf90186 YL |
4400 | ntohs(sctp_spec->hdr.src_port), |
4401 | ntohs(sctp_spec->hdr.dst_port)); | |
4402 | } else { | |
94740736 | 4403 | ds_put_cstr(&s, " Spec = null\n"); |
daf90186 YL |
4404 | } |
4405 | if (sctp_mask) { | |
94740736 IM |
4406 | ds_put_format(&s, |
4407 | " Mask: src_port=0x%"PRIx16", dst_port=0x%"PRIx16"\n", | |
daf90186 YL |
4408 | sctp_mask->hdr.src_port, |
4409 | sctp_mask->hdr.dst_port); | |
4410 | } else { | |
94740736 | 4411 | ds_put_cstr(&s, " Mask = null\n"); |
daf90186 YL |
4412 | } |
4413 | } | |
4414 | ||
4415 | if (item->type == RTE_FLOW_ITEM_TYPE_ICMP) { | |
4416 | const struct rte_flow_item_icmp *icmp_spec = item->spec; | |
4417 | const struct rte_flow_item_icmp *icmp_mask = item->mask; | |
4418 | ||
94740736 | 4419 | ds_put_cstr(&s, "rte flow icmp pattern:\n"); |
daf90186 | 4420 | if (icmp_spec) { |
94740736 IM |
4421 | ds_put_format(&s, |
4422 | " Spec: icmp_type=%"PRIu8", icmp_code=%"PRIu8"\n", | |
2b7b5dbb BP |
4423 | icmp_spec->hdr.icmp_type, |
4424 | icmp_spec->hdr.icmp_code); | |
daf90186 | 4425 | } else { |
94740736 | 4426 | ds_put_cstr(&s, " Spec = null\n"); |
daf90186 YL |
4427 | } |
4428 | if (icmp_mask) { | |
94740736 IM |
4429 | ds_put_format(&s, |
4430 | " Mask: icmp_type=0x%"PRIx8", icmp_code=0x%"PRIx8"\n", | |
daf90186 YL |
4431 | icmp_spec->hdr.icmp_type, |
4432 | icmp_spec->hdr.icmp_code); | |
4433 | } else { | |
94740736 | 4434 | ds_put_cstr(&s, " Mask = null\n"); |
daf90186 YL |
4435 | } |
4436 | } | |
4437 | ||
4438 | if (item->type == RTE_FLOW_ITEM_TYPE_TCP) { | |
4439 | const struct rte_flow_item_tcp *tcp_spec = item->spec; | |
4440 | const struct rte_flow_item_tcp *tcp_mask = item->mask; | |
4441 | ||
94740736 | 4442 | ds_put_cstr(&s, "rte flow tcp pattern:\n"); |
daf90186 | 4443 | if (tcp_spec) { |
94740736 IM |
4444 | ds_put_format(&s, |
4445 | " Spec: src_port=%"PRIu16", dst_port=%"PRIu16 | |
daf90186 YL |
4446 | ", data_off=0x%"PRIx8", tcp_flags=0x%"PRIx8"\n", |
4447 | ntohs(tcp_spec->hdr.src_port), | |
4448 | ntohs(tcp_spec->hdr.dst_port), | |
4449 | tcp_spec->hdr.data_off, | |
4450 | tcp_spec->hdr.tcp_flags); | |
4451 | } else { | |
94740736 | 4452 | ds_put_cstr(&s, " Spec = null\n"); |
daf90186 YL |
4453 | } |
4454 | if (tcp_mask) { | |
94740736 IM |
4455 | ds_put_format(&s, |
4456 | " Mask: src_port=%"PRIx16", dst_port=%"PRIx16 | |
daf90186 YL |
4457 | ", data_off=0x%"PRIx8", tcp_flags=0x%"PRIx8"\n", |
4458 | tcp_mask->hdr.src_port, | |
4459 | tcp_mask->hdr.dst_port, | |
4460 | tcp_mask->hdr.data_off, | |
4461 | tcp_mask->hdr.tcp_flags); | |
4462 | } else { | |
94740736 | 4463 | ds_put_cstr(&s, " Mask = null\n"); |
daf90186 YL |
4464 | } |
4465 | } | |
94740736 IM |
4466 | |
4467 | VLOG_DBG("%s", ds_cstr(&s)); | |
4468 | ds_destroy(&s); | |
daf90186 YL |
4469 | } |
4470 | ||
e8a2b5bf FC |
4471 | static void |
4472 | add_flow_pattern(struct flow_patterns *patterns, enum rte_flow_item_type type, | |
4473 | const void *spec, const void *mask) { | |
4474 | int cnt = patterns->cnt; | |
4475 | ||
4476 | if (cnt == 0) { | |
4477 | patterns->current_max = 8; | |
4478 | patterns->items = xcalloc(patterns->current_max, | |
4479 | sizeof(struct rte_flow_item)); | |
4480 | } else if (cnt == patterns->current_max) { | |
4481 | patterns->current_max *= 2; | |
4482 | patterns->items = xrealloc(patterns->items, patterns->current_max * | |
4483 | sizeof(struct rte_flow_item)); | |
4484 | } | |
4485 | ||
4486 | patterns->items[cnt].type = type; | |
4487 | patterns->items[cnt].spec = spec; | |
4488 | patterns->items[cnt].mask = mask; | |
4489 | patterns->items[cnt].last = NULL; | |
daf90186 | 4490 | dump_flow_pattern(&patterns->items[cnt]); |
e8a2b5bf FC |
4491 | patterns->cnt++; |
4492 | } | |
4493 | ||
4494 | static void | |
4495 | add_flow_action(struct flow_actions *actions, enum rte_flow_action_type type, | |
4496 | const void *conf) | |
4497 | { | |
4498 | int cnt = actions->cnt; | |
4499 | ||
4500 | if (cnt == 0) { | |
4501 | actions->current_max = 8; | |
4502 | actions->actions = xcalloc(actions->current_max, | |
4503 | sizeof(struct rte_flow_action)); | |
4504 | } else if (cnt == actions->current_max) { | |
4505 | actions->current_max *= 2; | |
4506 | actions->actions = xrealloc(actions->actions, actions->current_max * | |
4507 | sizeof(struct rte_flow_action)); | |
4508 | } | |
4509 | ||
4510 | actions->actions[cnt].type = type; | |
4511 | actions->actions[cnt].conf = conf; | |
4512 | actions->cnt++; | |
4513 | } | |
4514 | ||
03f3f9c0 OM |
4515 | struct action_rss_data { |
4516 | struct rte_flow_action_rss conf; | |
4517 | uint16_t queue[0]; | |
4518 | }; | |
4519 | ||
4520 | static struct action_rss_data * | |
e8a2b5bf FC |
4521 | add_flow_rss_action(struct flow_actions *actions, |
4522 | struct netdev *netdev) { | |
4523 | int i; | |
03f3f9c0 OM |
4524 | struct action_rss_data *rss_data; |
4525 | ||
4526 | rss_data = xmalloc(sizeof(struct action_rss_data) + | |
4527 | sizeof(uint16_t) * netdev->n_rxq); | |
4528 | *rss_data = (struct action_rss_data) { | |
4529 | .conf = (struct rte_flow_action_rss) { | |
4530 | .func = RTE_ETH_HASH_FUNCTION_DEFAULT, | |
4531 | .level = 0, | |
4532 | .types = 0, | |
4533 | .queue_num = netdev->n_rxq, | |
4534 | .queue = rss_data->queue, | |
4535 | .key_len = 0, | |
4536 | .key = NULL | |
4537 | }, | |
4538 | }; | |
e8a2b5bf | 4539 | |
03f3f9c0 OM |
4540 | /* Override queue array with default */ |
4541 | for (i = 0; i < netdev->n_rxq; i++) { | |
4542 | rss_data->queue[i] = i; | |
e8a2b5bf FC |
4543 | } |
4544 | ||
03f3f9c0 | 4545 | add_flow_action(actions, RTE_FLOW_ACTION_TYPE_RSS, &rss_data->conf); |
e8a2b5bf | 4546 | |
03f3f9c0 | 4547 | return rss_data; |
e8a2b5bf FC |
4548 | } |
4549 | ||
4550 | static int | |
4551 | netdev_dpdk_add_rte_flow_offload(struct netdev *netdev, | |
4552 | const struct match *match, | |
4553 | struct nlattr *nl_actions OVS_UNUSED, | |
4554 | size_t actions_len OVS_UNUSED, | |
4555 | const ovs_u128 *ufid, | |
4556 | struct offload_info *info) { | |
4557 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
4558 | const struct rte_flow_attr flow_attr = { | |
4559 | .group = 0, | |
4560 | .priority = 0, | |
4561 | .ingress = 1, | |
4562 | .egress = 0 | |
4563 | }; | |
4564 | struct flow_patterns patterns = { .items = NULL, .cnt = 0 }; | |
4565 | struct flow_actions actions = { .actions = NULL, .cnt = 0 }; | |
4566 | struct rte_flow *flow; | |
4567 | struct rte_flow_error error; | |
324215c6 | 4568 | uint8_t proto = 0; |
e8a2b5bf | 4569 | int ret = 0; |
324215c6 IM |
4570 | struct flow_items { |
4571 | struct rte_flow_item_eth eth; | |
4572 | struct rte_flow_item_vlan vlan; | |
4573 | struct rte_flow_item_ipv4 ipv4; | |
4574 | union { | |
4575 | struct rte_flow_item_tcp tcp; | |
4576 | struct rte_flow_item_udp udp; | |
4577 | struct rte_flow_item_sctp sctp; | |
4578 | struct rte_flow_item_icmp icmp; | |
4579 | }; | |
4580 | } spec, mask; | |
4581 | ||
4582 | memset(&spec, 0, sizeof spec); | |
4583 | memset(&mask, 0, sizeof mask); | |
e8a2b5bf FC |
4584 | |
4585 | /* Eth */ | |
e8a2b5bf FC |
4586 | if (!eth_addr_is_zero(match->wc.masks.dl_src) || |
4587 | !eth_addr_is_zero(match->wc.masks.dl_dst)) { | |
324215c6 IM |
4588 | memcpy(&spec.eth.dst, &match->flow.dl_dst, sizeof spec.eth.dst); |
4589 | memcpy(&spec.eth.src, &match->flow.dl_src, sizeof spec.eth.src); | |
4590 | spec.eth.type = match->flow.dl_type; | |
4591 | ||
4592 | memcpy(&mask.eth.dst, &match->wc.masks.dl_dst, sizeof mask.eth.dst); | |
4593 | memcpy(&mask.eth.src, &match->wc.masks.dl_src, sizeof mask.eth.src); | |
4594 | mask.eth.type = match->wc.masks.dl_type; | |
e8a2b5bf FC |
4595 | |
4596 | add_flow_pattern(&patterns, RTE_FLOW_ITEM_TYPE_ETH, | |
324215c6 | 4597 | &spec.eth, &mask.eth); |
e8a2b5bf FC |
4598 | } else { |
4599 | /* | |
4600 | * If user specifies a flow (like UDP flow) without L2 patterns, | |
4601 | * OVS will at least set the dl_type. Normally, it's enough to | |
4602 | * create an eth pattern just with it. Unluckily, some Intel's | |
4603 | * NIC (such as XL710) doesn't support that. Below is a workaround, | |
4604 | * which simply matches any L2 pkts. | |
4605 | */ | |
4606 | add_flow_pattern(&patterns, RTE_FLOW_ITEM_TYPE_ETH, NULL, NULL); | |
4607 | } | |
4608 | ||
4609 | /* VLAN */ | |
e8a2b5bf | 4610 | if (match->wc.masks.vlans[0].tci && match->flow.vlans[0].tci) { |
324215c6 IM |
4611 | spec.vlan.tci = match->flow.vlans[0].tci & ~htons(VLAN_CFI); |
4612 | mask.vlan.tci = match->wc.masks.vlans[0].tci & ~htons(VLAN_CFI); | |
e8a2b5bf FC |
4613 | |
4614 | /* match any protocols */ | |
324215c6 | 4615 | mask.vlan.inner_type = 0; |
e8a2b5bf FC |
4616 | |
4617 | add_flow_pattern(&patterns, RTE_FLOW_ITEM_TYPE_VLAN, | |
324215c6 | 4618 | &spec.vlan, &mask.vlan); |
e8a2b5bf FC |
4619 | } |
4620 | ||
4621 | /* IP v4 */ | |
31a033cb | 4622 | if (match->flow.dl_type == htons(ETH_TYPE_IP)) { |
324215c6 IM |
4623 | spec.ipv4.hdr.type_of_service = match->flow.nw_tos; |
4624 | spec.ipv4.hdr.time_to_live = match->flow.nw_ttl; | |
4625 | spec.ipv4.hdr.next_proto_id = match->flow.nw_proto; | |
4626 | spec.ipv4.hdr.src_addr = match->flow.nw_src; | |
4627 | spec.ipv4.hdr.dst_addr = match->flow.nw_dst; | |
4628 | ||
4629 | mask.ipv4.hdr.type_of_service = match->wc.masks.nw_tos; | |
4630 | mask.ipv4.hdr.time_to_live = match->wc.masks.nw_ttl; | |
4631 | mask.ipv4.hdr.next_proto_id = match->wc.masks.nw_proto; | |
4632 | mask.ipv4.hdr.src_addr = match->wc.masks.nw_src; | |
4633 | mask.ipv4.hdr.dst_addr = match->wc.masks.nw_dst; | |
e8a2b5bf FC |
4634 | |
4635 | add_flow_pattern(&patterns, RTE_FLOW_ITEM_TYPE_IPV4, | |
324215c6 | 4636 | &spec.ipv4, &mask.ipv4); |
e8a2b5bf FC |
4637 | |
4638 | /* Save proto for L4 protocol setup */ | |
324215c6 IM |
4639 | proto = spec.ipv4.hdr.next_proto_id & |
4640 | mask.ipv4.hdr.next_proto_id; | |
e8a2b5bf FC |
4641 | } |
4642 | ||
4643 | if (proto != IPPROTO_ICMP && proto != IPPROTO_UDP && | |
4644 | proto != IPPROTO_SCTP && proto != IPPROTO_TCP && | |
4645 | (match->wc.masks.tp_src || | |
4646 | match->wc.masks.tp_dst || | |
4647 | match->wc.masks.tcp_flags)) { | |
4648 | VLOG_DBG("L4 Protocol (%u) not supported", proto); | |
4649 | ret = -1; | |
4650 | goto out; | |
4651 | } | |
4652 | ||
31a033cb BP |
4653 | if ((match->wc.masks.tp_src && match->wc.masks.tp_src != OVS_BE16_MAX) || |
4654 | (match->wc.masks.tp_dst && match->wc.masks.tp_dst != OVS_BE16_MAX)) { | |
e8a2b5bf FC |
4655 | ret = -1; |
4656 | goto out; | |
4657 | } | |
4658 | ||
324215c6 IM |
4659 | switch (proto) { |
4660 | case IPPROTO_TCP: | |
4661 | spec.tcp.hdr.src_port = match->flow.tp_src; | |
4662 | spec.tcp.hdr.dst_port = match->flow.tp_dst; | |
4663 | spec.tcp.hdr.data_off = ntohs(match->flow.tcp_flags) >> 8; | |
4664 | spec.tcp.hdr.tcp_flags = ntohs(match->flow.tcp_flags) & 0xff; | |
e8a2b5bf | 4665 | |
324215c6 IM |
4666 | mask.tcp.hdr.src_port = match->wc.masks.tp_src; |
4667 | mask.tcp.hdr.dst_port = match->wc.masks.tp_dst; | |
4668 | mask.tcp.hdr.data_off = ntohs(match->wc.masks.tcp_flags) >> 8; | |
4669 | mask.tcp.hdr.tcp_flags = ntohs(match->wc.masks.tcp_flags) & 0xff; | |
e8a2b5bf FC |
4670 | |
4671 | add_flow_pattern(&patterns, RTE_FLOW_ITEM_TYPE_TCP, | |
324215c6 | 4672 | &spec.tcp, &mask.tcp); |
e8a2b5bf FC |
4673 | |
4674 | /* proto == TCP and ITEM_TYPE_TCP, thus no need for proto match */ | |
324215c6 IM |
4675 | mask.ipv4.hdr.next_proto_id = 0; |
4676 | break; | |
e8a2b5bf | 4677 | |
324215c6 IM |
4678 | case IPPROTO_UDP: |
4679 | spec.udp.hdr.src_port = match->flow.tp_src; | |
4680 | spec.udp.hdr.dst_port = match->flow.tp_dst; | |
e8a2b5bf | 4681 | |
324215c6 IM |
4682 | mask.udp.hdr.src_port = match->wc.masks.tp_src; |
4683 | mask.udp.hdr.dst_port = match->wc.masks.tp_dst; | |
e8a2b5bf FC |
4684 | |
4685 | add_flow_pattern(&patterns, RTE_FLOW_ITEM_TYPE_UDP, | |
324215c6 | 4686 | &spec.udp, &mask.udp); |
e8a2b5bf FC |
4687 | |
4688 | /* proto == UDP and ITEM_TYPE_UDP, thus no need for proto match */ | |
324215c6 IM |
4689 | mask.ipv4.hdr.next_proto_id = 0; |
4690 | break; | |
e8a2b5bf | 4691 | |
324215c6 IM |
4692 | case IPPROTO_SCTP: |
4693 | spec.sctp.hdr.src_port = match->flow.tp_src; | |
4694 | spec.sctp.hdr.dst_port = match->flow.tp_dst; | |
e8a2b5bf | 4695 | |
324215c6 IM |
4696 | mask.sctp.hdr.src_port = match->wc.masks.tp_src; |
4697 | mask.sctp.hdr.dst_port = match->wc.masks.tp_dst; | |
e8a2b5bf FC |
4698 | |
4699 | add_flow_pattern(&patterns, RTE_FLOW_ITEM_TYPE_SCTP, | |
324215c6 | 4700 | &spec.sctp, &mask.sctp); |
e8a2b5bf FC |
4701 | |
4702 | /* proto == SCTP and ITEM_TYPE_SCTP, thus no need for proto match */ | |
324215c6 IM |
4703 | mask.ipv4.hdr.next_proto_id = 0; |
4704 | break; | |
e8a2b5bf | 4705 | |
324215c6 IM |
4706 | case IPPROTO_ICMP: |
4707 | spec.icmp.hdr.icmp_type = (uint8_t) ntohs(match->flow.tp_src); | |
4708 | spec.icmp.hdr.icmp_code = (uint8_t) ntohs(match->flow.tp_dst); | |
e8a2b5bf | 4709 | |
324215c6 IM |
4710 | mask.icmp.hdr.icmp_type = (uint8_t) ntohs(match->wc.masks.tp_src); |
4711 | mask.icmp.hdr.icmp_code = (uint8_t) ntohs(match->wc.masks.tp_dst); | |
e8a2b5bf FC |
4712 | |
4713 | add_flow_pattern(&patterns, RTE_FLOW_ITEM_TYPE_ICMP, | |
324215c6 | 4714 | &spec.icmp, &mask.icmp); |
e8a2b5bf FC |
4715 | |
4716 | /* proto == ICMP and ITEM_TYPE_ICMP, thus no need for proto match */ | |
324215c6 IM |
4717 | mask.ipv4.hdr.next_proto_id = 0; |
4718 | break; | |
e8a2b5bf FC |
4719 | } |
4720 | ||
e8a2b5bf FC |
4721 | add_flow_pattern(&patterns, RTE_FLOW_ITEM_TYPE_END, NULL, NULL); |
4722 | ||
4723 | struct rte_flow_action_mark mark; | |
03f3f9c0 | 4724 | struct action_rss_data *rss; |
95ca79d5 | 4725 | |
e8a2b5bf FC |
4726 | mark.id = info->flow_mark; |
4727 | add_flow_action(&actions, RTE_FLOW_ACTION_TYPE_MARK, &mark); | |
4728 | ||
95ca79d5 IM |
4729 | ovs_mutex_lock(&dev->mutex); |
4730 | ||
e8a2b5bf FC |
4731 | rss = add_flow_rss_action(&actions, netdev); |
4732 | add_flow_action(&actions, RTE_FLOW_ACTION_TYPE_END, NULL); | |
4733 | ||
4734 | flow = rte_flow_create(dev->port_id, &flow_attr, patterns.items, | |
4735 | actions.actions, &error); | |
95ca79d5 IM |
4736 | |
4737 | ovs_mutex_unlock(&dev->mutex); | |
4738 | ||
e8a2b5bf FC |
4739 | free(rss); |
4740 | if (!flow) { | |
faf71e49 IM |
4741 | VLOG_ERR("%s: rte flow creat error: %u : message : %s\n", |
4742 | netdev_get_name(netdev), error.type, error.message); | |
e8a2b5bf FC |
4743 | ret = -1; |
4744 | goto out; | |
4745 | } | |
4746 | ufid_to_rte_flow_associate(ufid, flow); | |
faf71e49 IM |
4747 | VLOG_DBG("%s: installed flow %p by ufid "UUID_FMT"\n", |
4748 | netdev_get_name(netdev), flow, UUID_ARGS((struct uuid *)ufid)); | |
e8a2b5bf FC |
4749 | |
4750 | out: | |
4751 | free(patterns.items); | |
4752 | free(actions.actions); | |
4753 | return ret; | |
4754 | } | |
4755 | ||
e8a2b5bf FC |
4756 | /* |
4757 | * Check if any unsupported flow patterns are specified. | |
4758 | */ | |
4759 | static int | |
4760 | netdev_dpdk_validate_flow(const struct match *match) { | |
4761 | struct match match_zero_wc; | |
6ebc4b09 | 4762 | const struct flow *masks = &match->wc.masks; |
e8a2b5bf FC |
4763 | |
4764 | /* Create a wc-zeroed version of flow */ | |
4765 | match_init(&match_zero_wc, &match->flow, &match->wc); | |
4766 | ||
6ebc4b09 IM |
4767 | if (!is_all_zeros(&match_zero_wc.flow.tunnel, |
4768 | sizeof match_zero_wc.flow.tunnel)) { | |
e8a2b5bf FC |
4769 | goto err; |
4770 | } | |
4771 | ||
6ebc4b09 IM |
4772 | if (masks->metadata || masks->skb_priority || |
4773 | masks->pkt_mark || masks->dp_hash) { | |
e8a2b5bf FC |
4774 | goto err; |
4775 | } | |
4776 | ||
4777 | /* recirc id must be zero */ | |
4778 | if (match_zero_wc.flow.recirc_id) { | |
4779 | goto err; | |
4780 | } | |
4781 | ||
6ebc4b09 IM |
4782 | if (masks->ct_state || masks->ct_nw_proto || |
4783 | masks->ct_zone || masks->ct_mark || | |
4784 | !ovs_u128_is_zero(masks->ct_label)) { | |
e8a2b5bf FC |
4785 | goto err; |
4786 | } | |
4787 | ||
6ebc4b09 | 4788 | if (masks->conj_id || masks->actset_output) { |
e8a2b5bf FC |
4789 | goto err; |
4790 | } | |
4791 | ||
4792 | /* unsupported L2 */ | |
6ebc4b09 | 4793 | if (!is_all_zeros(masks->mpls_lse, sizeof masks->mpls_lse)) { |
e8a2b5bf FC |
4794 | goto err; |
4795 | } | |
4796 | ||
4797 | /* unsupported L3 */ | |
6ebc4b09 IM |
4798 | if (masks->ipv6_label || masks->ct_nw_src || masks->ct_nw_dst || |
4799 | !is_all_zeros(&masks->ipv6_src, sizeof masks->ipv6_src) || | |
4800 | !is_all_zeros(&masks->ipv6_dst, sizeof masks->ipv6_dst) || | |
4801 | !is_all_zeros(&masks->ct_ipv6_src, sizeof masks->ct_ipv6_src) || | |
4802 | !is_all_zeros(&masks->ct_ipv6_dst, sizeof masks->ct_ipv6_dst) || | |
4803 | !is_all_zeros(&masks->nd_target, sizeof masks->nd_target) || | |
4804 | !is_all_zeros(&masks->nsh, sizeof masks->nsh) || | |
4805 | !is_all_zeros(&masks->arp_sha, sizeof masks->arp_sha) || | |
4806 | !is_all_zeros(&masks->arp_tha, sizeof masks->arp_tha)) { | |
e8a2b5bf FC |
4807 | goto err; |
4808 | } | |
4809 | ||
4810 | /* If fragmented, then don't HW accelerate - for now */ | |
4811 | if (match_zero_wc.flow.nw_frag) { | |
4812 | goto err; | |
4813 | } | |
4814 | ||
4815 | /* unsupported L4 */ | |
6ebc4b09 | 4816 | if (masks->igmp_group_ip4 || masks->ct_tp_src || masks->ct_tp_dst) { |
e8a2b5bf FC |
4817 | goto err; |
4818 | } | |
4819 | ||
4820 | return 0; | |
4821 | ||
4822 | err: | |
4823 | VLOG_ERR("cannot HW accelerate this flow due to unsupported protocols"); | |
4824 | return -1; | |
4825 | } | |
4826 | ||
4827 | static int | |
faf71e49 | 4828 | netdev_dpdk_destroy_rte_flow(struct netdev *netdev, |
e8a2b5bf FC |
4829 | const ovs_u128 *ufid, |
4830 | struct rte_flow *rte_flow) { | |
faf71e49 | 4831 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
e8a2b5bf FC |
4832 | struct rte_flow_error error; |
4833 | int ret; | |
4834 | ||
95ca79d5 IM |
4835 | ovs_mutex_lock(&dev->mutex); |
4836 | ||
e8a2b5bf FC |
4837 | ret = rte_flow_destroy(dev->port_id, rte_flow, &error); |
4838 | if (ret == 0) { | |
4839 | ufid_to_rte_flow_disassociate(ufid); | |
faf71e49 IM |
4840 | VLOG_DBG("%s: removed rte flow %p associated with ufid " UUID_FMT "\n", |
4841 | netdev_get_name(netdev), rte_flow, | |
4842 | UUID_ARGS((struct uuid *)ufid)); | |
e8a2b5bf | 4843 | } else { |
faf71e49 IM |
4844 | VLOG_ERR("%s: rte flow destroy error: %u : message : %s\n", |
4845 | netdev_get_name(netdev), error.type, error.message); | |
e8a2b5bf FC |
4846 | } |
4847 | ||
95ca79d5 | 4848 | ovs_mutex_unlock(&dev->mutex); |
e8a2b5bf FC |
4849 | return ret; |
4850 | } | |
4851 | ||
4852 | static int | |
4853 | netdev_dpdk_flow_put(struct netdev *netdev, struct match *match, | |
4854 | struct nlattr *actions, size_t actions_len, | |
4855 | const ovs_u128 *ufid, struct offload_info *info, | |
4856 | struct dpif_flow_stats *stats OVS_UNUSED) { | |
4857 | struct rte_flow *rte_flow; | |
4858 | int ret; | |
4859 | ||
4860 | /* | |
4861 | * If an old rte_flow exists, it means it's a flow modification. | |
4862 | * Here destroy the old rte flow first before adding a new one. | |
4863 | */ | |
4864 | rte_flow = ufid_to_rte_flow_find(ufid); | |
4865 | if (rte_flow) { | |
faf71e49 | 4866 | ret = netdev_dpdk_destroy_rte_flow(netdev, ufid, rte_flow); |
e8a2b5bf FC |
4867 | if (ret < 0) { |
4868 | return ret; | |
4869 | } | |
4870 | } | |
4871 | ||
4872 | ret = netdev_dpdk_validate_flow(match); | |
4873 | if (ret < 0) { | |
4874 | return ret; | |
4875 | } | |
4876 | ||
4877 | return netdev_dpdk_add_rte_flow_offload(netdev, match, actions, | |
4878 | actions_len, ufid, info); | |
4879 | } | |
4880 | ||
4881 | static int | |
4882 | netdev_dpdk_flow_del(struct netdev *netdev, const ovs_u128 *ufid, | |
4883 | struct dpif_flow_stats *stats OVS_UNUSED) { | |
4884 | ||
4885 | struct rte_flow *rte_flow = ufid_to_rte_flow_find(ufid); | |
4886 | ||
4887 | if (!rte_flow) { | |
4888 | return -1; | |
4889 | } | |
4890 | ||
faf71e49 | 4891 | return netdev_dpdk_destroy_rte_flow(netdev, ufid, rte_flow); |
e8a2b5bf FC |
4892 | } |
4893 | ||
c0af6425 IM |
4894 | #define DPDK_FLOW_OFFLOAD_API \ |
4895 | .flow_put = netdev_dpdk_flow_put, \ | |
4896 | .flow_del = netdev_dpdk_flow_del | |
4897 | ||
89c09c1c BP |
4898 | #define NETDEV_DPDK_CLASS_COMMON \ |
4899 | .is_pmd = true, \ | |
4900 | .alloc = netdev_dpdk_alloc, \ | |
4901 | .dealloc = netdev_dpdk_dealloc, \ | |
4902 | .get_config = netdev_dpdk_get_config, \ | |
4903 | .get_numa_id = netdev_dpdk_get_numa_id, \ | |
4904 | .set_etheraddr = netdev_dpdk_set_etheraddr, \ | |
4905 | .get_etheraddr = netdev_dpdk_get_etheraddr, \ | |
4906 | .get_mtu = netdev_dpdk_get_mtu, \ | |
4907 | .set_mtu = netdev_dpdk_set_mtu, \ | |
4908 | .get_ifindex = netdev_dpdk_get_ifindex, \ | |
4909 | .get_carrier_resets = netdev_dpdk_get_carrier_resets, \ | |
4910 | .set_miimon_interval = netdev_dpdk_set_miimon, \ | |
4911 | .set_policing = netdev_dpdk_set_policing, \ | |
4912 | .get_qos_types = netdev_dpdk_get_qos_types, \ | |
4913 | .get_qos = netdev_dpdk_get_qos, \ | |
4914 | .set_qos = netdev_dpdk_set_qos, \ | |
4915 | .update_flags = netdev_dpdk_update_flags, \ | |
4916 | .rxq_alloc = netdev_dpdk_rxq_alloc, \ | |
4917 | .rxq_construct = netdev_dpdk_rxq_construct, \ | |
4918 | .rxq_destruct = netdev_dpdk_rxq_destruct, \ | |
c0af6425 | 4919 | .rxq_dealloc = netdev_dpdk_rxq_dealloc |
89c09c1c BP |
4920 | |
4921 | #define NETDEV_DPDK_CLASS_BASE \ | |
4922 | NETDEV_DPDK_CLASS_COMMON, \ | |
4923 | .init = netdev_dpdk_class_init, \ | |
4924 | .destruct = netdev_dpdk_destruct, \ | |
4925 | .set_tx_multiq = netdev_dpdk_set_tx_multiq, \ | |
4926 | .get_carrier = netdev_dpdk_get_carrier, \ | |
4927 | .get_stats = netdev_dpdk_get_stats, \ | |
4928 | .get_custom_stats = netdev_dpdk_get_custom_stats, \ | |
4929 | .get_features = netdev_dpdk_get_features, \ | |
4930 | .get_status = netdev_dpdk_get_status, \ | |
4931 | .reconfigure = netdev_dpdk_reconfigure, \ | |
c0af6425 IM |
4932 | .rxq_recv = netdev_dpdk_rxq_recv, \ |
4933 | DPDK_FLOW_OFFLOAD_API | |
89c09c1c BP |
4934 | |
4935 | static const struct netdev_class dpdk_class = { | |
4936 | .type = "dpdk", | |
4937 | NETDEV_DPDK_CLASS_BASE, | |
4938 | .construct = netdev_dpdk_construct, | |
4939 | .set_config = netdev_dpdk_set_config, | |
4940 | .send = netdev_dpdk_eth_send, | |
4941 | }; | |
4942 | ||
4943 | static const struct netdev_class dpdk_ring_class = { | |
4944 | .type = "dpdkr", | |
4945 | NETDEV_DPDK_CLASS_BASE, | |
4946 | .construct = netdev_dpdk_ring_construct, | |
4947 | .set_config = netdev_dpdk_ring_set_config, | |
4948 | .send = netdev_dpdk_ring_send, | |
4949 | }; | |
4950 | ||
4951 | static const struct netdev_class dpdk_vhost_class = { | |
4952 | .type = "dpdkvhostuser", | |
4953 | NETDEV_DPDK_CLASS_COMMON, | |
4954 | .construct = netdev_dpdk_vhost_construct, | |
4955 | .destruct = netdev_dpdk_vhost_destruct, | |
4956 | .send = netdev_dpdk_vhost_send, | |
4957 | .get_carrier = netdev_dpdk_vhost_get_carrier, | |
4958 | .get_stats = netdev_dpdk_vhost_get_stats, | |
4959 | .get_status = netdev_dpdk_vhost_user_get_status, | |
4960 | .reconfigure = netdev_dpdk_vhost_reconfigure, | |
4961 | .rxq_recv = netdev_dpdk_vhost_rxq_recv | |
4962 | }; | |
4963 | ||
4964 | static const struct netdev_class dpdk_vhost_client_class = { | |
4965 | .type = "dpdkvhostuserclient", | |
4966 | NETDEV_DPDK_CLASS_COMMON, | |
4967 | .construct = netdev_dpdk_vhost_client_construct, | |
4968 | .destruct = netdev_dpdk_vhost_destruct, | |
4969 | .set_config = netdev_dpdk_vhost_client_set_config, | |
4970 | .send = netdev_dpdk_vhost_send, | |
4971 | .get_carrier = netdev_dpdk_vhost_get_carrier, | |
4972 | .get_stats = netdev_dpdk_vhost_get_stats, | |
4973 | .get_status = netdev_dpdk_vhost_user_get_status, | |
4974 | .reconfigure = netdev_dpdk_vhost_client_reconfigure, | |
4975 | .rxq_recv = netdev_dpdk_vhost_rxq_recv | |
4976 | }; | |
95fb793a | 4977 | |
8a9562d2 PS |
4978 | void |
4979 | netdev_dpdk_register(void) | |
4980 | { | |
bab69409 AC |
4981 | netdev_register_provider(&dpdk_class); |
4982 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 4983 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 4984 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 4985 | } |