]>
Commit | Line | Data |
---|---|---|
8a9562d2 | 1 | /* |
12d0d124 | 2 | * Copyright (c) 2014, 2015, 2016, 2017 Nicira, Inc. |
8a9562d2 PS |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
01961bbd | 18 | #include "netdev-dpdk.h" |
8a9562d2 | 19 | |
8a9562d2 PS |
20 | #include <string.h> |
21 | #include <signal.h> | |
22 | #include <stdlib.h> | |
8a9562d2 | 23 | #include <errno.h> |
8a9562d2 | 24 | #include <unistd.h> |
f3e7ec25 MW |
25 | #include <linux/virtio_net.h> |
26 | #include <sys/socket.h> | |
27 | #include <linux/if.h> | |
01961bbd DDP |
28 | |
29 | #include <rte_config.h> | |
30 | #include <rte_cycles.h> | |
31 | #include <rte_errno.h> | |
32 | #include <rte_eth_ring.h> | |
33 | #include <rte_ethdev.h> | |
34 | #include <rte_malloc.h> | |
35 | #include <rte_mbuf.h> | |
36 | #include <rte_meter.h> | |
fc56f5e0 | 37 | #include <rte_pci.h> |
f3e7ec25 | 38 | #include <rte_vhost.h> |
8a9562d2 | 39 | |
7d1ced01 | 40 | #include "dirs.h" |
e14deea0 | 41 | #include "dp-packet.h" |
01961bbd | 42 | #include "dpdk.h" |
8a9562d2 | 43 | #include "dpif-netdev.h" |
e5c0f5a4 | 44 | #include "fatal-signal.h" |
8a9562d2 PS |
45 | #include "netdev-provider.h" |
46 | #include "netdev-vport.h" | |
47 | #include "odp-util.h" | |
eac84432 | 48 | #include "openvswitch/dynamic-string.h" |
25d436fb BW |
49 | #include "openvswitch/list.h" |
50 | #include "openvswitch/ofp-print.h" | |
51 | #include "openvswitch/vlog.h" | |
94143fc4 | 52 | #include "ovs-numa.h" |
8a9562d2 PS |
53 | #include "ovs-thread.h" |
54 | #include "ovs-rcu.h" | |
55 | #include "packets.h" | |
ee89ea7b | 56 | #include "openvswitch/shash.h" |
0bf765f7 | 57 | #include "smap.h" |
8a9562d2 PS |
58 | #include "sset.h" |
59 | #include "unaligned.h" | |
60 | #include "timeval.h" | |
61 | #include "unixctl.h" | |
8a9562d2 | 62 | |
f3e7ec25 MW |
63 | enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM}; |
64 | ||
05b49df6 | 65 | VLOG_DEFINE_THIS_MODULE(netdev_dpdk); |
8a9562d2 PS |
66 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
67 | ||
68 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
69 | ||
70 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
71 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
72 | ||
73 | /* | |
74 | * need to reserve tons of extra space in the mbufs so we can align the | |
75 | * DMA addresses to 4KB. | |
18f777b2 TP |
76 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
77 | * performance for standard Ethernet MTU. | |
8a9562d2 | 78 | */ |
58be5c0e MK |
79 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN \ |
80 | + (2 * VLAN_HEADER_LEN)) | |
4be4d22c MK |
81 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) |
82 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
58be5c0e MK |
83 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len) \ |
84 | - ETHER_HDR_LEN - ETHER_CRC_LEN) | |
31b88c97 SS |
85 | #define MBUF_SIZE(mtu) ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) \ |
86 | + sizeof(struct dp_packet) \ | |
87 | + RTE_PKTMBUF_HEADROOM), \ | |
88 | RTE_CACHE_LINE_SIZE) | |
4be4d22c | 89 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 90 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 91 | |
da79ce2b DDP |
92 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
93 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
94 | * enough hugepages) we keep halving the number until the allocation succeeds | |
95 | * or we reach MIN_NB_MBUF */ | |
96 | ||
97 | #define MAX_NB_MBUF (4096 * 64) | |
98 | #define MIN_NB_MBUF (4096 * 4) | |
99 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
100 | ||
101 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
102 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
103 | ||
104 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
105 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
106 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
107 | % MP_CACHE_SZ == 0); | |
108 | ||
d6e3feb5 | 109 | /* |
110 | * DPDK XSTATS Counter names definition | |
111 | */ | |
112 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
113 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
114 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
115 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
116 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
117 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
118 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
119 | ||
120 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
121 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
122 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
123 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
124 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
125 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
126 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
127 | ||
d57f777f | 128 | #define XSTAT_RX_MULTICAST_PACKETS "rx_multicast_packets" |
d6e3feb5 | 129 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" |
130 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
131 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
132 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
133 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
134 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
135 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
136 | ||
8a9562d2 PS |
137 | #define SOCKET0 0 |
138 | ||
b685696b CL |
139 | /* Default size of Physical NIC RXQ */ |
140 | #define NIC_PORT_DEFAULT_RXQ_SIZE 2048 | |
141 | /* Default size of Physical NIC TXQ */ | |
142 | #define NIC_PORT_DEFAULT_TXQ_SIZE 2048 | |
143 | /* Maximum size of Physical NIC Queues */ | |
144 | #define NIC_PORT_MAX_Q_SIZE 4096 | |
79f5354c | 145 | |
585a5bea | 146 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
147 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
148 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
149 | * yet mapped to another queue. */ | |
585a5bea | 150 | |
bb37956a IM |
151 | #define DPDK_ETH_PORT_ID_INVALID RTE_MAX_ETHPORTS |
152 | ||
153 | /* DPDK library uses uint8_t for port_id. */ | |
154 | typedef uint8_t dpdk_port_t; | |
155 | ||
31871ee3 | 156 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 157 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 158 | |
8a9562d2 | 159 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
160 | .rxmode = { |
161 | .mq_mode = ETH_MQ_RX_RSS, | |
162 | .split_hdr_size = 0, | |
163 | .header_split = 0, /* Header Split disabled */ | |
164 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
165 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
166 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
167 | .hw_strip_crc = 0, | |
168 | }, | |
169 | .rx_adv_conf = { | |
170 | .rss_conf = { | |
171 | .rss_key = NULL, | |
543342a4 | 172 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 173 | }, |
a28ddd11 DDP |
174 | }, |
175 | .txmode = { | |
176 | .mq_mode = ETH_MQ_TX_NONE, | |
177 | }, | |
8a9562d2 PS |
178 | }; |
179 | ||
f3e7ec25 MW |
180 | /* |
181 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
182 | * configuration has been fully completed. | |
183 | */ | |
184 | static int new_device(int vid); | |
185 | static void destroy_device(int vid); | |
186 | static int vring_state_changed(int vid, uint16_t queue_id, int enable); | |
187 | static const struct vhost_device_ops virtio_net_device_ops = | |
188 | { | |
189 | .new_device = new_device, | |
190 | .destroy_device = destroy_device, | |
191 | .vring_state_changed = vring_state_changed, | |
192 | .features_changed = NULL | |
193 | }; | |
194 | ||
58f7c37b DDP |
195 | enum { DPDK_RING_SIZE = 256 }; |
196 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
197 | enum { DRAIN_TSC = 200000ULL }; |
198 | ||
58397e6c KT |
199 | enum dpdk_dev_type { |
200 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 201 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
202 | }; |
203 | ||
0bf765f7 IS |
204 | /* Quality of Service */ |
205 | ||
206 | /* An instance of a QoS configuration. Always associated with a particular | |
207 | * network device. | |
208 | * | |
209 | * Each QoS implementation subclasses this with whatever additional data it | |
210 | * needs. | |
211 | */ | |
212 | struct qos_conf { | |
213 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 214 | rte_spinlock_t lock; |
0bf765f7 IS |
215 | }; |
216 | ||
217 | /* A particular implementation of dpdk QoS operations. | |
218 | * | |
219 | * The functions below return 0 if successful or a positive errno value on | |
220 | * failure, except where otherwise noted. All of them must be provided, except | |
221 | * where otherwise noted. | |
222 | */ | |
223 | struct dpdk_qos_ops { | |
224 | ||
225 | /* Name of the QoS type */ | |
226 | const char *qos_name; | |
227 | ||
78bd47cf DDP |
228 | /* Called to construct a qos_conf object. The implementation should make |
229 | * the appropriate calls to configure QoS according to 'details'. | |
0bf765f7 IS |
230 | * |
231 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
232 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
233 | * (which is built as ovs-vswitchd.conf.db(8)). | |
234 | * | |
78bd47cf DDP |
235 | * This function must return 0 if and only if it sets '*conf' to an |
236 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
237 | * |
238 | * For all QoS implementations it should always be non-null. | |
239 | */ | |
78bd47cf | 240 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
241 | |
242 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 243 | * 'qos_conf'. |
0bf765f7 IS |
244 | * |
245 | * For all QoS implementations it should always be non-null. | |
246 | */ | |
78bd47cf | 247 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 248 | |
78bd47cf | 249 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
250 | * |
251 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
252 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
253 | * (which is built as ovs-vswitchd.conf.db(8)). | |
254 | */ | |
78bd47cf | 255 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 256 | |
78bd47cf | 257 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
258 | * |
259 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
260 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
261 | * (which is built as ovs-vswitchd.conf.db(8)). | |
262 | * | |
78bd47cf | 263 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 264 | */ |
78bd47cf DDP |
265 | bool (*qos_is_equal)(const struct qos_conf *conf, |
266 | const struct smap *details); | |
0bf765f7 IS |
267 | |
268 | /* Modify an array of rte_mbufs. The modification is specific to | |
269 | * each qos implementation. | |
270 | * | |
271 | * The function should take and array of mbufs and an int representing | |
272 | * the current number of mbufs present in the array. | |
273 | * | |
274 | * After the function has performed a qos modification to the array of | |
275 | * mbufs it returns an int representing the number of mbufs now present in | |
276 | * the array. This value is can then be passed to the port send function | |
277 | * along with the modified array for transmission. | |
278 | * | |
279 | * For all QoS implementations it should always be non-null. | |
280 | */ | |
78bd47cf DDP |
281 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
282 | int pkt_cnt); | |
0bf765f7 IS |
283 | }; |
284 | ||
285 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
286 | static const struct dpdk_qos_ops egress_policer_ops; | |
287 | ||
288 | /* | |
289 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
290 | * operations. | |
291 | */ | |
292 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
293 | &egress_policer_ops, | |
294 | NULL | |
295 | }; | |
296 | ||
c2adb102 IM |
297 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
298 | ||
8a9562d2 | 299 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 300 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 301 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 302 | |
c2adb102 IM |
303 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
304 | = OVS_MUTEX_INITIALIZER; | |
305 | ||
306 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mp_mutex) | |
55951e15 | 307 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 308 | |
8a9562d2 PS |
309 | struct dpdk_mp { |
310 | struct rte_mempool *mp; | |
311 | int mtu; | |
312 | int socket_id; | |
313 | int refcount; | |
c2adb102 | 314 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mp_mutex); |
8a9562d2 PS |
315 | }; |
316 | ||
5a034064 AW |
317 | /* There should be one 'struct dpdk_tx_queue' created for |
318 | * each cpu core. */ | |
8a9562d2 | 319 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
320 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
321 | * from concurrent access. It is used only | |
322 | * if the queue is shared among different | |
324c8374 | 323 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
324 | int map; /* Mapping of configured vhost-user queues |
325 | * to enabled by guest. */ | |
8a9562d2 PS |
326 | }; |
327 | ||
95fb793a | 328 | /* dpdk has no way to remove dpdk ring ethernet devices |
329 | so we have to keep them around once they've been created | |
330 | */ | |
331 | ||
ca6ba700 | 332 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 333 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 334 | |
335 | struct dpdk_ring { | |
336 | /* For the client rings */ | |
337 | struct rte_ring *cring_tx; | |
338 | struct rte_ring *cring_rx; | |
b83a2df1 | 339 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
bb37956a | 340 | dpdk_port_t eth_port_id; /* ethernet device port id */ |
ca6ba700 | 341 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 342 | }; |
343 | ||
9509913a IS |
344 | struct ingress_policer { |
345 | struct rte_meter_srtcm_params app_srtcm_params; | |
346 | struct rte_meter_srtcm in_policer; | |
347 | rte_spinlock_t policer_lock; | |
348 | }; | |
349 | ||
1a2bb118 SC |
350 | enum dpdk_hw_ol_features { |
351 | NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0, | |
352 | }; | |
353 | ||
8a9562d2 PS |
354 | struct netdev_dpdk { |
355 | struct netdev up; | |
bb37956a | 356 | dpdk_port_t port_id; |
8a9562d2 | 357 | int max_packet_len; |
58397e6c | 358 | enum dpdk_dev_type type; |
8a9562d2 | 359 | |
5a034064 | 360 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
361 | |
362 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
363 | ||
364 | struct dpdk_mp *dpdk_mp; | |
365 | int mtu; | |
366 | int socket_id; | |
367 | int buf_size; | |
8a9562d2 | 368 | struct netdev_stats stats; |
45d947c4 DDP |
369 | /* Protects stats */ |
370 | rte_spinlock_t stats_lock; | |
8a9562d2 | 371 | |
74ff3298 | 372 | struct eth_addr hwaddr; |
8a9562d2 PS |
373 | enum netdev_flags flags; |
374 | ||
375 | struct rte_eth_link link; | |
376 | int link_reset_cnt; | |
377 | ||
0a0f39df CL |
378 | /* virtio identifier for vhost devices */ |
379 | ovsrcu_index vid; | |
380 | ||
381 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
382 | bool vhost_reconfigured; | |
58397e6c | 383 | |
2d24d165 CL |
384 | /* Identifier used to distinguish vhost devices from each other. */ |
385 | char vhost_id[PATH_MAX]; | |
7d1ced01 | 386 | |
55e075e6 CL |
387 | /* Device arguments for dpdk ports */ |
388 | char *devargs; | |
389 | ||
5dcde09c IM |
390 | /* If true, device was attached by rte_eth_dev_attach(). */ |
391 | bool attached; | |
392 | ||
8a9562d2 | 393 | /* In dpdk_list. */ |
ca6ba700 | 394 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
0bf765f7 IS |
395 | |
396 | /* QoS configuration and lock for the device */ | |
78bd47cf | 397 | OVSRCU_TYPE(struct qos_conf *) qos_conf; |
0bf765f7 | 398 | |
050c60bf DDP |
399 | /* The following properties cannot be changed when a device is running, |
400 | * so we remember the request and update them next time | |
401 | * netdev_dpdk*_reconfigure() is called */ | |
0072e931 | 402 | int requested_mtu; |
050c60bf DDP |
403 | int requested_n_txq; |
404 | int requested_n_rxq; | |
b685696b CL |
405 | int requested_rxq_size; |
406 | int requested_txq_size; | |
407 | ||
408 | /* Number of rx/tx descriptors for physical devices */ | |
409 | int rxq_size; | |
410 | int txq_size; | |
9509913a | 411 | |
db8f13b0 CL |
412 | /* Socket ID detected when vHost device is brought up */ |
413 | int requested_socket_id; | |
414 | ||
c1ff66ac CL |
415 | /* Denotes whether vHost port is client/server mode */ |
416 | uint64_t vhost_driver_flags; | |
417 | ||
9509913a IS |
418 | /* Ingress Policer */ |
419 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
420 | uint32_t policer_rate; | |
421 | uint32_t policer_burst; | |
9fd39370 SC |
422 | |
423 | /* DPDK-ETH Flow control */ | |
424 | struct rte_eth_fc_conf fc_conf; | |
1a2bb118 SC |
425 | |
426 | /* DPDK-ETH hardware offload features, | |
427 | * from the enum set 'dpdk_hw_ol_features' */ | |
428 | uint32_t hw_ol_features; | |
8a9562d2 PS |
429 | }; |
430 | ||
431 | struct netdev_rxq_dpdk { | |
432 | struct netdev_rxq up; | |
bb37956a | 433 | dpdk_port_t port_id; |
8a9562d2 PS |
434 | }; |
435 | ||
f3e7ec25 MW |
436 | static void netdev_dpdk_destruct(struct netdev *netdev); |
437 | static void netdev_dpdk_vhost_destruct(struct netdev *netdev); | |
8a9562d2 | 438 | |
0a0f39df | 439 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 440 | |
9509913a IS |
441 | struct ingress_policer * |
442 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
443 | ||
8a9562d2 PS |
444 | static bool |
445 | is_dpdk_class(const struct netdev_class *class) | |
446 | { | |
f3e7ec25 MW |
447 | return class->destruct == netdev_dpdk_destruct |
448 | || class->destruct == netdev_dpdk_vhost_destruct; | |
8a9562d2 PS |
449 | } |
450 | ||
4be4d22c MK |
451 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
452 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
453 | * value, insufficient buffers are allocated to accomodate the packet in its | |
454 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
455 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
456 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
457 | * behaviour, which reduces performance. To prevent this, use a buffer size |
458 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
459 | */ |
460 | static uint32_t | |
461 | dpdk_buf_size(int mtu) | |
462 | { | |
463 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
464 | NETDEV_DPDK_MBUF_ALIGN); | |
465 | } | |
466 | ||
eff23640 DDP |
467 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
468 | * | |
469 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
470 | static void * |
471 | dpdk_rte_mzalloc(size_t sz) | |
472 | { | |
eff23640 | 473 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
474 | } |
475 | ||
476 | void | |
e14deea0 | 477 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 478 | { |
db73f716 | 479 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 480 | |
b00b4a81 | 481 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
482 | } |
483 | ||
b3cd9f9d | 484 | static void |
401b70d6 | 485 | ovs_rte_pktmbuf_init(struct rte_mempool *mp OVS_UNUSED, |
b3cd9f9d | 486 | void *opaque_arg OVS_UNUSED, |
2391135c | 487 | void *_p, |
b3cd9f9d PS |
488 | unsigned i OVS_UNUSED) |
489 | { | |
2391135c | 490 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 491 | |
2391135c | 492 | dp_packet_init_dpdk((struct dp_packet *) pkt, pkt->buf_len); |
b3cd9f9d PS |
493 | } |
494 | ||
8a9562d2 | 495 | static struct dpdk_mp * |
2ae3d542 | 496 | dpdk_mp_create(int socket_id, int mtu) |
8a9562d2 | 497 | { |
2ae3d542 | 498 | struct dpdk_mp *dmp; |
da79ce2b | 499 | unsigned mp_size; |
0c6f39e5 | 500 | char *mp_name; |
8a9562d2 PS |
501 | |
502 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
eff23640 DDP |
503 | if (!dmp) { |
504 | return NULL; | |
505 | } | |
8a9562d2 PS |
506 | dmp->socket_id = socket_id; |
507 | dmp->mtu = mtu; | |
508 | dmp->refcount = 1; | |
0072e931 | 509 | /* XXX: this is a really rough method of provisioning memory. |
2ae3d542 | 510 | * It's impossible to determine what the exact memory requirements are |
58be5c0e MK |
511 | * when the number of ports and rxqs that utilize a particular mempool can |
512 | * change dynamically at runtime. For now, use this rough heurisitic. | |
0072e931 MK |
513 | */ |
514 | if (mtu >= ETHER_MTU) { | |
515 | mp_size = MAX_NB_MBUF; | |
516 | } else { | |
517 | mp_size = MIN_NB_MBUF; | |
518 | } | |
8a9562d2 | 519 | |
da79ce2b | 520 | do { |
0c6f39e5 DDP |
521 | mp_name = xasprintf("ovs_mp_%d_%d_%u", dmp->mtu, dmp->socket_id, |
522 | mp_size); | |
95fb793a | 523 | |
401b70d6 HA |
524 | dmp->mp = rte_pktmbuf_pool_create(mp_name, mp_size, |
525 | MP_CACHE_SZ, | |
526 | sizeof (struct dp_packet) | |
527 | - sizeof (struct rte_mbuf), | |
528 | MBUF_SIZE(mtu) | |
529 | - sizeof(struct dp_packet), | |
530 | socket_id); | |
0c6f39e5 DDP |
531 | if (dmp->mp) { |
532 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", | |
533 | mp_name, mp_size); | |
534 | } | |
535 | free(mp_name); | |
536 | if (dmp->mp) { | |
401b70d6 HA |
537 | /* rte_pktmbuf_pool_create has done some initialization of the |
538 | * rte_mbuf part of each dp_packet, while ovs_rte_pktmbuf_init | |
539 | * initializes some OVS specific fields of dp_packet. | |
540 | */ | |
541 | rte_mempool_obj_iter(dmp->mp, ovs_rte_pktmbuf_init, NULL); | |
0c6f39e5 DDP |
542 | return dmp; |
543 | } | |
544 | } while (rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
2ae3d542 | 545 | |
2ae3d542 DDP |
546 | rte_free(dmp); |
547 | return NULL; | |
548 | } | |
549 | ||
550 | static struct dpdk_mp * | |
551 | dpdk_mp_get(int socket_id, int mtu) | |
552 | { | |
553 | struct dpdk_mp *dmp; | |
554 | ||
555 | ovs_mutex_lock(&dpdk_mp_mutex); | |
556 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
557 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
558 | dmp->refcount++; | |
559 | goto out; | |
560 | } | |
561 | } | |
562 | ||
563 | dmp = dpdk_mp_create(socket_id, mtu); | |
96e9b168 IS |
564 | if (dmp) { |
565 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); | |
566 | } | |
5f88de0d | 567 | |
c2adb102 IM |
568 | out: |
569 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
570 | ||
c2adb102 | 571 | return dmp; |
8a9562d2 PS |
572 | } |
573 | ||
574 | static void | |
c2adb102 | 575 | dpdk_mp_put(struct dpdk_mp *dmp) |
8a9562d2 | 576 | { |
8a9562d2 PS |
577 | if (!dmp) { |
578 | return; | |
579 | } | |
580 | ||
c2adb102 | 581 | ovs_mutex_lock(&dpdk_mp_mutex); |
8d38823b | 582 | ovs_assert(dmp->refcount); |
8a9562d2 | 583 | |
8d38823b MK |
584 | if (!--dmp->refcount) { |
585 | ovs_list_remove(&dmp->list_node); | |
586 | rte_mempool_free(dmp->mp); | |
5f88de0d | 587 | rte_free(dmp); |
8a9562d2 | 588 | } |
c2adb102 | 589 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
590 | } |
591 | ||
0072e931 MK |
592 | /* Tries to allocate new mempool on requested_socket_id with |
593 | * mbuf size corresponding to requested_mtu. | |
594 | * On success new configuration will be applied. | |
595 | * On error, device will be left unchanged. */ | |
596 | static int | |
597 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
598 | OVS_REQUIRES(dev->mutex) |
599 | { | |
600 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
601 | struct dpdk_mp *mp; | |
602 | ||
603 | mp = dpdk_mp_get(dev->requested_socket_id, FRAME_LEN_TO_MTU(buf_size)); | |
604 | if (!mp) { | |
c67e46c0 MK |
605 | VLOG_ERR("Failed to create memory pool for netdev " |
606 | "%s, with MTU %d on socket %d: %s\n", | |
607 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
608 | rte_strerror(rte_errno)); | |
609 | return rte_errno; | |
0072e931 MK |
610 | } else { |
611 | dpdk_mp_put(dev->dpdk_mp); | |
612 | dev->dpdk_mp = mp; | |
613 | dev->mtu = dev->requested_mtu; | |
614 | dev->socket_id = dev->requested_socket_id; | |
615 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
616 | } | |
617 | ||
618 | return 0; | |
619 | } | |
620 | ||
8a9562d2 PS |
621 | static void |
622 | check_link_status(struct netdev_dpdk *dev) | |
623 | { | |
624 | struct rte_eth_link link; | |
625 | ||
626 | rte_eth_link_get_nowait(dev->port_id, &link); | |
627 | ||
628 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 629 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
630 | |
631 | dev->link_reset_cnt++; | |
632 | dev->link = link; | |
633 | if (dev->link.link_status) { | |
bb37956a | 634 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Up - speed %u Mbps - %s", |
58be5c0e | 635 | dev->port_id, (unsigned) dev->link.link_speed, |
8a9562d2 PS |
636 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? |
637 | ("full-duplex") : ("half-duplex")); | |
638 | } else { | |
bb37956a | 639 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Down", dev->port_id); |
8a9562d2 PS |
640 | } |
641 | } | |
642 | } | |
643 | ||
644 | static void * | |
645 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
646 | { | |
647 | struct netdev_dpdk *dev; | |
648 | ||
649 | pthread_detach(pthread_self()); | |
650 | ||
651 | for (;;) { | |
652 | ovs_mutex_lock(&dpdk_mutex); | |
653 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
654 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
655 | if (dev->type == DPDK_DEV_ETH) { |
656 | check_link_status(dev); | |
657 | } | |
8a9562d2 PS |
658 | ovs_mutex_unlock(&dev->mutex); |
659 | } | |
660 | ovs_mutex_unlock(&dpdk_mutex); | |
661 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
662 | } | |
663 | ||
664 | return NULL; | |
665 | } | |
666 | ||
b98d7669 DDP |
667 | static int |
668 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
669 | { | |
670 | int diag = 0; | |
671 | int i; | |
0072e931 | 672 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 673 | |
67fe6d63 MK |
674 | /* For some NICs (e.g. Niantic), scatter_rx mode needs to be explicitly |
675 | * enabled. */ | |
0072e931 | 676 | if (dev->mtu > ETHER_MTU) { |
67fe6d63 | 677 | conf.rxmode.enable_scatter = 1; |
0072e931 | 678 | } |
67fe6d63 | 679 | |
1a2bb118 SC |
680 | conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & |
681 | NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
b98d7669 DDP |
682 | /* A device may report more queues than it makes available (this has |
683 | * been observed for Intel xl710, which reserves some of them for | |
684 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
685 | * available. When this happens we can retry the configuration | |
686 | * and request less queues */ | |
687 | while (n_rxq && n_txq) { | |
688 | if (diag) { | |
689 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
690 | } | |
691 | ||
0072e931 | 692 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 693 | if (diag) { |
0072e931 MK |
694 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
695 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
696 | break; |
697 | } | |
698 | ||
67fe6d63 MK |
699 | diag = rte_eth_dev_set_mtu(dev->port_id, dev->mtu); |
700 | if (diag) { | |
701 | VLOG_ERR("Interface %s MTU (%d) setup error: %s", | |
702 | dev->up.name, dev->mtu, rte_strerror(-diag)); | |
703 | break; | |
704 | } | |
705 | ||
b98d7669 | 706 | for (i = 0; i < n_txq; i++) { |
b685696b | 707 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
708 | dev->socket_id, NULL); |
709 | if (diag) { | |
710 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
711 | dev->up.name, i, rte_strerror(-diag)); | |
712 | break; | |
713 | } | |
714 | } | |
715 | ||
716 | if (i != n_txq) { | |
717 | /* Retry with less tx queues */ | |
718 | n_txq = i; | |
719 | continue; | |
720 | } | |
721 | ||
722 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 723 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
b98d7669 DDP |
724 | dev->socket_id, NULL, |
725 | dev->dpdk_mp->mp); | |
726 | if (diag) { | |
727 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
728 | dev->up.name, i, rte_strerror(-diag)); | |
729 | break; | |
730 | } | |
731 | } | |
732 | ||
733 | if (i != n_rxq) { | |
734 | /* Retry with less rx queues */ | |
735 | n_rxq = i; | |
736 | continue; | |
737 | } | |
738 | ||
739 | dev->up.n_rxq = n_rxq; | |
81acebda | 740 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
741 | |
742 | return 0; | |
743 | } | |
744 | ||
745 | return diag; | |
746 | } | |
747 | ||
9fd39370 SC |
748 | static void |
749 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
750 | { | |
751 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
bb37956a IM |
752 | VLOG_WARN("Failed to enable flow control on device %"PRIu8, |
753 | dev->port_id); | |
9fd39370 SC |
754 | } |
755 | } | |
b98d7669 | 756 | |
8a9562d2 | 757 | static int |
c2adb102 IM |
758 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
759 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
760 | { |
761 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 762 | struct rte_eth_dev_info info; |
8a9562d2 PS |
763 | struct ether_addr eth_addr; |
764 | int diag; | |
b98d7669 | 765 | int n_rxq, n_txq; |
d4f5282c KT |
766 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | |
767 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
768 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
8a9562d2 | 769 | |
a0cb2d66 | 770 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 771 | |
d4f5282c KT |
772 | if ((info.rx_offload_capa & rx_chksm_offload_capa) != |
773 | rx_chksm_offload_capa) { | |
2cfe866f | 774 | VLOG_WARN("Rx checksum offload is not supported on port %"PRIu8, |
d4f5282c KT |
775 | dev->port_id); |
776 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; | |
777 | } else { | |
778 | dev->hw_ol_features |= NETDEV_RX_CHECKSUM_OFFLOAD; | |
779 | } | |
780 | ||
b98d7669 DDP |
781 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
782 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
783 | ||
784 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 785 | if (diag) { |
b98d7669 DDP |
786 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
787 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 788 | return -diag; |
8a9562d2 PS |
789 | } |
790 | ||
8a9562d2 PS |
791 | diag = rte_eth_dev_start(dev->port_id); |
792 | if (diag) { | |
b98d7669 DDP |
793 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
794 | rte_strerror(-diag)); | |
95fb793a | 795 | return -diag; |
8a9562d2 PS |
796 | } |
797 | ||
798 | rte_eth_promiscuous_enable(dev->port_id); | |
799 | rte_eth_allmulticast_enable(dev->port_id); | |
800 | ||
801 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
802 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
bb37956a | 803 | VLOG_INFO_RL(&rl, "Port %"PRIu8": "ETH_ADDR_FMT, |
ca92d173 | 804 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 805 | |
ca92d173 | 806 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
807 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
808 | ||
809 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
810 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
811 | ||
9fd39370 SC |
812 | /* Get the Flow control configuration for DPDK-ETH */ |
813 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
814 | if (diag) { | |
bb37956a | 815 | VLOG_DBG("cannot get flow control parameters on port=%"PRIu8", err=%d", |
9fd39370 SC |
816 | dev->port_id, diag); |
817 | } | |
818 | ||
8a9562d2 PS |
819 | return 0; |
820 | } | |
821 | ||
822 | static struct netdev_dpdk * | |
823 | netdev_dpdk_cast(const struct netdev *netdev) | |
824 | { | |
825 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
826 | } | |
827 | ||
828 | static struct netdev * | |
829 | netdev_dpdk_alloc(void) | |
830 | { | |
bab69409 AC |
831 | struct netdev_dpdk *dev; |
832 | ||
65e19e70 DDP |
833 | dev = dpdk_rte_mzalloc(sizeof *dev); |
834 | if (dev) { | |
835 | return &dev->up; | |
bab69409 | 836 | } |
65e19e70 | 837 | |
bab69409 | 838 | return NULL; |
8a9562d2 PS |
839 | } |
840 | ||
eff23640 DDP |
841 | static struct dpdk_tx_queue * |
842 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 843 | { |
eff23640 | 844 | struct dpdk_tx_queue *txqs; |
bd5131ba | 845 | unsigned i; |
5a034064 | 846 | |
eff23640 DDP |
847 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
848 | if (txqs) { | |
849 | for (i = 0; i < n_txqs; i++) { | |
850 | /* Initialize map for vhost devices. */ | |
851 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
852 | rte_spinlock_init(&txqs[i].tx_lock); | |
853 | } | |
5a034064 | 854 | } |
eff23640 DDP |
855 | |
856 | return txqs; | |
5a034064 AW |
857 | } |
858 | ||
8a9562d2 | 859 | static int |
bb37956a | 860 | common_construct(struct netdev *netdev, dpdk_port_t port_no, |
1ce30dfd | 861 | enum dpdk_dev_type type, int socket_id) |
5a034064 | 862 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 863 | { |
d46285a2 | 864 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 865 | |
d46285a2 | 866 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 867 | |
d46285a2 | 868 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 869 | |
1b7a04e0 AW |
870 | /* If the 'sid' is negative, it means that the kernel fails |
871 | * to obtain the pci numa info. In that situation, always | |
872 | * use 'SOCKET0'. */ | |
1ce30dfd | 873 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 874 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
875 | dev->port_id = port_no; |
876 | dev->type = type; | |
877 | dev->flags = 0; | |
7f381c2e | 878 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 879 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
880 | ovsrcu_index_init(&dev->vid, -1); |
881 | dev->vhost_reconfigured = false; | |
5dcde09c | 882 | dev->attached = false; |
8a9562d2 | 883 | |
78bd47cf | 884 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 885 | |
9509913a IS |
886 | ovsrcu_init(&dev->ingress_policer, NULL); |
887 | dev->policer_rate = 0; | |
888 | dev->policer_burst = 0; | |
889 | ||
7f381c2e DDP |
890 | netdev->n_rxq = 0; |
891 | netdev->n_txq = 0; | |
892 | dev->requested_n_rxq = NR_QUEUE; | |
893 | dev->requested_n_txq = NR_QUEUE; | |
894 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
895 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 896 | |
9fd39370 SC |
897 | /* Initialize the flow control to NULL */ |
898 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
899 | |
900 | /* Initilize the hardware offload flags to 0 */ | |
901 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
902 | |
903 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
904 | ||
d46285a2 | 905 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 906 | |
7f381c2e DDP |
907 | netdev_request_reconfigure(netdev); |
908 | ||
1ce30dfd | 909 | return 0; |
95fb793a | 910 | } |
911 | ||
b83a2df1 MV |
912 | /* dev_name must be the prefix followed by a positive decimal number. |
913 | * (no leading + or - signs are allowed) */ | |
95fb793a | 914 | static int |
915 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
916 | unsigned int *port_no) | |
917 | { | |
918 | const char *cport; | |
919 | ||
920 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
921 | return ENODEV; | |
922 | } | |
923 | ||
924 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
925 | |
926 | if (str_to_uint(cport, 10, port_no)) { | |
927 | return 0; | |
928 | } else { | |
929 | return ENODEV; | |
930 | } | |
95fb793a | 931 | } |
932 | ||
1ce30dfd DDP |
933 | static int |
934 | vhost_common_construct(struct netdev *netdev) | |
935 | OVS_REQUIRES(dpdk_mutex) | |
936 | { | |
937 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
938 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
939 | ||
940 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); | |
941 | if (!dev->tx_q) { | |
942 | return ENOMEM; | |
943 | } | |
944 | ||
bb37956a IM |
945 | return common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
946 | DPDK_DEV_VHOST, socket_id); | |
1ce30dfd DDP |
947 | } |
948 | ||
7d1ced01 | 949 | static int |
53f50d24 | 950 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 951 | { |
d46285a2 DDP |
952 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
953 | const char *name = netdev->name; | |
7d1ced01 | 954 | int err; |
a0cb2d66 | 955 | |
1af27e8a DDP |
956 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
957 | * the file system. '/' or '\' would traverse directories, so they're not | |
958 | * acceptable in 'name'. */ | |
959 | if (strchr(name, '/') || strchr(name, '\\')) { | |
960 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
961 | "A valid name must not include '/' or '\\'", | |
962 | name); | |
963 | return EINVAL; | |
964 | } | |
965 | ||
7d1ced01 CL |
966 | ovs_mutex_lock(&dpdk_mutex); |
967 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 968 | * the socket is to be created, then register the socket. |
7d1ced01 | 969 | */ |
2d24d165 | 970 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 971 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 972 | |
2d24d165 CL |
973 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
974 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
975 | if (err) { |
976 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 977 | dev->vhost_id); |
f3e7ec25 | 978 | goto out; |
e5c0f5a4 | 979 | } else { |
2d24d165 CL |
980 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
981 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
982 | dev->vhost_id, name); | |
983 | } | |
f3e7ec25 MW |
984 | |
985 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
986 | &virtio_net_device_ops); | |
987 | if (err) { | |
988 | VLOG_ERR("rte_vhost_driver_callback_register failed for vhost user " | |
989 | "port: %s\n", name); | |
990 | goto out; | |
991 | } | |
992 | ||
993 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
994 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
995 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
996 | | 1ULL << VIRTIO_NET_F_CSUM); | |
997 | if (err) { | |
998 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
999 | "port: %s\n", name); | |
1000 | goto out; | |
1001 | } | |
1002 | ||
1003 | err = rte_vhost_driver_start(dev->vhost_id); | |
1004 | if (err) { | |
1005 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
1006 | "port: %s\n", name); | |
1007 | goto out; | |
1008 | } | |
1009 | ||
1ce30dfd | 1010 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1011 | if (err) { |
1012 | VLOG_ERR("vhost_common_construct failed for vhost user " | |
1013 | "port: %s\n", name); | |
1014 | } | |
2d24d165 | 1015 | |
f3e7ec25 | 1016 | out: |
2d24d165 | 1017 | ovs_mutex_unlock(&dpdk_mutex); |
28ca969e AC |
1018 | VLOG_WARN_ONCE("dpdkvhostuser ports are considered deprecated; " |
1019 | "please migrate to dpdkvhostuserclient ports."); | |
2d24d165 CL |
1020 | return err; |
1021 | } | |
1022 | ||
1023 | static int | |
1024 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
1025 | { | |
1026 | int err; | |
1027 | ||
2d24d165 | 1028 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 1029 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1030 | if (err) { |
1031 | VLOG_ERR("vhost_common_construct failed for vhost user client" | |
1032 | "port: %s\n", netdev->name); | |
1033 | } | |
7d1ced01 | 1034 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
1035 | return err; |
1036 | } | |
1037 | ||
95fb793a | 1038 | static int |
1039 | netdev_dpdk_construct(struct netdev *netdev) | |
1040 | { | |
95fb793a | 1041 | int err; |
1042 | ||
95fb793a | 1043 | ovs_mutex_lock(&dpdk_mutex); |
bb37956a IM |
1044 | err = common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1045 | DPDK_DEV_ETH, SOCKET0); | |
8a9562d2 PS |
1046 | ovs_mutex_unlock(&dpdk_mutex); |
1047 | return err; | |
1048 | } | |
1049 | ||
1ce30dfd DDP |
1050 | static void |
1051 | common_destruct(struct netdev_dpdk *dev) | |
1052 | OVS_REQUIRES(dpdk_mutex) | |
1053 | OVS_EXCLUDED(dev->mutex) | |
1054 | { | |
1055 | rte_free(dev->tx_q); | |
1056 | dpdk_mp_put(dev->dpdk_mp); | |
1057 | ||
1058 | ovs_list_remove(&dev->list_node); | |
1059 | free(ovsrcu_get_protected(struct ingress_policer *, | |
1060 | &dev->ingress_policer)); | |
1061 | ovs_mutex_destroy(&dev->mutex); | |
1062 | } | |
1063 | ||
8a9562d2 | 1064 | static void |
d46285a2 | 1065 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 1066 | { |
d46285a2 | 1067 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5dcde09c | 1068 | char devname[RTE_ETH_NAME_MAX_LEN]; |
8a9562d2 | 1069 | |
8d38823b | 1070 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1071 | |
8a9562d2 | 1072 | rte_eth_dev_stop(dev->port_id); |
5dcde09c IM |
1073 | |
1074 | if (dev->attached) { | |
1075 | rte_eth_dev_close(dev->port_id); | |
1076 | if (rte_eth_dev_detach(dev->port_id, devname) < 0) { | |
1077 | VLOG_ERR("Device '%s' can not be detached", dev->devargs); | |
1078 | } else { | |
0ee821c2 | 1079 | VLOG_INFO("Device '%s' has been detached", devname); |
5dcde09c IM |
1080 | } |
1081 | } | |
1082 | ||
55e075e6 | 1083 | free(dev->devargs); |
1ce30dfd | 1084 | common_destruct(dev); |
8d38823b | 1085 | |
8a9562d2 | 1086 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1087 | } |
8a9562d2 | 1088 | |
3f891bbe DDP |
1089 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1090 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1091 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1092 | static int | |
c1ff66ac CL |
1093 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1094 | char *vhost_id) | |
3f891bbe DDP |
1095 | OVS_EXCLUDED(dpdk_mutex) |
1096 | OVS_EXCLUDED(dev->mutex) | |
1097 | { | |
c1ff66ac | 1098 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1099 | } |
1100 | ||
58397e6c | 1101 | static void |
d46285a2 | 1102 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1103 | { |
d46285a2 | 1104 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1105 | char *vhost_id; |
58397e6c | 1106 | |
8d38823b | 1107 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1108 | |
c62da695 | 1109 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1110 | if (netdev_dpdk_get_vid(dev) >= 0 |
1111 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1112 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1113 | netdev->name); |
58be5c0e MK |
1114 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1115 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1116 | } |
1117 | ||
2d24d165 | 1118 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1119 | |
1ce30dfd DDP |
1120 | common_destruct(dev); |
1121 | ||
58397e6c | 1122 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1123 | |
569c26da | 1124 | if (!vhost_id[0]) { |
821b8664 IM |
1125 | goto out; |
1126 | } | |
1127 | ||
c1ff66ac | 1128 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1129 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1130 | netdev->name, vhost_id); | |
c1ff66ac CL |
1131 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1132 | /* OVS server mode - remove this socket from list for deletion */ | |
1133 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1134 | } |
821b8664 | 1135 | out: |
c1ff66ac | 1136 | free(vhost_id); |
8a9562d2 PS |
1137 | } |
1138 | ||
1139 | static void | |
d46285a2 | 1140 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1141 | { |
d46285a2 | 1142 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1143 | |
d46285a2 | 1144 | rte_free(dev); |
8a9562d2 PS |
1145 | } |
1146 | ||
1147 | static int | |
a14b8947 | 1148 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1149 | { |
a14b8947 | 1150 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1151 | |
1152 | ovs_mutex_lock(&dev->mutex); | |
1153 | ||
050c60bf | 1154 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1155 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1156 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1157 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1158 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1159 | |
1160 | if (dev->type == DPDK_DEV_ETH) { | |
1161 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1162 | dev->requested_rxq_size); | |
1163 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1164 | dev->rxq_size); | |
1165 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1166 | dev->requested_txq_size); | |
1167 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1168 | dev->txq_size); | |
1a2bb118 SC |
1169 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1170 | smap_add(args, "rx_csum_offload", "true"); | |
8155ab7e KT |
1171 | } else { |
1172 | smap_add(args, "rx_csum_offload", "false"); | |
1a2bb118 | 1173 | } |
451f26fd | 1174 | } |
8a9562d2 PS |
1175 | ovs_mutex_unlock(&dev->mutex); |
1176 | ||
1177 | return 0; | |
1178 | } | |
1179 | ||
55e075e6 | 1180 | static struct netdev_dpdk * |
bb37956a | 1181 | netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id) |
55e075e6 CL |
1182 | OVS_REQUIRES(dpdk_mutex) |
1183 | { | |
1184 | struct netdev_dpdk *dev; | |
1185 | ||
1186 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1187 | if (dev->port_id == port_id) { | |
1188 | return dev; | |
1189 | } | |
1190 | } | |
1191 | ||
1192 | return NULL; | |
1193 | } | |
1194 | ||
bb37956a | 1195 | static dpdk_port_t |
5dcde09c IM |
1196 | netdev_dpdk_process_devargs(struct netdev_dpdk *dev, |
1197 | const char *devargs, char **errp) | |
55e075e6 | 1198 | { |
03d6399e IM |
1199 | /* Get the name up to the first comma. */ |
1200 | char *name = xmemdup0(devargs, strcspn(devargs, ",")); | |
bb37956a | 1201 | dpdk_port_t new_port_id = DPDK_ETH_PORT_ID_INVALID; |
55e075e6 | 1202 | |
69876ed7 | 1203 | if (!rte_eth_dev_count() |
03d6399e | 1204 | || rte_eth_dev_get_port_by_name(name, &new_port_id) |
69876ed7 CL |
1205 | || !rte_eth_dev_is_valid_port(new_port_id)) { |
1206 | /* Device not found in DPDK, attempt to attach it */ | |
1207 | if (!rte_eth_dev_attach(devargs, &new_port_id)) { | |
1208 | /* Attach successful */ | |
5dcde09c | 1209 | dev->attached = true; |
69876ed7 CL |
1210 | VLOG_INFO("Device '%s' attached to DPDK", devargs); |
1211 | } else { | |
1212 | /* Attach unsuccessful */ | |
bb37956a | 1213 | new_port_id = DPDK_ETH_PORT_ID_INVALID; |
47a45d86 KT |
1214 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", |
1215 | devargs); | |
55e075e6 CL |
1216 | } |
1217 | } | |
1218 | ||
03d6399e | 1219 | free(name); |
55e075e6 CL |
1220 | return new_port_id; |
1221 | } | |
1222 | ||
c3d062a7 CL |
1223 | static void |
1224 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1225 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1226 | { |
050c60bf | 1227 | int new_n_rxq; |
a14b8947 | 1228 | |
2a21e757 | 1229 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1230 | if (new_n_rxq != dev->requested_n_rxq) { |
1231 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1232 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1233 | } |
c3d062a7 CL |
1234 | } |
1235 | ||
b685696b CL |
1236 | static void |
1237 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1238 | const char *flag, int default_size, int *new_size) | |
1239 | { | |
1240 | int queue_size = smap_get_int(args, flag, default_size); | |
1241 | ||
1242 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1243 | || !is_pow2(queue_size)) { | |
1244 | queue_size = default_size; | |
1245 | } | |
1246 | ||
1247 | if (queue_size != *new_size) { | |
1248 | *new_size = queue_size; | |
1249 | netdev_request_reconfigure(netdev); | |
1250 | } | |
1251 | } | |
1252 | ||
c3d062a7 | 1253 | static int |
9fff138e DDP |
1254 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1255 | char **errp) | |
c3d062a7 CL |
1256 | { |
1257 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
b614c894 IM |
1258 | bool rx_fc_en, tx_fc_en, autoneg; |
1259 | enum rte_eth_fc_mode fc_mode; | |
1260 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1261 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1262 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1263 | }; | |
55e075e6 CL |
1264 | const char *new_devargs; |
1265 | int err = 0; | |
c3d062a7 | 1266 | |
55e075e6 | 1267 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1268 | ovs_mutex_lock(&dev->mutex); |
1269 | ||
1270 | dpdk_set_rxq_config(dev, args); | |
1271 | ||
b685696b CL |
1272 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1273 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1274 | &dev->requested_rxq_size); | |
1275 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1276 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1277 | &dev->requested_txq_size); | |
1278 | ||
55e075e6 CL |
1279 | new_devargs = smap_get(args, "dpdk-devargs"); |
1280 | ||
1281 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1282 | /* The user requested a new device. If we return error, the caller | |
1283 | * will delete this netdev and try to recreate it. */ | |
1284 | err = EAGAIN; | |
1285 | goto out; | |
1286 | } | |
1287 | ||
1288 | /* dpdk-devargs is required for device configuration */ | |
1289 | if (new_devargs && new_devargs[0]) { | |
1290 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1291 | * is valid */ | |
1292 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1293 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
bb37956a IM |
1294 | dpdk_port_t new_port_id = netdev_dpdk_process_devargs(dev, |
1295 | new_devargs, | |
1296 | errp); | |
55e075e6 CL |
1297 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1298 | err = EINVAL; | |
1299 | } else if (new_port_id == dev->port_id) { | |
1300 | /* Already configured, do not reconfigure again */ | |
1301 | err = 0; | |
1302 | } else { | |
1303 | struct netdev_dpdk *dup_dev; | |
bb37956a | 1304 | |
55e075e6 CL |
1305 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); |
1306 | if (dup_dev) { | |
9fff138e DDP |
1307 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
1308 | "which is already in use by '%s'", | |
1309 | netdev_get_name(netdev), new_devargs, | |
1310 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1311 | err = EADDRINUSE; |
1312 | } else { | |
bd4e172b | 1313 | int sid = rte_eth_dev_socket_id(new_port_id); |
bb37956a | 1314 | |
bd4e172b | 1315 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; |
55e075e6 CL |
1316 | dev->devargs = xstrdup(new_devargs); |
1317 | dev->port_id = new_port_id; | |
1318 | netdev_request_reconfigure(&dev->up); | |
1319 | err = 0; | |
1320 | } | |
1321 | } | |
1322 | } | |
1323 | } else { | |
9fff138e DDP |
1324 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1325 | "The old 'dpdk<port_id>' names are not supported", | |
1326 | netdev_get_name(netdev)); | |
55e075e6 CL |
1327 | err = EINVAL; |
1328 | } | |
1329 | ||
1330 | if (err) { | |
1331 | goto out; | |
1332 | } | |
1333 | ||
c3d062a7 CL |
1334 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1335 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1336 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1337 | |
b614c894 IM |
1338 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1339 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1340 | dev->fc_conf.mode = fc_mode; | |
1341 | dev->fc_conf.autoneg = autoneg; | |
1342 | dpdk_eth_flow_ctrl_setup(dev); | |
1343 | } | |
9fd39370 | 1344 | |
55e075e6 | 1345 | out: |
c3d062a7 | 1346 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1347 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1348 | |
55e075e6 | 1349 | return err; |
c3d062a7 CL |
1350 | } |
1351 | ||
1352 | static int | |
9fff138e DDP |
1353 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1354 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1355 | { |
1356 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1357 | ||
1358 | ovs_mutex_lock(&dev->mutex); | |
1359 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1360 | ovs_mutex_unlock(&dev->mutex); |
1361 | ||
1362 | return 0; | |
1363 | } | |
1364 | ||
c1ff66ac | 1365 | static int |
2d24d165 | 1366 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1367 | const struct smap *args, |
1368 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1369 | { |
1370 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1371 | const char *path; | |
1372 | ||
6881885a | 1373 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1374 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1375 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1376 | if (path && strcmp(path, dev->vhost_id)) { |
1377 | strcpy(dev->vhost_id, path); | |
c1ff66ac CL |
1378 | netdev_request_reconfigure(netdev); |
1379 | } | |
1380 | } | |
6881885a | 1381 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1382 | |
1383 | return 0; | |
1384 | } | |
1385 | ||
7dec44fe | 1386 | static int |
d46285a2 | 1387 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1388 | { |
d46285a2 | 1389 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1390 | |
d46285a2 | 1391 | return dev->socket_id; |
7dec44fe AW |
1392 | } |
1393 | ||
050c60bf | 1394 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1395 | static int |
050c60bf | 1396 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1397 | { |
d46285a2 | 1398 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1399 | |
d46285a2 | 1400 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1401 | |
050c60bf DDP |
1402 | if (dev->requested_n_txq == n_txq) { |
1403 | goto out; | |
4573fbd3 FL |
1404 | } |
1405 | ||
050c60bf DDP |
1406 | dev->requested_n_txq = n_txq; |
1407 | netdev_request_reconfigure(netdev); | |
58397e6c | 1408 | |
050c60bf | 1409 | out: |
d46285a2 | 1410 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1411 | return 0; |
58397e6c KT |
1412 | } |
1413 | ||
8a9562d2 PS |
1414 | static struct netdev_rxq * |
1415 | netdev_dpdk_rxq_alloc(void) | |
1416 | { | |
1417 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1418 | ||
eff23640 DDP |
1419 | if (rx) { |
1420 | return &rx->up; | |
1421 | } | |
1422 | ||
1423 | return NULL; | |
8a9562d2 PS |
1424 | } |
1425 | ||
1426 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1427 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1428 | { |
d46285a2 | 1429 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1430 | } |
1431 | ||
1432 | static int | |
d46285a2 | 1433 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1434 | { |
d46285a2 DDP |
1435 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1436 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1437 | |
d46285a2 DDP |
1438 | ovs_mutex_lock(&dev->mutex); |
1439 | rx->port_id = dev->port_id; | |
1440 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1441 | |
1442 | return 0; | |
1443 | } | |
1444 | ||
1445 | static void | |
d46285a2 | 1446 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1447 | { |
1448 | } | |
1449 | ||
1450 | static void | |
d46285a2 | 1451 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1452 | { |
d46285a2 | 1453 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1454 | |
1455 | rte_free(rx); | |
1456 | } | |
1457 | ||
819f13bd DDP |
1458 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1459 | * 'pkts', even in case of failure. | |
1460 | * | |
1461 | * Returns the number of packets that weren't transmitted. */ | |
1462 | static inline int | |
b59cc14e | 1463 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 1464 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 1465 | { |
1304f1f8 DDP |
1466 | uint32_t nb_tx = 0; |
1467 | ||
b59cc14e | 1468 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1469 | uint32_t ret; |
1470 | ||
b59cc14e | 1471 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1472 | if (!ret) { |
1473 | break; | |
1474 | } | |
1475 | ||
1476 | nb_tx += ret; | |
1477 | } | |
8a9562d2 | 1478 | |
b59cc14e | 1479 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 1480 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
1481 | * packet could come from a different mempool) */ |
1482 | int i; | |
1483 | ||
b59cc14e IM |
1484 | for (i = nb_tx; i < cnt; i++) { |
1485 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1486 | } |
8a9562d2 | 1487 | } |
819f13bd DDP |
1488 | |
1489 | return cnt - nb_tx; | |
8a9562d2 PS |
1490 | } |
1491 | ||
f3926f29 IS |
1492 | static inline bool |
1493 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1494 | struct rte_mbuf *pkt, uint64_t time) | |
1495 | { | |
1496 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1497 | ||
1498 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1499 | e_RTE_METER_GREEN; | |
1500 | } | |
1501 | ||
1502 | static int | |
1503 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
1504 | struct rte_mbuf **pkts, int pkt_cnt) | |
1505 | { | |
1506 | int i = 0; | |
1507 | int cnt = 0; | |
1508 | struct rte_mbuf *pkt = NULL; | |
1509 | uint64_t current_time = rte_rdtsc(); | |
1510 | ||
1511 | for (i = 0; i < pkt_cnt; i++) { | |
1512 | pkt = pkts[i]; | |
1513 | /* Handle current packet */ | |
1514 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1515 | if (cnt != i) { | |
1516 | pkts[cnt] = pkt; | |
1517 | } | |
1518 | cnt++; | |
1519 | } else { | |
1520 | rte_pktmbuf_free(pkt); | |
1521 | } | |
1522 | } | |
1523 | ||
1524 | return cnt; | |
1525 | } | |
1526 | ||
9509913a IS |
1527 | static int |
1528 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
1529 | int pkt_cnt) | |
1530 | { | |
1531 | int cnt = 0; | |
1532 | ||
1533 | rte_spinlock_lock(&policer->policer_lock); | |
1534 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, pkt_cnt); | |
1535 | rte_spinlock_unlock(&policer->policer_lock); | |
1536 | ||
1537 | return cnt; | |
1538 | } | |
1539 | ||
58397e6c | 1540 | static bool |
0a0f39df | 1541 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1542 | { |
0a0f39df | 1543 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1544 | } |
1545 | ||
d6e3feb5 | 1546 | static inline void |
1547 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1548 | unsigned int packet_size) | |
1549 | { | |
1550 | /* Hard-coded search for the size bucket. */ | |
1551 | if (packet_size < 256) { | |
1552 | if (packet_size >= 128) { | |
1553 | stats->rx_128_to_255_packets++; | |
1554 | } else if (packet_size <= 64) { | |
1555 | stats->rx_1_to_64_packets++; | |
1556 | } else { | |
1557 | stats->rx_65_to_127_packets++; | |
1558 | } | |
1559 | } else { | |
1560 | if (packet_size >= 1523) { | |
1561 | stats->rx_1523_to_max_packets++; | |
1562 | } else if (packet_size >= 1024) { | |
1563 | stats->rx_1024_to_1522_packets++; | |
1564 | } else if (packet_size < 512) { | |
1565 | stats->rx_256_to_511_packets++; | |
1566 | } else { | |
1567 | stats->rx_512_to_1023_packets++; | |
1568 | } | |
1569 | } | |
1570 | } | |
1571 | ||
9e3ddd45 TP |
1572 | static inline void |
1573 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1574 | struct dp_packet **packets, int count, |
1575 | int dropped) | |
9e3ddd45 TP |
1576 | { |
1577 | int i; | |
d6e3feb5 | 1578 | unsigned int packet_size; |
9e3ddd45 TP |
1579 | struct dp_packet *packet; |
1580 | ||
1581 | stats->rx_packets += count; | |
9509913a | 1582 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1583 | for (i = 0; i < count; i++) { |
1584 | packet = packets[i]; | |
d6e3feb5 | 1585 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1586 | |
d6e3feb5 | 1587 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1588 | /* This only protects the following multicast counting from |
1589 | * too short packets, but it does not stop the packet from | |
1590 | * further processing. */ | |
1591 | stats->rx_errors++; | |
1592 | stats->rx_length_errors++; | |
1593 | continue; | |
1594 | } | |
1595 | ||
d6e3feb5 | 1596 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1597 | ||
9e3ddd45 TP |
1598 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1599 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1600 | stats->multicast++; | |
1601 | } | |
1602 | ||
d6e3feb5 | 1603 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1604 | } |
1605 | } | |
1606 | ||
58397e6c KT |
1607 | /* |
1608 | * The receive path for the vhost port is the TX path out from guest. | |
1609 | */ | |
1610 | static int | |
d46285a2 | 1611 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1612 | struct dp_packet_batch *batch) |
58397e6c | 1613 | { |
d46285a2 | 1614 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
d46285a2 | 1615 | int qid = rxq->queue_id; |
9509913a | 1616 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1617 | uint16_t nb_rx = 0; |
9509913a | 1618 | uint16_t dropped = 0; |
58397e6c | 1619 | |
0a0f39df | 1620 | if (OVS_UNLIKELY(!is_vhost_running(dev) |
e543851d | 1621 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1622 | return EAGAIN; |
1623 | } | |
1624 | ||
0a0f39df CL |
1625 | nb_rx = rte_vhost_dequeue_burst(netdev_dpdk_get_vid(dev), |
1626 | qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
d46285a2 | 1627 | dev->dpdk_mp->mp, |
64839cf4 | 1628 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1629 | NETDEV_MAX_BURST); |
58397e6c KT |
1630 | if (!nb_rx) { |
1631 | return EAGAIN; | |
1632 | } | |
1633 | ||
9509913a IS |
1634 | if (policer) { |
1635 | dropped = nb_rx; | |
64839cf4 WT |
1636 | nb_rx = ingress_policer_run(policer, |
1637 | (struct rte_mbuf **) batch->packets, | |
1638 | nb_rx); | |
9509913a IS |
1639 | dropped -= nb_rx; |
1640 | } | |
1641 | ||
d46285a2 | 1642 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1643 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1644 | nb_rx, dropped); | |
d46285a2 | 1645 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1646 | |
11d4c7a8 | 1647 | dp_packet_batch_init_cutlen(batch); |
64839cf4 | 1648 | batch->count = (int) nb_rx; |
58397e6c KT |
1649 | return 0; |
1650 | } | |
1651 | ||
8a9562d2 | 1652 | static int |
64839cf4 | 1653 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1654 | { |
d46285a2 DDP |
1655 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1656 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1657 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1658 | int nb_rx; |
9509913a | 1659 | int dropped = 0; |
8a9562d2 | 1660 | |
3b1fb077 DDP |
1661 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1662 | return EAGAIN; | |
1663 | } | |
1664 | ||
d46285a2 | 1665 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1666 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1667 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1668 | if (!nb_rx) { |
1669 | return EAGAIN; | |
1670 | } | |
1671 | ||
9509913a IS |
1672 | if (policer) { |
1673 | dropped = nb_rx; | |
64839cf4 | 1674 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 1675 | (struct rte_mbuf **) batch->packets, |
64839cf4 | 1676 | nb_rx); |
9509913a IS |
1677 | dropped -= nb_rx; |
1678 | } | |
1679 | ||
1680 | /* Update stats to reflect dropped packets */ | |
1681 | if (OVS_UNLIKELY(dropped)) { | |
1682 | rte_spinlock_lock(&dev->stats_lock); | |
1683 | dev->stats.rx_dropped += dropped; | |
1684 | rte_spinlock_unlock(&dev->stats_lock); | |
1685 | } | |
1686 | ||
11d4c7a8 | 1687 | dp_packet_batch_init_cutlen(batch); |
64839cf4 | 1688 | batch->count = nb_rx; |
8a9562d2 PS |
1689 | |
1690 | return 0; | |
1691 | } | |
1692 | ||
0bf765f7 | 1693 | static inline int |
78bd47cf DDP |
1694 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
1695 | int cnt) | |
0bf765f7 | 1696 | { |
78bd47cf | 1697 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 1698 | |
78bd47cf DDP |
1699 | if (qos_conf) { |
1700 | rte_spinlock_lock(&qos_conf->lock); | |
1701 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt); | |
1702 | rte_spinlock_unlock(&qos_conf->lock); | |
0bf765f7 IS |
1703 | } |
1704 | ||
1705 | return cnt; | |
1706 | } | |
1707 | ||
c6ec9d17 IM |
1708 | static int |
1709 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1710 | int pkt_cnt) | |
1711 | { | |
1712 | int i = 0; | |
1713 | int cnt = 0; | |
1714 | struct rte_mbuf *pkt; | |
1715 | ||
1716 | for (i = 0; i < pkt_cnt; i++) { | |
1717 | pkt = pkts[i]; | |
1718 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
1719 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
1720 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
1721 | rte_pktmbuf_free(pkt); | |
1722 | continue; | |
1723 | } | |
1724 | ||
1725 | if (OVS_UNLIKELY(i != cnt)) { | |
1726 | pkts[cnt] = pkt; | |
1727 | } | |
1728 | cnt++; | |
1729 | } | |
1730 | ||
1731 | return cnt; | |
1732 | } | |
1733 | ||
9e3ddd45 TP |
1734 | static inline void |
1735 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1736 | struct dp_packet **packets, | |
1737 | int attempted, | |
1738 | int dropped) | |
1739 | { | |
1740 | int i; | |
1741 | int sent = attempted - dropped; | |
1742 | ||
1743 | stats->tx_packets += sent; | |
1744 | stats->tx_dropped += dropped; | |
1745 | ||
1746 | for (i = 0; i < sent; i++) { | |
1747 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1748 | } | |
1749 | } | |
1750 | ||
58397e6c | 1751 | static void |
4573fbd3 | 1752 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 1753 | struct dp_packet **pkts, int cnt) |
58397e6c | 1754 | { |
d46285a2 | 1755 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1756 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1757 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 1758 | unsigned int dropped = 0; |
dd52de45 | 1759 | int i, retries = 0; |
58397e6c | 1760 | |
81acebda | 1761 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1762 | |
0a0f39df | 1763 | if (OVS_UNLIKELY(!is_vhost_running(dev) || qid < 0 |
e543851d | 1764 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1765 | rte_spinlock_lock(&dev->stats_lock); |
1766 | dev->stats.tx_dropped+= cnt; | |
1767 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1768 | goto out; |
58397e6c KT |
1769 | } |
1770 | ||
d46285a2 | 1771 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1772 | |
c6ec9d17 | 1773 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 1774 | /* Check has QoS has been configured for the netdev */ |
78bd47cf | 1775 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt); |
c6ec9d17 | 1776 | dropped = total_pkts - cnt; |
0bf765f7 | 1777 | |
95e9881f | 1778 | do { |
4573fbd3 | 1779 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1780 | unsigned int tx_pkts; |
1781 | ||
0a0f39df | 1782 | tx_pkts = rte_vhost_enqueue_burst(netdev_dpdk_get_vid(dev), |
c6ec9d17 | 1783 | vhost_qid, cur_pkts, cnt); |
95e9881f KT |
1784 | if (OVS_LIKELY(tx_pkts)) { |
1785 | /* Packets have been sent.*/ | |
1786 | cnt -= tx_pkts; | |
31871ee3 | 1787 | /* Prepare for possible retry.*/ |
95e9881f KT |
1788 | cur_pkts = &cur_pkts[tx_pkts]; |
1789 | } else { | |
31871ee3 KT |
1790 | /* No packets sent - do not retry.*/ |
1791 | break; | |
95e9881f | 1792 | } |
c6ec9d17 | 1793 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1794 | |
d46285a2 | 1795 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1796 | |
d46285a2 | 1797 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 1798 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 1799 | cnt + dropped); |
d46285a2 | 1800 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1801 | |
1802 | out: | |
c6ec9d17 | 1803 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 1804 | dp_packet_delete(pkts[i]); |
58397e6c KT |
1805 | } |
1806 | } | |
1807 | ||
8a9562d2 PS |
1808 | /* Tx function. Transmit packets indefinitely */ |
1809 | static void | |
64839cf4 | 1810 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1811 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1812 | { |
bce01e3a | 1813 | #if !defined(__CHECKER__) && !defined(_WIN32) |
64839cf4 | 1814 | const size_t PKT_ARRAY_SIZE = batch->count; |
bce01e3a EJ |
1815 | #else |
1816 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1817 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1818 | #endif |
8a9562d2 | 1819 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 1820 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
175cf4de RW |
1821 | int dropped = 0; |
1822 | int newcnt = 0; | |
1823 | int i; | |
8a9562d2 | 1824 | |
7d6d1a40 WT |
1825 | dp_packet_batch_apply_cutlen(batch); |
1826 | ||
64839cf4 WT |
1827 | for (i = 0; i < batch->count; i++) { |
1828 | int size = dp_packet_size(batch->packets[i]); | |
95fb793a | 1829 | |
f98d7864 | 1830 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c | 1831 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
58be5c0e | 1832 | (int) size, dev->max_packet_len); |
f4fd623c | 1833 | |
175cf4de | 1834 | dropped++; |
f4fd623c DDP |
1835 | continue; |
1836 | } | |
8a9562d2 | 1837 | |
2391135c | 1838 | pkts[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1839 | |
2391135c | 1840 | if (!pkts[newcnt]) { |
64839cf4 | 1841 | dropped += batch->count - i; |
175cf4de | 1842 | break; |
f4fd623c DDP |
1843 | } |
1844 | ||
1845 | /* We have to do a copy for now */ | |
2391135c | 1846 | memcpy(rte_pktmbuf_mtod(pkts[newcnt], void *), |
64839cf4 | 1847 | dp_packet_data(batch->packets[i]), size); |
f4fd623c | 1848 | |
2391135c MK |
1849 | rte_pktmbuf_data_len(pkts[newcnt]) = size; |
1850 | rte_pktmbuf_pkt_len(pkts[newcnt]) = size; | |
f4fd623c DDP |
1851 | |
1852 | newcnt++; | |
1853 | } | |
8a9562d2 | 1854 | |
58397e6c | 1855 | if (dev->type == DPDK_DEV_VHOST) { |
2391135c | 1856 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, |
dd52de45 | 1857 | newcnt); |
58397e6c | 1858 | } else { |
0bf765f7 IS |
1859 | unsigned int qos_pkts = newcnt; |
1860 | ||
1861 | /* Check if QoS has been configured for this netdev. */ | |
78bd47cf | 1862 | newcnt = netdev_dpdk_qos_run(dev, pkts, newcnt); |
0bf765f7 IS |
1863 | |
1864 | dropped += qos_pkts - newcnt; | |
819f13bd | 1865 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, newcnt); |
58397e6c | 1866 | } |
db73f716 | 1867 | |
0bf765f7 IS |
1868 | if (OVS_UNLIKELY(dropped)) { |
1869 | rte_spinlock_lock(&dev->stats_lock); | |
1870 | dev->stats.tx_dropped += dropped; | |
1871 | rte_spinlock_unlock(&dev->stats_lock); | |
1872 | } | |
8a9562d2 PS |
1873 | } |
1874 | ||
58397e6c | 1875 | static int |
64839cf4 WT |
1876 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1877 | struct dp_packet_batch *batch, | |
324c8374 | 1878 | bool may_steal, bool concurrent_txq OVS_UNUSED) |
58397e6c | 1879 | { |
58397e6c | 1880 | |
dd52de45 | 1881 | if (OVS_UNLIKELY(!may_steal || batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 WT |
1882 | dpdk_do_tx_copy(netdev, qid, batch); |
1883 | dp_packet_delete_batch(batch, may_steal); | |
58397e6c | 1884 | } else { |
64839cf4 | 1885 | dp_packet_batch_apply_cutlen(batch); |
dd52de45 | 1886 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
1887 | } |
1888 | return 0; | |
1889 | } | |
1890 | ||
7251515e DV |
1891 | static inline void |
1892 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
324c8374 IM |
1893 | struct dp_packet_batch *batch, bool may_steal, |
1894 | bool concurrent_txq) | |
8a9562d2 | 1895 | { |
3b1fb077 DDP |
1896 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1897 | dp_packet_delete_batch(batch, may_steal); | |
1898 | return; | |
1899 | } | |
1900 | ||
324c8374 | 1901 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 1902 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
1903 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
1904 | } | |
1905 | ||
7251515e | 1906 | if (OVS_UNLIKELY(!may_steal || |
64839cf4 | 1907 | batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1908 | struct netdev *netdev = &dev->up; |
1909 | ||
64839cf4 WT |
1910 | dpdk_do_tx_copy(netdev, qid, batch); |
1911 | dp_packet_delete_batch(batch, may_steal); | |
8a9562d2 | 1912 | } else { |
6b094bf4 | 1913 | int dropped; |
64839cf4 | 1914 | int cnt = batch->count; |
2391135c | 1915 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 1916 | |
7d6d1a40 WT |
1917 | dp_packet_batch_apply_cutlen(batch); |
1918 | ||
2391135c | 1919 | cnt = netdev_dpdk_filter_packet_len(dev, pkts, cnt); |
78bd47cf | 1920 | cnt = netdev_dpdk_qos_run(dev, pkts, cnt); |
6b094bf4 | 1921 | dropped = batch->count - cnt; |
1b99bb05 | 1922 | |
819f13bd | 1923 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, cnt); |
8a9562d2 | 1924 | |
f4fd623c | 1925 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1926 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1927 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1928 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1929 | } |
8a9562d2 | 1930 | } |
a0cb2d66 | 1931 | |
324c8374 | 1932 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
1933 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
1934 | } | |
7251515e DV |
1935 | } |
1936 | ||
1937 | static int | |
1938 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
324c8374 IM |
1939 | struct dp_packet_batch *batch, bool may_steal, |
1940 | bool concurrent_txq) | |
7251515e DV |
1941 | { |
1942 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1943 | |
324c8374 | 1944 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e | 1945 | return 0; |
8a9562d2 PS |
1946 | } |
1947 | ||
1948 | static int | |
74ff3298 | 1949 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1950 | { |
1951 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1952 | ||
1953 | ovs_mutex_lock(&dev->mutex); | |
1954 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1955 | dev->hwaddr = mac; |
045c0d1a | 1956 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1957 | } |
1958 | ovs_mutex_unlock(&dev->mutex); | |
1959 | ||
1960 | return 0; | |
1961 | } | |
1962 | ||
1963 | static int | |
74ff3298 | 1964 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1965 | { |
1966 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1967 | ||
1968 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1969 | *mac = dev->hwaddr; |
8a9562d2 PS |
1970 | ovs_mutex_unlock(&dev->mutex); |
1971 | ||
1972 | return 0; | |
1973 | } | |
1974 | ||
1975 | static int | |
1976 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1977 | { | |
1978 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1979 | ||
1980 | ovs_mutex_lock(&dev->mutex); | |
1981 | *mtup = dev->mtu; | |
1982 | ovs_mutex_unlock(&dev->mutex); | |
1983 | ||
1984 | return 0; | |
1985 | } | |
1986 | ||
0072e931 MK |
1987 | static int |
1988 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
1989 | { | |
1990 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1991 | ||
1992 | if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
1993 | || mtu < ETHER_MIN_MTU) { | |
1994 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
1995 | return EINVAL; | |
1996 | } | |
1997 | ||
1998 | ovs_mutex_lock(&dev->mutex); | |
1999 | if (dev->requested_mtu != mtu) { | |
2000 | dev->requested_mtu = mtu; | |
2001 | netdev_request_reconfigure(netdev); | |
2002 | } | |
2003 | ovs_mutex_unlock(&dev->mutex); | |
2004 | ||
2005 | return 0; | |
2006 | } | |
2007 | ||
8a9562d2 | 2008 | static int |
d46285a2 | 2009 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 2010 | |
58397e6c KT |
2011 | static int |
2012 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
2013 | struct netdev_stats *stats) | |
2014 | { | |
2015 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2016 | ||
2017 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 2018 | |
45d947c4 | 2019 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
2020 | /* Supported Stats */ |
2021 | stats->rx_packets += dev->stats.rx_packets; | |
2022 | stats->tx_packets += dev->stats.tx_packets; | |
9509913a | 2023 | stats->rx_dropped = dev->stats.rx_dropped; |
58397e6c | 2024 | stats->tx_dropped += dev->stats.tx_dropped; |
9e3ddd45 TP |
2025 | stats->multicast = dev->stats.multicast; |
2026 | stats->rx_bytes = dev->stats.rx_bytes; | |
2027 | stats->tx_bytes = dev->stats.tx_bytes; | |
2028 | stats->rx_errors = dev->stats.rx_errors; | |
2029 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 2030 | |
2031 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
2032 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
2033 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
2034 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
2035 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
2036 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
2037 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
2038 | ||
45d947c4 | 2039 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 2040 | |
58397e6c KT |
2041 | ovs_mutex_unlock(&dev->mutex); |
2042 | ||
2043 | return 0; | |
2044 | } | |
2045 | ||
d6e3feb5 | 2046 | static void |
2047 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
2048 | const struct rte_eth_xstat *xstats, |
2049 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 2050 | const unsigned int size) |
2051 | { | |
d6e3feb5 | 2052 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 2053 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2054 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2055 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2056 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2057 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2058 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2059 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2060 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2061 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2062 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2063 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2064 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2065 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2066 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 2067 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2068 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2069 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2070 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2071 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2072 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2073 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2074 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2075 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2076 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2077 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2078 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2079 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2080 | stats->tx_1523_to_max_packets = xstats[i].value; |
d57f777f PS |
2081 | } else if (strcmp(XSTAT_RX_MULTICAST_PACKETS, names[i].name) == 0) { |
2082 | stats->multicast = xstats[i].value; | |
0a0f39df | 2083 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2084 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 2085 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2086 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2087 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2088 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2089 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2090 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2091 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2092 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2093 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2094 | stats->rx_jabber_errors = xstats[i].value; |
2095 | } | |
2096 | } | |
2097 | } | |
2098 | ||
8a9562d2 PS |
2099 | static int |
2100 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2101 | { | |
2102 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2103 | struct rte_eth_stats rte_stats; | |
2104 | bool gg; | |
2105 | ||
2106 | netdev_dpdk_get_carrier(netdev, &gg); | |
2107 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2108 | |
0a0f39df CL |
2109 | struct rte_eth_xstat *rte_xstats = NULL; |
2110 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2111 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2112 | |
2113 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
bb37956a | 2114 | VLOG_ERR("Can't get ETH statistics for port: %"PRIu8, dev->port_id); |
f9256822 | 2115 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2116 | return EPROTO; |
2117 | } | |
2118 | ||
0a0f39df CL |
2119 | /* Get length of statistics */ |
2120 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2121 | if (rte_xstats_len < 0) { | |
bb37956a | 2122 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2123 | goto out; |
2124 | } | |
2125 | /* Reserve memory for xstats names and values */ | |
2126 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2127 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2128 | ||
2129 | /* Retreive xstats names */ | |
2130 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2131 | rte_xstats_names, | |
2132 | rte_xstats_len); | |
2133 | if (rte_xstats_new_len != rte_xstats_len) { | |
bb37956a | 2134 | VLOG_WARN("Cannot get XSTATS names for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2135 | goto out; |
2136 | } | |
2137 | /* Retreive xstats values */ | |
2138 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2139 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2140 | rte_xstats_len); | |
2141 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2142 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2143 | rte_xstats_len); | |
d6e3feb5 | 2144 | } else { |
bb37956a | 2145 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
d6e3feb5 | 2146 | } |
8a9562d2 | 2147 | |
0a0f39df CL |
2148 | out: |
2149 | free(rte_xstats); | |
2150 | free(rte_xstats_names); | |
2151 | ||
2f9dd77f PS |
2152 | stats->rx_packets = rte_stats.ipackets; |
2153 | stats->tx_packets = rte_stats.opackets; | |
2154 | stats->rx_bytes = rte_stats.ibytes; | |
2155 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2156 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2157 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2158 | |
45d947c4 | 2159 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2160 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2161 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2162 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2163 | |
2164 | /* These are the available DPDK counters for packets not received due to | |
2165 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2166 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2167 | stats->rx_missed_errors = rte_stats.imissed; |
2168 | ||
8a9562d2 PS |
2169 | ovs_mutex_unlock(&dev->mutex); |
2170 | ||
2171 | return 0; | |
2172 | } | |
2173 | ||
2174 | static int | |
d46285a2 | 2175 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2176 | enum netdev_features *current, |
ca3d4f55 BX |
2177 | enum netdev_features *advertised, |
2178 | enum netdev_features *supported, | |
2179 | enum netdev_features *peer) | |
8a9562d2 | 2180 | { |
d46285a2 | 2181 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2182 | struct rte_eth_link link; |
2183 | ||
2184 | ovs_mutex_lock(&dev->mutex); | |
2185 | link = dev->link; | |
2186 | ovs_mutex_unlock(&dev->mutex); | |
2187 | ||
362ca396 | 2188 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2189 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
2190 | *current = NETDEV_F_10MB_HD; |
2191 | } | |
362ca396 | 2192 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2193 | *current = NETDEV_F_100MB_HD; |
2194 | } | |
362ca396 | 2195 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2196 | *current = NETDEV_F_1GB_HD; |
2197 | } | |
2198 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 2199 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
2200 | *current = NETDEV_F_10MB_FD; |
2201 | } | |
362ca396 | 2202 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2203 | *current = NETDEV_F_100MB_FD; |
2204 | } | |
362ca396 | 2205 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2206 | *current = NETDEV_F_1GB_FD; |
2207 | } | |
362ca396 | 2208 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
2209 | *current = NETDEV_F_10GB_FD; |
2210 | } | |
2211 | } | |
2212 | ||
362ca396 | 2213 | if (link.link_autoneg) { |
2214 | *current |= NETDEV_F_AUTONEG; | |
2215 | } | |
2216 | ||
ca3d4f55 BX |
2217 | *advertised = *supported = *peer = 0; |
2218 | ||
8a9562d2 PS |
2219 | return 0; |
2220 | } | |
2221 | ||
9509913a IS |
2222 | static struct ingress_policer * |
2223 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2224 | { | |
2225 | struct ingress_policer *policer = NULL; | |
2226 | uint64_t rate_bytes; | |
2227 | uint64_t burst_bytes; | |
2228 | int err = 0; | |
2229 | ||
2230 | policer = xmalloc(sizeof *policer); | |
2231 | rte_spinlock_init(&policer->policer_lock); | |
2232 | ||
2233 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
602c8668 LR |
2234 | rate_bytes = rate * 1000ULL / 8; |
2235 | burst_bytes = burst * 1000ULL / 8; | |
9509913a IS |
2236 | |
2237 | policer->app_srtcm_params.cir = rate_bytes; | |
2238 | policer->app_srtcm_params.cbs = burst_bytes; | |
2239 | policer->app_srtcm_params.ebs = 0; | |
2240 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2241 | &policer->app_srtcm_params); | |
58be5c0e | 2242 | if (err) { |
9509913a IS |
2243 | VLOG_ERR("Could not create rte meter for ingress policer"); |
2244 | return NULL; | |
2245 | } | |
2246 | ||
2247 | return policer; | |
2248 | } | |
2249 | ||
2250 | static int | |
2251 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2252 | uint32_t policer_burst) | |
2253 | { | |
2254 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2255 | struct ingress_policer *policer; | |
2256 | ||
2257 | /* Force to 0 if no rate specified, | |
2258 | * default to 8000 kbits if burst is 0, | |
2259 | * else stick with user-specified value. | |
2260 | */ | |
2261 | policer_burst = (!policer_rate ? 0 | |
2262 | : !policer_burst ? 8000 | |
2263 | : policer_burst); | |
2264 | ||
2265 | ovs_mutex_lock(&dev->mutex); | |
2266 | ||
2267 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2268 | &dev->ingress_policer); | |
2269 | ||
2270 | if (dev->policer_rate == policer_rate && | |
2271 | dev->policer_burst == policer_burst) { | |
2272 | /* Assume that settings haven't changed since we last set them. */ | |
2273 | ovs_mutex_unlock(&dev->mutex); | |
2274 | return 0; | |
2275 | } | |
2276 | ||
2277 | /* Destroy any existing ingress policer for the device if one exists */ | |
2278 | if (policer) { | |
2279 | ovsrcu_postpone(free, policer); | |
2280 | } | |
2281 | ||
2282 | if (policer_rate != 0) { | |
2283 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2284 | } else { | |
2285 | policer = NULL; | |
2286 | } | |
2287 | ovsrcu_set(&dev->ingress_policer, policer); | |
2288 | dev->policer_rate = policer_rate; | |
2289 | dev->policer_burst = policer_burst; | |
2290 | ovs_mutex_unlock(&dev->mutex); | |
2291 | ||
2292 | return 0; | |
2293 | } | |
2294 | ||
8a9562d2 PS |
2295 | static int |
2296 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2297 | { | |
2298 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
2299 | |
2300 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
2301 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
2302 | * postive integer to meet RFC 2863 recommendations. | |
2303 | */ | |
2304 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
2305 | ovs_mutex_unlock(&dev->mutex); |
2306 | ||
2307 | return ifindex; | |
2308 | } | |
2309 | ||
2310 | static int | |
d46285a2 | 2311 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2312 | { |
d46285a2 | 2313 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2314 | |
2315 | ovs_mutex_lock(&dev->mutex); | |
2316 | check_link_status(dev); | |
2317 | *carrier = dev->link.link_status; | |
58397e6c KT |
2318 | |
2319 | ovs_mutex_unlock(&dev->mutex); | |
2320 | ||
2321 | return 0; | |
2322 | } | |
2323 | ||
2324 | static int | |
d46285a2 | 2325 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2326 | { |
d46285a2 | 2327 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2328 | |
2329 | ovs_mutex_lock(&dev->mutex); | |
2330 | ||
0a0f39df | 2331 | if (is_vhost_running(dev)) { |
58397e6c KT |
2332 | *carrier = 1; |
2333 | } else { | |
2334 | *carrier = 0; | |
2335 | } | |
2336 | ||
8a9562d2 PS |
2337 | ovs_mutex_unlock(&dev->mutex); |
2338 | ||
2339 | return 0; | |
2340 | } | |
2341 | ||
2342 | static long long int | |
d46285a2 | 2343 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2344 | { |
d46285a2 | 2345 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2346 | long long int carrier_resets; |
2347 | ||
2348 | ovs_mutex_lock(&dev->mutex); | |
2349 | carrier_resets = dev->link_reset_cnt; | |
2350 | ovs_mutex_unlock(&dev->mutex); | |
2351 | ||
2352 | return carrier_resets; | |
2353 | } | |
2354 | ||
2355 | static int | |
d46285a2 | 2356 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2357 | long long int interval OVS_UNUSED) |
2358 | { | |
ee32150e | 2359 | return EOPNOTSUPP; |
8a9562d2 PS |
2360 | } |
2361 | ||
2362 | static int | |
2363 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2364 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2365 | enum netdev_flags *old_flagsp) |
2366 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 2367 | { |
8a9562d2 PS |
2368 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
2369 | return EINVAL; | |
2370 | } | |
2371 | ||
2372 | *old_flagsp = dev->flags; | |
2373 | dev->flags |= on; | |
2374 | dev->flags &= ~off; | |
2375 | ||
2376 | if (dev->flags == *old_flagsp) { | |
2377 | return 0; | |
2378 | } | |
2379 | ||
58397e6c | 2380 | if (dev->type == DPDK_DEV_ETH) { |
58397e6c KT |
2381 | if (dev->flags & NETDEV_PROMISC) { |
2382 | rte_eth_promiscuous_enable(dev->port_id); | |
2383 | } | |
8a9562d2 | 2384 | |
314fb5ad | 2385 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
2386 | } else { |
2387 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2388 | * running then change netdev's change_seq to trigger link state | |
2389 | * update. */ | |
e543851d ZB |
2390 | |
2391 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2392 | && is_vhost_running(dev)) { |
e543851d ZB |
2393 | netdev_change_seq_changed(&dev->up); |
2394 | ||
2395 | /* Clear statistics if device is getting up. */ | |
2396 | if (NETDEV_UP & on) { | |
2397 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 2398 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
2399 | rte_spinlock_unlock(&dev->stats_lock); |
2400 | } | |
2401 | } | |
8a9562d2 PS |
2402 | } |
2403 | ||
2404 | return 0; | |
2405 | } | |
2406 | ||
2407 | static int | |
d46285a2 | 2408 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2409 | enum netdev_flags off, enum netdev_flags on, |
2410 | enum netdev_flags *old_flagsp) | |
2411 | { | |
d46285a2 | 2412 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2413 | int error; |
2414 | ||
d46285a2 DDP |
2415 | ovs_mutex_lock(&dev->mutex); |
2416 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2417 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2418 | |
2419 | return error; | |
2420 | } | |
2421 | ||
2422 | static int | |
d46285a2 | 2423 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2424 | { |
d46285a2 | 2425 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2426 | struct rte_eth_dev_info dev_info; |
2427 | ||
7cd1261d | 2428 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2429 | return ENODEV; |
7cd1261d | 2430 | } |
8a9562d2 PS |
2431 | |
2432 | ovs_mutex_lock(&dev->mutex); | |
2433 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2434 | ovs_mutex_unlock(&dev->mutex); | |
2435 | ||
95fb793a | 2436 | smap_add_format(args, "port_no", "%d", dev->port_id); |
58be5c0e MK |
2437 | smap_add_format(args, "numa_id", "%d", |
2438 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
2439 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
2440 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2441 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2442 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2443 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2444 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
2445 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
2446 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
2447 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
2448 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2449 | ||
39c2baa9 | 2450 | if (dev_info.pci_dev) { |
2451 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2452 | dev_info.pci_dev->id.vendor_id); | |
2453 | smap_add_format(args, "pci-device_id", "0x%x", | |
2454 | dev_info.pci_dev->id.device_id); | |
2455 | } | |
8a9562d2 PS |
2456 | |
2457 | return 0; | |
2458 | } | |
2459 | ||
2460 | static void | |
2461 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2462 | OVS_REQUIRES(dev->mutex) | |
2463 | { | |
2464 | enum netdev_flags old_flags; | |
2465 | ||
2466 | if (admin_state) { | |
2467 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2468 | } else { | |
2469 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2470 | } | |
2471 | } | |
2472 | ||
2473 | static void | |
2474 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2475 | const char *argv[], void *aux OVS_UNUSED) | |
2476 | { | |
2477 | bool up; | |
2478 | ||
2479 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2480 | up = true; | |
2481 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2482 | up = false; | |
2483 | } else { | |
2484 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2485 | return; | |
2486 | } | |
2487 | ||
2488 | if (argc > 2) { | |
2489 | struct netdev *netdev = netdev_from_name(argv[1]); | |
2490 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
2491 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
2492 | ||
2493 | ovs_mutex_lock(&dpdk_dev->mutex); | |
2494 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
2495 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
2496 | ||
2497 | netdev_close(netdev); | |
2498 | } else { | |
2499 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2500 | netdev_close(netdev); | |
2501 | return; | |
2502 | } | |
2503 | } else { | |
2504 | struct netdev_dpdk *netdev; | |
2505 | ||
2506 | ovs_mutex_lock(&dpdk_mutex); | |
2507 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
2508 | ovs_mutex_lock(&netdev->mutex); | |
2509 | netdev_dpdk_set_admin_state__(netdev, up); | |
2510 | ovs_mutex_unlock(&netdev->mutex); | |
2511 | } | |
2512 | ovs_mutex_unlock(&dpdk_mutex); | |
2513 | } | |
2514 | unixctl_command_reply(conn, "OK"); | |
2515 | } | |
2516 | ||
0ee821c2 DB |
2517 | static void |
2518 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2519 | const char *argv[], void *aux OVS_UNUSED) | |
2520 | { | |
2521 | int ret; | |
2522 | char *response; | |
2523 | uint8_t port_id; | |
2524 | char devname[RTE_ETH_NAME_MAX_LEN]; | |
2525 | struct netdev_dpdk *dev; | |
2526 | ||
2527 | ovs_mutex_lock(&dpdk_mutex); | |
2528 | ||
2529 | if (!rte_eth_dev_count() || rte_eth_dev_get_port_by_name(argv[1], | |
2530 | &port_id)) { | |
2531 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); | |
2532 | goto error; | |
2533 | } | |
2534 | ||
2535 | dev = netdev_dpdk_lookup_by_port_id(port_id); | |
2536 | if (dev) { | |
2537 | response = xasprintf("Device '%s' is being used by interface '%s'. " | |
2538 | "Remove it before detaching", | |
2539 | argv[1], netdev_get_name(&dev->up)); | |
2540 | goto error; | |
2541 | } | |
2542 | ||
2543 | rte_eth_dev_close(port_id); | |
2544 | ||
2545 | ret = rte_eth_dev_detach(port_id, devname); | |
2546 | if (ret < 0) { | |
2547 | response = xasprintf("Device '%s' can not be detached", argv[1]); | |
2548 | goto error; | |
2549 | } | |
2550 | ||
2551 | response = xasprintf("Device '%s' has been detached", argv[1]); | |
2552 | ||
2553 | ovs_mutex_unlock(&dpdk_mutex); | |
2554 | unixctl_command_reply(conn, response); | |
2555 | free(response); | |
2556 | return; | |
2557 | ||
2558 | error: | |
2559 | ovs_mutex_unlock(&dpdk_mutex); | |
2560 | unixctl_command_reply_error(conn, response); | |
2561 | free(response); | |
2562 | } | |
2563 | ||
58397e6c KT |
2564 | /* |
2565 | * Set virtqueue flags so that we do not receive interrupts. | |
2566 | */ | |
2567 | static void | |
0a0f39df | 2568 | set_irq_status(int vid) |
58397e6c | 2569 | { |
4573fbd3 | 2570 | uint32_t i; |
4573fbd3 | 2571 | |
f3e7ec25 MW |
2572 | for (i = 0; i < rte_vhost_get_vring_num(vid); i++) { |
2573 | rte_vhost_enable_guest_notification(vid, i, 0); | |
4573fbd3 FL |
2574 | } |
2575 | } | |
2576 | ||
585a5bea IM |
2577 | /* |
2578 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2579 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2580 | */ |
2581 | static void | |
d46285a2 DDP |
2582 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2583 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2584 | { |
2585 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2586 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 2587 | |
eff23640 | 2588 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
2589 | |
2590 | for (i = 0; i < total_txqs; i++) { | |
2591 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2592 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2593 | enabled_queues[n_enabled++] = i; |
2594 | } | |
2595 | } | |
2596 | ||
2597 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2598 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2599 | n_enabled = 1; |
2600 | } | |
2601 | ||
2602 | k = 0; | |
2603 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2604 | if (dev->tx_q[i].map != i) { |
2605 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2606 | k = (k + 1) % n_enabled; |
2607 | } | |
2608 | } | |
2609 | ||
2d24d165 | 2610 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2611 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2612 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2613 | } |
2614 | ||
eff23640 | 2615 | free(enabled_queues); |
585a5bea | 2616 | } |
4573fbd3 | 2617 | |
58397e6c KT |
2618 | /* |
2619 | * A new virtio-net device is added to a vhost port. | |
2620 | */ | |
2621 | static int | |
0a0f39df | 2622 | new_device(int vid) |
58397e6c | 2623 | { |
d46285a2 | 2624 | struct netdev_dpdk *dev; |
58397e6c | 2625 | bool exists = false; |
db8f13b0 | 2626 | int newnode = 0; |
0a0f39df CL |
2627 | char ifname[IF_NAME_SZ]; |
2628 | ||
58be5c0e | 2629 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2630 | |
2631 | ovs_mutex_lock(&dpdk_mutex); | |
2632 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2633 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 2634 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2635 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
f3e7ec25 | 2636 | uint32_t qp_num = rte_vhost_get_vring_num(vid)/VIRTIO_QNUM; |
db8f13b0 CL |
2637 | |
2638 | /* Get NUMA information */ | |
0a0f39df CL |
2639 | newnode = rte_vhost_get_numa_node(vid); |
2640 | if (newnode == -1) { | |
5b9bf9e0 | 2641 | #ifdef VHOST_NUMA |
db8f13b0 | 2642 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2643 | ifname); |
5b9bf9e0 | 2644 | #endif |
db8f13b0 | 2645 | newnode = dev->socket_id; |
db8f13b0 CL |
2646 | } |
2647 | ||
7f5f2bd0 IM |
2648 | if (dev->requested_n_txq != qp_num |
2649 | || dev->requested_n_rxq != qp_num | |
2650 | || dev->requested_socket_id != newnode) { | |
2651 | dev->requested_socket_id = newnode; | |
2652 | dev->requested_n_rxq = qp_num; | |
2653 | dev->requested_n_txq = qp_num; | |
2654 | netdev_request_reconfigure(&dev->up); | |
2655 | } else { | |
2656 | /* Reconfiguration not required. */ | |
2657 | dev->vhost_reconfigured = true; | |
2658 | } | |
81acebda | 2659 | |
0a0f39df | 2660 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2661 | exists = true; |
2662 | ||
58397e6c | 2663 | /* Disable notifications. */ |
0a0f39df | 2664 | set_irq_status(vid); |
e543851d | 2665 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2666 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2667 | break; |
2668 | } | |
c1ff66ac | 2669 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2670 | } |
2671 | ovs_mutex_unlock(&dpdk_mutex); | |
2672 | ||
2673 | if (!exists) { | |
0a0f39df | 2674 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2675 | |
2676 | return -1; | |
2677 | } | |
2678 | ||
0a0f39df CL |
2679 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2680 | ifname, newnode); | |
2681 | ||
58397e6c KT |
2682 | return 0; |
2683 | } | |
2684 | ||
f3ea2ad2 IM |
2685 | /* Clears mapping for all available queues of vhost interface. */ |
2686 | static void | |
2687 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2688 | OVS_REQUIRES(dev->mutex) | |
2689 | { | |
2690 | int i; | |
2691 | ||
81acebda | 2692 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
2693 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
2694 | } | |
2695 | } | |
2696 | ||
58397e6c KT |
2697 | /* |
2698 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2699 | * flag to stop any more packets from being sent or received to/from a VM and | |
2700 | * ensure all currently queued packets have been sent/received before removing | |
2701 | * the device. | |
2702 | */ | |
2703 | static void | |
0a0f39df | 2704 | destroy_device(int vid) |
58397e6c | 2705 | { |
d46285a2 | 2706 | struct netdev_dpdk *dev; |
afee281f | 2707 | bool exists = false; |
0a0f39df CL |
2708 | char ifname[IF_NAME_SZ]; |
2709 | ||
58be5c0e | 2710 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2711 | |
2712 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2713 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2714 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 2715 | |
d46285a2 | 2716 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
2717 | dev->vhost_reconfigured = false; |
2718 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 2719 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 2720 | |
e543851d | 2721 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2722 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 2723 | exists = true; |
afee281f | 2724 | break; |
58397e6c KT |
2725 | } |
2726 | } | |
afee281f | 2727 | |
58397e6c KT |
2728 | ovs_mutex_unlock(&dpdk_mutex); |
2729 | ||
0a0f39df | 2730 | if (exists) { |
afee281f KT |
2731 | /* |
2732 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2733 | * to NULL, before returning. | |
2734 | */ | |
2735 | ovsrcu_synchronize(); | |
2736 | /* | |
2737 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2738 | * put thread back into quiescent state before returning. | |
2739 | */ | |
2740 | ovsrcu_quiesce_start(); | |
0a0f39df | 2741 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 2742 | } else { |
0a0f39df | 2743 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 2744 | } |
58397e6c KT |
2745 | } |
2746 | ||
585a5bea | 2747 | static int |
0a0f39df | 2748 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 2749 | { |
d46285a2 | 2750 | struct netdev_dpdk *dev; |
585a5bea IM |
2751 | bool exists = false; |
2752 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
2753 | char ifname[IF_NAME_SZ]; |
2754 | ||
58be5c0e | 2755 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
2756 | |
2757 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2758 | return 0; | |
2759 | } | |
2760 | ||
2761 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2762 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 2763 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2764 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 2765 | if (enable) { |
d46285a2 | 2766 | dev->tx_q[qid].map = qid; |
585a5bea | 2767 | } else { |
d46285a2 | 2768 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2769 | } |
d46285a2 | 2770 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2771 | exists = true; |
d46285a2 | 2772 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2773 | break; |
2774 | } | |
c1ff66ac | 2775 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2776 | } |
2777 | ovs_mutex_unlock(&dpdk_mutex); | |
2778 | ||
2779 | if (exists) { | |
0a0f39df CL |
2780 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
2781 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 2782 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 2783 | } else { |
0a0f39df | 2784 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
2785 | return -1; |
2786 | } | |
2787 | ||
2788 | return 0; | |
2789 | } | |
2790 | ||
0a0f39df CL |
2791 | int |
2792 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 2793 | { |
0a0f39df | 2794 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
2795 | } |
2796 | ||
9509913a IS |
2797 | struct ingress_policer * |
2798 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
2799 | { | |
2800 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
2801 | } | |
2802 | ||
58397e6c | 2803 | static int |
ecc1a34e | 2804 | netdev_dpdk_class_init(void) |
7d1ced01 | 2805 | { |
ecc1a34e DDP |
2806 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2807 | ||
2808 | /* This function can be called for different classes. The initialization | |
2809 | * needs to be done only once */ | |
2810 | if (ovsthread_once_start(&once)) { | |
2811 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
2812 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2813 | "[netdev] up|down", 1, 2, | |
2814 | netdev_dpdk_set_admin_state, NULL); | |
2815 | ||
0ee821c2 DB |
2816 | unixctl_command_register("netdev-dpdk/detach", |
2817 | "pci address of device", 1, 1, | |
2818 | netdev_dpdk_detach, NULL); | |
2819 | ||
ecc1a34e DDP |
2820 | ovsthread_once_done(&once); |
2821 | } | |
362ca396 | 2822 | |
7d1ced01 CL |
2823 | return 0; |
2824 | } | |
2825 | ||
033e9df2 | 2826 | |
95fb793a | 2827 | /* Client Rings */ |
2828 | ||
95fb793a | 2829 | static int |
2830 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
bb37956a | 2831 | dpdk_port_t *eth_port_id) |
95fb793a | 2832 | { |
48fffdee | 2833 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 2834 | char *ring_name; |
b8374d0d | 2835 | int port_id; |
95fb793a | 2836 | |
48fffdee KT |
2837 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
2838 | if (!ring_pair) { | |
95fb793a | 2839 | return ENOMEM; |
2840 | } | |
2841 | ||
7251515e | 2842 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 2843 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 2844 | |
8f0a76c9 | 2845 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 2846 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2847 | RING_F_SP_ENQ); |
0c6f39e5 | 2848 | free(ring_name); |
48fffdee KT |
2849 | if (ring_pair->cring_tx == NULL) { |
2850 | rte_free(ring_pair); | |
95fb793a | 2851 | return ENOMEM; |
2852 | } | |
2853 | ||
0c6f39e5 | 2854 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 2855 | |
8f0a76c9 | 2856 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 2857 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2858 | RING_F_SC_DEQ); |
0c6f39e5 | 2859 | free(ring_name); |
48fffdee KT |
2860 | if (ring_pair->cring_rx == NULL) { |
2861 | rte_free(ring_pair); | |
95fb793a | 2862 | return ENOMEM; |
2863 | } | |
2864 | ||
b8374d0d MV |
2865 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
2866 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 2867 | |
b8374d0d | 2868 | if (port_id < 0) { |
48fffdee | 2869 | rte_free(ring_pair); |
95fb793a | 2870 | return ENODEV; |
2871 | } | |
2872 | ||
48fffdee | 2873 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
2874 | ring_pair->eth_port_id = port_id; |
2875 | *eth_port_id = port_id; | |
2876 | ||
48fffdee | 2877 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 2878 | |
95fb793a | 2879 | return 0; |
2880 | } | |
2881 | ||
2882 | static int | |
bb37956a | 2883 | dpdk_ring_open(const char dev_name[], dpdk_port_t *eth_port_id) |
64839cf4 | 2884 | OVS_REQUIRES(dpdk_mutex) |
95fb793a | 2885 | { |
48fffdee | 2886 | struct dpdk_ring *ring_pair; |
95fb793a | 2887 | unsigned int port_no; |
2888 | int err = 0; | |
2889 | ||
2890 | /* Names always start with "dpdkr" */ | |
2891 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2892 | if (err) { | |
2893 | return err; | |
2894 | } | |
2895 | ||
58be5c0e | 2896 | /* Look through our list to find the device */ |
48fffdee KT |
2897 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
2898 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 2899 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 2900 | /* Really all that is needed */ |
48fffdee | 2901 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 2902 | return 0; |
2903 | } | |
2904 | } | |
2905 | /* Need to create the device rings */ | |
2906 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2907 | } | |
2908 | ||
7251515e | 2909 | static int |
d46285a2 | 2910 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
324c8374 IM |
2911 | struct dp_packet_batch *batch, bool may_steal, |
2912 | bool concurrent_txq) | |
7251515e | 2913 | { |
d46285a2 | 2914 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b99bb05 MG |
2915 | unsigned i; |
2916 | ||
58be5c0e MK |
2917 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
2918 | * the rss hash field is clear. This is because the same mbuf may be | |
2919 | * modified by the consumer of the ring and return into the datapath | |
2920 | * without recalculating the RSS hash. */ | |
64839cf4 | 2921 | for (i = 0; i < batch->count; i++) { |
f8121b39 | 2922 | dp_packet_mbuf_rss_flag_reset(batch->packets[i]); |
1b99bb05 | 2923 | } |
7251515e | 2924 | |
324c8374 | 2925 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e DV |
2926 | return 0; |
2927 | } | |
2928 | ||
95fb793a | 2929 | static int |
2930 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2931 | { | |
bb37956a | 2932 | dpdk_port_t port_no = 0; |
95fb793a | 2933 | int err = 0; |
2934 | ||
95fb793a | 2935 | ovs_mutex_lock(&dpdk_mutex); |
2936 | ||
2937 | err = dpdk_ring_open(netdev->name, &port_no); | |
2938 | if (err) { | |
2939 | goto unlock_dpdk; | |
2940 | } | |
2941 | ||
1ce30dfd DDP |
2942 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
2943 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 2944 | unlock_dpdk: |
2945 | ovs_mutex_unlock(&dpdk_mutex); | |
2946 | return err; | |
2947 | } | |
2948 | ||
0bf765f7 IS |
2949 | /* QoS Functions */ |
2950 | ||
2951 | /* | |
2952 | * Initialize QoS configuration operations. | |
2953 | */ | |
2954 | static void | |
2955 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
2956 | { | |
2957 | conf->ops = ops; | |
78bd47cf | 2958 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
2959 | } |
2960 | ||
2961 | /* | |
2962 | * Search existing QoS operations in qos_ops and compare each set of | |
2963 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
2964 | * else return NULL | |
2965 | */ | |
2966 | static const struct dpdk_qos_ops * | |
2967 | qos_lookup_name(const char *name) | |
2968 | { | |
2969 | const struct dpdk_qos_ops *const *opsp; | |
2970 | ||
2971 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2972 | const struct dpdk_qos_ops *ops = *opsp; | |
2973 | if (!strcmp(name, ops->qos_name)) { | |
2974 | return ops; | |
2975 | } | |
2976 | } | |
2977 | return NULL; | |
2978 | } | |
2979 | ||
0bf765f7 IS |
2980 | static int |
2981 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
2982 | struct sset *types) | |
2983 | { | |
2984 | const struct dpdk_qos_ops *const *opsp; | |
2985 | ||
2986 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2987 | const struct dpdk_qos_ops *ops = *opsp; | |
2988 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
2989 | sset_add(types, ops->qos_name); | |
2990 | } | |
2991 | } | |
2992 | return 0; | |
2993 | } | |
2994 | ||
2995 | static int | |
d46285a2 | 2996 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
2997 | const char **typep, struct smap *details) |
2998 | { | |
d46285a2 | 2999 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 3000 | struct qos_conf *qos_conf; |
0bf765f7 IS |
3001 | int error = 0; |
3002 | ||
d46285a2 | 3003 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
3004 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
3005 | if (qos_conf) { | |
3006 | *typep = qos_conf->ops->qos_name; | |
3007 | error = (qos_conf->ops->qos_get | |
3008 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
3009 | } else { |
3010 | /* No QoS configuration set, return an empty string */ | |
3011 | *typep = ""; | |
0bf765f7 | 3012 | } |
d46285a2 | 3013 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
3014 | |
3015 | return error; | |
3016 | } | |
3017 | ||
3018 | static int | |
78bd47cf DDP |
3019 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
3020 | const struct smap *details) | |
0bf765f7 | 3021 | { |
d46285a2 | 3022 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 3023 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 3024 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
3025 | int error = 0; |
3026 | ||
d46285a2 | 3027 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 3028 | |
78bd47cf | 3029 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 3030 | |
78bd47cf DDP |
3031 | new_ops = qos_lookup_name(type); |
3032 | ||
3033 | if (!new_ops || !new_ops->qos_construct) { | |
3034 | new_qos_conf = NULL; | |
3035 | if (type && type[0]) { | |
3036 | error = EOPNOTSUPP; | |
0bf765f7 | 3037 | } |
44975bb0 | 3038 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
3039 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
3040 | new_qos_conf = qos_conf; | |
0bf765f7 | 3041 | } else { |
78bd47cf | 3042 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
3043 | } |
3044 | ||
7ea266e9 | 3045 | if (error) { |
78bd47cf DDP |
3046 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
3047 | type, netdev->name, rte_strerror(error)); | |
3048 | } | |
3049 | ||
3050 | if (new_qos_conf != qos_conf) { | |
3051 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
3052 | if (qos_conf) { | |
3053 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
3054 | } | |
0bf765f7 IS |
3055 | } |
3056 | ||
d46285a2 | 3057 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 3058 | |
0bf765f7 IS |
3059 | return error; |
3060 | } | |
3061 | ||
3062 | /* egress-policer details */ | |
3063 | ||
3064 | struct egress_policer { | |
3065 | struct qos_conf qos_conf; | |
3066 | struct rte_meter_srtcm_params app_srtcm_params; | |
3067 | struct rte_meter_srtcm egress_meter; | |
3068 | }; | |
3069 | ||
78bd47cf DDP |
3070 | static void |
3071 | egress_policer_details_to_param(const struct smap *details, | |
3072 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 3073 | { |
78bd47cf DDP |
3074 | memset(params, 0, sizeof *params); |
3075 | params->cir = smap_get_ullong(details, "cir", 0); | |
3076 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
3077 | params->ebs = 0; | |
0bf765f7 IS |
3078 | } |
3079 | ||
3080 | static int | |
78bd47cf DDP |
3081 | egress_policer_qos_construct(const struct smap *details, |
3082 | struct qos_conf **conf) | |
0bf765f7 | 3083 | { |
0bf765f7 | 3084 | struct egress_policer *policer; |
0bf765f7 IS |
3085 | int err = 0; |
3086 | ||
0bf765f7 IS |
3087 | policer = xmalloc(sizeof *policer); |
3088 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 3089 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
0bf765f7 | 3090 | err = rte_meter_srtcm_config(&policer->egress_meter, |
78bd47cf DDP |
3091 | &policer->app_srtcm_params); |
3092 | if (!err) { | |
3093 | *conf = &policer->qos_conf; | |
3094 | } else { | |
7ea266e9 | 3095 | free(policer); |
78bd47cf | 3096 | *conf = NULL; |
7ea266e9 IS |
3097 | err = -err; |
3098 | } | |
0bf765f7 IS |
3099 | |
3100 | return err; | |
3101 | } | |
3102 | ||
3103 | static void | |
78bd47cf | 3104 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
3105 | { |
3106 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 3107 | qos_conf); |
0bf765f7 IS |
3108 | free(policer); |
3109 | } | |
3110 | ||
3111 | static int | |
78bd47cf | 3112 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 3113 | { |
78bd47cf DDP |
3114 | struct egress_policer *policer = |
3115 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3116 | ||
3117 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
3118 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 3119 | |
0bf765f7 IS |
3120 | return 0; |
3121 | } | |
3122 | ||
78bd47cf | 3123 | static bool |
47a45d86 KT |
3124 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
3125 | const struct smap *details) | |
0bf765f7 | 3126 | { |
78bd47cf DDP |
3127 | struct egress_policer *policer = |
3128 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3129 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 3130 | |
78bd47cf | 3131 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 3132 | |
78bd47cf | 3133 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
3134 | } |
3135 | ||
0bf765f7 | 3136 | static int |
78bd47cf | 3137 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt) |
0bf765f7 | 3138 | { |
0bf765f7 | 3139 | int cnt = 0; |
78bd47cf DDP |
3140 | struct egress_policer *policer = |
3141 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 3142 | |
f3926f29 | 3143 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, pkt_cnt); |
0bf765f7 IS |
3144 | |
3145 | return cnt; | |
3146 | } | |
3147 | ||
3148 | static const struct dpdk_qos_ops egress_policer_ops = { | |
3149 | "egress-policer", /* qos_name */ | |
3150 | egress_policer_qos_construct, | |
3151 | egress_policer_qos_destruct, | |
3152 | egress_policer_qos_get, | |
78bd47cf | 3153 | egress_policer_qos_is_equal, |
0bf765f7 IS |
3154 | egress_policer_run |
3155 | }; | |
3156 | ||
050c60bf DDP |
3157 | static int |
3158 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
3159 | { | |
3160 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3161 | int err = 0; | |
3162 | ||
050c60bf DDP |
3163 | ovs_mutex_lock(&dev->mutex); |
3164 | ||
3165 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 3166 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b CL |
3167 | && dev->mtu == dev->requested_mtu |
3168 | && dev->rxq_size == dev->requested_rxq_size | |
bd4e172b | 3169 | && dev->txq_size == dev->requested_txq_size |
3170 | && dev->socket_id == dev->requested_socket_id) { | |
050c60bf DDP |
3171 | /* Reconfiguration is unnecessary */ |
3172 | ||
3173 | goto out; | |
3174 | } | |
3175 | ||
3176 | rte_eth_dev_stop(dev->port_id); | |
3177 | ||
bd4e172b | 3178 | if (dev->mtu != dev->requested_mtu |
3179 | || dev->socket_id != dev->requested_socket_id) { | |
96e9b168 IS |
3180 | err = netdev_dpdk_mempool_configure(dev); |
3181 | if (err) { | |
3182 | goto out; | |
3183 | } | |
0072e931 MK |
3184 | } |
3185 | ||
050c60bf DDP |
3186 | netdev->n_txq = dev->requested_n_txq; |
3187 | netdev->n_rxq = dev->requested_n_rxq; | |
3188 | ||
b685696b CL |
3189 | dev->rxq_size = dev->requested_rxq_size; |
3190 | dev->txq_size = dev->requested_txq_size; | |
3191 | ||
050c60bf DDP |
3192 | rte_free(dev->tx_q); |
3193 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
3194 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
3195 | if (!dev->tx_q) { | |
3196 | err = ENOMEM; | |
3197 | } | |
050c60bf | 3198 | |
0072e931 MK |
3199 | netdev_change_seq_changed(netdev); |
3200 | ||
050c60bf | 3201 | out: |
050c60bf | 3202 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
3203 | return err; |
3204 | } | |
3205 | ||
7f381c2e | 3206 | static int |
2d24d165 | 3207 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 3208 | OVS_REQUIRES(dev->mutex) |
050c60bf | 3209 | { |
2d24d165 CL |
3210 | dev->up.n_txq = dev->requested_n_txq; |
3211 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 3212 | int err; |
050c60bf | 3213 | |
81acebda IM |
3214 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
3215 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
3216 | dev->tx_q[0].map = 0; | |
3217 | } | |
3218 | ||
3219 | netdev_dpdk_remap_txqs(dev); | |
3220 | ||
0072e931 MK |
3221 | if (dev->requested_socket_id != dev->socket_id |
3222 | || dev->requested_mtu != dev->mtu) { | |
96e9b168 IS |
3223 | err = netdev_dpdk_mempool_configure(dev); |
3224 | if (err) { | |
3225 | return err; | |
3226 | } else { | |
2d24d165 | 3227 | netdev_change_seq_changed(&dev->up); |
db8f13b0 CL |
3228 | } |
3229 | } | |
3230 | ||
0a0f39df CL |
3231 | if (netdev_dpdk_get_vid(dev) >= 0) { |
3232 | dev->vhost_reconfigured = true; | |
81acebda | 3233 | } |
7f381c2e DDP |
3234 | |
3235 | return 0; | |
2d24d165 CL |
3236 | } |
3237 | ||
3238 | static int | |
3239 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
3240 | { | |
3241 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3242 | int err; |
2d24d165 | 3243 | |
2d24d165 | 3244 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 3245 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 3246 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
3247 | |
3248 | return err; | |
2d24d165 CL |
3249 | } |
3250 | ||
3251 | static int | |
3252 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3253 | { | |
3254 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3255 | int err; |
2d24d165 | 3256 | |
2d24d165 CL |
3257 | ovs_mutex_lock(&dev->mutex); |
3258 | ||
c1ff66ac CL |
3259 | /* Configure vHost client mode if requested and if the following criteria |
3260 | * are met: | |
2d24d165 CL |
3261 | * 1. Device hasn't been registered yet. |
3262 | * 2. A path has been specified. | |
c1ff66ac CL |
3263 | */ |
3264 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 CL |
3265 | && strlen(dev->vhost_id)) { |
3266 | /* Register client-mode device */ | |
3267 | err = rte_vhost_driver_register(dev->vhost_id, | |
3268 | RTE_VHOST_USER_CLIENT); | |
c1ff66ac | 3269 | if (err) { |
2d24d165 CL |
3270 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3271 | dev->vhost_id); | |
7f381c2e | 3272 | goto unlock; |
c1ff66ac | 3273 | } else { |
2d24d165 CL |
3274 | /* Configuration successful */ |
3275 | dev->vhost_driver_flags |= RTE_VHOST_USER_CLIENT; | |
3276 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " | |
3277 | "using client socket '%s'", | |
3278 | dev->up.name, dev->vhost_id); | |
c1ff66ac | 3279 | } |
f3e7ec25 MW |
3280 | |
3281 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
3282 | &virtio_net_device_ops); | |
3283 | if (err) { | |
3284 | VLOG_ERR("rte_vhost_driver_callback_register failed for " | |
3285 | "vhost user client port: %s\n", dev->up.name); | |
3286 | goto unlock; | |
3287 | } | |
3288 | ||
3289 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
3290 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
3291 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
3292 | | 1ULL << VIRTIO_NET_F_CSUM); | |
3293 | if (err) { | |
3294 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
3295 | "client port: %s\n", dev->up.name); | |
3296 | goto unlock; | |
3297 | } | |
3298 | ||
3299 | err = rte_vhost_driver_start(dev->vhost_id); | |
3300 | if (err) { | |
3301 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
3302 | "client port: %s\n", dev->up.name); | |
3303 | goto unlock; | |
3304 | } | |
c1ff66ac CL |
3305 | } |
3306 | ||
7f381c2e DDP |
3307 | err = dpdk_vhost_reconfigure_helper(dev); |
3308 | ||
3309 | unlock: | |
050c60bf | 3310 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 3311 | |
7f381c2e | 3312 | return err; |
050c60bf DDP |
3313 | } |
3314 | ||
ecc1a34e | 3315 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
81acebda IM |
3316 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
3317 | GET_CARRIER, GET_STATS, \ | |
3318 | GET_FEATURES, GET_STATUS, \ | |
3319 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3320 | { \ |
3321 | NAME, \ | |
118c77b1 | 3322 | true, /* is_pmd */ \ |
ecc1a34e | 3323 | INIT, /* init */ \ |
95fb793a | 3324 | NULL, /* netdev_dpdk_run */ \ |
3325 | NULL, /* netdev_dpdk_wait */ \ | |
3326 | \ | |
3327 | netdev_dpdk_alloc, \ | |
3328 | CONSTRUCT, \ | |
58397e6c | 3329 | DESTRUCT, \ |
95fb793a | 3330 | netdev_dpdk_dealloc, \ |
3331 | netdev_dpdk_get_config, \ | |
81acebda | 3332 | SET_CONFIG, \ |
95fb793a | 3333 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3334 | NULL, /* build header */ \ |
3335 | NULL, /* push header */ \ | |
3336 | NULL, /* pop header */ \ | |
7dec44fe | 3337 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3338 | SET_TX_MULTIQ, \ |
95fb793a | 3339 | \ |
7251515e | 3340 | SEND, /* send */ \ |
95fb793a | 3341 | NULL, /* send_wait */ \ |
3342 | \ | |
3343 | netdev_dpdk_set_etheraddr, \ | |
3344 | netdev_dpdk_get_etheraddr, \ | |
3345 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3346 | netdev_dpdk_set_mtu, \ |
95fb793a | 3347 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3348 | GET_CARRIER, \ |
95fb793a | 3349 | netdev_dpdk_get_carrier_resets, \ |
3350 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
3351 | GET_STATS, \ |
3352 | GET_FEATURES, \ | |
95fb793a | 3353 | NULL, /* set_advertisements */ \ |
875ab130 | 3354 | NULL, /* get_pt_mode */ \ |
95fb793a | 3355 | \ |
9509913a | 3356 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3357 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3358 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3359 | netdev_dpdk_get_qos, \ |
3360 | netdev_dpdk_set_qos, \ | |
95fb793a | 3361 | NULL, /* get_queue */ \ |
3362 | NULL, /* set_queue */ \ | |
3363 | NULL, /* delete_queue */ \ | |
3364 | NULL, /* get_queue_stats */ \ | |
3365 | NULL, /* queue_dump_start */ \ | |
3366 | NULL, /* queue_dump_next */ \ | |
3367 | NULL, /* queue_dump_done */ \ | |
3368 | NULL, /* dump_queue_stats */ \ | |
3369 | \ | |
95fb793a | 3370 | NULL, /* set_in4 */ \ |
a8704b50 | 3371 | NULL, /* get_addr_list */ \ |
95fb793a | 3372 | NULL, /* add_router */ \ |
3373 | NULL, /* get_next_hop */ \ | |
58397e6c | 3374 | GET_STATUS, \ |
95fb793a | 3375 | NULL, /* arp_lookup */ \ |
3376 | \ | |
3377 | netdev_dpdk_update_flags, \ | |
050c60bf | 3378 | RECONFIGURE, \ |
95fb793a | 3379 | \ |
3380 | netdev_dpdk_rxq_alloc, \ | |
3381 | netdev_dpdk_rxq_construct, \ | |
3382 | netdev_dpdk_rxq_destruct, \ | |
3383 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3384 | RXQ_RECV, \ |
95fb793a | 3385 | NULL, /* rx_wait */ \ |
3386 | NULL, /* rxq_drain */ \ | |
18ebd48c | 3387 | NO_OFFLOAD_API \ |
95fb793a | 3388 | } |
8a9562d2 | 3389 | |
bce01e3a | 3390 | static const struct netdev_class dpdk_class = |
95fb793a | 3391 | NETDEV_DPDK_CLASS( |
3392 | "dpdk", | |
ecc1a34e | 3393 | netdev_dpdk_class_init, |
5496878c | 3394 | netdev_dpdk_construct, |
58397e6c | 3395 | netdev_dpdk_destruct, |
81acebda IM |
3396 | netdev_dpdk_set_config, |
3397 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3398 | netdev_dpdk_eth_send, |
3399 | netdev_dpdk_get_carrier, | |
3400 | netdev_dpdk_get_stats, | |
3401 | netdev_dpdk_get_features, | |
3402 | netdev_dpdk_get_status, | |
050c60bf | 3403 | netdev_dpdk_reconfigure, |
58397e6c | 3404 | netdev_dpdk_rxq_recv); |
95fb793a | 3405 | |
bce01e3a | 3406 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3407 | NETDEV_DPDK_CLASS( |
3408 | "dpdkr", | |
ecc1a34e | 3409 | netdev_dpdk_class_init, |
5496878c | 3410 | netdev_dpdk_ring_construct, |
58397e6c | 3411 | netdev_dpdk_destruct, |
c3d062a7 | 3412 | netdev_dpdk_ring_set_config, |
81acebda | 3413 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3414 | netdev_dpdk_ring_send, |
3415 | netdev_dpdk_get_carrier, | |
3416 | netdev_dpdk_get_stats, | |
3417 | netdev_dpdk_get_features, | |
3418 | netdev_dpdk_get_status, | |
050c60bf | 3419 | netdev_dpdk_reconfigure, |
58397e6c KT |
3420 | netdev_dpdk_rxq_recv); |
3421 | ||
53f50d24 | 3422 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3423 | NETDEV_DPDK_CLASS( |
3424 | "dpdkvhostuser", | |
f3e7ec25 | 3425 | NULL, |
53f50d24 | 3426 | netdev_dpdk_vhost_construct, |
58397e6c | 3427 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3428 | NULL, |
81acebda | 3429 | NULL, |
58397e6c KT |
3430 | netdev_dpdk_vhost_send, |
3431 | netdev_dpdk_vhost_get_carrier, | |
3432 | netdev_dpdk_vhost_get_stats, | |
3433 | NULL, | |
7251515e | 3434 | NULL, |
53f50d24 | 3435 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3436 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3437 | static const struct netdev_class dpdk_vhost_client_class = |
3438 | NETDEV_DPDK_CLASS( | |
3439 | "dpdkvhostuserclient", | |
f3e7ec25 | 3440 | NULL, |
2d24d165 CL |
3441 | netdev_dpdk_vhost_client_construct, |
3442 | netdev_dpdk_vhost_destruct, | |
3443 | netdev_dpdk_vhost_client_set_config, | |
3444 | NULL, | |
3445 | netdev_dpdk_vhost_send, | |
3446 | netdev_dpdk_vhost_get_carrier, | |
3447 | netdev_dpdk_vhost_get_stats, | |
3448 | NULL, | |
3449 | NULL, | |
3450 | netdev_dpdk_vhost_client_reconfigure, | |
3451 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 3452 | |
8a9562d2 PS |
3453 | void |
3454 | netdev_dpdk_register(void) | |
3455 | { | |
bab69409 AC |
3456 | netdev_register_provider(&dpdk_class); |
3457 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3458 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 3459 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 3460 | } |