]>
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> |
01961bbd DDP |
25 | |
26 | #include <rte_config.h> | |
27 | #include <rte_cycles.h> | |
28 | #include <rte_errno.h> | |
29 | #include <rte_eth_ring.h> | |
30 | #include <rte_ethdev.h> | |
31 | #include <rte_malloc.h> | |
32 | #include <rte_mbuf.h> | |
33 | #include <rte_meter.h> | |
01961bbd | 34 | #include <rte_virtio_net.h> |
8a9562d2 | 35 | |
7d1ced01 | 36 | #include "dirs.h" |
e14deea0 | 37 | #include "dp-packet.h" |
01961bbd | 38 | #include "dpdk.h" |
8a9562d2 | 39 | #include "dpif-netdev.h" |
e5c0f5a4 | 40 | #include "fatal-signal.h" |
8a9562d2 PS |
41 | #include "netdev-provider.h" |
42 | #include "netdev-vport.h" | |
43 | #include "odp-util.h" | |
eac84432 | 44 | #include "openvswitch/dynamic-string.h" |
25d436fb BW |
45 | #include "openvswitch/list.h" |
46 | #include "openvswitch/ofp-print.h" | |
47 | #include "openvswitch/vlog.h" | |
94143fc4 | 48 | #include "ovs-numa.h" |
8a9562d2 PS |
49 | #include "ovs-thread.h" |
50 | #include "ovs-rcu.h" | |
51 | #include "packets.h" | |
ee89ea7b | 52 | #include "openvswitch/shash.h" |
0bf765f7 | 53 | #include "smap.h" |
8a9562d2 PS |
54 | #include "sset.h" |
55 | #include "unaligned.h" | |
56 | #include "timeval.h" | |
57 | #include "unixctl.h" | |
8a9562d2 | 58 | |
05b49df6 | 59 | VLOG_DEFINE_THIS_MODULE(netdev_dpdk); |
8a9562d2 PS |
60 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
61 | ||
62 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
63 | ||
64 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
65 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
66 | ||
67 | /* | |
68 | * need to reserve tons of extra space in the mbufs so we can align the | |
69 | * DMA addresses to 4KB. | |
18f777b2 TP |
70 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
71 | * performance for standard Ethernet MTU. | |
8a9562d2 | 72 | */ |
58be5c0e MK |
73 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN \ |
74 | + (2 * VLAN_HEADER_LEN)) | |
4be4d22c MK |
75 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) |
76 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
58be5c0e MK |
77 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len) \ |
78 | - ETHER_HDR_LEN - ETHER_CRC_LEN) | |
31b88c97 SS |
79 | #define MBUF_SIZE(mtu) ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) \ |
80 | + sizeof(struct dp_packet) \ | |
81 | + RTE_PKTMBUF_HEADROOM), \ | |
82 | RTE_CACHE_LINE_SIZE) | |
4be4d22c | 83 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 84 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 85 | |
da79ce2b DDP |
86 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
87 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
88 | * enough hugepages) we keep halving the number until the allocation succeeds | |
89 | * or we reach MIN_NB_MBUF */ | |
90 | ||
91 | #define MAX_NB_MBUF (4096 * 64) | |
92 | #define MIN_NB_MBUF (4096 * 4) | |
93 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
94 | ||
95 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
96 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
97 | ||
98 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
99 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
100 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
101 | % MP_CACHE_SZ == 0); | |
102 | ||
d6e3feb5 | 103 | /* |
104 | * DPDK XSTATS Counter names definition | |
105 | */ | |
106 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
107 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
108 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
109 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
110 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
111 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
112 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
113 | ||
114 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
115 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
116 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
117 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
118 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
119 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
120 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
121 | ||
d57f777f | 122 | #define XSTAT_RX_MULTICAST_PACKETS "rx_multicast_packets" |
d6e3feb5 | 123 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" |
124 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
125 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
126 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
127 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
128 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
129 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
130 | ||
8a9562d2 PS |
131 | #define SOCKET0 0 |
132 | ||
b685696b CL |
133 | /* Default size of Physical NIC RXQ */ |
134 | #define NIC_PORT_DEFAULT_RXQ_SIZE 2048 | |
135 | /* Default size of Physical NIC TXQ */ | |
136 | #define NIC_PORT_DEFAULT_TXQ_SIZE 2048 | |
137 | /* Maximum size of Physical NIC Queues */ | |
138 | #define NIC_PORT_MAX_Q_SIZE 4096 | |
79f5354c | 139 | |
585a5bea | 140 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
141 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
142 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
143 | * yet mapped to another queue. */ | |
585a5bea | 144 | |
bb37956a IM |
145 | #define DPDK_ETH_PORT_ID_INVALID RTE_MAX_ETHPORTS |
146 | ||
147 | /* DPDK library uses uint8_t for port_id. */ | |
148 | typedef uint8_t dpdk_port_t; | |
149 | ||
31871ee3 | 150 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 151 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 152 | |
8a9562d2 | 153 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
154 | .rxmode = { |
155 | .mq_mode = ETH_MQ_RX_RSS, | |
156 | .split_hdr_size = 0, | |
157 | .header_split = 0, /* Header Split disabled */ | |
158 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
159 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
160 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
161 | .hw_strip_crc = 0, | |
162 | }, | |
163 | .rx_adv_conf = { | |
164 | .rss_conf = { | |
165 | .rss_key = NULL, | |
543342a4 | 166 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 167 | }, |
a28ddd11 DDP |
168 | }, |
169 | .txmode = { | |
170 | .mq_mode = ETH_MQ_TX_NONE, | |
171 | }, | |
8a9562d2 PS |
172 | }; |
173 | ||
58f7c37b DDP |
174 | enum { DPDK_RING_SIZE = 256 }; |
175 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
176 | enum { DRAIN_TSC = 200000ULL }; |
177 | ||
58397e6c KT |
178 | enum dpdk_dev_type { |
179 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 180 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
181 | }; |
182 | ||
0bf765f7 IS |
183 | /* Quality of Service */ |
184 | ||
185 | /* An instance of a QoS configuration. Always associated with a particular | |
186 | * network device. | |
187 | * | |
188 | * Each QoS implementation subclasses this with whatever additional data it | |
189 | * needs. | |
190 | */ | |
191 | struct qos_conf { | |
192 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 193 | rte_spinlock_t lock; |
0bf765f7 IS |
194 | }; |
195 | ||
196 | /* A particular implementation of dpdk QoS operations. | |
197 | * | |
198 | * The functions below return 0 if successful or a positive errno value on | |
199 | * failure, except where otherwise noted. All of them must be provided, except | |
200 | * where otherwise noted. | |
201 | */ | |
202 | struct dpdk_qos_ops { | |
203 | ||
204 | /* Name of the QoS type */ | |
205 | const char *qos_name; | |
206 | ||
78bd47cf DDP |
207 | /* Called to construct a qos_conf object. The implementation should make |
208 | * the appropriate calls to configure QoS according to 'details'. | |
0bf765f7 IS |
209 | * |
210 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
211 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
212 | * (which is built as ovs-vswitchd.conf.db(8)). | |
213 | * | |
78bd47cf DDP |
214 | * This function must return 0 if and only if it sets '*conf' to an |
215 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
216 | * |
217 | * For all QoS implementations it should always be non-null. | |
218 | */ | |
78bd47cf | 219 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
220 | |
221 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 222 | * 'qos_conf'. |
0bf765f7 IS |
223 | * |
224 | * For all QoS implementations it should always be non-null. | |
225 | */ | |
78bd47cf | 226 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 227 | |
78bd47cf | 228 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
229 | * |
230 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
231 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
232 | * (which is built as ovs-vswitchd.conf.db(8)). | |
233 | */ | |
78bd47cf | 234 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 235 | |
78bd47cf | 236 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
237 | * |
238 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
239 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
240 | * (which is built as ovs-vswitchd.conf.db(8)). | |
241 | * | |
78bd47cf | 242 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 243 | */ |
78bd47cf DDP |
244 | bool (*qos_is_equal)(const struct qos_conf *conf, |
245 | const struct smap *details); | |
0bf765f7 IS |
246 | |
247 | /* Modify an array of rte_mbufs. The modification is specific to | |
248 | * each qos implementation. | |
249 | * | |
250 | * The function should take and array of mbufs and an int representing | |
251 | * the current number of mbufs present in the array. | |
252 | * | |
253 | * After the function has performed a qos modification to the array of | |
254 | * mbufs it returns an int representing the number of mbufs now present in | |
255 | * the array. This value is can then be passed to the port send function | |
256 | * along with the modified array for transmission. | |
257 | * | |
258 | * For all QoS implementations it should always be non-null. | |
259 | */ | |
78bd47cf DDP |
260 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
261 | int pkt_cnt); | |
0bf765f7 IS |
262 | }; |
263 | ||
264 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
265 | static const struct dpdk_qos_ops egress_policer_ops; | |
266 | ||
267 | /* | |
268 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
269 | * operations. | |
270 | */ | |
271 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
272 | &egress_policer_ops, | |
273 | NULL | |
274 | }; | |
275 | ||
c2adb102 IM |
276 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
277 | ||
8a9562d2 | 278 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 279 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 280 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 281 | |
c2adb102 IM |
282 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
283 | = OVS_MUTEX_INITIALIZER; | |
284 | ||
285 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mp_mutex) | |
55951e15 | 286 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 287 | |
8a9562d2 PS |
288 | struct dpdk_mp { |
289 | struct rte_mempool *mp; | |
290 | int mtu; | |
291 | int socket_id; | |
292 | int refcount; | |
c2adb102 | 293 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mp_mutex); |
8a9562d2 PS |
294 | }; |
295 | ||
5a034064 AW |
296 | /* There should be one 'struct dpdk_tx_queue' created for |
297 | * each cpu core. */ | |
8a9562d2 | 298 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
299 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
300 | * from concurrent access. It is used only | |
301 | * if the queue is shared among different | |
324c8374 | 302 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
303 | int map; /* Mapping of configured vhost-user queues |
304 | * to enabled by guest. */ | |
8a9562d2 PS |
305 | }; |
306 | ||
95fb793a | 307 | /* dpdk has no way to remove dpdk ring ethernet devices |
308 | so we have to keep them around once they've been created | |
309 | */ | |
310 | ||
ca6ba700 | 311 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 312 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 313 | |
314 | struct dpdk_ring { | |
315 | /* For the client rings */ | |
316 | struct rte_ring *cring_tx; | |
317 | struct rte_ring *cring_rx; | |
b83a2df1 | 318 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
bb37956a | 319 | dpdk_port_t eth_port_id; /* ethernet device port id */ |
ca6ba700 | 320 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 321 | }; |
322 | ||
9509913a IS |
323 | struct ingress_policer { |
324 | struct rte_meter_srtcm_params app_srtcm_params; | |
325 | struct rte_meter_srtcm in_policer; | |
326 | rte_spinlock_t policer_lock; | |
327 | }; | |
328 | ||
1a2bb118 SC |
329 | enum dpdk_hw_ol_features { |
330 | NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0, | |
331 | }; | |
332 | ||
8a9562d2 PS |
333 | struct netdev_dpdk { |
334 | struct netdev up; | |
bb37956a | 335 | dpdk_port_t port_id; |
8a9562d2 | 336 | int max_packet_len; |
58397e6c | 337 | enum dpdk_dev_type type; |
8a9562d2 | 338 | |
5a034064 | 339 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
340 | |
341 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
342 | ||
343 | struct dpdk_mp *dpdk_mp; | |
344 | int mtu; | |
345 | int socket_id; | |
346 | int buf_size; | |
8a9562d2 | 347 | struct netdev_stats stats; |
45d947c4 DDP |
348 | /* Protects stats */ |
349 | rte_spinlock_t stats_lock; | |
8a9562d2 | 350 | |
74ff3298 | 351 | struct eth_addr hwaddr; |
8a9562d2 PS |
352 | enum netdev_flags flags; |
353 | ||
354 | struct rte_eth_link link; | |
355 | int link_reset_cnt; | |
356 | ||
0a0f39df CL |
357 | /* virtio identifier for vhost devices */ |
358 | ovsrcu_index vid; | |
359 | ||
360 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
361 | bool vhost_reconfigured; | |
58397e6c | 362 | |
2d24d165 CL |
363 | /* Identifier used to distinguish vhost devices from each other. */ |
364 | char vhost_id[PATH_MAX]; | |
7d1ced01 | 365 | |
55e075e6 CL |
366 | /* Device arguments for dpdk ports */ |
367 | char *devargs; | |
368 | ||
5dcde09c IM |
369 | /* If true, device was attached by rte_eth_dev_attach(). */ |
370 | bool attached; | |
371 | ||
8a9562d2 | 372 | /* In dpdk_list. */ |
ca6ba700 | 373 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
0bf765f7 IS |
374 | |
375 | /* QoS configuration and lock for the device */ | |
78bd47cf | 376 | OVSRCU_TYPE(struct qos_conf *) qos_conf; |
0bf765f7 | 377 | |
050c60bf DDP |
378 | /* The following properties cannot be changed when a device is running, |
379 | * so we remember the request and update them next time | |
380 | * netdev_dpdk*_reconfigure() is called */ | |
0072e931 | 381 | int requested_mtu; |
050c60bf DDP |
382 | int requested_n_txq; |
383 | int requested_n_rxq; | |
b685696b CL |
384 | int requested_rxq_size; |
385 | int requested_txq_size; | |
386 | ||
387 | /* Number of rx/tx descriptors for physical devices */ | |
388 | int rxq_size; | |
389 | int txq_size; | |
9509913a | 390 | |
db8f13b0 CL |
391 | /* Socket ID detected when vHost device is brought up */ |
392 | int requested_socket_id; | |
393 | ||
c1ff66ac CL |
394 | /* Denotes whether vHost port is client/server mode */ |
395 | uint64_t vhost_driver_flags; | |
396 | ||
9509913a IS |
397 | /* Ingress Policer */ |
398 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
399 | uint32_t policer_rate; | |
400 | uint32_t policer_burst; | |
9fd39370 SC |
401 | |
402 | /* DPDK-ETH Flow control */ | |
403 | struct rte_eth_fc_conf fc_conf; | |
1a2bb118 SC |
404 | |
405 | /* DPDK-ETH hardware offload features, | |
406 | * from the enum set 'dpdk_hw_ol_features' */ | |
407 | uint32_t hw_ol_features; | |
8a9562d2 PS |
408 | }; |
409 | ||
410 | struct netdev_rxq_dpdk { | |
411 | struct netdev_rxq up; | |
bb37956a | 412 | dpdk_port_t port_id; |
8a9562d2 PS |
413 | }; |
414 | ||
b3722dca | 415 | static int netdev_dpdk_class_init(void); |
416 | static int netdev_dpdk_vhost_class_init(void); | |
8a9562d2 | 417 | |
0a0f39df | 418 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 419 | |
9509913a IS |
420 | struct ingress_policer * |
421 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
422 | ||
8a9562d2 PS |
423 | static bool |
424 | is_dpdk_class(const struct netdev_class *class) | |
425 | { | |
b3722dca | 426 | return class->init == netdev_dpdk_class_init |
427 | || class->init == netdev_dpdk_vhost_class_init; | |
8a9562d2 PS |
428 | } |
429 | ||
4be4d22c MK |
430 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
431 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
432 | * value, insufficient buffers are allocated to accomodate the packet in its | |
433 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
434 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
435 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
436 | * behaviour, which reduces performance. To prevent this, use a buffer size |
437 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
438 | */ |
439 | static uint32_t | |
440 | dpdk_buf_size(int mtu) | |
441 | { | |
442 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
443 | NETDEV_DPDK_MBUF_ALIGN); | |
444 | } | |
445 | ||
eff23640 DDP |
446 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
447 | * | |
448 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
449 | static void * |
450 | dpdk_rte_mzalloc(size_t sz) | |
451 | { | |
eff23640 | 452 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
453 | } |
454 | ||
455 | void | |
e14deea0 | 456 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 457 | { |
db73f716 | 458 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 459 | |
b00b4a81 | 460 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
461 | } |
462 | ||
b3cd9f9d | 463 | static void |
401b70d6 | 464 | ovs_rte_pktmbuf_init(struct rte_mempool *mp OVS_UNUSED, |
b3cd9f9d | 465 | void *opaque_arg OVS_UNUSED, |
2391135c | 466 | void *_p, |
b3cd9f9d PS |
467 | unsigned i OVS_UNUSED) |
468 | { | |
2391135c | 469 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 470 | |
2391135c | 471 | dp_packet_init_dpdk((struct dp_packet *) pkt, pkt->buf_len); |
b3cd9f9d PS |
472 | } |
473 | ||
8a9562d2 | 474 | static struct dpdk_mp * |
2ae3d542 | 475 | dpdk_mp_create(int socket_id, int mtu) |
8a9562d2 | 476 | { |
2ae3d542 | 477 | struct dpdk_mp *dmp; |
da79ce2b | 478 | unsigned mp_size; |
0c6f39e5 | 479 | char *mp_name; |
8a9562d2 PS |
480 | |
481 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
eff23640 DDP |
482 | if (!dmp) { |
483 | return NULL; | |
484 | } | |
8a9562d2 PS |
485 | dmp->socket_id = socket_id; |
486 | dmp->mtu = mtu; | |
487 | dmp->refcount = 1; | |
0072e931 | 488 | /* XXX: this is a really rough method of provisioning memory. |
2ae3d542 | 489 | * It's impossible to determine what the exact memory requirements are |
58be5c0e MK |
490 | * when the number of ports and rxqs that utilize a particular mempool can |
491 | * change dynamically at runtime. For now, use this rough heurisitic. | |
0072e931 MK |
492 | */ |
493 | if (mtu >= ETHER_MTU) { | |
494 | mp_size = MAX_NB_MBUF; | |
495 | } else { | |
496 | mp_size = MIN_NB_MBUF; | |
497 | } | |
8a9562d2 | 498 | |
da79ce2b | 499 | do { |
0c6f39e5 DDP |
500 | mp_name = xasprintf("ovs_mp_%d_%d_%u", dmp->mtu, dmp->socket_id, |
501 | mp_size); | |
95fb793a | 502 | |
401b70d6 HA |
503 | dmp->mp = rte_pktmbuf_pool_create(mp_name, mp_size, |
504 | MP_CACHE_SZ, | |
505 | sizeof (struct dp_packet) | |
506 | - sizeof (struct rte_mbuf), | |
507 | MBUF_SIZE(mtu) | |
508 | - sizeof(struct dp_packet), | |
509 | socket_id); | |
0c6f39e5 DDP |
510 | if (dmp->mp) { |
511 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", | |
512 | mp_name, mp_size); | |
513 | } | |
514 | free(mp_name); | |
515 | if (dmp->mp) { | |
401b70d6 HA |
516 | /* rte_pktmbuf_pool_create has done some initialization of the |
517 | * rte_mbuf part of each dp_packet, while ovs_rte_pktmbuf_init | |
518 | * initializes some OVS specific fields of dp_packet. | |
519 | */ | |
520 | rte_mempool_obj_iter(dmp->mp, ovs_rte_pktmbuf_init, NULL); | |
0c6f39e5 DDP |
521 | return dmp; |
522 | } | |
523 | } while (rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
2ae3d542 | 524 | |
2ae3d542 DDP |
525 | rte_free(dmp); |
526 | return NULL; | |
527 | } | |
528 | ||
529 | static struct dpdk_mp * | |
530 | dpdk_mp_get(int socket_id, int mtu) | |
531 | { | |
532 | struct dpdk_mp *dmp; | |
533 | ||
534 | ovs_mutex_lock(&dpdk_mp_mutex); | |
535 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
536 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
537 | dmp->refcount++; | |
538 | goto out; | |
539 | } | |
540 | } | |
541 | ||
542 | dmp = dpdk_mp_create(socket_id, mtu); | |
96e9b168 IS |
543 | if (dmp) { |
544 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); | |
545 | } | |
5f88de0d | 546 | |
c2adb102 IM |
547 | out: |
548 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
549 | ||
c2adb102 | 550 | return dmp; |
8a9562d2 PS |
551 | } |
552 | ||
553 | static void | |
c2adb102 | 554 | dpdk_mp_put(struct dpdk_mp *dmp) |
8a9562d2 | 555 | { |
8a9562d2 PS |
556 | if (!dmp) { |
557 | return; | |
558 | } | |
559 | ||
c2adb102 | 560 | ovs_mutex_lock(&dpdk_mp_mutex); |
8d38823b | 561 | ovs_assert(dmp->refcount); |
8a9562d2 | 562 | |
8d38823b MK |
563 | if (!--dmp->refcount) { |
564 | ovs_list_remove(&dmp->list_node); | |
565 | rte_mempool_free(dmp->mp); | |
5f88de0d | 566 | rte_free(dmp); |
8a9562d2 | 567 | } |
c2adb102 | 568 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
569 | } |
570 | ||
0072e931 MK |
571 | /* Tries to allocate new mempool on requested_socket_id with |
572 | * mbuf size corresponding to requested_mtu. | |
573 | * On success new configuration will be applied. | |
574 | * On error, device will be left unchanged. */ | |
575 | static int | |
576 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
577 | OVS_REQUIRES(dev->mutex) |
578 | { | |
579 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
580 | struct dpdk_mp *mp; | |
581 | ||
582 | mp = dpdk_mp_get(dev->requested_socket_id, FRAME_LEN_TO_MTU(buf_size)); | |
583 | if (!mp) { | |
c67e46c0 MK |
584 | VLOG_ERR("Failed to create memory pool for netdev " |
585 | "%s, with MTU %d on socket %d: %s\n", | |
586 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
587 | rte_strerror(rte_errno)); | |
588 | return rte_errno; | |
0072e931 MK |
589 | } else { |
590 | dpdk_mp_put(dev->dpdk_mp); | |
591 | dev->dpdk_mp = mp; | |
592 | dev->mtu = dev->requested_mtu; | |
593 | dev->socket_id = dev->requested_socket_id; | |
594 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
595 | } | |
596 | ||
597 | return 0; | |
598 | } | |
599 | ||
8a9562d2 PS |
600 | static void |
601 | check_link_status(struct netdev_dpdk *dev) | |
602 | { | |
603 | struct rte_eth_link link; | |
604 | ||
605 | rte_eth_link_get_nowait(dev->port_id, &link); | |
606 | ||
607 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 608 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
609 | |
610 | dev->link_reset_cnt++; | |
611 | dev->link = link; | |
612 | if (dev->link.link_status) { | |
bb37956a | 613 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Up - speed %u Mbps - %s", |
58be5c0e | 614 | dev->port_id, (unsigned) dev->link.link_speed, |
8a9562d2 PS |
615 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? |
616 | ("full-duplex") : ("half-duplex")); | |
617 | } else { | |
bb37956a | 618 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Down", dev->port_id); |
8a9562d2 PS |
619 | } |
620 | } | |
621 | } | |
622 | ||
623 | static void * | |
624 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
625 | { | |
626 | struct netdev_dpdk *dev; | |
627 | ||
628 | pthread_detach(pthread_self()); | |
629 | ||
630 | for (;;) { | |
631 | ovs_mutex_lock(&dpdk_mutex); | |
632 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
633 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
634 | if (dev->type == DPDK_DEV_ETH) { |
635 | check_link_status(dev); | |
636 | } | |
8a9562d2 PS |
637 | ovs_mutex_unlock(&dev->mutex); |
638 | } | |
639 | ovs_mutex_unlock(&dpdk_mutex); | |
640 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
641 | } | |
642 | ||
643 | return NULL; | |
644 | } | |
645 | ||
b98d7669 DDP |
646 | static int |
647 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
648 | { | |
649 | int diag = 0; | |
650 | int i; | |
0072e931 | 651 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 652 | |
0072e931 MK |
653 | if (dev->mtu > ETHER_MTU) { |
654 | conf.rxmode.jumbo_frame = 1; | |
655 | conf.rxmode.max_rx_pkt_len = dev->max_packet_len; | |
656 | } else { | |
657 | conf.rxmode.jumbo_frame = 0; | |
658 | conf.rxmode.max_rx_pkt_len = 0; | |
659 | } | |
1a2bb118 SC |
660 | conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & |
661 | NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
b98d7669 DDP |
662 | /* A device may report more queues than it makes available (this has |
663 | * been observed for Intel xl710, which reserves some of them for | |
664 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
665 | * available. When this happens we can retry the configuration | |
666 | * and request less queues */ | |
667 | while (n_rxq && n_txq) { | |
668 | if (diag) { | |
669 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
670 | } | |
671 | ||
0072e931 | 672 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 673 | if (diag) { |
0072e931 MK |
674 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
675 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
676 | break; |
677 | } | |
678 | ||
679 | for (i = 0; i < n_txq; i++) { | |
b685696b | 680 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
681 | dev->socket_id, NULL); |
682 | if (diag) { | |
683 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
684 | dev->up.name, i, rte_strerror(-diag)); | |
685 | break; | |
686 | } | |
687 | } | |
688 | ||
689 | if (i != n_txq) { | |
690 | /* Retry with less tx queues */ | |
691 | n_txq = i; | |
692 | continue; | |
693 | } | |
694 | ||
695 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 696 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
b98d7669 DDP |
697 | dev->socket_id, NULL, |
698 | dev->dpdk_mp->mp); | |
699 | if (diag) { | |
700 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
701 | dev->up.name, i, rte_strerror(-diag)); | |
702 | break; | |
703 | } | |
704 | } | |
705 | ||
706 | if (i != n_rxq) { | |
707 | /* Retry with less rx queues */ | |
708 | n_rxq = i; | |
709 | continue; | |
710 | } | |
711 | ||
712 | dev->up.n_rxq = n_rxq; | |
81acebda | 713 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
714 | |
715 | return 0; | |
716 | } | |
717 | ||
718 | return diag; | |
719 | } | |
720 | ||
1a2bb118 SC |
721 | static void |
722 | dpdk_eth_checksum_offload_configure(struct netdev_dpdk *dev) | |
723 | OVS_REQUIRES(dev->mutex) | |
724 | { | |
725 | struct rte_eth_dev_info info; | |
726 | bool rx_csum_ol_flag = false; | |
727 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | | |
728 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
729 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
730 | rte_eth_dev_info_get(dev->port_id, &info); | |
731 | rx_csum_ol_flag = (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
732 | ||
733 | if (rx_csum_ol_flag && | |
734 | (info.rx_offload_capa & rx_chksm_offload_capa) != | |
735 | rx_chksm_offload_capa) { | |
bb37956a IM |
736 | VLOG_WARN_ONCE("Rx checksum offload is not supported on device %"PRIu8, |
737 | dev->port_id); | |
1a2bb118 SC |
738 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; |
739 | return; | |
740 | } | |
741 | netdev_request_reconfigure(&dev->up); | |
742 | } | |
743 | ||
9fd39370 SC |
744 | static void |
745 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
746 | { | |
747 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
bb37956a IM |
748 | VLOG_WARN("Failed to enable flow control on device %"PRIu8, |
749 | dev->port_id); | |
9fd39370 SC |
750 | } |
751 | } | |
b98d7669 | 752 | |
8a9562d2 | 753 | static int |
c2adb102 IM |
754 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
755 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
756 | { |
757 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 758 | struct rte_eth_dev_info info; |
8a9562d2 PS |
759 | struct ether_addr eth_addr; |
760 | int diag; | |
b98d7669 | 761 | int n_rxq, n_txq; |
8a9562d2 | 762 | |
a0cb2d66 | 763 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 764 | |
b98d7669 DDP |
765 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
766 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
767 | ||
768 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 769 | if (diag) { |
b98d7669 DDP |
770 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
771 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 772 | return -diag; |
8a9562d2 PS |
773 | } |
774 | ||
8a9562d2 PS |
775 | diag = rte_eth_dev_start(dev->port_id); |
776 | if (diag) { | |
b98d7669 DDP |
777 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
778 | rte_strerror(-diag)); | |
95fb793a | 779 | return -diag; |
8a9562d2 PS |
780 | } |
781 | ||
782 | rte_eth_promiscuous_enable(dev->port_id); | |
783 | rte_eth_allmulticast_enable(dev->port_id); | |
784 | ||
785 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
786 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
bb37956a | 787 | VLOG_INFO_RL(&rl, "Port %"PRIu8": "ETH_ADDR_FMT, |
ca92d173 | 788 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 789 | |
ca92d173 | 790 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
791 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
792 | ||
793 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
794 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
795 | ||
9fd39370 SC |
796 | /* Get the Flow control configuration for DPDK-ETH */ |
797 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
798 | if (diag) { | |
bb37956a | 799 | VLOG_DBG("cannot get flow control parameters on port=%"PRIu8", err=%d", |
9fd39370 SC |
800 | dev->port_id, diag); |
801 | } | |
802 | ||
8a9562d2 PS |
803 | return 0; |
804 | } | |
805 | ||
806 | static struct netdev_dpdk * | |
807 | netdev_dpdk_cast(const struct netdev *netdev) | |
808 | { | |
809 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
810 | } | |
811 | ||
812 | static struct netdev * | |
813 | netdev_dpdk_alloc(void) | |
814 | { | |
bab69409 AC |
815 | struct netdev_dpdk *dev; |
816 | ||
65e19e70 DDP |
817 | dev = dpdk_rte_mzalloc(sizeof *dev); |
818 | if (dev) { | |
819 | return &dev->up; | |
bab69409 | 820 | } |
65e19e70 | 821 | |
bab69409 | 822 | return NULL; |
8a9562d2 PS |
823 | } |
824 | ||
eff23640 DDP |
825 | static struct dpdk_tx_queue * |
826 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 827 | { |
eff23640 | 828 | struct dpdk_tx_queue *txqs; |
bd5131ba | 829 | unsigned i; |
5a034064 | 830 | |
eff23640 DDP |
831 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
832 | if (txqs) { | |
833 | for (i = 0; i < n_txqs; i++) { | |
834 | /* Initialize map for vhost devices. */ | |
835 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
836 | rte_spinlock_init(&txqs[i].tx_lock); | |
837 | } | |
5a034064 | 838 | } |
eff23640 DDP |
839 | |
840 | return txqs; | |
5a034064 AW |
841 | } |
842 | ||
8a9562d2 | 843 | static int |
bb37956a | 844 | common_construct(struct netdev *netdev, dpdk_port_t port_no, |
1ce30dfd | 845 | enum dpdk_dev_type type, int socket_id) |
5a034064 | 846 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 847 | { |
d46285a2 | 848 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 849 | |
d46285a2 | 850 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 851 | |
d46285a2 | 852 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 853 | |
1b7a04e0 AW |
854 | /* If the 'sid' is negative, it means that the kernel fails |
855 | * to obtain the pci numa info. In that situation, always | |
856 | * use 'SOCKET0'. */ | |
1ce30dfd | 857 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 858 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
859 | dev->port_id = port_no; |
860 | dev->type = type; | |
861 | dev->flags = 0; | |
7f381c2e | 862 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 863 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
864 | ovsrcu_index_init(&dev->vid, -1); |
865 | dev->vhost_reconfigured = false; | |
5dcde09c | 866 | dev->attached = false; |
8a9562d2 | 867 | |
78bd47cf | 868 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 869 | |
9509913a IS |
870 | ovsrcu_init(&dev->ingress_policer, NULL); |
871 | dev->policer_rate = 0; | |
872 | dev->policer_burst = 0; | |
873 | ||
7f381c2e DDP |
874 | netdev->n_rxq = 0; |
875 | netdev->n_txq = 0; | |
876 | dev->requested_n_rxq = NR_QUEUE; | |
877 | dev->requested_n_txq = NR_QUEUE; | |
878 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
879 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 880 | |
9fd39370 SC |
881 | /* Initialize the flow control to NULL */ |
882 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
883 | |
884 | /* Initilize the hardware offload flags to 0 */ | |
885 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
886 | |
887 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
888 | ||
d46285a2 | 889 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 890 | |
7f381c2e DDP |
891 | netdev_request_reconfigure(netdev); |
892 | ||
1ce30dfd | 893 | return 0; |
95fb793a | 894 | } |
895 | ||
b83a2df1 MV |
896 | /* dev_name must be the prefix followed by a positive decimal number. |
897 | * (no leading + or - signs are allowed) */ | |
95fb793a | 898 | static int |
899 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
900 | unsigned int *port_no) | |
901 | { | |
902 | const char *cport; | |
903 | ||
904 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
905 | return ENODEV; | |
906 | } | |
907 | ||
908 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
909 | |
910 | if (str_to_uint(cport, 10, port_no)) { | |
911 | return 0; | |
912 | } else { | |
913 | return ENODEV; | |
914 | } | |
95fb793a | 915 | } |
916 | ||
1ce30dfd DDP |
917 | static int |
918 | vhost_common_construct(struct netdev *netdev) | |
919 | OVS_REQUIRES(dpdk_mutex) | |
920 | { | |
921 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
922 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
923 | ||
924 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); | |
925 | if (!dev->tx_q) { | |
926 | return ENOMEM; | |
927 | } | |
928 | ||
bb37956a IM |
929 | return common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
930 | DPDK_DEV_VHOST, socket_id); | |
1ce30dfd DDP |
931 | } |
932 | ||
7d1ced01 | 933 | static int |
53f50d24 | 934 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 935 | { |
d46285a2 DDP |
936 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
937 | const char *name = netdev->name; | |
7d1ced01 | 938 | int err; |
a0cb2d66 | 939 | |
1af27e8a DDP |
940 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
941 | * the file system. '/' or '\' would traverse directories, so they're not | |
942 | * acceptable in 'name'. */ | |
943 | if (strchr(name, '/') || strchr(name, '\\')) { | |
944 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
945 | "A valid name must not include '/' or '\\'", | |
946 | name); | |
947 | return EINVAL; | |
948 | } | |
949 | ||
7d1ced01 CL |
950 | ovs_mutex_lock(&dpdk_mutex); |
951 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 952 | * the socket is to be created, then register the socket. |
7d1ced01 | 953 | */ |
2d24d165 | 954 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 955 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 956 | |
2d24d165 CL |
957 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
958 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
959 | if (err) { |
960 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 961 | dev->vhost_id); |
e5c0f5a4 | 962 | } else { |
2d24d165 CL |
963 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
964 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
965 | dev->vhost_id, name); | |
966 | } | |
1ce30dfd | 967 | err = vhost_common_construct(netdev); |
2d24d165 CL |
968 | |
969 | ovs_mutex_unlock(&dpdk_mutex); | |
28ca969e AC |
970 | VLOG_WARN_ONCE("dpdkvhostuser ports are considered deprecated; " |
971 | "please migrate to dpdkvhostuserclient ports."); | |
2d24d165 CL |
972 | return err; |
973 | } | |
974 | ||
975 | static int | |
976 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
977 | { | |
978 | int err; | |
979 | ||
2d24d165 | 980 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 981 | err = vhost_common_construct(netdev); |
7d1ced01 | 982 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
983 | return err; |
984 | } | |
985 | ||
95fb793a | 986 | static int |
987 | netdev_dpdk_construct(struct netdev *netdev) | |
988 | { | |
95fb793a | 989 | int err; |
990 | ||
95fb793a | 991 | ovs_mutex_lock(&dpdk_mutex); |
bb37956a IM |
992 | err = common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
993 | DPDK_DEV_ETH, SOCKET0); | |
8a9562d2 PS |
994 | ovs_mutex_unlock(&dpdk_mutex); |
995 | return err; | |
996 | } | |
997 | ||
1ce30dfd DDP |
998 | static void |
999 | common_destruct(struct netdev_dpdk *dev) | |
1000 | OVS_REQUIRES(dpdk_mutex) | |
1001 | OVS_EXCLUDED(dev->mutex) | |
1002 | { | |
1003 | rte_free(dev->tx_q); | |
1004 | dpdk_mp_put(dev->dpdk_mp); | |
1005 | ||
1006 | ovs_list_remove(&dev->list_node); | |
1007 | free(ovsrcu_get_protected(struct ingress_policer *, | |
1008 | &dev->ingress_policer)); | |
1009 | ovs_mutex_destroy(&dev->mutex); | |
1010 | } | |
1011 | ||
8a9562d2 | 1012 | static void |
d46285a2 | 1013 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 1014 | { |
d46285a2 | 1015 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5dcde09c | 1016 | char devname[RTE_ETH_NAME_MAX_LEN]; |
8a9562d2 | 1017 | |
8d38823b | 1018 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1019 | |
8a9562d2 | 1020 | rte_eth_dev_stop(dev->port_id); |
5dcde09c IM |
1021 | |
1022 | if (dev->attached) { | |
1023 | rte_eth_dev_close(dev->port_id); | |
1024 | if (rte_eth_dev_detach(dev->port_id, devname) < 0) { | |
1025 | VLOG_ERR("Device '%s' can not be detached", dev->devargs); | |
1026 | } else { | |
1027 | VLOG_INFO("Device '%s' detached", devname); | |
1028 | } | |
1029 | } | |
1030 | ||
55e075e6 | 1031 | free(dev->devargs); |
1ce30dfd | 1032 | common_destruct(dev); |
8d38823b | 1033 | |
8a9562d2 | 1034 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1035 | } |
8a9562d2 | 1036 | |
3f891bbe DDP |
1037 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1038 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1039 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1040 | static int | |
c1ff66ac CL |
1041 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1042 | char *vhost_id) | |
3f891bbe DDP |
1043 | OVS_EXCLUDED(dpdk_mutex) |
1044 | OVS_EXCLUDED(dev->mutex) | |
1045 | { | |
c1ff66ac | 1046 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1047 | } |
1048 | ||
58397e6c | 1049 | static void |
d46285a2 | 1050 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1051 | { |
d46285a2 | 1052 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1053 | char *vhost_id; |
58397e6c | 1054 | |
8d38823b | 1055 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1056 | |
c62da695 | 1057 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1058 | if (netdev_dpdk_get_vid(dev) >= 0 |
1059 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1060 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1061 | netdev->name); |
58be5c0e MK |
1062 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1063 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1064 | } |
1065 | ||
2d24d165 | 1066 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1067 | |
1ce30dfd DDP |
1068 | common_destruct(dev); |
1069 | ||
58397e6c | 1070 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1071 | |
569c26da | 1072 | if (!vhost_id[0]) { |
821b8664 IM |
1073 | goto out; |
1074 | } | |
1075 | ||
c1ff66ac | 1076 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1077 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1078 | netdev->name, vhost_id); | |
c1ff66ac CL |
1079 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1080 | /* OVS server mode - remove this socket from list for deletion */ | |
1081 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1082 | } |
821b8664 | 1083 | out: |
c1ff66ac | 1084 | free(vhost_id); |
8a9562d2 PS |
1085 | } |
1086 | ||
1087 | static void | |
d46285a2 | 1088 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1089 | { |
d46285a2 | 1090 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1091 | |
d46285a2 | 1092 | rte_free(dev); |
8a9562d2 PS |
1093 | } |
1094 | ||
1095 | static int | |
a14b8947 | 1096 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1097 | { |
a14b8947 | 1098 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1099 | |
1100 | ovs_mutex_lock(&dev->mutex); | |
1101 | ||
050c60bf | 1102 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1103 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1104 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1105 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1106 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1107 | |
1108 | if (dev->type == DPDK_DEV_ETH) { | |
1109 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1110 | dev->requested_rxq_size); | |
1111 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1112 | dev->rxq_size); | |
1113 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1114 | dev->requested_txq_size); | |
1115 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1116 | dev->txq_size); | |
1a2bb118 SC |
1117 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1118 | smap_add(args, "rx_csum_offload", "true"); | |
8155ab7e KT |
1119 | } else { |
1120 | smap_add(args, "rx_csum_offload", "false"); | |
1a2bb118 | 1121 | } |
451f26fd | 1122 | } |
8a9562d2 PS |
1123 | ovs_mutex_unlock(&dev->mutex); |
1124 | ||
1125 | return 0; | |
1126 | } | |
1127 | ||
55e075e6 | 1128 | static struct netdev_dpdk * |
bb37956a | 1129 | netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id) |
55e075e6 CL |
1130 | OVS_REQUIRES(dpdk_mutex) |
1131 | { | |
1132 | struct netdev_dpdk *dev; | |
1133 | ||
1134 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1135 | if (dev->port_id == port_id) { | |
1136 | return dev; | |
1137 | } | |
1138 | } | |
1139 | ||
1140 | return NULL; | |
1141 | } | |
1142 | ||
bb37956a | 1143 | static dpdk_port_t |
5dcde09c IM |
1144 | netdev_dpdk_process_devargs(struct netdev_dpdk *dev, |
1145 | const char *devargs, char **errp) | |
55e075e6 | 1146 | { |
03d6399e IM |
1147 | /* Get the name up to the first comma. */ |
1148 | char *name = xmemdup0(devargs, strcspn(devargs, ",")); | |
bb37956a | 1149 | dpdk_port_t new_port_id = DPDK_ETH_PORT_ID_INVALID; |
55e075e6 | 1150 | |
69876ed7 | 1151 | if (!rte_eth_dev_count() |
03d6399e | 1152 | || rte_eth_dev_get_port_by_name(name, &new_port_id) |
69876ed7 CL |
1153 | || !rte_eth_dev_is_valid_port(new_port_id)) { |
1154 | /* Device not found in DPDK, attempt to attach it */ | |
1155 | if (!rte_eth_dev_attach(devargs, &new_port_id)) { | |
1156 | /* Attach successful */ | |
5dcde09c | 1157 | dev->attached = true; |
69876ed7 CL |
1158 | VLOG_INFO("Device '%s' attached to DPDK", devargs); |
1159 | } else { | |
1160 | /* Attach unsuccessful */ | |
bb37956a | 1161 | new_port_id = DPDK_ETH_PORT_ID_INVALID; |
47a45d86 KT |
1162 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", |
1163 | devargs); | |
55e075e6 CL |
1164 | } |
1165 | } | |
1166 | ||
03d6399e | 1167 | free(name); |
55e075e6 CL |
1168 | return new_port_id; |
1169 | } | |
1170 | ||
c3d062a7 CL |
1171 | static void |
1172 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1173 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1174 | { |
050c60bf | 1175 | int new_n_rxq; |
a14b8947 | 1176 | |
2a21e757 | 1177 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1178 | if (new_n_rxq != dev->requested_n_rxq) { |
1179 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1180 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1181 | } |
c3d062a7 CL |
1182 | } |
1183 | ||
b685696b CL |
1184 | static void |
1185 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1186 | const char *flag, int default_size, int *new_size) | |
1187 | { | |
1188 | int queue_size = smap_get_int(args, flag, default_size); | |
1189 | ||
1190 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1191 | || !is_pow2(queue_size)) { | |
1192 | queue_size = default_size; | |
1193 | } | |
1194 | ||
1195 | if (queue_size != *new_size) { | |
1196 | *new_size = queue_size; | |
1197 | netdev_request_reconfigure(netdev); | |
1198 | } | |
1199 | } | |
1200 | ||
c3d062a7 | 1201 | static int |
9fff138e DDP |
1202 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1203 | char **errp) | |
c3d062a7 CL |
1204 | { |
1205 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
b614c894 IM |
1206 | bool rx_fc_en, tx_fc_en, autoneg; |
1207 | enum rte_eth_fc_mode fc_mode; | |
1208 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1209 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1210 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1211 | }; | |
1a2bb118 SC |
1212 | bool rx_chksm_ofld; |
1213 | bool temp_flag; | |
55e075e6 CL |
1214 | const char *new_devargs; |
1215 | int err = 0; | |
c3d062a7 | 1216 | |
55e075e6 | 1217 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1218 | ovs_mutex_lock(&dev->mutex); |
1219 | ||
1220 | dpdk_set_rxq_config(dev, args); | |
1221 | ||
b685696b CL |
1222 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1223 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1224 | &dev->requested_rxq_size); | |
1225 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1226 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1227 | &dev->requested_txq_size); | |
1228 | ||
55e075e6 CL |
1229 | new_devargs = smap_get(args, "dpdk-devargs"); |
1230 | ||
1231 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1232 | /* The user requested a new device. If we return error, the caller | |
1233 | * will delete this netdev and try to recreate it. */ | |
1234 | err = EAGAIN; | |
1235 | goto out; | |
1236 | } | |
1237 | ||
1238 | /* dpdk-devargs is required for device configuration */ | |
1239 | if (new_devargs && new_devargs[0]) { | |
1240 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1241 | * is valid */ | |
1242 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1243 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
bb37956a IM |
1244 | dpdk_port_t new_port_id = netdev_dpdk_process_devargs(dev, |
1245 | new_devargs, | |
1246 | errp); | |
55e075e6 CL |
1247 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1248 | err = EINVAL; | |
1249 | } else if (new_port_id == dev->port_id) { | |
1250 | /* Already configured, do not reconfigure again */ | |
1251 | err = 0; | |
1252 | } else { | |
1253 | struct netdev_dpdk *dup_dev; | |
bb37956a | 1254 | |
55e075e6 CL |
1255 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); |
1256 | if (dup_dev) { | |
9fff138e DDP |
1257 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
1258 | "which is already in use by '%s'", | |
1259 | netdev_get_name(netdev), new_devargs, | |
1260 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1261 | err = EADDRINUSE; |
1262 | } else { | |
bd4e172b | 1263 | int sid = rte_eth_dev_socket_id(new_port_id); |
bb37956a | 1264 | |
bd4e172b | 1265 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; |
55e075e6 CL |
1266 | dev->devargs = xstrdup(new_devargs); |
1267 | dev->port_id = new_port_id; | |
1268 | netdev_request_reconfigure(&dev->up); | |
1269 | err = 0; | |
1270 | } | |
1271 | } | |
1272 | } | |
1273 | } else { | |
9fff138e DDP |
1274 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1275 | "The old 'dpdk<port_id>' names are not supported", | |
1276 | netdev_get_name(netdev)); | |
55e075e6 CL |
1277 | err = EINVAL; |
1278 | } | |
1279 | ||
1280 | if (err) { | |
1281 | goto out; | |
1282 | } | |
1283 | ||
c3d062a7 CL |
1284 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1285 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1286 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1287 | |
b614c894 IM |
1288 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1289 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1290 | dev->fc_conf.mode = fc_mode; | |
1291 | dev->fc_conf.autoneg = autoneg; | |
1292 | dpdk_eth_flow_ctrl_setup(dev); | |
1293 | } | |
9fd39370 | 1294 | |
1a2bb118 SC |
1295 | /* Rx checksum offload configuration */ |
1296 | /* By default the Rx checksum offload is ON */ | |
1297 | rx_chksm_ofld = smap_get_bool(args, "rx-checksum-offload", true); | |
1298 | temp_flag = (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) | |
1299 | != 0; | |
1300 | if (temp_flag != rx_chksm_ofld) { | |
1301 | dev->hw_ol_features ^= NETDEV_RX_CHECKSUM_OFFLOAD; | |
1302 | dpdk_eth_checksum_offload_configure(dev); | |
1303 | } | |
55e075e6 CL |
1304 | |
1305 | out: | |
c3d062a7 | 1306 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1307 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1308 | |
55e075e6 | 1309 | return err; |
c3d062a7 CL |
1310 | } |
1311 | ||
1312 | static int | |
9fff138e DDP |
1313 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1314 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1315 | { |
1316 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1317 | ||
1318 | ovs_mutex_lock(&dev->mutex); | |
1319 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1320 | ovs_mutex_unlock(&dev->mutex); |
1321 | ||
1322 | return 0; | |
1323 | } | |
1324 | ||
c1ff66ac | 1325 | static int |
2d24d165 | 1326 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1327 | const struct smap *args, |
1328 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1329 | { |
1330 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1331 | const char *path; | |
1332 | ||
6881885a | 1333 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1334 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1335 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1336 | if (path && strcmp(path, dev->vhost_id)) { |
1337 | strcpy(dev->vhost_id, path); | |
c1ff66ac CL |
1338 | netdev_request_reconfigure(netdev); |
1339 | } | |
1340 | } | |
6881885a | 1341 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1342 | |
1343 | return 0; | |
1344 | } | |
1345 | ||
7dec44fe | 1346 | static int |
d46285a2 | 1347 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1348 | { |
d46285a2 | 1349 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1350 | |
d46285a2 | 1351 | return dev->socket_id; |
7dec44fe AW |
1352 | } |
1353 | ||
050c60bf | 1354 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1355 | static int |
050c60bf | 1356 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1357 | { |
d46285a2 | 1358 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1359 | |
d46285a2 | 1360 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1361 | |
050c60bf DDP |
1362 | if (dev->requested_n_txq == n_txq) { |
1363 | goto out; | |
4573fbd3 FL |
1364 | } |
1365 | ||
050c60bf DDP |
1366 | dev->requested_n_txq = n_txq; |
1367 | netdev_request_reconfigure(netdev); | |
58397e6c | 1368 | |
050c60bf | 1369 | out: |
d46285a2 | 1370 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1371 | return 0; |
58397e6c KT |
1372 | } |
1373 | ||
8a9562d2 PS |
1374 | static struct netdev_rxq * |
1375 | netdev_dpdk_rxq_alloc(void) | |
1376 | { | |
1377 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1378 | ||
eff23640 DDP |
1379 | if (rx) { |
1380 | return &rx->up; | |
1381 | } | |
1382 | ||
1383 | return NULL; | |
8a9562d2 PS |
1384 | } |
1385 | ||
1386 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1387 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1388 | { |
d46285a2 | 1389 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1390 | } |
1391 | ||
1392 | static int | |
d46285a2 | 1393 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1394 | { |
d46285a2 DDP |
1395 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1396 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1397 | |
d46285a2 DDP |
1398 | ovs_mutex_lock(&dev->mutex); |
1399 | rx->port_id = dev->port_id; | |
1400 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1401 | |
1402 | return 0; | |
1403 | } | |
1404 | ||
1405 | static void | |
d46285a2 | 1406 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1407 | { |
1408 | } | |
1409 | ||
1410 | static void | |
d46285a2 | 1411 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1412 | { |
d46285a2 | 1413 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1414 | |
1415 | rte_free(rx); | |
1416 | } | |
1417 | ||
819f13bd DDP |
1418 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1419 | * 'pkts', even in case of failure. | |
1420 | * | |
1421 | * Returns the number of packets that weren't transmitted. */ | |
1422 | static inline int | |
b59cc14e | 1423 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 1424 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 1425 | { |
1304f1f8 DDP |
1426 | uint32_t nb_tx = 0; |
1427 | ||
b59cc14e | 1428 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1429 | uint32_t ret; |
1430 | ||
b59cc14e | 1431 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1432 | if (!ret) { |
1433 | break; | |
1434 | } | |
1435 | ||
1436 | nb_tx += ret; | |
1437 | } | |
8a9562d2 | 1438 | |
b59cc14e | 1439 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 1440 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
1441 | * packet could come from a different mempool) */ |
1442 | int i; | |
1443 | ||
b59cc14e IM |
1444 | for (i = nb_tx; i < cnt; i++) { |
1445 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1446 | } |
8a9562d2 | 1447 | } |
819f13bd DDP |
1448 | |
1449 | return cnt - nb_tx; | |
8a9562d2 PS |
1450 | } |
1451 | ||
f3926f29 IS |
1452 | static inline bool |
1453 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1454 | struct rte_mbuf *pkt, uint64_t time) | |
1455 | { | |
1456 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1457 | ||
1458 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1459 | e_RTE_METER_GREEN; | |
1460 | } | |
1461 | ||
1462 | static int | |
1463 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
1464 | struct rte_mbuf **pkts, int pkt_cnt) | |
1465 | { | |
1466 | int i = 0; | |
1467 | int cnt = 0; | |
1468 | struct rte_mbuf *pkt = NULL; | |
1469 | uint64_t current_time = rte_rdtsc(); | |
1470 | ||
1471 | for (i = 0; i < pkt_cnt; i++) { | |
1472 | pkt = pkts[i]; | |
1473 | /* Handle current packet */ | |
1474 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1475 | if (cnt != i) { | |
1476 | pkts[cnt] = pkt; | |
1477 | } | |
1478 | cnt++; | |
1479 | } else { | |
1480 | rte_pktmbuf_free(pkt); | |
1481 | } | |
1482 | } | |
1483 | ||
1484 | return cnt; | |
1485 | } | |
1486 | ||
9509913a IS |
1487 | static int |
1488 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
1489 | int pkt_cnt) | |
1490 | { | |
1491 | int cnt = 0; | |
1492 | ||
1493 | rte_spinlock_lock(&policer->policer_lock); | |
1494 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, pkt_cnt); | |
1495 | rte_spinlock_unlock(&policer->policer_lock); | |
1496 | ||
1497 | return cnt; | |
1498 | } | |
1499 | ||
58397e6c | 1500 | static bool |
0a0f39df | 1501 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1502 | { |
0a0f39df | 1503 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1504 | } |
1505 | ||
d6e3feb5 | 1506 | static inline void |
1507 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1508 | unsigned int packet_size) | |
1509 | { | |
1510 | /* Hard-coded search for the size bucket. */ | |
1511 | if (packet_size < 256) { | |
1512 | if (packet_size >= 128) { | |
1513 | stats->rx_128_to_255_packets++; | |
1514 | } else if (packet_size <= 64) { | |
1515 | stats->rx_1_to_64_packets++; | |
1516 | } else { | |
1517 | stats->rx_65_to_127_packets++; | |
1518 | } | |
1519 | } else { | |
1520 | if (packet_size >= 1523) { | |
1521 | stats->rx_1523_to_max_packets++; | |
1522 | } else if (packet_size >= 1024) { | |
1523 | stats->rx_1024_to_1522_packets++; | |
1524 | } else if (packet_size < 512) { | |
1525 | stats->rx_256_to_511_packets++; | |
1526 | } else { | |
1527 | stats->rx_512_to_1023_packets++; | |
1528 | } | |
1529 | } | |
1530 | } | |
1531 | ||
9e3ddd45 TP |
1532 | static inline void |
1533 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1534 | struct dp_packet **packets, int count, |
1535 | int dropped) | |
9e3ddd45 TP |
1536 | { |
1537 | int i; | |
d6e3feb5 | 1538 | unsigned int packet_size; |
9e3ddd45 TP |
1539 | struct dp_packet *packet; |
1540 | ||
1541 | stats->rx_packets += count; | |
9509913a | 1542 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1543 | for (i = 0; i < count; i++) { |
1544 | packet = packets[i]; | |
d6e3feb5 | 1545 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1546 | |
d6e3feb5 | 1547 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1548 | /* This only protects the following multicast counting from |
1549 | * too short packets, but it does not stop the packet from | |
1550 | * further processing. */ | |
1551 | stats->rx_errors++; | |
1552 | stats->rx_length_errors++; | |
1553 | continue; | |
1554 | } | |
1555 | ||
d6e3feb5 | 1556 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1557 | ||
9e3ddd45 TP |
1558 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1559 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1560 | stats->multicast++; | |
1561 | } | |
1562 | ||
d6e3feb5 | 1563 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1564 | } |
1565 | } | |
1566 | ||
58397e6c KT |
1567 | /* |
1568 | * The receive path for the vhost port is the TX path out from guest. | |
1569 | */ | |
1570 | static int | |
d46285a2 | 1571 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1572 | struct dp_packet_batch *batch) |
58397e6c | 1573 | { |
d46285a2 | 1574 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
d46285a2 | 1575 | int qid = rxq->queue_id; |
9509913a | 1576 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1577 | uint16_t nb_rx = 0; |
9509913a | 1578 | uint16_t dropped = 0; |
58397e6c | 1579 | |
0a0f39df | 1580 | if (OVS_UNLIKELY(!is_vhost_running(dev) |
e543851d | 1581 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1582 | return EAGAIN; |
1583 | } | |
1584 | ||
0a0f39df CL |
1585 | nb_rx = rte_vhost_dequeue_burst(netdev_dpdk_get_vid(dev), |
1586 | qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
d46285a2 | 1587 | dev->dpdk_mp->mp, |
64839cf4 | 1588 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1589 | NETDEV_MAX_BURST); |
58397e6c KT |
1590 | if (!nb_rx) { |
1591 | return EAGAIN; | |
1592 | } | |
1593 | ||
9509913a IS |
1594 | if (policer) { |
1595 | dropped = nb_rx; | |
64839cf4 WT |
1596 | nb_rx = ingress_policer_run(policer, |
1597 | (struct rte_mbuf **) batch->packets, | |
1598 | nb_rx); | |
9509913a IS |
1599 | dropped -= nb_rx; |
1600 | } | |
1601 | ||
d46285a2 | 1602 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1603 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1604 | nb_rx, dropped); | |
d46285a2 | 1605 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1606 | |
64839cf4 | 1607 | batch->count = (int) nb_rx; |
58397e6c KT |
1608 | return 0; |
1609 | } | |
1610 | ||
8a9562d2 | 1611 | static int |
64839cf4 | 1612 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1613 | { |
d46285a2 DDP |
1614 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1615 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1616 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1617 | int nb_rx; |
9509913a | 1618 | int dropped = 0; |
8a9562d2 | 1619 | |
3b1fb077 DDP |
1620 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1621 | return EAGAIN; | |
1622 | } | |
1623 | ||
d46285a2 | 1624 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1625 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1626 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1627 | if (!nb_rx) { |
1628 | return EAGAIN; | |
1629 | } | |
1630 | ||
9509913a IS |
1631 | if (policer) { |
1632 | dropped = nb_rx; | |
64839cf4 | 1633 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 1634 | (struct rte_mbuf **) batch->packets, |
64839cf4 | 1635 | nb_rx); |
9509913a IS |
1636 | dropped -= nb_rx; |
1637 | } | |
1638 | ||
1639 | /* Update stats to reflect dropped packets */ | |
1640 | if (OVS_UNLIKELY(dropped)) { | |
1641 | rte_spinlock_lock(&dev->stats_lock); | |
1642 | dev->stats.rx_dropped += dropped; | |
1643 | rte_spinlock_unlock(&dev->stats_lock); | |
1644 | } | |
1645 | ||
64839cf4 | 1646 | batch->count = nb_rx; |
8a9562d2 PS |
1647 | |
1648 | return 0; | |
1649 | } | |
1650 | ||
0bf765f7 | 1651 | static inline int |
78bd47cf DDP |
1652 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
1653 | int cnt) | |
0bf765f7 | 1654 | { |
78bd47cf | 1655 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 1656 | |
78bd47cf DDP |
1657 | if (qos_conf) { |
1658 | rte_spinlock_lock(&qos_conf->lock); | |
1659 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt); | |
1660 | rte_spinlock_unlock(&qos_conf->lock); | |
0bf765f7 IS |
1661 | } |
1662 | ||
1663 | return cnt; | |
1664 | } | |
1665 | ||
c6ec9d17 IM |
1666 | static int |
1667 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1668 | int pkt_cnt) | |
1669 | { | |
1670 | int i = 0; | |
1671 | int cnt = 0; | |
1672 | struct rte_mbuf *pkt; | |
1673 | ||
1674 | for (i = 0; i < pkt_cnt; i++) { | |
1675 | pkt = pkts[i]; | |
1676 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
1677 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
1678 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
1679 | rte_pktmbuf_free(pkt); | |
1680 | continue; | |
1681 | } | |
1682 | ||
1683 | if (OVS_UNLIKELY(i != cnt)) { | |
1684 | pkts[cnt] = pkt; | |
1685 | } | |
1686 | cnt++; | |
1687 | } | |
1688 | ||
1689 | return cnt; | |
1690 | } | |
1691 | ||
9e3ddd45 TP |
1692 | static inline void |
1693 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1694 | struct dp_packet **packets, | |
1695 | int attempted, | |
1696 | int dropped) | |
1697 | { | |
1698 | int i; | |
1699 | int sent = attempted - dropped; | |
1700 | ||
1701 | stats->tx_packets += sent; | |
1702 | stats->tx_dropped += dropped; | |
1703 | ||
1704 | for (i = 0; i < sent; i++) { | |
1705 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1706 | } | |
1707 | } | |
1708 | ||
58397e6c | 1709 | static void |
4573fbd3 | 1710 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 1711 | struct dp_packet **pkts, int cnt) |
58397e6c | 1712 | { |
d46285a2 | 1713 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1714 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1715 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 1716 | unsigned int dropped = 0; |
dd52de45 | 1717 | int i, retries = 0; |
58397e6c | 1718 | |
81acebda | 1719 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1720 | |
0a0f39df | 1721 | if (OVS_UNLIKELY(!is_vhost_running(dev) || qid < 0 |
e543851d | 1722 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1723 | rte_spinlock_lock(&dev->stats_lock); |
1724 | dev->stats.tx_dropped+= cnt; | |
1725 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1726 | goto out; |
58397e6c KT |
1727 | } |
1728 | ||
d46285a2 | 1729 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1730 | |
c6ec9d17 | 1731 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 1732 | /* Check has QoS has been configured for the netdev */ |
78bd47cf | 1733 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt); |
c6ec9d17 | 1734 | dropped = total_pkts - cnt; |
0bf765f7 | 1735 | |
95e9881f | 1736 | do { |
4573fbd3 | 1737 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1738 | unsigned int tx_pkts; |
1739 | ||
0a0f39df | 1740 | tx_pkts = rte_vhost_enqueue_burst(netdev_dpdk_get_vid(dev), |
c6ec9d17 | 1741 | vhost_qid, cur_pkts, cnt); |
95e9881f KT |
1742 | if (OVS_LIKELY(tx_pkts)) { |
1743 | /* Packets have been sent.*/ | |
1744 | cnt -= tx_pkts; | |
31871ee3 | 1745 | /* Prepare for possible retry.*/ |
95e9881f KT |
1746 | cur_pkts = &cur_pkts[tx_pkts]; |
1747 | } else { | |
31871ee3 KT |
1748 | /* No packets sent - do not retry.*/ |
1749 | break; | |
95e9881f | 1750 | } |
c6ec9d17 | 1751 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1752 | |
d46285a2 | 1753 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1754 | |
d46285a2 | 1755 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 1756 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 1757 | cnt + dropped); |
d46285a2 | 1758 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1759 | |
1760 | out: | |
c6ec9d17 | 1761 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 1762 | dp_packet_delete(pkts[i]); |
58397e6c KT |
1763 | } |
1764 | } | |
1765 | ||
8a9562d2 PS |
1766 | /* Tx function. Transmit packets indefinitely */ |
1767 | static void | |
64839cf4 | 1768 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1769 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1770 | { |
bce01e3a | 1771 | #if !defined(__CHECKER__) && !defined(_WIN32) |
64839cf4 | 1772 | const size_t PKT_ARRAY_SIZE = batch->count; |
bce01e3a EJ |
1773 | #else |
1774 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1775 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1776 | #endif |
8a9562d2 | 1777 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 1778 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
175cf4de RW |
1779 | int dropped = 0; |
1780 | int newcnt = 0; | |
1781 | int i; | |
8a9562d2 | 1782 | |
7d6d1a40 WT |
1783 | dp_packet_batch_apply_cutlen(batch); |
1784 | ||
64839cf4 WT |
1785 | for (i = 0; i < batch->count; i++) { |
1786 | int size = dp_packet_size(batch->packets[i]); | |
95fb793a | 1787 | |
f98d7864 | 1788 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c | 1789 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
58be5c0e | 1790 | (int) size, dev->max_packet_len); |
f4fd623c | 1791 | |
175cf4de | 1792 | dropped++; |
f4fd623c DDP |
1793 | continue; |
1794 | } | |
8a9562d2 | 1795 | |
2391135c | 1796 | pkts[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1797 | |
2391135c | 1798 | if (!pkts[newcnt]) { |
64839cf4 | 1799 | dropped += batch->count - i; |
175cf4de | 1800 | break; |
f4fd623c DDP |
1801 | } |
1802 | ||
1803 | /* We have to do a copy for now */ | |
2391135c | 1804 | memcpy(rte_pktmbuf_mtod(pkts[newcnt], void *), |
64839cf4 | 1805 | dp_packet_data(batch->packets[i]), size); |
f4fd623c | 1806 | |
2391135c MK |
1807 | rte_pktmbuf_data_len(pkts[newcnt]) = size; |
1808 | rte_pktmbuf_pkt_len(pkts[newcnt]) = size; | |
f4fd623c DDP |
1809 | |
1810 | newcnt++; | |
1811 | } | |
8a9562d2 | 1812 | |
58397e6c | 1813 | if (dev->type == DPDK_DEV_VHOST) { |
2391135c | 1814 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, |
dd52de45 | 1815 | newcnt); |
58397e6c | 1816 | } else { |
0bf765f7 IS |
1817 | unsigned int qos_pkts = newcnt; |
1818 | ||
1819 | /* Check if QoS has been configured for this netdev. */ | |
78bd47cf | 1820 | newcnt = netdev_dpdk_qos_run(dev, pkts, newcnt); |
0bf765f7 IS |
1821 | |
1822 | dropped += qos_pkts - newcnt; | |
819f13bd | 1823 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, newcnt); |
58397e6c | 1824 | } |
db73f716 | 1825 | |
0bf765f7 IS |
1826 | if (OVS_UNLIKELY(dropped)) { |
1827 | rte_spinlock_lock(&dev->stats_lock); | |
1828 | dev->stats.tx_dropped += dropped; | |
1829 | rte_spinlock_unlock(&dev->stats_lock); | |
1830 | } | |
8a9562d2 PS |
1831 | } |
1832 | ||
58397e6c | 1833 | static int |
64839cf4 WT |
1834 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1835 | struct dp_packet_batch *batch, | |
324c8374 | 1836 | bool may_steal, bool concurrent_txq OVS_UNUSED) |
58397e6c | 1837 | { |
58397e6c | 1838 | |
dd52de45 | 1839 | if (OVS_UNLIKELY(!may_steal || batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 WT |
1840 | dpdk_do_tx_copy(netdev, qid, batch); |
1841 | dp_packet_delete_batch(batch, may_steal); | |
58397e6c | 1842 | } else { |
64839cf4 | 1843 | dp_packet_batch_apply_cutlen(batch); |
dd52de45 | 1844 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
1845 | } |
1846 | return 0; | |
1847 | } | |
1848 | ||
7251515e DV |
1849 | static inline void |
1850 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
324c8374 IM |
1851 | struct dp_packet_batch *batch, bool may_steal, |
1852 | bool concurrent_txq) | |
8a9562d2 | 1853 | { |
3b1fb077 DDP |
1854 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1855 | dp_packet_delete_batch(batch, may_steal); | |
1856 | return; | |
1857 | } | |
1858 | ||
324c8374 | 1859 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 1860 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
1861 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
1862 | } | |
1863 | ||
7251515e | 1864 | if (OVS_UNLIKELY(!may_steal || |
64839cf4 | 1865 | batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1866 | struct netdev *netdev = &dev->up; |
1867 | ||
64839cf4 WT |
1868 | dpdk_do_tx_copy(netdev, qid, batch); |
1869 | dp_packet_delete_batch(batch, may_steal); | |
8a9562d2 | 1870 | } else { |
6b094bf4 | 1871 | int dropped; |
64839cf4 | 1872 | int cnt = batch->count; |
2391135c | 1873 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 1874 | |
7d6d1a40 WT |
1875 | dp_packet_batch_apply_cutlen(batch); |
1876 | ||
2391135c | 1877 | cnt = netdev_dpdk_filter_packet_len(dev, pkts, cnt); |
78bd47cf | 1878 | cnt = netdev_dpdk_qos_run(dev, pkts, cnt); |
6b094bf4 | 1879 | dropped = batch->count - cnt; |
1b99bb05 | 1880 | |
819f13bd | 1881 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, cnt); |
8a9562d2 | 1882 | |
f4fd623c | 1883 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1884 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1885 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1886 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1887 | } |
8a9562d2 | 1888 | } |
a0cb2d66 | 1889 | |
324c8374 | 1890 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
1891 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
1892 | } | |
7251515e DV |
1893 | } |
1894 | ||
1895 | static int | |
1896 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
324c8374 IM |
1897 | struct dp_packet_batch *batch, bool may_steal, |
1898 | bool concurrent_txq) | |
7251515e DV |
1899 | { |
1900 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1901 | |
324c8374 | 1902 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e | 1903 | return 0; |
8a9562d2 PS |
1904 | } |
1905 | ||
1906 | static int | |
74ff3298 | 1907 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1908 | { |
1909 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1910 | ||
1911 | ovs_mutex_lock(&dev->mutex); | |
1912 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1913 | dev->hwaddr = mac; |
045c0d1a | 1914 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1915 | } |
1916 | ovs_mutex_unlock(&dev->mutex); | |
1917 | ||
1918 | return 0; | |
1919 | } | |
1920 | ||
1921 | static int | |
74ff3298 | 1922 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1923 | { |
1924 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1925 | ||
1926 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1927 | *mac = dev->hwaddr; |
8a9562d2 PS |
1928 | ovs_mutex_unlock(&dev->mutex); |
1929 | ||
1930 | return 0; | |
1931 | } | |
1932 | ||
1933 | static int | |
1934 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1935 | { | |
1936 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1937 | ||
1938 | ovs_mutex_lock(&dev->mutex); | |
1939 | *mtup = dev->mtu; | |
1940 | ovs_mutex_unlock(&dev->mutex); | |
1941 | ||
1942 | return 0; | |
1943 | } | |
1944 | ||
0072e931 MK |
1945 | static int |
1946 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
1947 | { | |
1948 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1949 | ||
1950 | if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
1951 | || mtu < ETHER_MIN_MTU) { | |
1952 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
1953 | return EINVAL; | |
1954 | } | |
1955 | ||
1956 | ovs_mutex_lock(&dev->mutex); | |
1957 | if (dev->requested_mtu != mtu) { | |
1958 | dev->requested_mtu = mtu; | |
1959 | netdev_request_reconfigure(netdev); | |
1960 | } | |
1961 | ovs_mutex_unlock(&dev->mutex); | |
1962 | ||
1963 | return 0; | |
1964 | } | |
1965 | ||
8a9562d2 | 1966 | static int |
d46285a2 | 1967 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 1968 | |
58397e6c KT |
1969 | static int |
1970 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1971 | struct netdev_stats *stats) | |
1972 | { | |
1973 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1974 | ||
1975 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 1976 | |
45d947c4 | 1977 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
1978 | /* Supported Stats */ |
1979 | stats->rx_packets += dev->stats.rx_packets; | |
1980 | stats->tx_packets += dev->stats.tx_packets; | |
9509913a | 1981 | stats->rx_dropped = dev->stats.rx_dropped; |
58397e6c | 1982 | stats->tx_dropped += dev->stats.tx_dropped; |
9e3ddd45 TP |
1983 | stats->multicast = dev->stats.multicast; |
1984 | stats->rx_bytes = dev->stats.rx_bytes; | |
1985 | stats->tx_bytes = dev->stats.tx_bytes; | |
1986 | stats->rx_errors = dev->stats.rx_errors; | |
1987 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 1988 | |
1989 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
1990 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
1991 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
1992 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
1993 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
1994 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
1995 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
1996 | ||
45d947c4 | 1997 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 1998 | |
58397e6c KT |
1999 | ovs_mutex_unlock(&dev->mutex); |
2000 | ||
2001 | return 0; | |
2002 | } | |
2003 | ||
d6e3feb5 | 2004 | static void |
2005 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
2006 | const struct rte_eth_xstat *xstats, |
2007 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 2008 | const unsigned int size) |
2009 | { | |
d6e3feb5 | 2010 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 2011 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2012 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2013 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2014 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2015 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2016 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2017 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2018 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2019 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2020 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2021 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2022 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2023 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2024 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 2025 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2026 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2027 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2028 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2029 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2030 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2031 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2032 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2033 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2034 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2035 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2036 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2037 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2038 | stats->tx_1523_to_max_packets = xstats[i].value; |
d57f777f PS |
2039 | } else if (strcmp(XSTAT_RX_MULTICAST_PACKETS, names[i].name) == 0) { |
2040 | stats->multicast = xstats[i].value; | |
0a0f39df | 2041 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2042 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 2043 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2044 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2045 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2046 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2047 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2048 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2049 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2050 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2051 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2052 | stats->rx_jabber_errors = xstats[i].value; |
2053 | } | |
2054 | } | |
2055 | } | |
2056 | ||
8a9562d2 PS |
2057 | static int |
2058 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2059 | { | |
2060 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2061 | struct rte_eth_stats rte_stats; | |
2062 | bool gg; | |
2063 | ||
2064 | netdev_dpdk_get_carrier(netdev, &gg); | |
2065 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2066 | |
0a0f39df CL |
2067 | struct rte_eth_xstat *rte_xstats = NULL; |
2068 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2069 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2070 | |
2071 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
bb37956a | 2072 | VLOG_ERR("Can't get ETH statistics for port: %"PRIu8, dev->port_id); |
f9256822 | 2073 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2074 | return EPROTO; |
2075 | } | |
2076 | ||
0a0f39df CL |
2077 | /* Get length of statistics */ |
2078 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2079 | if (rte_xstats_len < 0) { | |
bb37956a | 2080 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2081 | goto out; |
2082 | } | |
2083 | /* Reserve memory for xstats names and values */ | |
2084 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2085 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2086 | ||
2087 | /* Retreive xstats names */ | |
2088 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2089 | rte_xstats_names, | |
2090 | rte_xstats_len); | |
2091 | if (rte_xstats_new_len != rte_xstats_len) { | |
bb37956a | 2092 | VLOG_WARN("Cannot get XSTATS names for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2093 | goto out; |
2094 | } | |
2095 | /* Retreive xstats values */ | |
2096 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2097 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2098 | rte_xstats_len); | |
2099 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2100 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2101 | rte_xstats_len); | |
d6e3feb5 | 2102 | } else { |
bb37956a | 2103 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
d6e3feb5 | 2104 | } |
8a9562d2 | 2105 | |
0a0f39df CL |
2106 | out: |
2107 | free(rte_xstats); | |
2108 | free(rte_xstats_names); | |
2109 | ||
2f9dd77f PS |
2110 | stats->rx_packets = rte_stats.ipackets; |
2111 | stats->tx_packets = rte_stats.opackets; | |
2112 | stats->rx_bytes = rte_stats.ibytes; | |
2113 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2114 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2115 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2116 | |
45d947c4 | 2117 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2118 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2119 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2120 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2121 | |
2122 | /* These are the available DPDK counters for packets not received due to | |
2123 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2124 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2125 | stats->rx_missed_errors = rte_stats.imissed; |
2126 | ||
8a9562d2 PS |
2127 | ovs_mutex_unlock(&dev->mutex); |
2128 | ||
2129 | return 0; | |
2130 | } | |
2131 | ||
2132 | static int | |
d46285a2 | 2133 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2134 | enum netdev_features *current, |
ca3d4f55 BX |
2135 | enum netdev_features *advertised, |
2136 | enum netdev_features *supported, | |
2137 | enum netdev_features *peer) | |
8a9562d2 | 2138 | { |
d46285a2 | 2139 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2140 | struct rte_eth_link link; |
2141 | ||
2142 | ovs_mutex_lock(&dev->mutex); | |
2143 | link = dev->link; | |
2144 | ovs_mutex_unlock(&dev->mutex); | |
2145 | ||
362ca396 | 2146 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2147 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
2148 | *current = NETDEV_F_10MB_HD; |
2149 | } | |
362ca396 | 2150 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2151 | *current = NETDEV_F_100MB_HD; |
2152 | } | |
362ca396 | 2153 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2154 | *current = NETDEV_F_1GB_HD; |
2155 | } | |
2156 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 2157 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
2158 | *current = NETDEV_F_10MB_FD; |
2159 | } | |
362ca396 | 2160 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2161 | *current = NETDEV_F_100MB_FD; |
2162 | } | |
362ca396 | 2163 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2164 | *current = NETDEV_F_1GB_FD; |
2165 | } | |
362ca396 | 2166 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
2167 | *current = NETDEV_F_10GB_FD; |
2168 | } | |
2169 | } | |
2170 | ||
362ca396 | 2171 | if (link.link_autoneg) { |
2172 | *current |= NETDEV_F_AUTONEG; | |
2173 | } | |
2174 | ||
ca3d4f55 BX |
2175 | *advertised = *supported = *peer = 0; |
2176 | ||
8a9562d2 PS |
2177 | return 0; |
2178 | } | |
2179 | ||
9509913a IS |
2180 | static struct ingress_policer * |
2181 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2182 | { | |
2183 | struct ingress_policer *policer = NULL; | |
2184 | uint64_t rate_bytes; | |
2185 | uint64_t burst_bytes; | |
2186 | int err = 0; | |
2187 | ||
2188 | policer = xmalloc(sizeof *policer); | |
2189 | rte_spinlock_init(&policer->policer_lock); | |
2190 | ||
2191 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
2192 | rate_bytes = rate * 1000/8; | |
2193 | burst_bytes = burst * 1000/8; | |
2194 | ||
2195 | policer->app_srtcm_params.cir = rate_bytes; | |
2196 | policer->app_srtcm_params.cbs = burst_bytes; | |
2197 | policer->app_srtcm_params.ebs = 0; | |
2198 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2199 | &policer->app_srtcm_params); | |
58be5c0e | 2200 | if (err) { |
9509913a IS |
2201 | VLOG_ERR("Could not create rte meter for ingress policer"); |
2202 | return NULL; | |
2203 | } | |
2204 | ||
2205 | return policer; | |
2206 | } | |
2207 | ||
2208 | static int | |
2209 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2210 | uint32_t policer_burst) | |
2211 | { | |
2212 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2213 | struct ingress_policer *policer; | |
2214 | ||
2215 | /* Force to 0 if no rate specified, | |
2216 | * default to 8000 kbits if burst is 0, | |
2217 | * else stick with user-specified value. | |
2218 | */ | |
2219 | policer_burst = (!policer_rate ? 0 | |
2220 | : !policer_burst ? 8000 | |
2221 | : policer_burst); | |
2222 | ||
2223 | ovs_mutex_lock(&dev->mutex); | |
2224 | ||
2225 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2226 | &dev->ingress_policer); | |
2227 | ||
2228 | if (dev->policer_rate == policer_rate && | |
2229 | dev->policer_burst == policer_burst) { | |
2230 | /* Assume that settings haven't changed since we last set them. */ | |
2231 | ovs_mutex_unlock(&dev->mutex); | |
2232 | return 0; | |
2233 | } | |
2234 | ||
2235 | /* Destroy any existing ingress policer for the device if one exists */ | |
2236 | if (policer) { | |
2237 | ovsrcu_postpone(free, policer); | |
2238 | } | |
2239 | ||
2240 | if (policer_rate != 0) { | |
2241 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2242 | } else { | |
2243 | policer = NULL; | |
2244 | } | |
2245 | ovsrcu_set(&dev->ingress_policer, policer); | |
2246 | dev->policer_rate = policer_rate; | |
2247 | dev->policer_burst = policer_burst; | |
2248 | ovs_mutex_unlock(&dev->mutex); | |
2249 | ||
2250 | return 0; | |
2251 | } | |
2252 | ||
8a9562d2 PS |
2253 | static int |
2254 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2255 | { | |
2256 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
2257 | |
2258 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
2259 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
2260 | * postive integer to meet RFC 2863 recommendations. | |
2261 | */ | |
2262 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
2263 | ovs_mutex_unlock(&dev->mutex); |
2264 | ||
2265 | return ifindex; | |
2266 | } | |
2267 | ||
2268 | static int | |
d46285a2 | 2269 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2270 | { |
d46285a2 | 2271 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2272 | |
2273 | ovs_mutex_lock(&dev->mutex); | |
2274 | check_link_status(dev); | |
2275 | *carrier = dev->link.link_status; | |
58397e6c KT |
2276 | |
2277 | ovs_mutex_unlock(&dev->mutex); | |
2278 | ||
2279 | return 0; | |
2280 | } | |
2281 | ||
2282 | static int | |
d46285a2 | 2283 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2284 | { |
d46285a2 | 2285 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2286 | |
2287 | ovs_mutex_lock(&dev->mutex); | |
2288 | ||
0a0f39df | 2289 | if (is_vhost_running(dev)) { |
58397e6c KT |
2290 | *carrier = 1; |
2291 | } else { | |
2292 | *carrier = 0; | |
2293 | } | |
2294 | ||
8a9562d2 PS |
2295 | ovs_mutex_unlock(&dev->mutex); |
2296 | ||
2297 | return 0; | |
2298 | } | |
2299 | ||
2300 | static long long int | |
d46285a2 | 2301 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2302 | { |
d46285a2 | 2303 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2304 | long long int carrier_resets; |
2305 | ||
2306 | ovs_mutex_lock(&dev->mutex); | |
2307 | carrier_resets = dev->link_reset_cnt; | |
2308 | ovs_mutex_unlock(&dev->mutex); | |
2309 | ||
2310 | return carrier_resets; | |
2311 | } | |
2312 | ||
2313 | static int | |
d46285a2 | 2314 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2315 | long long int interval OVS_UNUSED) |
2316 | { | |
ee32150e | 2317 | return EOPNOTSUPP; |
8a9562d2 PS |
2318 | } |
2319 | ||
2320 | static int | |
2321 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2322 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2323 | enum netdev_flags *old_flagsp) |
2324 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 2325 | { |
8a9562d2 PS |
2326 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
2327 | return EINVAL; | |
2328 | } | |
2329 | ||
2330 | *old_flagsp = dev->flags; | |
2331 | dev->flags |= on; | |
2332 | dev->flags &= ~off; | |
2333 | ||
2334 | if (dev->flags == *old_flagsp) { | |
2335 | return 0; | |
2336 | } | |
2337 | ||
58397e6c | 2338 | if (dev->type == DPDK_DEV_ETH) { |
58397e6c KT |
2339 | if (dev->flags & NETDEV_PROMISC) { |
2340 | rte_eth_promiscuous_enable(dev->port_id); | |
2341 | } | |
8a9562d2 | 2342 | |
314fb5ad | 2343 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
2344 | } else { |
2345 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2346 | * running then change netdev's change_seq to trigger link state | |
2347 | * update. */ | |
e543851d ZB |
2348 | |
2349 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2350 | && is_vhost_running(dev)) { |
e543851d ZB |
2351 | netdev_change_seq_changed(&dev->up); |
2352 | ||
2353 | /* Clear statistics if device is getting up. */ | |
2354 | if (NETDEV_UP & on) { | |
2355 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 2356 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
2357 | rte_spinlock_unlock(&dev->stats_lock); |
2358 | } | |
2359 | } | |
8a9562d2 PS |
2360 | } |
2361 | ||
2362 | return 0; | |
2363 | } | |
2364 | ||
2365 | static int | |
d46285a2 | 2366 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2367 | enum netdev_flags off, enum netdev_flags on, |
2368 | enum netdev_flags *old_flagsp) | |
2369 | { | |
d46285a2 | 2370 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2371 | int error; |
2372 | ||
d46285a2 DDP |
2373 | ovs_mutex_lock(&dev->mutex); |
2374 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2375 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2376 | |
2377 | return error; | |
2378 | } | |
2379 | ||
2380 | static int | |
d46285a2 | 2381 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2382 | { |
d46285a2 | 2383 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2384 | struct rte_eth_dev_info dev_info; |
2385 | ||
7cd1261d | 2386 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2387 | return ENODEV; |
7cd1261d | 2388 | } |
8a9562d2 PS |
2389 | |
2390 | ovs_mutex_lock(&dev->mutex); | |
2391 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2392 | ovs_mutex_unlock(&dev->mutex); | |
2393 | ||
95fb793a | 2394 | smap_add_format(args, "port_no", "%d", dev->port_id); |
58be5c0e MK |
2395 | smap_add_format(args, "numa_id", "%d", |
2396 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
2397 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
2398 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2399 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2400 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2401 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2402 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
2403 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
2404 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
2405 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
2406 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2407 | ||
39c2baa9 | 2408 | if (dev_info.pci_dev) { |
2409 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2410 | dev_info.pci_dev->id.vendor_id); | |
2411 | smap_add_format(args, "pci-device_id", "0x%x", | |
2412 | dev_info.pci_dev->id.device_id); | |
2413 | } | |
8a9562d2 PS |
2414 | |
2415 | return 0; | |
2416 | } | |
2417 | ||
2418 | static void | |
2419 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2420 | OVS_REQUIRES(dev->mutex) | |
2421 | { | |
2422 | enum netdev_flags old_flags; | |
2423 | ||
2424 | if (admin_state) { | |
2425 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2426 | } else { | |
2427 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2428 | } | |
2429 | } | |
2430 | ||
2431 | static void | |
2432 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2433 | const char *argv[], void *aux OVS_UNUSED) | |
2434 | { | |
2435 | bool up; | |
2436 | ||
2437 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2438 | up = true; | |
2439 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2440 | up = false; | |
2441 | } else { | |
2442 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2443 | return; | |
2444 | } | |
2445 | ||
2446 | if (argc > 2) { | |
2447 | struct netdev *netdev = netdev_from_name(argv[1]); | |
2448 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
2449 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
2450 | ||
2451 | ovs_mutex_lock(&dpdk_dev->mutex); | |
2452 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
2453 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
2454 | ||
2455 | netdev_close(netdev); | |
2456 | } else { | |
2457 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2458 | netdev_close(netdev); | |
2459 | return; | |
2460 | } | |
2461 | } else { | |
2462 | struct netdev_dpdk *netdev; | |
2463 | ||
2464 | ovs_mutex_lock(&dpdk_mutex); | |
2465 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
2466 | ovs_mutex_lock(&netdev->mutex); | |
2467 | netdev_dpdk_set_admin_state__(netdev, up); | |
2468 | ovs_mutex_unlock(&netdev->mutex); | |
2469 | } | |
2470 | ovs_mutex_unlock(&dpdk_mutex); | |
2471 | } | |
2472 | unixctl_command_reply(conn, "OK"); | |
2473 | } | |
2474 | ||
58397e6c KT |
2475 | /* |
2476 | * Set virtqueue flags so that we do not receive interrupts. | |
2477 | */ | |
2478 | static void | |
0a0f39df | 2479 | set_irq_status(int vid) |
58397e6c | 2480 | { |
4573fbd3 FL |
2481 | uint32_t i; |
2482 | uint64_t idx; | |
2483 | ||
0a0f39df | 2484 | for (i = 0; i < rte_vhost_get_queue_num(vid); i++) { |
4573fbd3 | 2485 | idx = i * VIRTIO_QNUM; |
0a0f39df CL |
2486 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_RXQ, 0); |
2487 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_TXQ, 0); | |
4573fbd3 FL |
2488 | } |
2489 | } | |
2490 | ||
585a5bea IM |
2491 | /* |
2492 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2493 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2494 | */ |
2495 | static void | |
d46285a2 DDP |
2496 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2497 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2498 | { |
2499 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2500 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 2501 | |
eff23640 | 2502 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
2503 | |
2504 | for (i = 0; i < total_txqs; i++) { | |
2505 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2506 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2507 | enabled_queues[n_enabled++] = i; |
2508 | } | |
2509 | } | |
2510 | ||
2511 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2512 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2513 | n_enabled = 1; |
2514 | } | |
2515 | ||
2516 | k = 0; | |
2517 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2518 | if (dev->tx_q[i].map != i) { |
2519 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2520 | k = (k + 1) % n_enabled; |
2521 | } | |
2522 | } | |
2523 | ||
2d24d165 | 2524 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2525 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2526 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2527 | } |
2528 | ||
eff23640 | 2529 | free(enabled_queues); |
585a5bea | 2530 | } |
4573fbd3 | 2531 | |
58397e6c KT |
2532 | /* |
2533 | * A new virtio-net device is added to a vhost port. | |
2534 | */ | |
2535 | static int | |
0a0f39df | 2536 | new_device(int vid) |
58397e6c | 2537 | { |
d46285a2 | 2538 | struct netdev_dpdk *dev; |
58397e6c | 2539 | bool exists = false; |
db8f13b0 | 2540 | int newnode = 0; |
0a0f39df CL |
2541 | char ifname[IF_NAME_SZ]; |
2542 | ||
58be5c0e | 2543 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2544 | |
2545 | ovs_mutex_lock(&dpdk_mutex); | |
2546 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2547 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 2548 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2549 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
0a0f39df | 2550 | uint32_t qp_num = rte_vhost_get_queue_num(vid); |
db8f13b0 CL |
2551 | |
2552 | /* Get NUMA information */ | |
0a0f39df CL |
2553 | newnode = rte_vhost_get_numa_node(vid); |
2554 | if (newnode == -1) { | |
5b9bf9e0 | 2555 | #ifdef VHOST_NUMA |
db8f13b0 | 2556 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2557 | ifname); |
5b9bf9e0 | 2558 | #endif |
db8f13b0 | 2559 | newnode = dev->socket_id; |
db8f13b0 CL |
2560 | } |
2561 | ||
7f5f2bd0 IM |
2562 | if (dev->requested_n_txq != qp_num |
2563 | || dev->requested_n_rxq != qp_num | |
2564 | || dev->requested_socket_id != newnode) { | |
2565 | dev->requested_socket_id = newnode; | |
2566 | dev->requested_n_rxq = qp_num; | |
2567 | dev->requested_n_txq = qp_num; | |
2568 | netdev_request_reconfigure(&dev->up); | |
2569 | } else { | |
2570 | /* Reconfiguration not required. */ | |
2571 | dev->vhost_reconfigured = true; | |
2572 | } | |
81acebda | 2573 | |
0a0f39df | 2574 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2575 | exists = true; |
2576 | ||
58397e6c | 2577 | /* Disable notifications. */ |
0a0f39df | 2578 | set_irq_status(vid); |
e543851d | 2579 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2580 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2581 | break; |
2582 | } | |
c1ff66ac | 2583 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2584 | } |
2585 | ovs_mutex_unlock(&dpdk_mutex); | |
2586 | ||
2587 | if (!exists) { | |
0a0f39df | 2588 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2589 | |
2590 | return -1; | |
2591 | } | |
2592 | ||
0a0f39df CL |
2593 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2594 | ifname, newnode); | |
2595 | ||
58397e6c KT |
2596 | return 0; |
2597 | } | |
2598 | ||
f3ea2ad2 IM |
2599 | /* Clears mapping for all available queues of vhost interface. */ |
2600 | static void | |
2601 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2602 | OVS_REQUIRES(dev->mutex) | |
2603 | { | |
2604 | int i; | |
2605 | ||
81acebda | 2606 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
2607 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
2608 | } | |
2609 | } | |
2610 | ||
58397e6c KT |
2611 | /* |
2612 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2613 | * flag to stop any more packets from being sent or received to/from a VM and | |
2614 | * ensure all currently queued packets have been sent/received before removing | |
2615 | * the device. | |
2616 | */ | |
2617 | static void | |
0a0f39df | 2618 | destroy_device(int vid) |
58397e6c | 2619 | { |
d46285a2 | 2620 | struct netdev_dpdk *dev; |
afee281f | 2621 | bool exists = false; |
0a0f39df CL |
2622 | char ifname[IF_NAME_SZ]; |
2623 | ||
58be5c0e | 2624 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2625 | |
2626 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2627 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2628 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 2629 | |
d46285a2 | 2630 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
2631 | dev->vhost_reconfigured = false; |
2632 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 2633 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 2634 | |
e543851d | 2635 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2636 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 2637 | exists = true; |
afee281f | 2638 | break; |
58397e6c KT |
2639 | } |
2640 | } | |
afee281f | 2641 | |
58397e6c KT |
2642 | ovs_mutex_unlock(&dpdk_mutex); |
2643 | ||
0a0f39df | 2644 | if (exists) { |
afee281f KT |
2645 | /* |
2646 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2647 | * to NULL, before returning. | |
2648 | */ | |
2649 | ovsrcu_synchronize(); | |
2650 | /* | |
2651 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2652 | * put thread back into quiescent state before returning. | |
2653 | */ | |
2654 | ovsrcu_quiesce_start(); | |
0a0f39df | 2655 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 2656 | } else { |
0a0f39df | 2657 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 2658 | } |
58397e6c KT |
2659 | } |
2660 | ||
585a5bea | 2661 | static int |
0a0f39df | 2662 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 2663 | { |
d46285a2 | 2664 | struct netdev_dpdk *dev; |
585a5bea IM |
2665 | bool exists = false; |
2666 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
2667 | char ifname[IF_NAME_SZ]; |
2668 | ||
58be5c0e | 2669 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
2670 | |
2671 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2672 | return 0; | |
2673 | } | |
2674 | ||
2675 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2676 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 2677 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2678 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 2679 | if (enable) { |
d46285a2 | 2680 | dev->tx_q[qid].map = qid; |
585a5bea | 2681 | } else { |
d46285a2 | 2682 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2683 | } |
d46285a2 | 2684 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2685 | exists = true; |
d46285a2 | 2686 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2687 | break; |
2688 | } | |
c1ff66ac | 2689 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2690 | } |
2691 | ovs_mutex_unlock(&dpdk_mutex); | |
2692 | ||
2693 | if (exists) { | |
0a0f39df CL |
2694 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
2695 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 2696 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 2697 | } else { |
0a0f39df | 2698 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
2699 | return -1; |
2700 | } | |
2701 | ||
2702 | return 0; | |
2703 | } | |
2704 | ||
0a0f39df CL |
2705 | int |
2706 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 2707 | { |
0a0f39df | 2708 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
2709 | } |
2710 | ||
9509913a IS |
2711 | struct ingress_policer * |
2712 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
2713 | { | |
2714 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
2715 | } | |
2716 | ||
58397e6c KT |
2717 | /* |
2718 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
2719 | * configuration has been fully complete. | |
2720 | */ | |
bce01e3a | 2721 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
2722 | { |
2723 | .new_device = new_device, | |
2724 | .destroy_device = destroy_device, | |
585a5bea | 2725 | .vring_state_changed = vring_state_changed |
58397e6c KT |
2726 | }; |
2727 | ||
2728 | static void * | |
7d1ced01 | 2729 | start_vhost_loop(void *dummy OVS_UNUSED) |
58397e6c KT |
2730 | { |
2731 | pthread_detach(pthread_self()); | |
41987644 | 2732 | /* Put the vhost thread into quiescent state. */ |
618f44f7 | 2733 | ovsrcu_quiesce_start(); |
58397e6c KT |
2734 | rte_vhost_driver_session_start(); |
2735 | return NULL; | |
2736 | } | |
2737 | ||
2738 | static int | |
ecc1a34e | 2739 | netdev_dpdk_class_init(void) |
7d1ced01 | 2740 | { |
ecc1a34e DDP |
2741 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2742 | ||
2743 | /* This function can be called for different classes. The initialization | |
2744 | * needs to be done only once */ | |
2745 | if (ovsthread_once_start(&once)) { | |
2746 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
2747 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2748 | "[netdev] up|down", 1, 2, | |
2749 | netdev_dpdk_set_admin_state, NULL); | |
2750 | ||
2751 | ovsthread_once_done(&once); | |
2752 | } | |
362ca396 | 2753 | |
7d1ced01 CL |
2754 | return 0; |
2755 | } | |
2756 | ||
ecc1a34e DDP |
2757 | static int |
2758 | netdev_dpdk_vhost_class_init(void) | |
033e9df2 | 2759 | { |
ecc1a34e | 2760 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
033e9df2 | 2761 | |
ecc1a34e DDP |
2762 | /* This function can be called for different classes. The initialization |
2763 | * needs to be done only once */ | |
2764 | if (ovsthread_once_start(&once)) { | |
2765 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
2766 | rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4 | |
2767 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
86bd9ed9 | 2768 | | 1ULL << VIRTIO_NET_F_CSUM); |
ecc1a34e DDP |
2769 | ovs_thread_create("vhost_thread", start_vhost_loop, NULL); |
2770 | ||
2771 | ovsthread_once_done(&once); | |
2772 | } | |
2773 | ||
2774 | return 0; | |
033e9df2 DDP |
2775 | } |
2776 | ||
95fb793a | 2777 | /* Client Rings */ |
2778 | ||
95fb793a | 2779 | static int |
2780 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
bb37956a | 2781 | dpdk_port_t *eth_port_id) |
95fb793a | 2782 | { |
48fffdee | 2783 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 2784 | char *ring_name; |
b8374d0d | 2785 | int port_id; |
95fb793a | 2786 | |
48fffdee KT |
2787 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
2788 | if (!ring_pair) { | |
95fb793a | 2789 | return ENOMEM; |
2790 | } | |
2791 | ||
7251515e | 2792 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 2793 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 2794 | |
8f0a76c9 | 2795 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 2796 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2797 | RING_F_SP_ENQ); |
0c6f39e5 | 2798 | free(ring_name); |
48fffdee KT |
2799 | if (ring_pair->cring_tx == NULL) { |
2800 | rte_free(ring_pair); | |
95fb793a | 2801 | return ENOMEM; |
2802 | } | |
2803 | ||
0c6f39e5 | 2804 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 2805 | |
8f0a76c9 | 2806 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 2807 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2808 | RING_F_SC_DEQ); |
0c6f39e5 | 2809 | free(ring_name); |
48fffdee KT |
2810 | if (ring_pair->cring_rx == NULL) { |
2811 | rte_free(ring_pair); | |
95fb793a | 2812 | return ENOMEM; |
2813 | } | |
2814 | ||
b8374d0d MV |
2815 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
2816 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 2817 | |
b8374d0d | 2818 | if (port_id < 0) { |
48fffdee | 2819 | rte_free(ring_pair); |
95fb793a | 2820 | return ENODEV; |
2821 | } | |
2822 | ||
48fffdee | 2823 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
2824 | ring_pair->eth_port_id = port_id; |
2825 | *eth_port_id = port_id; | |
2826 | ||
48fffdee | 2827 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 2828 | |
95fb793a | 2829 | return 0; |
2830 | } | |
2831 | ||
2832 | static int | |
bb37956a | 2833 | dpdk_ring_open(const char dev_name[], dpdk_port_t *eth_port_id) |
64839cf4 | 2834 | OVS_REQUIRES(dpdk_mutex) |
95fb793a | 2835 | { |
48fffdee | 2836 | struct dpdk_ring *ring_pair; |
95fb793a | 2837 | unsigned int port_no; |
2838 | int err = 0; | |
2839 | ||
2840 | /* Names always start with "dpdkr" */ | |
2841 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2842 | if (err) { | |
2843 | return err; | |
2844 | } | |
2845 | ||
58be5c0e | 2846 | /* Look through our list to find the device */ |
48fffdee KT |
2847 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
2848 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 2849 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 2850 | /* Really all that is needed */ |
48fffdee | 2851 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 2852 | return 0; |
2853 | } | |
2854 | } | |
2855 | /* Need to create the device rings */ | |
2856 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2857 | } | |
2858 | ||
7251515e | 2859 | static int |
d46285a2 | 2860 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
324c8374 IM |
2861 | struct dp_packet_batch *batch, bool may_steal, |
2862 | bool concurrent_txq) | |
7251515e | 2863 | { |
d46285a2 | 2864 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b99bb05 MG |
2865 | unsigned i; |
2866 | ||
58be5c0e MK |
2867 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
2868 | * the rss hash field is clear. This is because the same mbuf may be | |
2869 | * modified by the consumer of the ring and return into the datapath | |
2870 | * without recalculating the RSS hash. */ | |
64839cf4 WT |
2871 | for (i = 0; i < batch->count; i++) { |
2872 | dp_packet_rss_invalidate(batch->packets[i]); | |
1b99bb05 | 2873 | } |
7251515e | 2874 | |
324c8374 | 2875 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e DV |
2876 | return 0; |
2877 | } | |
2878 | ||
95fb793a | 2879 | static int |
2880 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2881 | { | |
bb37956a | 2882 | dpdk_port_t port_no = 0; |
95fb793a | 2883 | int err = 0; |
2884 | ||
95fb793a | 2885 | ovs_mutex_lock(&dpdk_mutex); |
2886 | ||
2887 | err = dpdk_ring_open(netdev->name, &port_no); | |
2888 | if (err) { | |
2889 | goto unlock_dpdk; | |
2890 | } | |
2891 | ||
1ce30dfd DDP |
2892 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
2893 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 2894 | unlock_dpdk: |
2895 | ovs_mutex_unlock(&dpdk_mutex); | |
2896 | return err; | |
2897 | } | |
2898 | ||
0bf765f7 IS |
2899 | /* QoS Functions */ |
2900 | ||
2901 | /* | |
2902 | * Initialize QoS configuration operations. | |
2903 | */ | |
2904 | static void | |
2905 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
2906 | { | |
2907 | conf->ops = ops; | |
78bd47cf | 2908 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
2909 | } |
2910 | ||
2911 | /* | |
2912 | * Search existing QoS operations in qos_ops and compare each set of | |
2913 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
2914 | * else return NULL | |
2915 | */ | |
2916 | static const struct dpdk_qos_ops * | |
2917 | qos_lookup_name(const char *name) | |
2918 | { | |
2919 | const struct dpdk_qos_ops *const *opsp; | |
2920 | ||
2921 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2922 | const struct dpdk_qos_ops *ops = *opsp; | |
2923 | if (!strcmp(name, ops->qos_name)) { | |
2924 | return ops; | |
2925 | } | |
2926 | } | |
2927 | return NULL; | |
2928 | } | |
2929 | ||
0bf765f7 IS |
2930 | static int |
2931 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
2932 | struct sset *types) | |
2933 | { | |
2934 | const struct dpdk_qos_ops *const *opsp; | |
2935 | ||
2936 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2937 | const struct dpdk_qos_ops *ops = *opsp; | |
2938 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
2939 | sset_add(types, ops->qos_name); | |
2940 | } | |
2941 | } | |
2942 | return 0; | |
2943 | } | |
2944 | ||
2945 | static int | |
d46285a2 | 2946 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
2947 | const char **typep, struct smap *details) |
2948 | { | |
d46285a2 | 2949 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 2950 | struct qos_conf *qos_conf; |
0bf765f7 IS |
2951 | int error = 0; |
2952 | ||
d46285a2 | 2953 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
2954 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
2955 | if (qos_conf) { | |
2956 | *typep = qos_conf->ops->qos_name; | |
2957 | error = (qos_conf->ops->qos_get | |
2958 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
2959 | } else { |
2960 | /* No QoS configuration set, return an empty string */ | |
2961 | *typep = ""; | |
0bf765f7 | 2962 | } |
d46285a2 | 2963 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2964 | |
2965 | return error; | |
2966 | } | |
2967 | ||
2968 | static int | |
78bd47cf DDP |
2969 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
2970 | const struct smap *details) | |
0bf765f7 | 2971 | { |
d46285a2 | 2972 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 2973 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 2974 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
2975 | int error = 0; |
2976 | ||
d46285a2 | 2977 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 2978 | |
78bd47cf | 2979 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 2980 | |
78bd47cf DDP |
2981 | new_ops = qos_lookup_name(type); |
2982 | ||
2983 | if (!new_ops || !new_ops->qos_construct) { | |
2984 | new_qos_conf = NULL; | |
2985 | if (type && type[0]) { | |
2986 | error = EOPNOTSUPP; | |
0bf765f7 | 2987 | } |
44975bb0 | 2988 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
2989 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
2990 | new_qos_conf = qos_conf; | |
0bf765f7 | 2991 | } else { |
78bd47cf | 2992 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
2993 | } |
2994 | ||
7ea266e9 | 2995 | if (error) { |
78bd47cf DDP |
2996 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
2997 | type, netdev->name, rte_strerror(error)); | |
2998 | } | |
2999 | ||
3000 | if (new_qos_conf != qos_conf) { | |
3001 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
3002 | if (qos_conf) { | |
3003 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
3004 | } | |
0bf765f7 IS |
3005 | } |
3006 | ||
d46285a2 | 3007 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 3008 | |
0bf765f7 IS |
3009 | return error; |
3010 | } | |
3011 | ||
3012 | /* egress-policer details */ | |
3013 | ||
3014 | struct egress_policer { | |
3015 | struct qos_conf qos_conf; | |
3016 | struct rte_meter_srtcm_params app_srtcm_params; | |
3017 | struct rte_meter_srtcm egress_meter; | |
3018 | }; | |
3019 | ||
78bd47cf DDP |
3020 | static void |
3021 | egress_policer_details_to_param(const struct smap *details, | |
3022 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 3023 | { |
78bd47cf DDP |
3024 | memset(params, 0, sizeof *params); |
3025 | params->cir = smap_get_ullong(details, "cir", 0); | |
3026 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
3027 | params->ebs = 0; | |
0bf765f7 IS |
3028 | } |
3029 | ||
3030 | static int | |
78bd47cf DDP |
3031 | egress_policer_qos_construct(const struct smap *details, |
3032 | struct qos_conf **conf) | |
0bf765f7 | 3033 | { |
0bf765f7 | 3034 | struct egress_policer *policer; |
0bf765f7 IS |
3035 | int err = 0; |
3036 | ||
0bf765f7 IS |
3037 | policer = xmalloc(sizeof *policer); |
3038 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 3039 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
0bf765f7 | 3040 | err = rte_meter_srtcm_config(&policer->egress_meter, |
78bd47cf DDP |
3041 | &policer->app_srtcm_params); |
3042 | if (!err) { | |
3043 | *conf = &policer->qos_conf; | |
3044 | } else { | |
7ea266e9 | 3045 | free(policer); |
78bd47cf | 3046 | *conf = NULL; |
7ea266e9 IS |
3047 | err = -err; |
3048 | } | |
0bf765f7 IS |
3049 | |
3050 | return err; | |
3051 | } | |
3052 | ||
3053 | static void | |
78bd47cf | 3054 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
3055 | { |
3056 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 3057 | qos_conf); |
0bf765f7 IS |
3058 | free(policer); |
3059 | } | |
3060 | ||
3061 | static int | |
78bd47cf | 3062 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 3063 | { |
78bd47cf DDP |
3064 | struct egress_policer *policer = |
3065 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3066 | ||
3067 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
3068 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 3069 | |
0bf765f7 IS |
3070 | return 0; |
3071 | } | |
3072 | ||
78bd47cf | 3073 | static bool |
47a45d86 KT |
3074 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
3075 | const struct smap *details) | |
0bf765f7 | 3076 | { |
78bd47cf DDP |
3077 | struct egress_policer *policer = |
3078 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3079 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 3080 | |
78bd47cf | 3081 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 3082 | |
78bd47cf | 3083 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
3084 | } |
3085 | ||
0bf765f7 | 3086 | static int |
78bd47cf | 3087 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt) |
0bf765f7 | 3088 | { |
0bf765f7 | 3089 | int cnt = 0; |
78bd47cf DDP |
3090 | struct egress_policer *policer = |
3091 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 3092 | |
f3926f29 | 3093 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, pkt_cnt); |
0bf765f7 IS |
3094 | |
3095 | return cnt; | |
3096 | } | |
3097 | ||
3098 | static const struct dpdk_qos_ops egress_policer_ops = { | |
3099 | "egress-policer", /* qos_name */ | |
3100 | egress_policer_qos_construct, | |
3101 | egress_policer_qos_destruct, | |
3102 | egress_policer_qos_get, | |
78bd47cf | 3103 | egress_policer_qos_is_equal, |
0bf765f7 IS |
3104 | egress_policer_run |
3105 | }; | |
3106 | ||
050c60bf DDP |
3107 | static int |
3108 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
3109 | { | |
3110 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3111 | int err = 0; | |
3112 | ||
050c60bf DDP |
3113 | ovs_mutex_lock(&dev->mutex); |
3114 | ||
3115 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 3116 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b CL |
3117 | && dev->mtu == dev->requested_mtu |
3118 | && dev->rxq_size == dev->requested_rxq_size | |
bd4e172b | 3119 | && dev->txq_size == dev->requested_txq_size |
3120 | && dev->socket_id == dev->requested_socket_id) { | |
050c60bf DDP |
3121 | /* Reconfiguration is unnecessary */ |
3122 | ||
3123 | goto out; | |
3124 | } | |
3125 | ||
3126 | rte_eth_dev_stop(dev->port_id); | |
3127 | ||
bd4e172b | 3128 | if (dev->mtu != dev->requested_mtu |
3129 | || dev->socket_id != dev->requested_socket_id) { | |
96e9b168 IS |
3130 | err = netdev_dpdk_mempool_configure(dev); |
3131 | if (err) { | |
3132 | goto out; | |
3133 | } | |
0072e931 MK |
3134 | } |
3135 | ||
050c60bf DDP |
3136 | netdev->n_txq = dev->requested_n_txq; |
3137 | netdev->n_rxq = dev->requested_n_rxq; | |
3138 | ||
b685696b CL |
3139 | dev->rxq_size = dev->requested_rxq_size; |
3140 | dev->txq_size = dev->requested_txq_size; | |
3141 | ||
050c60bf DDP |
3142 | rte_free(dev->tx_q); |
3143 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
3144 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
3145 | if (!dev->tx_q) { | |
3146 | err = ENOMEM; | |
3147 | } | |
050c60bf | 3148 | |
0072e931 MK |
3149 | netdev_change_seq_changed(netdev); |
3150 | ||
050c60bf | 3151 | out: |
050c60bf | 3152 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
3153 | return err; |
3154 | } | |
3155 | ||
7f381c2e | 3156 | static int |
2d24d165 | 3157 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 3158 | OVS_REQUIRES(dev->mutex) |
050c60bf | 3159 | { |
2d24d165 CL |
3160 | dev->up.n_txq = dev->requested_n_txq; |
3161 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 3162 | int err; |
050c60bf | 3163 | |
81acebda IM |
3164 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
3165 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
3166 | dev->tx_q[0].map = 0; | |
3167 | } | |
3168 | ||
3169 | netdev_dpdk_remap_txqs(dev); | |
3170 | ||
0072e931 MK |
3171 | if (dev->requested_socket_id != dev->socket_id |
3172 | || dev->requested_mtu != dev->mtu) { | |
96e9b168 IS |
3173 | err = netdev_dpdk_mempool_configure(dev); |
3174 | if (err) { | |
3175 | return err; | |
3176 | } else { | |
2d24d165 | 3177 | netdev_change_seq_changed(&dev->up); |
db8f13b0 CL |
3178 | } |
3179 | } | |
3180 | ||
0a0f39df CL |
3181 | if (netdev_dpdk_get_vid(dev) >= 0) { |
3182 | dev->vhost_reconfigured = true; | |
81acebda | 3183 | } |
7f381c2e DDP |
3184 | |
3185 | return 0; | |
2d24d165 CL |
3186 | } |
3187 | ||
3188 | static int | |
3189 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
3190 | { | |
3191 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3192 | int err; |
2d24d165 | 3193 | |
2d24d165 | 3194 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 3195 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 3196 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
3197 | |
3198 | return err; | |
2d24d165 CL |
3199 | } |
3200 | ||
3201 | static int | |
3202 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3203 | { | |
3204 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3205 | int err; |
2d24d165 | 3206 | |
2d24d165 CL |
3207 | ovs_mutex_lock(&dev->mutex); |
3208 | ||
c1ff66ac CL |
3209 | /* Configure vHost client mode if requested and if the following criteria |
3210 | * are met: | |
2d24d165 CL |
3211 | * 1. Device hasn't been registered yet. |
3212 | * 2. A path has been specified. | |
c1ff66ac CL |
3213 | */ |
3214 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 CL |
3215 | && strlen(dev->vhost_id)) { |
3216 | /* Register client-mode device */ | |
3217 | err = rte_vhost_driver_register(dev->vhost_id, | |
3218 | RTE_VHOST_USER_CLIENT); | |
c1ff66ac | 3219 | if (err) { |
2d24d165 CL |
3220 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3221 | dev->vhost_id); | |
7f381c2e | 3222 | goto unlock; |
c1ff66ac | 3223 | } else { |
2d24d165 CL |
3224 | /* Configuration successful */ |
3225 | dev->vhost_driver_flags |= RTE_VHOST_USER_CLIENT; | |
3226 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " | |
3227 | "using client socket '%s'", | |
3228 | dev->up.name, dev->vhost_id); | |
c1ff66ac CL |
3229 | } |
3230 | } | |
3231 | ||
7f381c2e DDP |
3232 | err = dpdk_vhost_reconfigure_helper(dev); |
3233 | ||
3234 | unlock: | |
050c60bf | 3235 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 3236 | |
7f381c2e | 3237 | return err; |
050c60bf DDP |
3238 | } |
3239 | ||
ecc1a34e | 3240 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
81acebda IM |
3241 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
3242 | GET_CARRIER, GET_STATS, \ | |
3243 | GET_FEATURES, GET_STATUS, \ | |
3244 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3245 | { \ |
3246 | NAME, \ | |
118c77b1 | 3247 | true, /* is_pmd */ \ |
ecc1a34e | 3248 | INIT, /* init */ \ |
95fb793a | 3249 | NULL, /* netdev_dpdk_run */ \ |
3250 | NULL, /* netdev_dpdk_wait */ \ | |
3251 | \ | |
3252 | netdev_dpdk_alloc, \ | |
3253 | CONSTRUCT, \ | |
58397e6c | 3254 | DESTRUCT, \ |
95fb793a | 3255 | netdev_dpdk_dealloc, \ |
3256 | netdev_dpdk_get_config, \ | |
81acebda | 3257 | SET_CONFIG, \ |
95fb793a | 3258 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3259 | NULL, /* build header */ \ |
3260 | NULL, /* push header */ \ | |
3261 | NULL, /* pop header */ \ | |
7dec44fe | 3262 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3263 | SET_TX_MULTIQ, \ |
95fb793a | 3264 | \ |
7251515e | 3265 | SEND, /* send */ \ |
95fb793a | 3266 | NULL, /* send_wait */ \ |
3267 | \ | |
3268 | netdev_dpdk_set_etheraddr, \ | |
3269 | netdev_dpdk_get_etheraddr, \ | |
3270 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3271 | netdev_dpdk_set_mtu, \ |
95fb793a | 3272 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3273 | GET_CARRIER, \ |
95fb793a | 3274 | netdev_dpdk_get_carrier_resets, \ |
3275 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
3276 | GET_STATS, \ |
3277 | GET_FEATURES, \ | |
95fb793a | 3278 | NULL, /* set_advertisements */ \ |
3279 | \ | |
9509913a | 3280 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3281 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3282 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3283 | netdev_dpdk_get_qos, \ |
3284 | netdev_dpdk_set_qos, \ | |
95fb793a | 3285 | NULL, /* get_queue */ \ |
3286 | NULL, /* set_queue */ \ | |
3287 | NULL, /* delete_queue */ \ | |
3288 | NULL, /* get_queue_stats */ \ | |
3289 | NULL, /* queue_dump_start */ \ | |
3290 | NULL, /* queue_dump_next */ \ | |
3291 | NULL, /* queue_dump_done */ \ | |
3292 | NULL, /* dump_queue_stats */ \ | |
3293 | \ | |
95fb793a | 3294 | NULL, /* set_in4 */ \ |
a8704b50 | 3295 | NULL, /* get_addr_list */ \ |
95fb793a | 3296 | NULL, /* add_router */ \ |
3297 | NULL, /* get_next_hop */ \ | |
58397e6c | 3298 | GET_STATUS, \ |
95fb793a | 3299 | NULL, /* arp_lookup */ \ |
3300 | \ | |
3301 | netdev_dpdk_update_flags, \ | |
050c60bf | 3302 | RECONFIGURE, \ |
95fb793a | 3303 | \ |
3304 | netdev_dpdk_rxq_alloc, \ | |
3305 | netdev_dpdk_rxq_construct, \ | |
3306 | netdev_dpdk_rxq_destruct, \ | |
3307 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3308 | RXQ_RECV, \ |
95fb793a | 3309 | NULL, /* rx_wait */ \ |
3310 | NULL, /* rxq_drain */ \ | |
18ebd48c | 3311 | NO_OFFLOAD_API \ |
95fb793a | 3312 | } |
8a9562d2 | 3313 | |
bce01e3a | 3314 | static const struct netdev_class dpdk_class = |
95fb793a | 3315 | NETDEV_DPDK_CLASS( |
3316 | "dpdk", | |
ecc1a34e | 3317 | netdev_dpdk_class_init, |
5496878c | 3318 | netdev_dpdk_construct, |
58397e6c | 3319 | netdev_dpdk_destruct, |
81acebda IM |
3320 | netdev_dpdk_set_config, |
3321 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3322 | netdev_dpdk_eth_send, |
3323 | netdev_dpdk_get_carrier, | |
3324 | netdev_dpdk_get_stats, | |
3325 | netdev_dpdk_get_features, | |
3326 | netdev_dpdk_get_status, | |
050c60bf | 3327 | netdev_dpdk_reconfigure, |
58397e6c | 3328 | netdev_dpdk_rxq_recv); |
95fb793a | 3329 | |
bce01e3a | 3330 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3331 | NETDEV_DPDK_CLASS( |
3332 | "dpdkr", | |
ecc1a34e | 3333 | netdev_dpdk_class_init, |
5496878c | 3334 | netdev_dpdk_ring_construct, |
58397e6c | 3335 | netdev_dpdk_destruct, |
c3d062a7 | 3336 | netdev_dpdk_ring_set_config, |
81acebda | 3337 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3338 | netdev_dpdk_ring_send, |
3339 | netdev_dpdk_get_carrier, | |
3340 | netdev_dpdk_get_stats, | |
3341 | netdev_dpdk_get_features, | |
3342 | netdev_dpdk_get_status, | |
050c60bf | 3343 | netdev_dpdk_reconfigure, |
58397e6c KT |
3344 | netdev_dpdk_rxq_recv); |
3345 | ||
53f50d24 | 3346 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3347 | NETDEV_DPDK_CLASS( |
3348 | "dpdkvhostuser", | |
ecc1a34e | 3349 | netdev_dpdk_vhost_class_init, |
53f50d24 | 3350 | netdev_dpdk_vhost_construct, |
58397e6c | 3351 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3352 | NULL, |
81acebda | 3353 | NULL, |
58397e6c KT |
3354 | netdev_dpdk_vhost_send, |
3355 | netdev_dpdk_vhost_get_carrier, | |
3356 | netdev_dpdk_vhost_get_stats, | |
3357 | NULL, | |
7251515e | 3358 | NULL, |
53f50d24 | 3359 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3360 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3361 | static const struct netdev_class dpdk_vhost_client_class = |
3362 | NETDEV_DPDK_CLASS( | |
3363 | "dpdkvhostuserclient", | |
ecc1a34e | 3364 | netdev_dpdk_vhost_class_init, |
2d24d165 CL |
3365 | netdev_dpdk_vhost_client_construct, |
3366 | netdev_dpdk_vhost_destruct, | |
3367 | netdev_dpdk_vhost_client_set_config, | |
3368 | NULL, | |
3369 | netdev_dpdk_vhost_send, | |
3370 | netdev_dpdk_vhost_get_carrier, | |
3371 | netdev_dpdk_vhost_get_stats, | |
3372 | NULL, | |
3373 | NULL, | |
3374 | netdev_dpdk_vhost_client_reconfigure, | |
3375 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 3376 | |
8a9562d2 PS |
3377 | void |
3378 | netdev_dpdk_register(void) | |
3379 | { | |
bab69409 AC |
3380 | netdev_register_provider(&dpdk_class); |
3381 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3382 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 3383 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 3384 | } |