]>
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) | |
79 | #define MBUF_SIZE(mtu) (MTU_TO_MAX_FRAME_LEN(mtu) \ | |
80 | + sizeof(struct dp_packet) \ | |
81 | + RTE_PKTMBUF_HEADROOM) | |
4be4d22c | 82 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 83 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 84 | |
da79ce2b DDP |
85 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
86 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
87 | * enough hugepages) we keep halving the number until the allocation succeeds | |
88 | * or we reach MIN_NB_MBUF */ | |
89 | ||
90 | #define MAX_NB_MBUF (4096 * 64) | |
91 | #define MIN_NB_MBUF (4096 * 4) | |
92 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
93 | ||
94 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
95 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
96 | ||
97 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
98 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
99 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
100 | % MP_CACHE_SZ == 0); | |
101 | ||
d6e3feb5 | 102 | /* |
103 | * DPDK XSTATS Counter names definition | |
104 | */ | |
105 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
106 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
107 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
108 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
109 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
110 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
111 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
112 | ||
113 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
114 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
115 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
116 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
117 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
118 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
119 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
120 | ||
d57f777f | 121 | #define XSTAT_RX_MULTICAST_PACKETS "rx_multicast_packets" |
d6e3feb5 | 122 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" |
123 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
124 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
125 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
126 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
127 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
128 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
129 | ||
8a9562d2 PS |
130 | #define SOCKET0 0 |
131 | ||
b685696b CL |
132 | /* Default size of Physical NIC RXQ */ |
133 | #define NIC_PORT_DEFAULT_RXQ_SIZE 2048 | |
134 | /* Default size of Physical NIC TXQ */ | |
135 | #define NIC_PORT_DEFAULT_TXQ_SIZE 2048 | |
136 | /* Maximum size of Physical NIC Queues */ | |
137 | #define NIC_PORT_MAX_Q_SIZE 4096 | |
79f5354c | 138 | |
585a5bea | 139 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
140 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
141 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
142 | * yet mapped to another queue. */ | |
585a5bea | 143 | |
31871ee3 | 144 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 145 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 146 | |
8a9562d2 | 147 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
148 | .rxmode = { |
149 | .mq_mode = ETH_MQ_RX_RSS, | |
150 | .split_hdr_size = 0, | |
151 | .header_split = 0, /* Header Split disabled */ | |
152 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
153 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
154 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
155 | .hw_strip_crc = 0, | |
156 | }, | |
157 | .rx_adv_conf = { | |
158 | .rss_conf = { | |
159 | .rss_key = NULL, | |
543342a4 | 160 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 161 | }, |
a28ddd11 DDP |
162 | }, |
163 | .txmode = { | |
164 | .mq_mode = ETH_MQ_TX_NONE, | |
165 | }, | |
8a9562d2 PS |
166 | }; |
167 | ||
58f7c37b DDP |
168 | enum { DPDK_RING_SIZE = 256 }; |
169 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
170 | enum { DRAIN_TSC = 200000ULL }; |
171 | ||
58397e6c KT |
172 | enum dpdk_dev_type { |
173 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 174 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
175 | }; |
176 | ||
0bf765f7 IS |
177 | /* Quality of Service */ |
178 | ||
179 | /* An instance of a QoS configuration. Always associated with a particular | |
180 | * network device. | |
181 | * | |
182 | * Each QoS implementation subclasses this with whatever additional data it | |
183 | * needs. | |
184 | */ | |
185 | struct qos_conf { | |
186 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 187 | rte_spinlock_t lock; |
0bf765f7 IS |
188 | }; |
189 | ||
190 | /* A particular implementation of dpdk QoS operations. | |
191 | * | |
192 | * The functions below return 0 if successful or a positive errno value on | |
193 | * failure, except where otherwise noted. All of them must be provided, except | |
194 | * where otherwise noted. | |
195 | */ | |
196 | struct dpdk_qos_ops { | |
197 | ||
198 | /* Name of the QoS type */ | |
199 | const char *qos_name; | |
200 | ||
78bd47cf DDP |
201 | /* Called to construct a qos_conf object. The implementation should make |
202 | * the appropriate calls to configure QoS according to 'details'. | |
0bf765f7 IS |
203 | * |
204 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
205 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
206 | * (which is built as ovs-vswitchd.conf.db(8)). | |
207 | * | |
78bd47cf DDP |
208 | * This function must return 0 if and only if it sets '*conf' to an |
209 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
210 | * |
211 | * For all QoS implementations it should always be non-null. | |
212 | */ | |
78bd47cf | 213 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
214 | |
215 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 216 | * 'qos_conf'. |
0bf765f7 IS |
217 | * |
218 | * For all QoS implementations it should always be non-null. | |
219 | */ | |
78bd47cf | 220 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 221 | |
78bd47cf | 222 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
223 | * |
224 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
225 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
226 | * (which is built as ovs-vswitchd.conf.db(8)). | |
227 | */ | |
78bd47cf | 228 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 229 | |
78bd47cf | 230 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
231 | * |
232 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
233 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
234 | * (which is built as ovs-vswitchd.conf.db(8)). | |
235 | * | |
78bd47cf | 236 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 237 | */ |
78bd47cf DDP |
238 | bool (*qos_is_equal)(const struct qos_conf *conf, |
239 | const struct smap *details); | |
0bf765f7 IS |
240 | |
241 | /* Modify an array of rte_mbufs. The modification is specific to | |
242 | * each qos implementation. | |
243 | * | |
244 | * The function should take and array of mbufs and an int representing | |
245 | * the current number of mbufs present in the array. | |
246 | * | |
247 | * After the function has performed a qos modification to the array of | |
248 | * mbufs it returns an int representing the number of mbufs now present in | |
249 | * the array. This value is can then be passed to the port send function | |
250 | * along with the modified array for transmission. | |
251 | * | |
252 | * For all QoS implementations it should always be non-null. | |
253 | */ | |
78bd47cf DDP |
254 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
255 | int pkt_cnt); | |
0bf765f7 IS |
256 | }; |
257 | ||
258 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
259 | static const struct dpdk_qos_ops egress_policer_ops; | |
260 | ||
261 | /* | |
262 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
263 | * operations. | |
264 | */ | |
265 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
266 | &egress_policer_ops, | |
267 | NULL | |
268 | }; | |
269 | ||
c2adb102 IM |
270 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
271 | ||
8a9562d2 | 272 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 273 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 274 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 275 | |
c2adb102 IM |
276 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
277 | = OVS_MUTEX_INITIALIZER; | |
278 | ||
279 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mp_mutex) | |
55951e15 | 280 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 281 | |
8a9562d2 PS |
282 | struct dpdk_mp { |
283 | struct rte_mempool *mp; | |
284 | int mtu; | |
285 | int socket_id; | |
286 | int refcount; | |
c2adb102 | 287 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mp_mutex); |
8a9562d2 PS |
288 | }; |
289 | ||
5a034064 AW |
290 | /* There should be one 'struct dpdk_tx_queue' created for |
291 | * each cpu core. */ | |
8a9562d2 | 292 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
293 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
294 | * from concurrent access. It is used only | |
295 | * if the queue is shared among different | |
324c8374 | 296 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
297 | int map; /* Mapping of configured vhost-user queues |
298 | * to enabled by guest. */ | |
8a9562d2 PS |
299 | }; |
300 | ||
95fb793a | 301 | /* dpdk has no way to remove dpdk ring ethernet devices |
302 | so we have to keep them around once they've been created | |
303 | */ | |
304 | ||
ca6ba700 | 305 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 306 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 307 | |
308 | struct dpdk_ring { | |
309 | /* For the client rings */ | |
310 | struct rte_ring *cring_tx; | |
311 | struct rte_ring *cring_rx; | |
b83a2df1 | 312 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
95fb793a | 313 | int eth_port_id; /* ethernet device port id */ |
ca6ba700 | 314 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 315 | }; |
316 | ||
9509913a IS |
317 | struct ingress_policer { |
318 | struct rte_meter_srtcm_params app_srtcm_params; | |
319 | struct rte_meter_srtcm in_policer; | |
320 | rte_spinlock_t policer_lock; | |
321 | }; | |
322 | ||
1a2bb118 SC |
323 | enum dpdk_hw_ol_features { |
324 | NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0, | |
325 | }; | |
326 | ||
8a9562d2 PS |
327 | struct netdev_dpdk { |
328 | struct netdev up; | |
329 | int port_id; | |
330 | int max_packet_len; | |
58397e6c | 331 | enum dpdk_dev_type type; |
8a9562d2 | 332 | |
5a034064 | 333 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
334 | |
335 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
336 | ||
337 | struct dpdk_mp *dpdk_mp; | |
338 | int mtu; | |
339 | int socket_id; | |
340 | int buf_size; | |
8a9562d2 | 341 | struct netdev_stats stats; |
45d947c4 DDP |
342 | /* Protects stats */ |
343 | rte_spinlock_t stats_lock; | |
8a9562d2 | 344 | |
74ff3298 | 345 | struct eth_addr hwaddr; |
8a9562d2 PS |
346 | enum netdev_flags flags; |
347 | ||
348 | struct rte_eth_link link; | |
349 | int link_reset_cnt; | |
350 | ||
0a0f39df CL |
351 | /* virtio identifier for vhost devices */ |
352 | ovsrcu_index vid; | |
353 | ||
354 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
355 | bool vhost_reconfigured; | |
58397e6c | 356 | |
2d24d165 CL |
357 | /* Identifier used to distinguish vhost devices from each other. */ |
358 | char vhost_id[PATH_MAX]; | |
7d1ced01 | 359 | |
55e075e6 CL |
360 | /* Device arguments for dpdk ports */ |
361 | char *devargs; | |
362 | ||
8a9562d2 | 363 | /* In dpdk_list. */ |
ca6ba700 | 364 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
0bf765f7 IS |
365 | |
366 | /* QoS configuration and lock for the device */ | |
78bd47cf | 367 | OVSRCU_TYPE(struct qos_conf *) qos_conf; |
0bf765f7 | 368 | |
050c60bf DDP |
369 | /* The following properties cannot be changed when a device is running, |
370 | * so we remember the request and update them next time | |
371 | * netdev_dpdk*_reconfigure() is called */ | |
0072e931 | 372 | int requested_mtu; |
050c60bf DDP |
373 | int requested_n_txq; |
374 | int requested_n_rxq; | |
b685696b CL |
375 | int requested_rxq_size; |
376 | int requested_txq_size; | |
377 | ||
378 | /* Number of rx/tx descriptors for physical devices */ | |
379 | int rxq_size; | |
380 | int txq_size; | |
9509913a | 381 | |
db8f13b0 CL |
382 | /* Socket ID detected when vHost device is brought up */ |
383 | int requested_socket_id; | |
384 | ||
c1ff66ac CL |
385 | /* Denotes whether vHost port is client/server mode */ |
386 | uint64_t vhost_driver_flags; | |
387 | ||
9509913a IS |
388 | /* Ingress Policer */ |
389 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
390 | uint32_t policer_rate; | |
391 | uint32_t policer_burst; | |
9fd39370 SC |
392 | |
393 | /* DPDK-ETH Flow control */ | |
394 | struct rte_eth_fc_conf fc_conf; | |
1a2bb118 SC |
395 | |
396 | /* DPDK-ETH hardware offload features, | |
397 | * from the enum set 'dpdk_hw_ol_features' */ | |
398 | uint32_t hw_ol_features; | |
8a9562d2 PS |
399 | }; |
400 | ||
401 | struct netdev_rxq_dpdk { | |
402 | struct netdev_rxq up; | |
403 | int port_id; | |
404 | }; | |
405 | ||
b3722dca | 406 | static int netdev_dpdk_class_init(void); |
407 | static int netdev_dpdk_vhost_class_init(void); | |
8a9562d2 | 408 | |
0a0f39df | 409 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 410 | |
9509913a IS |
411 | struct ingress_policer * |
412 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
413 | ||
8a9562d2 PS |
414 | static bool |
415 | is_dpdk_class(const struct netdev_class *class) | |
416 | { | |
b3722dca | 417 | return class->init == netdev_dpdk_class_init |
418 | || class->init == netdev_dpdk_vhost_class_init; | |
8a9562d2 PS |
419 | } |
420 | ||
4be4d22c MK |
421 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
422 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
423 | * value, insufficient buffers are allocated to accomodate the packet in its | |
424 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
425 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
426 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
427 | * behaviour, which reduces performance. To prevent this, use a buffer size |
428 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
429 | */ |
430 | static uint32_t | |
431 | dpdk_buf_size(int mtu) | |
432 | { | |
433 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
434 | NETDEV_DPDK_MBUF_ALIGN); | |
435 | } | |
436 | ||
eff23640 DDP |
437 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
438 | * | |
439 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
440 | static void * |
441 | dpdk_rte_mzalloc(size_t sz) | |
442 | { | |
eff23640 | 443 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
444 | } |
445 | ||
446 | void | |
e14deea0 | 447 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 448 | { |
db73f716 | 449 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 450 | |
b00b4a81 | 451 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
452 | } |
453 | ||
b3cd9f9d | 454 | static void |
401b70d6 | 455 | ovs_rte_pktmbuf_init(struct rte_mempool *mp OVS_UNUSED, |
b3cd9f9d | 456 | void *opaque_arg OVS_UNUSED, |
2391135c | 457 | void *_p, |
b3cd9f9d PS |
458 | unsigned i OVS_UNUSED) |
459 | { | |
2391135c | 460 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 461 | |
2391135c | 462 | dp_packet_init_dpdk((struct dp_packet *) pkt, pkt->buf_len); |
b3cd9f9d PS |
463 | } |
464 | ||
8a9562d2 | 465 | static struct dpdk_mp * |
2ae3d542 | 466 | dpdk_mp_create(int socket_id, int mtu) |
8a9562d2 | 467 | { |
2ae3d542 | 468 | struct dpdk_mp *dmp; |
da79ce2b | 469 | unsigned mp_size; |
0c6f39e5 | 470 | char *mp_name; |
8a9562d2 PS |
471 | |
472 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
eff23640 DDP |
473 | if (!dmp) { |
474 | return NULL; | |
475 | } | |
8a9562d2 PS |
476 | dmp->socket_id = socket_id; |
477 | dmp->mtu = mtu; | |
478 | dmp->refcount = 1; | |
0072e931 | 479 | /* XXX: this is a really rough method of provisioning memory. |
2ae3d542 | 480 | * It's impossible to determine what the exact memory requirements are |
58be5c0e MK |
481 | * when the number of ports and rxqs that utilize a particular mempool can |
482 | * change dynamically at runtime. For now, use this rough heurisitic. | |
0072e931 MK |
483 | */ |
484 | if (mtu >= ETHER_MTU) { | |
485 | mp_size = MAX_NB_MBUF; | |
486 | } else { | |
487 | mp_size = MIN_NB_MBUF; | |
488 | } | |
8a9562d2 | 489 | |
da79ce2b | 490 | do { |
0c6f39e5 DDP |
491 | mp_name = xasprintf("ovs_mp_%d_%d_%u", dmp->mtu, dmp->socket_id, |
492 | mp_size); | |
95fb793a | 493 | |
401b70d6 HA |
494 | dmp->mp = rte_pktmbuf_pool_create(mp_name, mp_size, |
495 | MP_CACHE_SZ, | |
496 | sizeof (struct dp_packet) | |
497 | - sizeof (struct rte_mbuf), | |
498 | MBUF_SIZE(mtu) | |
499 | - sizeof(struct dp_packet), | |
500 | socket_id); | |
0c6f39e5 DDP |
501 | if (dmp->mp) { |
502 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", | |
503 | mp_name, mp_size); | |
504 | } | |
505 | free(mp_name); | |
506 | if (dmp->mp) { | |
401b70d6 HA |
507 | /* rte_pktmbuf_pool_create has done some initialization of the |
508 | * rte_mbuf part of each dp_packet, while ovs_rte_pktmbuf_init | |
509 | * initializes some OVS specific fields of dp_packet. | |
510 | */ | |
511 | rte_mempool_obj_iter(dmp->mp, ovs_rte_pktmbuf_init, NULL); | |
0c6f39e5 DDP |
512 | return dmp; |
513 | } | |
514 | } while (rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
2ae3d542 | 515 | |
2ae3d542 DDP |
516 | rte_free(dmp); |
517 | return NULL; | |
518 | } | |
519 | ||
520 | static struct dpdk_mp * | |
521 | dpdk_mp_get(int socket_id, int mtu) | |
522 | { | |
523 | struct dpdk_mp *dmp; | |
524 | ||
525 | ovs_mutex_lock(&dpdk_mp_mutex); | |
526 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
527 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
528 | dmp->refcount++; | |
529 | goto out; | |
530 | } | |
531 | } | |
532 | ||
533 | dmp = dpdk_mp_create(socket_id, mtu); | |
96e9b168 IS |
534 | if (dmp) { |
535 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); | |
536 | } | |
5f88de0d | 537 | |
c2adb102 IM |
538 | out: |
539 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
540 | ||
c2adb102 | 541 | return dmp; |
8a9562d2 PS |
542 | } |
543 | ||
544 | static void | |
c2adb102 | 545 | dpdk_mp_put(struct dpdk_mp *dmp) |
8a9562d2 | 546 | { |
8a9562d2 PS |
547 | if (!dmp) { |
548 | return; | |
549 | } | |
550 | ||
c2adb102 | 551 | ovs_mutex_lock(&dpdk_mp_mutex); |
8d38823b | 552 | ovs_assert(dmp->refcount); |
8a9562d2 | 553 | |
8d38823b MK |
554 | if (!--dmp->refcount) { |
555 | ovs_list_remove(&dmp->list_node); | |
556 | rte_mempool_free(dmp->mp); | |
5f88de0d | 557 | rte_free(dmp); |
8a9562d2 | 558 | } |
c2adb102 | 559 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
560 | } |
561 | ||
0072e931 MK |
562 | /* Tries to allocate new mempool on requested_socket_id with |
563 | * mbuf size corresponding to requested_mtu. | |
564 | * On success new configuration will be applied. | |
565 | * On error, device will be left unchanged. */ | |
566 | static int | |
567 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
568 | OVS_REQUIRES(dev->mutex) |
569 | { | |
570 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
571 | struct dpdk_mp *mp; | |
572 | ||
573 | mp = dpdk_mp_get(dev->requested_socket_id, FRAME_LEN_TO_MTU(buf_size)); | |
574 | if (!mp) { | |
c67e46c0 MK |
575 | VLOG_ERR("Failed to create memory pool for netdev " |
576 | "%s, with MTU %d on socket %d: %s\n", | |
577 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
578 | rte_strerror(rte_errno)); | |
579 | return rte_errno; | |
0072e931 MK |
580 | } else { |
581 | dpdk_mp_put(dev->dpdk_mp); | |
582 | dev->dpdk_mp = mp; | |
583 | dev->mtu = dev->requested_mtu; | |
584 | dev->socket_id = dev->requested_socket_id; | |
585 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
586 | } | |
587 | ||
588 | return 0; | |
589 | } | |
590 | ||
8a9562d2 PS |
591 | static void |
592 | check_link_status(struct netdev_dpdk *dev) | |
593 | { | |
594 | struct rte_eth_link link; | |
595 | ||
596 | rte_eth_link_get_nowait(dev->port_id, &link); | |
597 | ||
598 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 599 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
600 | |
601 | dev->link_reset_cnt++; | |
602 | dev->link = link; | |
603 | if (dev->link.link_status) { | |
604 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
58be5c0e | 605 | dev->port_id, (unsigned) dev->link.link_speed, |
8a9562d2 PS |
606 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? |
607 | ("full-duplex") : ("half-duplex")); | |
608 | } else { | |
609 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
610 | } | |
611 | } | |
612 | } | |
613 | ||
614 | static void * | |
615 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
616 | { | |
617 | struct netdev_dpdk *dev; | |
618 | ||
619 | pthread_detach(pthread_self()); | |
620 | ||
621 | for (;;) { | |
622 | ovs_mutex_lock(&dpdk_mutex); | |
623 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
624 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
625 | if (dev->type == DPDK_DEV_ETH) { |
626 | check_link_status(dev); | |
627 | } | |
8a9562d2 PS |
628 | ovs_mutex_unlock(&dev->mutex); |
629 | } | |
630 | ovs_mutex_unlock(&dpdk_mutex); | |
631 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
632 | } | |
633 | ||
634 | return NULL; | |
635 | } | |
636 | ||
b98d7669 DDP |
637 | static int |
638 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
639 | { | |
640 | int diag = 0; | |
641 | int i; | |
0072e931 | 642 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 643 | |
0072e931 MK |
644 | if (dev->mtu > ETHER_MTU) { |
645 | conf.rxmode.jumbo_frame = 1; | |
646 | conf.rxmode.max_rx_pkt_len = dev->max_packet_len; | |
647 | } else { | |
648 | conf.rxmode.jumbo_frame = 0; | |
649 | conf.rxmode.max_rx_pkt_len = 0; | |
650 | } | |
1a2bb118 SC |
651 | conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & |
652 | NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
b98d7669 DDP |
653 | /* A device may report more queues than it makes available (this has |
654 | * been observed for Intel xl710, which reserves some of them for | |
655 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
656 | * available. When this happens we can retry the configuration | |
657 | * and request less queues */ | |
658 | while (n_rxq && n_txq) { | |
659 | if (diag) { | |
660 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
661 | } | |
662 | ||
0072e931 | 663 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 664 | if (diag) { |
0072e931 MK |
665 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
666 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
667 | break; |
668 | } | |
669 | ||
670 | for (i = 0; i < n_txq; i++) { | |
b685696b | 671 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
672 | dev->socket_id, NULL); |
673 | if (diag) { | |
674 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
675 | dev->up.name, i, rte_strerror(-diag)); | |
676 | break; | |
677 | } | |
678 | } | |
679 | ||
680 | if (i != n_txq) { | |
681 | /* Retry with less tx queues */ | |
682 | n_txq = i; | |
683 | continue; | |
684 | } | |
685 | ||
686 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 687 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
b98d7669 DDP |
688 | dev->socket_id, NULL, |
689 | dev->dpdk_mp->mp); | |
690 | if (diag) { | |
691 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
692 | dev->up.name, i, rte_strerror(-diag)); | |
693 | break; | |
694 | } | |
695 | } | |
696 | ||
697 | if (i != n_rxq) { | |
698 | /* Retry with less rx queues */ | |
699 | n_rxq = i; | |
700 | continue; | |
701 | } | |
702 | ||
703 | dev->up.n_rxq = n_rxq; | |
81acebda | 704 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
705 | |
706 | return 0; | |
707 | } | |
708 | ||
709 | return diag; | |
710 | } | |
711 | ||
1a2bb118 SC |
712 | static void |
713 | dpdk_eth_checksum_offload_configure(struct netdev_dpdk *dev) | |
714 | OVS_REQUIRES(dev->mutex) | |
715 | { | |
716 | struct rte_eth_dev_info info; | |
717 | bool rx_csum_ol_flag = false; | |
718 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | | |
719 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
720 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
721 | rte_eth_dev_info_get(dev->port_id, &info); | |
722 | rx_csum_ol_flag = (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
723 | ||
724 | if (rx_csum_ol_flag && | |
725 | (info.rx_offload_capa & rx_chksm_offload_capa) != | |
726 | rx_chksm_offload_capa) { | |
727 | VLOG_WARN_ONCE("Rx checksum offload is not supported on device %d", | |
728 | dev->port_id); | |
729 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; | |
730 | return; | |
731 | } | |
732 | netdev_request_reconfigure(&dev->up); | |
733 | } | |
734 | ||
9fd39370 SC |
735 | static void |
736 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
737 | { | |
738 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
739 | VLOG_WARN("Failed to enable flow control on device %d", dev->port_id); | |
740 | } | |
741 | } | |
b98d7669 | 742 | |
8a9562d2 | 743 | static int |
c2adb102 IM |
744 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
745 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
746 | { |
747 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 748 | struct rte_eth_dev_info info; |
8a9562d2 PS |
749 | struct ether_addr eth_addr; |
750 | int diag; | |
b98d7669 | 751 | int n_rxq, n_txq; |
8a9562d2 | 752 | |
a0cb2d66 | 753 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 754 | |
b98d7669 DDP |
755 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
756 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
757 | ||
758 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 759 | if (diag) { |
b98d7669 DDP |
760 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
761 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 762 | return -diag; |
8a9562d2 PS |
763 | } |
764 | ||
8a9562d2 PS |
765 | diag = rte_eth_dev_start(dev->port_id); |
766 | if (diag) { | |
b98d7669 DDP |
767 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
768 | rte_strerror(-diag)); | |
95fb793a | 769 | return -diag; |
8a9562d2 PS |
770 | } |
771 | ||
772 | rte_eth_promiscuous_enable(dev->port_id); | |
773 | rte_eth_allmulticast_enable(dev->port_id); | |
774 | ||
775 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
776 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
182ca1ae | 777 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT, |
ca92d173 | 778 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 779 | |
ca92d173 | 780 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
781 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
782 | ||
783 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
784 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
785 | ||
9fd39370 SC |
786 | /* Get the Flow control configuration for DPDK-ETH */ |
787 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
788 | if (diag) { | |
789 | VLOG_DBG("cannot get flow control parameters on port=%d, err=%d", | |
790 | dev->port_id, diag); | |
791 | } | |
792 | ||
8a9562d2 PS |
793 | return 0; |
794 | } | |
795 | ||
796 | static struct netdev_dpdk * | |
797 | netdev_dpdk_cast(const struct netdev *netdev) | |
798 | { | |
799 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
800 | } | |
801 | ||
802 | static struct netdev * | |
803 | netdev_dpdk_alloc(void) | |
804 | { | |
bab69409 AC |
805 | struct netdev_dpdk *dev; |
806 | ||
65e19e70 DDP |
807 | dev = dpdk_rte_mzalloc(sizeof *dev); |
808 | if (dev) { | |
809 | return &dev->up; | |
bab69409 | 810 | } |
65e19e70 | 811 | |
bab69409 | 812 | return NULL; |
8a9562d2 PS |
813 | } |
814 | ||
eff23640 DDP |
815 | static struct dpdk_tx_queue * |
816 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 817 | { |
eff23640 | 818 | struct dpdk_tx_queue *txqs; |
bd5131ba | 819 | unsigned i; |
5a034064 | 820 | |
eff23640 DDP |
821 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
822 | if (txqs) { | |
823 | for (i = 0; i < n_txqs; i++) { | |
824 | /* Initialize map for vhost devices. */ | |
825 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
826 | rte_spinlock_init(&txqs[i].tx_lock); | |
827 | } | |
5a034064 | 828 | } |
eff23640 DDP |
829 | |
830 | return txqs; | |
5a034064 AW |
831 | } |
832 | ||
8a9562d2 | 833 | static int |
1ce30dfd DDP |
834 | common_construct(struct netdev *netdev, unsigned int port_no, |
835 | enum dpdk_dev_type type, int socket_id) | |
5a034064 | 836 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 837 | { |
d46285a2 | 838 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 839 | |
d46285a2 | 840 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 841 | |
d46285a2 | 842 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 843 | |
1b7a04e0 AW |
844 | /* If the 'sid' is negative, it means that the kernel fails |
845 | * to obtain the pci numa info. In that situation, always | |
846 | * use 'SOCKET0'. */ | |
1ce30dfd | 847 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 848 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
849 | dev->port_id = port_no; |
850 | dev->type = type; | |
851 | dev->flags = 0; | |
7f381c2e | 852 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 853 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
854 | ovsrcu_index_init(&dev->vid, -1); |
855 | dev->vhost_reconfigured = false; | |
8a9562d2 | 856 | |
78bd47cf | 857 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 858 | |
9509913a IS |
859 | ovsrcu_init(&dev->ingress_policer, NULL); |
860 | dev->policer_rate = 0; | |
861 | dev->policer_burst = 0; | |
862 | ||
7f381c2e DDP |
863 | netdev->n_rxq = 0; |
864 | netdev->n_txq = 0; | |
865 | dev->requested_n_rxq = NR_QUEUE; | |
866 | dev->requested_n_txq = NR_QUEUE; | |
867 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
868 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 869 | |
9fd39370 SC |
870 | /* Initialize the flow control to NULL */ |
871 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
872 | |
873 | /* Initilize the hardware offload flags to 0 */ | |
874 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
875 | |
876 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
877 | ||
d46285a2 | 878 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 879 | |
7f381c2e DDP |
880 | netdev_request_reconfigure(netdev); |
881 | ||
1ce30dfd | 882 | return 0; |
95fb793a | 883 | } |
884 | ||
b83a2df1 MV |
885 | /* dev_name must be the prefix followed by a positive decimal number. |
886 | * (no leading + or - signs are allowed) */ | |
95fb793a | 887 | static int |
888 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
889 | unsigned int *port_no) | |
890 | { | |
891 | const char *cport; | |
892 | ||
893 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
894 | return ENODEV; | |
895 | } | |
896 | ||
897 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
898 | |
899 | if (str_to_uint(cport, 10, port_no)) { | |
900 | return 0; | |
901 | } else { | |
902 | return ENODEV; | |
903 | } | |
95fb793a | 904 | } |
905 | ||
1ce30dfd DDP |
906 | static int |
907 | vhost_common_construct(struct netdev *netdev) | |
908 | OVS_REQUIRES(dpdk_mutex) | |
909 | { | |
910 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
911 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
912 | ||
913 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); | |
914 | if (!dev->tx_q) { | |
915 | return ENOMEM; | |
916 | } | |
917 | ||
918 | return common_construct(netdev, -1, DPDK_DEV_VHOST, socket_id); | |
919 | } | |
920 | ||
7d1ced01 | 921 | static int |
53f50d24 | 922 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 923 | { |
d46285a2 DDP |
924 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
925 | const char *name = netdev->name; | |
7d1ced01 | 926 | int err; |
a0cb2d66 | 927 | |
1af27e8a DDP |
928 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
929 | * the file system. '/' or '\' would traverse directories, so they're not | |
930 | * acceptable in 'name'. */ | |
931 | if (strchr(name, '/') || strchr(name, '\\')) { | |
932 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
933 | "A valid name must not include '/' or '\\'", | |
934 | name); | |
935 | return EINVAL; | |
936 | } | |
937 | ||
7d1ced01 CL |
938 | ovs_mutex_lock(&dpdk_mutex); |
939 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 940 | * the socket is to be created, then register the socket. |
7d1ced01 | 941 | */ |
2d24d165 | 942 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 943 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 944 | |
2d24d165 CL |
945 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
946 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
947 | if (err) { |
948 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 949 | dev->vhost_id); |
e5c0f5a4 | 950 | } else { |
2d24d165 CL |
951 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
952 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
953 | dev->vhost_id, name); | |
954 | } | |
1ce30dfd | 955 | err = vhost_common_construct(netdev); |
2d24d165 CL |
956 | |
957 | ovs_mutex_unlock(&dpdk_mutex); | |
958 | return err; | |
959 | } | |
960 | ||
961 | static int | |
962 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
963 | { | |
964 | int err; | |
965 | ||
2d24d165 | 966 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 967 | err = vhost_common_construct(netdev); |
7d1ced01 | 968 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
969 | return err; |
970 | } | |
971 | ||
95fb793a | 972 | static int |
973 | netdev_dpdk_construct(struct netdev *netdev) | |
974 | { | |
95fb793a | 975 | int err; |
976 | ||
95fb793a | 977 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 978 | err = common_construct(netdev, -1, DPDK_DEV_ETH, SOCKET0); |
8a9562d2 PS |
979 | ovs_mutex_unlock(&dpdk_mutex); |
980 | return err; | |
981 | } | |
982 | ||
1ce30dfd DDP |
983 | static void |
984 | common_destruct(struct netdev_dpdk *dev) | |
985 | OVS_REQUIRES(dpdk_mutex) | |
986 | OVS_EXCLUDED(dev->mutex) | |
987 | { | |
988 | rte_free(dev->tx_q); | |
989 | dpdk_mp_put(dev->dpdk_mp); | |
990 | ||
991 | ovs_list_remove(&dev->list_node); | |
992 | free(ovsrcu_get_protected(struct ingress_policer *, | |
993 | &dev->ingress_policer)); | |
994 | ovs_mutex_destroy(&dev->mutex); | |
995 | } | |
996 | ||
8a9562d2 | 997 | static void |
d46285a2 | 998 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 999 | { |
d46285a2 | 1000 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1001 | |
8d38823b | 1002 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1003 | |
8a9562d2 | 1004 | rte_eth_dev_stop(dev->port_id); |
55e075e6 | 1005 | free(dev->devargs); |
1ce30dfd | 1006 | common_destruct(dev); |
8d38823b | 1007 | |
8a9562d2 | 1008 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1009 | } |
8a9562d2 | 1010 | |
3f891bbe DDP |
1011 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1012 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1013 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1014 | static int | |
c1ff66ac CL |
1015 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1016 | char *vhost_id) | |
3f891bbe DDP |
1017 | OVS_EXCLUDED(dpdk_mutex) |
1018 | OVS_EXCLUDED(dev->mutex) | |
1019 | { | |
c1ff66ac | 1020 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1021 | } |
1022 | ||
58397e6c | 1023 | static void |
d46285a2 | 1024 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1025 | { |
d46285a2 | 1026 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1027 | char *vhost_id; |
58397e6c | 1028 | |
8d38823b | 1029 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1030 | |
c62da695 | 1031 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1032 | if (netdev_dpdk_get_vid(dev) >= 0 |
1033 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1034 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1035 | netdev->name); |
58be5c0e MK |
1036 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1037 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1038 | } |
1039 | ||
2d24d165 | 1040 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1041 | |
1ce30dfd DDP |
1042 | common_destruct(dev); |
1043 | ||
58397e6c | 1044 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1045 | |
569c26da | 1046 | if (!vhost_id[0]) { |
821b8664 IM |
1047 | goto out; |
1048 | } | |
1049 | ||
c1ff66ac | 1050 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1051 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1052 | netdev->name, vhost_id); | |
c1ff66ac CL |
1053 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1054 | /* OVS server mode - remove this socket from list for deletion */ | |
1055 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1056 | } |
821b8664 | 1057 | out: |
c1ff66ac | 1058 | free(vhost_id); |
8a9562d2 PS |
1059 | } |
1060 | ||
1061 | static void | |
d46285a2 | 1062 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1063 | { |
d46285a2 | 1064 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1065 | |
d46285a2 | 1066 | rte_free(dev); |
8a9562d2 PS |
1067 | } |
1068 | ||
1069 | static int | |
a14b8947 | 1070 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1071 | { |
a14b8947 | 1072 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1073 | |
1074 | ovs_mutex_lock(&dev->mutex); | |
1075 | ||
050c60bf | 1076 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1077 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1078 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1079 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1080 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1081 | |
1082 | if (dev->type == DPDK_DEV_ETH) { | |
1083 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1084 | dev->requested_rxq_size); | |
1085 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1086 | dev->rxq_size); | |
1087 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1088 | dev->requested_txq_size); | |
1089 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1090 | dev->txq_size); | |
1a2bb118 SC |
1091 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1092 | smap_add(args, "rx_csum_offload", "true"); | |
1093 | } | |
451f26fd | 1094 | } |
8a9562d2 PS |
1095 | ovs_mutex_unlock(&dev->mutex); |
1096 | ||
1097 | return 0; | |
1098 | } | |
1099 | ||
55e075e6 CL |
1100 | static struct netdev_dpdk * |
1101 | netdev_dpdk_lookup_by_port_id(int port_id) | |
1102 | OVS_REQUIRES(dpdk_mutex) | |
1103 | { | |
1104 | struct netdev_dpdk *dev; | |
1105 | ||
1106 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1107 | if (dev->port_id == port_id) { | |
1108 | return dev; | |
1109 | } | |
1110 | } | |
1111 | ||
1112 | return NULL; | |
1113 | } | |
1114 | ||
1115 | static int | |
9fff138e | 1116 | netdev_dpdk_process_devargs(const char *devargs, char **errp) |
55e075e6 | 1117 | { |
55e075e6 CL |
1118 | uint8_t new_port_id = UINT8_MAX; |
1119 | ||
69876ed7 CL |
1120 | if (!rte_eth_dev_count() |
1121 | || rte_eth_dev_get_port_by_name(devargs, &new_port_id) | |
1122 | || !rte_eth_dev_is_valid_port(new_port_id)) { | |
1123 | /* Device not found in DPDK, attempt to attach it */ | |
1124 | if (!rte_eth_dev_attach(devargs, &new_port_id)) { | |
1125 | /* Attach successful */ | |
1126 | VLOG_INFO("Device '%s' attached to DPDK", devargs); | |
1127 | } else { | |
1128 | /* Attach unsuccessful */ | |
47a45d86 KT |
1129 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", |
1130 | devargs); | |
69876ed7 | 1131 | return -1; |
55e075e6 CL |
1132 | } |
1133 | } | |
1134 | ||
1135 | return new_port_id; | |
1136 | } | |
1137 | ||
c3d062a7 CL |
1138 | static void |
1139 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1140 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1141 | { |
050c60bf | 1142 | int new_n_rxq; |
a14b8947 | 1143 | |
2a21e757 | 1144 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1145 | if (new_n_rxq != dev->requested_n_rxq) { |
1146 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1147 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1148 | } |
c3d062a7 CL |
1149 | } |
1150 | ||
b685696b CL |
1151 | static void |
1152 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1153 | const char *flag, int default_size, int *new_size) | |
1154 | { | |
1155 | int queue_size = smap_get_int(args, flag, default_size); | |
1156 | ||
1157 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1158 | || !is_pow2(queue_size)) { | |
1159 | queue_size = default_size; | |
1160 | } | |
1161 | ||
1162 | if (queue_size != *new_size) { | |
1163 | *new_size = queue_size; | |
1164 | netdev_request_reconfigure(netdev); | |
1165 | } | |
1166 | } | |
1167 | ||
c3d062a7 | 1168 | static int |
9fff138e DDP |
1169 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1170 | char **errp) | |
c3d062a7 CL |
1171 | { |
1172 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
b614c894 IM |
1173 | bool rx_fc_en, tx_fc_en, autoneg; |
1174 | enum rte_eth_fc_mode fc_mode; | |
1175 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1176 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1177 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1178 | }; | |
1a2bb118 SC |
1179 | bool rx_chksm_ofld; |
1180 | bool temp_flag; | |
55e075e6 CL |
1181 | const char *new_devargs; |
1182 | int err = 0; | |
c3d062a7 | 1183 | |
55e075e6 | 1184 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1185 | ovs_mutex_lock(&dev->mutex); |
1186 | ||
1187 | dpdk_set_rxq_config(dev, args); | |
1188 | ||
b685696b CL |
1189 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1190 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1191 | &dev->requested_rxq_size); | |
1192 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1193 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1194 | &dev->requested_txq_size); | |
1195 | ||
55e075e6 CL |
1196 | new_devargs = smap_get(args, "dpdk-devargs"); |
1197 | ||
1198 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1199 | /* The user requested a new device. If we return error, the caller | |
1200 | * will delete this netdev and try to recreate it. */ | |
1201 | err = EAGAIN; | |
1202 | goto out; | |
1203 | } | |
1204 | ||
1205 | /* dpdk-devargs is required for device configuration */ | |
1206 | if (new_devargs && new_devargs[0]) { | |
1207 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1208 | * is valid */ | |
1209 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1210 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
9fff138e | 1211 | int new_port_id = netdev_dpdk_process_devargs(new_devargs, errp); |
55e075e6 CL |
1212 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1213 | err = EINVAL; | |
1214 | } else if (new_port_id == dev->port_id) { | |
1215 | /* Already configured, do not reconfigure again */ | |
1216 | err = 0; | |
1217 | } else { | |
1218 | struct netdev_dpdk *dup_dev; | |
1219 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); | |
1220 | if (dup_dev) { | |
9fff138e DDP |
1221 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
1222 | "which is already in use by '%s'", | |
1223 | netdev_get_name(netdev), new_devargs, | |
1224 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1225 | err = EADDRINUSE; |
1226 | } else { | |
bd4e172b | 1227 | int sid = rte_eth_dev_socket_id(new_port_id); |
1228 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; | |
55e075e6 CL |
1229 | dev->devargs = xstrdup(new_devargs); |
1230 | dev->port_id = new_port_id; | |
1231 | netdev_request_reconfigure(&dev->up); | |
1232 | err = 0; | |
1233 | } | |
1234 | } | |
1235 | } | |
1236 | } else { | |
9fff138e DDP |
1237 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1238 | "The old 'dpdk<port_id>' names are not supported", | |
1239 | netdev_get_name(netdev)); | |
55e075e6 CL |
1240 | err = EINVAL; |
1241 | } | |
1242 | ||
1243 | if (err) { | |
1244 | goto out; | |
1245 | } | |
1246 | ||
c3d062a7 CL |
1247 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1248 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1249 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1250 | |
b614c894 IM |
1251 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1252 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1253 | dev->fc_conf.mode = fc_mode; | |
1254 | dev->fc_conf.autoneg = autoneg; | |
1255 | dpdk_eth_flow_ctrl_setup(dev); | |
1256 | } | |
9fd39370 | 1257 | |
1a2bb118 SC |
1258 | /* Rx checksum offload configuration */ |
1259 | /* By default the Rx checksum offload is ON */ | |
1260 | rx_chksm_ofld = smap_get_bool(args, "rx-checksum-offload", true); | |
1261 | temp_flag = (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) | |
1262 | != 0; | |
1263 | if (temp_flag != rx_chksm_ofld) { | |
1264 | dev->hw_ol_features ^= NETDEV_RX_CHECKSUM_OFFLOAD; | |
1265 | dpdk_eth_checksum_offload_configure(dev); | |
1266 | } | |
55e075e6 CL |
1267 | |
1268 | out: | |
c3d062a7 | 1269 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1270 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1271 | |
55e075e6 | 1272 | return err; |
c3d062a7 CL |
1273 | } |
1274 | ||
1275 | static int | |
9fff138e DDP |
1276 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1277 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1278 | { |
1279 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1280 | ||
1281 | ovs_mutex_lock(&dev->mutex); | |
1282 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1283 | ovs_mutex_unlock(&dev->mutex); |
1284 | ||
1285 | return 0; | |
1286 | } | |
1287 | ||
c1ff66ac | 1288 | static int |
2d24d165 | 1289 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1290 | const struct smap *args, |
1291 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1292 | { |
1293 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1294 | const char *path; | |
1295 | ||
6881885a | 1296 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1297 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1298 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1299 | if (path && strcmp(path, dev->vhost_id)) { |
1300 | strcpy(dev->vhost_id, path); | |
c1ff66ac CL |
1301 | netdev_request_reconfigure(netdev); |
1302 | } | |
1303 | } | |
6881885a | 1304 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1305 | |
1306 | return 0; | |
1307 | } | |
1308 | ||
7dec44fe | 1309 | static int |
d46285a2 | 1310 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1311 | { |
d46285a2 | 1312 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1313 | |
d46285a2 | 1314 | return dev->socket_id; |
7dec44fe AW |
1315 | } |
1316 | ||
050c60bf | 1317 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1318 | static int |
050c60bf | 1319 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1320 | { |
d46285a2 | 1321 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1322 | |
d46285a2 | 1323 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1324 | |
050c60bf DDP |
1325 | if (dev->requested_n_txq == n_txq) { |
1326 | goto out; | |
4573fbd3 FL |
1327 | } |
1328 | ||
050c60bf DDP |
1329 | dev->requested_n_txq = n_txq; |
1330 | netdev_request_reconfigure(netdev); | |
58397e6c | 1331 | |
050c60bf | 1332 | out: |
d46285a2 | 1333 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1334 | return 0; |
58397e6c KT |
1335 | } |
1336 | ||
8a9562d2 PS |
1337 | static struct netdev_rxq * |
1338 | netdev_dpdk_rxq_alloc(void) | |
1339 | { | |
1340 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1341 | ||
eff23640 DDP |
1342 | if (rx) { |
1343 | return &rx->up; | |
1344 | } | |
1345 | ||
1346 | return NULL; | |
8a9562d2 PS |
1347 | } |
1348 | ||
1349 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1350 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1351 | { |
d46285a2 | 1352 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1353 | } |
1354 | ||
1355 | static int | |
d46285a2 | 1356 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1357 | { |
d46285a2 DDP |
1358 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1359 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1360 | |
d46285a2 DDP |
1361 | ovs_mutex_lock(&dev->mutex); |
1362 | rx->port_id = dev->port_id; | |
1363 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1364 | |
1365 | return 0; | |
1366 | } | |
1367 | ||
1368 | static void | |
d46285a2 | 1369 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1370 | { |
1371 | } | |
1372 | ||
1373 | static void | |
d46285a2 | 1374 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1375 | { |
d46285a2 | 1376 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1377 | |
1378 | rte_free(rx); | |
1379 | } | |
1380 | ||
819f13bd DDP |
1381 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1382 | * 'pkts', even in case of failure. | |
1383 | * | |
1384 | * Returns the number of packets that weren't transmitted. */ | |
1385 | static inline int | |
b59cc14e | 1386 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 1387 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 1388 | { |
1304f1f8 DDP |
1389 | uint32_t nb_tx = 0; |
1390 | ||
b59cc14e | 1391 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1392 | uint32_t ret; |
1393 | ||
b59cc14e | 1394 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1395 | if (!ret) { |
1396 | break; | |
1397 | } | |
1398 | ||
1399 | nb_tx += ret; | |
1400 | } | |
8a9562d2 | 1401 | |
b59cc14e | 1402 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 1403 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
1404 | * packet could come from a different mempool) */ |
1405 | int i; | |
1406 | ||
b59cc14e IM |
1407 | for (i = nb_tx; i < cnt; i++) { |
1408 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1409 | } |
8a9562d2 | 1410 | } |
819f13bd DDP |
1411 | |
1412 | return cnt - nb_tx; | |
8a9562d2 PS |
1413 | } |
1414 | ||
f3926f29 IS |
1415 | static inline bool |
1416 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1417 | struct rte_mbuf *pkt, uint64_t time) | |
1418 | { | |
1419 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1420 | ||
1421 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1422 | e_RTE_METER_GREEN; | |
1423 | } | |
1424 | ||
1425 | static int | |
1426 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
1427 | struct rte_mbuf **pkts, int pkt_cnt) | |
1428 | { | |
1429 | int i = 0; | |
1430 | int cnt = 0; | |
1431 | struct rte_mbuf *pkt = NULL; | |
1432 | uint64_t current_time = rte_rdtsc(); | |
1433 | ||
1434 | for (i = 0; i < pkt_cnt; i++) { | |
1435 | pkt = pkts[i]; | |
1436 | /* Handle current packet */ | |
1437 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1438 | if (cnt != i) { | |
1439 | pkts[cnt] = pkt; | |
1440 | } | |
1441 | cnt++; | |
1442 | } else { | |
1443 | rte_pktmbuf_free(pkt); | |
1444 | } | |
1445 | } | |
1446 | ||
1447 | return cnt; | |
1448 | } | |
1449 | ||
9509913a IS |
1450 | static int |
1451 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
1452 | int pkt_cnt) | |
1453 | { | |
1454 | int cnt = 0; | |
1455 | ||
1456 | rte_spinlock_lock(&policer->policer_lock); | |
1457 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, pkt_cnt); | |
1458 | rte_spinlock_unlock(&policer->policer_lock); | |
1459 | ||
1460 | return cnt; | |
1461 | } | |
1462 | ||
58397e6c | 1463 | static bool |
0a0f39df | 1464 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1465 | { |
0a0f39df | 1466 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1467 | } |
1468 | ||
d6e3feb5 | 1469 | static inline void |
1470 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1471 | unsigned int packet_size) | |
1472 | { | |
1473 | /* Hard-coded search for the size bucket. */ | |
1474 | if (packet_size < 256) { | |
1475 | if (packet_size >= 128) { | |
1476 | stats->rx_128_to_255_packets++; | |
1477 | } else if (packet_size <= 64) { | |
1478 | stats->rx_1_to_64_packets++; | |
1479 | } else { | |
1480 | stats->rx_65_to_127_packets++; | |
1481 | } | |
1482 | } else { | |
1483 | if (packet_size >= 1523) { | |
1484 | stats->rx_1523_to_max_packets++; | |
1485 | } else if (packet_size >= 1024) { | |
1486 | stats->rx_1024_to_1522_packets++; | |
1487 | } else if (packet_size < 512) { | |
1488 | stats->rx_256_to_511_packets++; | |
1489 | } else { | |
1490 | stats->rx_512_to_1023_packets++; | |
1491 | } | |
1492 | } | |
1493 | } | |
1494 | ||
9e3ddd45 TP |
1495 | static inline void |
1496 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1497 | struct dp_packet **packets, int count, |
1498 | int dropped) | |
9e3ddd45 TP |
1499 | { |
1500 | int i; | |
d6e3feb5 | 1501 | unsigned int packet_size; |
9e3ddd45 TP |
1502 | struct dp_packet *packet; |
1503 | ||
1504 | stats->rx_packets += count; | |
9509913a | 1505 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1506 | for (i = 0; i < count; i++) { |
1507 | packet = packets[i]; | |
d6e3feb5 | 1508 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1509 | |
d6e3feb5 | 1510 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1511 | /* This only protects the following multicast counting from |
1512 | * too short packets, but it does not stop the packet from | |
1513 | * further processing. */ | |
1514 | stats->rx_errors++; | |
1515 | stats->rx_length_errors++; | |
1516 | continue; | |
1517 | } | |
1518 | ||
d6e3feb5 | 1519 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1520 | ||
9e3ddd45 TP |
1521 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1522 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1523 | stats->multicast++; | |
1524 | } | |
1525 | ||
d6e3feb5 | 1526 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1527 | } |
1528 | } | |
1529 | ||
58397e6c KT |
1530 | /* |
1531 | * The receive path for the vhost port is the TX path out from guest. | |
1532 | */ | |
1533 | static int | |
d46285a2 | 1534 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1535 | struct dp_packet_batch *batch) |
58397e6c | 1536 | { |
d46285a2 | 1537 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
d46285a2 | 1538 | int qid = rxq->queue_id; |
9509913a | 1539 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1540 | uint16_t nb_rx = 0; |
9509913a | 1541 | uint16_t dropped = 0; |
58397e6c | 1542 | |
0a0f39df | 1543 | if (OVS_UNLIKELY(!is_vhost_running(dev) |
e543851d | 1544 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1545 | return EAGAIN; |
1546 | } | |
1547 | ||
0a0f39df CL |
1548 | nb_rx = rte_vhost_dequeue_burst(netdev_dpdk_get_vid(dev), |
1549 | qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
d46285a2 | 1550 | dev->dpdk_mp->mp, |
64839cf4 | 1551 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1552 | NETDEV_MAX_BURST); |
58397e6c KT |
1553 | if (!nb_rx) { |
1554 | return EAGAIN; | |
1555 | } | |
1556 | ||
9509913a IS |
1557 | if (policer) { |
1558 | dropped = nb_rx; | |
64839cf4 WT |
1559 | nb_rx = ingress_policer_run(policer, |
1560 | (struct rte_mbuf **) batch->packets, | |
1561 | nb_rx); | |
9509913a IS |
1562 | dropped -= nb_rx; |
1563 | } | |
1564 | ||
d46285a2 | 1565 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1566 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1567 | nb_rx, dropped); | |
d46285a2 | 1568 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1569 | |
64839cf4 | 1570 | batch->count = (int) nb_rx; |
58397e6c KT |
1571 | return 0; |
1572 | } | |
1573 | ||
8a9562d2 | 1574 | static int |
64839cf4 | 1575 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1576 | { |
d46285a2 DDP |
1577 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1578 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1579 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1580 | int nb_rx; |
9509913a | 1581 | int dropped = 0; |
8a9562d2 | 1582 | |
3b1fb077 DDP |
1583 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1584 | return EAGAIN; | |
1585 | } | |
1586 | ||
d46285a2 | 1587 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1588 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1589 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1590 | if (!nb_rx) { |
1591 | return EAGAIN; | |
1592 | } | |
1593 | ||
9509913a IS |
1594 | if (policer) { |
1595 | dropped = nb_rx; | |
64839cf4 | 1596 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 1597 | (struct rte_mbuf **) batch->packets, |
64839cf4 | 1598 | nb_rx); |
9509913a IS |
1599 | dropped -= nb_rx; |
1600 | } | |
1601 | ||
1602 | /* Update stats to reflect dropped packets */ | |
1603 | if (OVS_UNLIKELY(dropped)) { | |
1604 | rte_spinlock_lock(&dev->stats_lock); | |
1605 | dev->stats.rx_dropped += dropped; | |
1606 | rte_spinlock_unlock(&dev->stats_lock); | |
1607 | } | |
1608 | ||
64839cf4 | 1609 | batch->count = nb_rx; |
8a9562d2 PS |
1610 | |
1611 | return 0; | |
1612 | } | |
1613 | ||
0bf765f7 | 1614 | static inline int |
78bd47cf DDP |
1615 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
1616 | int cnt) | |
0bf765f7 | 1617 | { |
78bd47cf | 1618 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 1619 | |
78bd47cf DDP |
1620 | if (qos_conf) { |
1621 | rte_spinlock_lock(&qos_conf->lock); | |
1622 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt); | |
1623 | rte_spinlock_unlock(&qos_conf->lock); | |
0bf765f7 IS |
1624 | } |
1625 | ||
1626 | return cnt; | |
1627 | } | |
1628 | ||
c6ec9d17 IM |
1629 | static int |
1630 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1631 | int pkt_cnt) | |
1632 | { | |
1633 | int i = 0; | |
1634 | int cnt = 0; | |
1635 | struct rte_mbuf *pkt; | |
1636 | ||
1637 | for (i = 0; i < pkt_cnt; i++) { | |
1638 | pkt = pkts[i]; | |
1639 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
1640 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
1641 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
1642 | rte_pktmbuf_free(pkt); | |
1643 | continue; | |
1644 | } | |
1645 | ||
1646 | if (OVS_UNLIKELY(i != cnt)) { | |
1647 | pkts[cnt] = pkt; | |
1648 | } | |
1649 | cnt++; | |
1650 | } | |
1651 | ||
1652 | return cnt; | |
1653 | } | |
1654 | ||
9e3ddd45 TP |
1655 | static inline void |
1656 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1657 | struct dp_packet **packets, | |
1658 | int attempted, | |
1659 | int dropped) | |
1660 | { | |
1661 | int i; | |
1662 | int sent = attempted - dropped; | |
1663 | ||
1664 | stats->tx_packets += sent; | |
1665 | stats->tx_dropped += dropped; | |
1666 | ||
1667 | for (i = 0; i < sent; i++) { | |
1668 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1669 | } | |
1670 | } | |
1671 | ||
58397e6c | 1672 | static void |
4573fbd3 | 1673 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 1674 | struct dp_packet **pkts, int cnt) |
58397e6c | 1675 | { |
d46285a2 | 1676 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1677 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1678 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 1679 | unsigned int dropped = 0; |
dd52de45 | 1680 | int i, retries = 0; |
58397e6c | 1681 | |
81acebda | 1682 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1683 | |
0a0f39df | 1684 | if (OVS_UNLIKELY(!is_vhost_running(dev) || qid < 0 |
e543851d | 1685 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1686 | rte_spinlock_lock(&dev->stats_lock); |
1687 | dev->stats.tx_dropped+= cnt; | |
1688 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1689 | goto out; |
58397e6c KT |
1690 | } |
1691 | ||
d46285a2 | 1692 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1693 | |
c6ec9d17 | 1694 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 1695 | /* Check has QoS has been configured for the netdev */ |
78bd47cf | 1696 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt); |
c6ec9d17 | 1697 | dropped = total_pkts - cnt; |
0bf765f7 | 1698 | |
95e9881f | 1699 | do { |
4573fbd3 | 1700 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1701 | unsigned int tx_pkts; |
1702 | ||
0a0f39df | 1703 | tx_pkts = rte_vhost_enqueue_burst(netdev_dpdk_get_vid(dev), |
c6ec9d17 | 1704 | vhost_qid, cur_pkts, cnt); |
95e9881f KT |
1705 | if (OVS_LIKELY(tx_pkts)) { |
1706 | /* Packets have been sent.*/ | |
1707 | cnt -= tx_pkts; | |
31871ee3 | 1708 | /* Prepare for possible retry.*/ |
95e9881f KT |
1709 | cur_pkts = &cur_pkts[tx_pkts]; |
1710 | } else { | |
31871ee3 KT |
1711 | /* No packets sent - do not retry.*/ |
1712 | break; | |
95e9881f | 1713 | } |
c6ec9d17 | 1714 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1715 | |
d46285a2 | 1716 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1717 | |
d46285a2 | 1718 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 1719 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 1720 | cnt + dropped); |
d46285a2 | 1721 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1722 | |
1723 | out: | |
c6ec9d17 | 1724 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 1725 | dp_packet_delete(pkts[i]); |
58397e6c KT |
1726 | } |
1727 | } | |
1728 | ||
8a9562d2 PS |
1729 | /* Tx function. Transmit packets indefinitely */ |
1730 | static void | |
64839cf4 | 1731 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1732 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1733 | { |
bce01e3a | 1734 | #if !defined(__CHECKER__) && !defined(_WIN32) |
64839cf4 | 1735 | const size_t PKT_ARRAY_SIZE = batch->count; |
bce01e3a EJ |
1736 | #else |
1737 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1738 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1739 | #endif |
8a9562d2 | 1740 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 1741 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
175cf4de RW |
1742 | int dropped = 0; |
1743 | int newcnt = 0; | |
1744 | int i; | |
8a9562d2 | 1745 | |
7d6d1a40 WT |
1746 | dp_packet_batch_apply_cutlen(batch); |
1747 | ||
64839cf4 WT |
1748 | for (i = 0; i < batch->count; i++) { |
1749 | int size = dp_packet_size(batch->packets[i]); | |
95fb793a | 1750 | |
f98d7864 | 1751 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c | 1752 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
58be5c0e | 1753 | (int) size, dev->max_packet_len); |
f4fd623c | 1754 | |
175cf4de | 1755 | dropped++; |
f4fd623c DDP |
1756 | continue; |
1757 | } | |
8a9562d2 | 1758 | |
2391135c | 1759 | pkts[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1760 | |
2391135c | 1761 | if (!pkts[newcnt]) { |
64839cf4 | 1762 | dropped += batch->count - i; |
175cf4de | 1763 | break; |
f4fd623c DDP |
1764 | } |
1765 | ||
1766 | /* We have to do a copy for now */ | |
2391135c | 1767 | memcpy(rte_pktmbuf_mtod(pkts[newcnt], void *), |
64839cf4 | 1768 | dp_packet_data(batch->packets[i]), size); |
f4fd623c | 1769 | |
2391135c MK |
1770 | rte_pktmbuf_data_len(pkts[newcnt]) = size; |
1771 | rte_pktmbuf_pkt_len(pkts[newcnt]) = size; | |
f4fd623c DDP |
1772 | |
1773 | newcnt++; | |
1774 | } | |
8a9562d2 | 1775 | |
58397e6c | 1776 | if (dev->type == DPDK_DEV_VHOST) { |
2391135c | 1777 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, |
dd52de45 | 1778 | newcnt); |
58397e6c | 1779 | } else { |
0bf765f7 IS |
1780 | unsigned int qos_pkts = newcnt; |
1781 | ||
1782 | /* Check if QoS has been configured for this netdev. */ | |
78bd47cf | 1783 | newcnt = netdev_dpdk_qos_run(dev, pkts, newcnt); |
0bf765f7 IS |
1784 | |
1785 | dropped += qos_pkts - newcnt; | |
819f13bd | 1786 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, newcnt); |
58397e6c | 1787 | } |
db73f716 | 1788 | |
0bf765f7 IS |
1789 | if (OVS_UNLIKELY(dropped)) { |
1790 | rte_spinlock_lock(&dev->stats_lock); | |
1791 | dev->stats.tx_dropped += dropped; | |
1792 | rte_spinlock_unlock(&dev->stats_lock); | |
1793 | } | |
8a9562d2 PS |
1794 | } |
1795 | ||
58397e6c | 1796 | static int |
64839cf4 WT |
1797 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1798 | struct dp_packet_batch *batch, | |
324c8374 | 1799 | bool may_steal, bool concurrent_txq OVS_UNUSED) |
58397e6c | 1800 | { |
58397e6c | 1801 | |
dd52de45 | 1802 | if (OVS_UNLIKELY(!may_steal || batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 WT |
1803 | dpdk_do_tx_copy(netdev, qid, batch); |
1804 | dp_packet_delete_batch(batch, may_steal); | |
58397e6c | 1805 | } else { |
64839cf4 | 1806 | dp_packet_batch_apply_cutlen(batch); |
dd52de45 | 1807 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
1808 | } |
1809 | return 0; | |
1810 | } | |
1811 | ||
7251515e DV |
1812 | static inline void |
1813 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
324c8374 IM |
1814 | struct dp_packet_batch *batch, bool may_steal, |
1815 | bool concurrent_txq) | |
8a9562d2 | 1816 | { |
3b1fb077 DDP |
1817 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1818 | dp_packet_delete_batch(batch, may_steal); | |
1819 | return; | |
1820 | } | |
1821 | ||
324c8374 | 1822 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 1823 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
1824 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
1825 | } | |
1826 | ||
7251515e | 1827 | if (OVS_UNLIKELY(!may_steal || |
64839cf4 | 1828 | batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1829 | struct netdev *netdev = &dev->up; |
1830 | ||
64839cf4 WT |
1831 | dpdk_do_tx_copy(netdev, qid, batch); |
1832 | dp_packet_delete_batch(batch, may_steal); | |
8a9562d2 | 1833 | } else { |
6b094bf4 | 1834 | int dropped; |
64839cf4 | 1835 | int cnt = batch->count; |
2391135c | 1836 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 1837 | |
7d6d1a40 WT |
1838 | dp_packet_batch_apply_cutlen(batch); |
1839 | ||
2391135c | 1840 | cnt = netdev_dpdk_filter_packet_len(dev, pkts, cnt); |
78bd47cf | 1841 | cnt = netdev_dpdk_qos_run(dev, pkts, cnt); |
6b094bf4 | 1842 | dropped = batch->count - cnt; |
1b99bb05 | 1843 | |
819f13bd | 1844 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, cnt); |
8a9562d2 | 1845 | |
f4fd623c | 1846 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1847 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1848 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1849 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1850 | } |
8a9562d2 | 1851 | } |
a0cb2d66 | 1852 | |
324c8374 | 1853 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
1854 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
1855 | } | |
7251515e DV |
1856 | } |
1857 | ||
1858 | static int | |
1859 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
324c8374 IM |
1860 | struct dp_packet_batch *batch, bool may_steal, |
1861 | bool concurrent_txq) | |
7251515e DV |
1862 | { |
1863 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1864 | |
324c8374 | 1865 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e | 1866 | return 0; |
8a9562d2 PS |
1867 | } |
1868 | ||
1869 | static int | |
74ff3298 | 1870 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1871 | { |
1872 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1873 | ||
1874 | ovs_mutex_lock(&dev->mutex); | |
1875 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1876 | dev->hwaddr = mac; |
045c0d1a | 1877 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1878 | } |
1879 | ovs_mutex_unlock(&dev->mutex); | |
1880 | ||
1881 | return 0; | |
1882 | } | |
1883 | ||
1884 | static int | |
74ff3298 | 1885 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1886 | { |
1887 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1888 | ||
1889 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1890 | *mac = dev->hwaddr; |
8a9562d2 PS |
1891 | ovs_mutex_unlock(&dev->mutex); |
1892 | ||
1893 | return 0; | |
1894 | } | |
1895 | ||
1896 | static int | |
1897 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1898 | { | |
1899 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1900 | ||
1901 | ovs_mutex_lock(&dev->mutex); | |
1902 | *mtup = dev->mtu; | |
1903 | ovs_mutex_unlock(&dev->mutex); | |
1904 | ||
1905 | return 0; | |
1906 | } | |
1907 | ||
0072e931 MK |
1908 | static int |
1909 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
1910 | { | |
1911 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1912 | ||
1913 | if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
1914 | || mtu < ETHER_MIN_MTU) { | |
1915 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
1916 | return EINVAL; | |
1917 | } | |
1918 | ||
1919 | ovs_mutex_lock(&dev->mutex); | |
1920 | if (dev->requested_mtu != mtu) { | |
1921 | dev->requested_mtu = mtu; | |
1922 | netdev_request_reconfigure(netdev); | |
1923 | } | |
1924 | ovs_mutex_unlock(&dev->mutex); | |
1925 | ||
1926 | return 0; | |
1927 | } | |
1928 | ||
8a9562d2 | 1929 | static int |
d46285a2 | 1930 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 1931 | |
58397e6c KT |
1932 | static int |
1933 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1934 | struct netdev_stats *stats) | |
1935 | { | |
1936 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1937 | ||
1938 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 1939 | |
45d947c4 | 1940 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
1941 | /* Supported Stats */ |
1942 | stats->rx_packets += dev->stats.rx_packets; | |
1943 | stats->tx_packets += dev->stats.tx_packets; | |
9509913a | 1944 | stats->rx_dropped = dev->stats.rx_dropped; |
58397e6c | 1945 | stats->tx_dropped += dev->stats.tx_dropped; |
9e3ddd45 TP |
1946 | stats->multicast = dev->stats.multicast; |
1947 | stats->rx_bytes = dev->stats.rx_bytes; | |
1948 | stats->tx_bytes = dev->stats.tx_bytes; | |
1949 | stats->rx_errors = dev->stats.rx_errors; | |
1950 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 1951 | |
1952 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
1953 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
1954 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
1955 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
1956 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
1957 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
1958 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
1959 | ||
45d947c4 | 1960 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 1961 | |
58397e6c KT |
1962 | ovs_mutex_unlock(&dev->mutex); |
1963 | ||
1964 | return 0; | |
1965 | } | |
1966 | ||
d6e3feb5 | 1967 | static void |
1968 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
1969 | const struct rte_eth_xstat *xstats, |
1970 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 1971 | const unsigned int size) |
1972 | { | |
d6e3feb5 | 1973 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 1974 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1975 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1976 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1977 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1978 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1979 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1980 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1981 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1982 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1983 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1984 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1985 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 1986 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1987 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 1988 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1989 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1990 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1991 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1992 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1993 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1994 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1995 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1996 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1997 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1998 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1999 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2000 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2001 | stats->tx_1523_to_max_packets = xstats[i].value; |
d57f777f PS |
2002 | } else if (strcmp(XSTAT_RX_MULTICAST_PACKETS, names[i].name) == 0) { |
2003 | stats->multicast = xstats[i].value; | |
0a0f39df | 2004 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2005 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 2006 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2007 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2008 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2009 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2010 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2011 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2012 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2013 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2014 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2015 | stats->rx_jabber_errors = xstats[i].value; |
2016 | } | |
2017 | } | |
2018 | } | |
2019 | ||
8a9562d2 PS |
2020 | static int |
2021 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2022 | { | |
2023 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2024 | struct rte_eth_stats rte_stats; | |
2025 | bool gg; | |
2026 | ||
2027 | netdev_dpdk_get_carrier(netdev, &gg); | |
2028 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2029 | |
0a0f39df CL |
2030 | struct rte_eth_xstat *rte_xstats = NULL; |
2031 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2032 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2033 | |
2034 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
2035 | VLOG_ERR("Can't get ETH statistics for port: %i.", dev->port_id); | |
f9256822 | 2036 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2037 | return EPROTO; |
2038 | } | |
2039 | ||
0a0f39df CL |
2040 | /* Get length of statistics */ |
2041 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2042 | if (rte_xstats_len < 0) { | |
2043 | VLOG_WARN("Cannot get XSTATS values for port: %i", dev->port_id); | |
2044 | goto out; | |
2045 | } | |
2046 | /* Reserve memory for xstats names and values */ | |
2047 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2048 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2049 | ||
2050 | /* Retreive xstats names */ | |
2051 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2052 | rte_xstats_names, | |
2053 | rte_xstats_len); | |
2054 | if (rte_xstats_new_len != rte_xstats_len) { | |
2055 | VLOG_WARN("Cannot get XSTATS names for port: %i.", dev->port_id); | |
2056 | goto out; | |
2057 | } | |
2058 | /* Retreive xstats values */ | |
2059 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2060 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2061 | rte_xstats_len); | |
2062 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2063 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2064 | rte_xstats_len); | |
d6e3feb5 | 2065 | } else { |
0a0f39df | 2066 | VLOG_WARN("Cannot get XSTATS values for port: %i.", dev->port_id); |
d6e3feb5 | 2067 | } |
8a9562d2 | 2068 | |
0a0f39df CL |
2069 | out: |
2070 | free(rte_xstats); | |
2071 | free(rte_xstats_names); | |
2072 | ||
2f9dd77f PS |
2073 | stats->rx_packets = rte_stats.ipackets; |
2074 | stats->tx_packets = rte_stats.opackets; | |
2075 | stats->rx_bytes = rte_stats.ibytes; | |
2076 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2077 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2078 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2079 | |
45d947c4 | 2080 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2081 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2082 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2083 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2084 | |
2085 | /* These are the available DPDK counters for packets not received due to | |
2086 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2087 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2088 | stats->rx_missed_errors = rte_stats.imissed; |
2089 | ||
8a9562d2 PS |
2090 | ovs_mutex_unlock(&dev->mutex); |
2091 | ||
2092 | return 0; | |
2093 | } | |
2094 | ||
2095 | static int | |
d46285a2 | 2096 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2097 | enum netdev_features *current, |
ca3d4f55 BX |
2098 | enum netdev_features *advertised, |
2099 | enum netdev_features *supported, | |
2100 | enum netdev_features *peer) | |
8a9562d2 | 2101 | { |
d46285a2 | 2102 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2103 | struct rte_eth_link link; |
2104 | ||
2105 | ovs_mutex_lock(&dev->mutex); | |
2106 | link = dev->link; | |
2107 | ovs_mutex_unlock(&dev->mutex); | |
2108 | ||
362ca396 | 2109 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2110 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
2111 | *current = NETDEV_F_10MB_HD; |
2112 | } | |
362ca396 | 2113 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2114 | *current = NETDEV_F_100MB_HD; |
2115 | } | |
362ca396 | 2116 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2117 | *current = NETDEV_F_1GB_HD; |
2118 | } | |
2119 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 2120 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
2121 | *current = NETDEV_F_10MB_FD; |
2122 | } | |
362ca396 | 2123 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2124 | *current = NETDEV_F_100MB_FD; |
2125 | } | |
362ca396 | 2126 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2127 | *current = NETDEV_F_1GB_FD; |
2128 | } | |
362ca396 | 2129 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
2130 | *current = NETDEV_F_10GB_FD; |
2131 | } | |
2132 | } | |
2133 | ||
362ca396 | 2134 | if (link.link_autoneg) { |
2135 | *current |= NETDEV_F_AUTONEG; | |
2136 | } | |
2137 | ||
ca3d4f55 BX |
2138 | *advertised = *supported = *peer = 0; |
2139 | ||
8a9562d2 PS |
2140 | return 0; |
2141 | } | |
2142 | ||
9509913a IS |
2143 | static struct ingress_policer * |
2144 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2145 | { | |
2146 | struct ingress_policer *policer = NULL; | |
2147 | uint64_t rate_bytes; | |
2148 | uint64_t burst_bytes; | |
2149 | int err = 0; | |
2150 | ||
2151 | policer = xmalloc(sizeof *policer); | |
2152 | rte_spinlock_init(&policer->policer_lock); | |
2153 | ||
2154 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
2155 | rate_bytes = rate * 1000/8; | |
2156 | burst_bytes = burst * 1000/8; | |
2157 | ||
2158 | policer->app_srtcm_params.cir = rate_bytes; | |
2159 | policer->app_srtcm_params.cbs = burst_bytes; | |
2160 | policer->app_srtcm_params.ebs = 0; | |
2161 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2162 | &policer->app_srtcm_params); | |
58be5c0e | 2163 | if (err) { |
9509913a IS |
2164 | VLOG_ERR("Could not create rte meter for ingress policer"); |
2165 | return NULL; | |
2166 | } | |
2167 | ||
2168 | return policer; | |
2169 | } | |
2170 | ||
2171 | static int | |
2172 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2173 | uint32_t policer_burst) | |
2174 | { | |
2175 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2176 | struct ingress_policer *policer; | |
2177 | ||
2178 | /* Force to 0 if no rate specified, | |
2179 | * default to 8000 kbits if burst is 0, | |
2180 | * else stick with user-specified value. | |
2181 | */ | |
2182 | policer_burst = (!policer_rate ? 0 | |
2183 | : !policer_burst ? 8000 | |
2184 | : policer_burst); | |
2185 | ||
2186 | ovs_mutex_lock(&dev->mutex); | |
2187 | ||
2188 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2189 | &dev->ingress_policer); | |
2190 | ||
2191 | if (dev->policer_rate == policer_rate && | |
2192 | dev->policer_burst == policer_burst) { | |
2193 | /* Assume that settings haven't changed since we last set them. */ | |
2194 | ovs_mutex_unlock(&dev->mutex); | |
2195 | return 0; | |
2196 | } | |
2197 | ||
2198 | /* Destroy any existing ingress policer for the device if one exists */ | |
2199 | if (policer) { | |
2200 | ovsrcu_postpone(free, policer); | |
2201 | } | |
2202 | ||
2203 | if (policer_rate != 0) { | |
2204 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2205 | } else { | |
2206 | policer = NULL; | |
2207 | } | |
2208 | ovsrcu_set(&dev->ingress_policer, policer); | |
2209 | dev->policer_rate = policer_rate; | |
2210 | dev->policer_burst = policer_burst; | |
2211 | ovs_mutex_unlock(&dev->mutex); | |
2212 | ||
2213 | return 0; | |
2214 | } | |
2215 | ||
8a9562d2 PS |
2216 | static int |
2217 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2218 | { | |
2219 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
2220 | |
2221 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
2222 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
2223 | * postive integer to meet RFC 2863 recommendations. | |
2224 | */ | |
2225 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
2226 | ovs_mutex_unlock(&dev->mutex); |
2227 | ||
2228 | return ifindex; | |
2229 | } | |
2230 | ||
2231 | static int | |
d46285a2 | 2232 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2233 | { |
d46285a2 | 2234 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2235 | |
2236 | ovs_mutex_lock(&dev->mutex); | |
2237 | check_link_status(dev); | |
2238 | *carrier = dev->link.link_status; | |
58397e6c KT |
2239 | |
2240 | ovs_mutex_unlock(&dev->mutex); | |
2241 | ||
2242 | return 0; | |
2243 | } | |
2244 | ||
2245 | static int | |
d46285a2 | 2246 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2247 | { |
d46285a2 | 2248 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2249 | |
2250 | ovs_mutex_lock(&dev->mutex); | |
2251 | ||
0a0f39df | 2252 | if (is_vhost_running(dev)) { |
58397e6c KT |
2253 | *carrier = 1; |
2254 | } else { | |
2255 | *carrier = 0; | |
2256 | } | |
2257 | ||
8a9562d2 PS |
2258 | ovs_mutex_unlock(&dev->mutex); |
2259 | ||
2260 | return 0; | |
2261 | } | |
2262 | ||
2263 | static long long int | |
d46285a2 | 2264 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2265 | { |
d46285a2 | 2266 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2267 | long long int carrier_resets; |
2268 | ||
2269 | ovs_mutex_lock(&dev->mutex); | |
2270 | carrier_resets = dev->link_reset_cnt; | |
2271 | ovs_mutex_unlock(&dev->mutex); | |
2272 | ||
2273 | return carrier_resets; | |
2274 | } | |
2275 | ||
2276 | static int | |
d46285a2 | 2277 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2278 | long long int interval OVS_UNUSED) |
2279 | { | |
ee32150e | 2280 | return EOPNOTSUPP; |
8a9562d2 PS |
2281 | } |
2282 | ||
2283 | static int | |
2284 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2285 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2286 | enum netdev_flags *old_flagsp) |
2287 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 2288 | { |
8a9562d2 PS |
2289 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
2290 | return EINVAL; | |
2291 | } | |
2292 | ||
2293 | *old_flagsp = dev->flags; | |
2294 | dev->flags |= on; | |
2295 | dev->flags &= ~off; | |
2296 | ||
2297 | if (dev->flags == *old_flagsp) { | |
2298 | return 0; | |
2299 | } | |
2300 | ||
58397e6c | 2301 | if (dev->type == DPDK_DEV_ETH) { |
58397e6c KT |
2302 | if (dev->flags & NETDEV_PROMISC) { |
2303 | rte_eth_promiscuous_enable(dev->port_id); | |
2304 | } | |
8a9562d2 | 2305 | |
314fb5ad | 2306 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
2307 | } else { |
2308 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2309 | * running then change netdev's change_seq to trigger link state | |
2310 | * update. */ | |
e543851d ZB |
2311 | |
2312 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2313 | && is_vhost_running(dev)) { |
e543851d ZB |
2314 | netdev_change_seq_changed(&dev->up); |
2315 | ||
2316 | /* Clear statistics if device is getting up. */ | |
2317 | if (NETDEV_UP & on) { | |
2318 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 2319 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
2320 | rte_spinlock_unlock(&dev->stats_lock); |
2321 | } | |
2322 | } | |
8a9562d2 PS |
2323 | } |
2324 | ||
2325 | return 0; | |
2326 | } | |
2327 | ||
2328 | static int | |
d46285a2 | 2329 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2330 | enum netdev_flags off, enum netdev_flags on, |
2331 | enum netdev_flags *old_flagsp) | |
2332 | { | |
d46285a2 | 2333 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2334 | int error; |
2335 | ||
d46285a2 DDP |
2336 | ovs_mutex_lock(&dev->mutex); |
2337 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2338 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2339 | |
2340 | return error; | |
2341 | } | |
2342 | ||
2343 | static int | |
d46285a2 | 2344 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2345 | { |
d46285a2 | 2346 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2347 | struct rte_eth_dev_info dev_info; |
2348 | ||
7cd1261d | 2349 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2350 | return ENODEV; |
7cd1261d | 2351 | } |
8a9562d2 PS |
2352 | |
2353 | ovs_mutex_lock(&dev->mutex); | |
2354 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2355 | ovs_mutex_unlock(&dev->mutex); | |
2356 | ||
95fb793a | 2357 | smap_add_format(args, "port_no", "%d", dev->port_id); |
58be5c0e MK |
2358 | smap_add_format(args, "numa_id", "%d", |
2359 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
2360 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
2361 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2362 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2363 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2364 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2365 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
2366 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
2367 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
2368 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
2369 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2370 | ||
39c2baa9 | 2371 | if (dev_info.pci_dev) { |
2372 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2373 | dev_info.pci_dev->id.vendor_id); | |
2374 | smap_add_format(args, "pci-device_id", "0x%x", | |
2375 | dev_info.pci_dev->id.device_id); | |
2376 | } | |
8a9562d2 PS |
2377 | |
2378 | return 0; | |
2379 | } | |
2380 | ||
2381 | static void | |
2382 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2383 | OVS_REQUIRES(dev->mutex) | |
2384 | { | |
2385 | enum netdev_flags old_flags; | |
2386 | ||
2387 | if (admin_state) { | |
2388 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2389 | } else { | |
2390 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2391 | } | |
2392 | } | |
2393 | ||
2394 | static void | |
2395 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2396 | const char *argv[], void *aux OVS_UNUSED) | |
2397 | { | |
2398 | bool up; | |
2399 | ||
2400 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2401 | up = true; | |
2402 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2403 | up = false; | |
2404 | } else { | |
2405 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2406 | return; | |
2407 | } | |
2408 | ||
2409 | if (argc > 2) { | |
2410 | struct netdev *netdev = netdev_from_name(argv[1]); | |
2411 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
2412 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
2413 | ||
2414 | ovs_mutex_lock(&dpdk_dev->mutex); | |
2415 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
2416 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
2417 | ||
2418 | netdev_close(netdev); | |
2419 | } else { | |
2420 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2421 | netdev_close(netdev); | |
2422 | return; | |
2423 | } | |
2424 | } else { | |
2425 | struct netdev_dpdk *netdev; | |
2426 | ||
2427 | ovs_mutex_lock(&dpdk_mutex); | |
2428 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
2429 | ovs_mutex_lock(&netdev->mutex); | |
2430 | netdev_dpdk_set_admin_state__(netdev, up); | |
2431 | ovs_mutex_unlock(&netdev->mutex); | |
2432 | } | |
2433 | ovs_mutex_unlock(&dpdk_mutex); | |
2434 | } | |
2435 | unixctl_command_reply(conn, "OK"); | |
2436 | } | |
2437 | ||
b8374d0d MV |
2438 | static void |
2439 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2440 | const char *argv[], void *aux OVS_UNUSED) | |
2441 | { | |
2442 | int ret; | |
2443 | char *response; | |
b8374d0d MV |
2444 | uint8_t port_id; |
2445 | char devname[RTE_ETH_NAME_MAX_LEN]; | |
55e075e6 | 2446 | struct netdev_dpdk *dev; |
b8374d0d MV |
2447 | |
2448 | ovs_mutex_lock(&dpdk_mutex); | |
2449 | ||
55e075e6 CL |
2450 | if (!rte_eth_dev_count() || rte_eth_dev_get_port_by_name(argv[1], |
2451 | &port_id)) { | |
2452 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); | |
2453 | goto error; | |
2454 | } | |
b8374d0d | 2455 | |
55e075e6 CL |
2456 | dev = netdev_dpdk_lookup_by_port_id(port_id); |
2457 | if (dev) { | |
2458 | response = xasprintf("Device '%s' is being used by interface '%s'. " | |
2459 | "Remove it before detaching", | |
2460 | argv[1], netdev_get_name(&dev->up)); | |
b8374d0d MV |
2461 | goto error; |
2462 | } | |
2463 | ||
2464 | rte_eth_dev_close(port_id); | |
2465 | ||
2466 | ret = rte_eth_dev_detach(port_id, devname); | |
2467 | if (ret < 0) { | |
55e075e6 | 2468 | response = xasprintf("Device '%s' can not be detached", argv[1]); |
b8374d0d MV |
2469 | goto error; |
2470 | } | |
2471 | ||
55e075e6 | 2472 | response = xasprintf("Device '%s' has been detached", argv[1]); |
b8374d0d MV |
2473 | |
2474 | ovs_mutex_unlock(&dpdk_mutex); | |
2475 | unixctl_command_reply(conn, response); | |
2476 | free(response); | |
2477 | return; | |
2478 | ||
2479 | error: | |
2480 | ovs_mutex_unlock(&dpdk_mutex); | |
2481 | unixctl_command_reply_error(conn, response); | |
2482 | free(response); | |
2483 | } | |
2484 | ||
58397e6c KT |
2485 | /* |
2486 | * Set virtqueue flags so that we do not receive interrupts. | |
2487 | */ | |
2488 | static void | |
0a0f39df | 2489 | set_irq_status(int vid) |
58397e6c | 2490 | { |
4573fbd3 FL |
2491 | uint32_t i; |
2492 | uint64_t idx; | |
2493 | ||
0a0f39df | 2494 | for (i = 0; i < rte_vhost_get_queue_num(vid); i++) { |
4573fbd3 | 2495 | idx = i * VIRTIO_QNUM; |
0a0f39df CL |
2496 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_RXQ, 0); |
2497 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_TXQ, 0); | |
4573fbd3 FL |
2498 | } |
2499 | } | |
2500 | ||
585a5bea IM |
2501 | /* |
2502 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2503 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2504 | */ |
2505 | static void | |
d46285a2 DDP |
2506 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2507 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2508 | { |
2509 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2510 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 2511 | |
eff23640 | 2512 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
2513 | |
2514 | for (i = 0; i < total_txqs; i++) { | |
2515 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2516 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2517 | enabled_queues[n_enabled++] = i; |
2518 | } | |
2519 | } | |
2520 | ||
2521 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2522 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2523 | n_enabled = 1; |
2524 | } | |
2525 | ||
2526 | k = 0; | |
2527 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2528 | if (dev->tx_q[i].map != i) { |
2529 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2530 | k = (k + 1) % n_enabled; |
2531 | } | |
2532 | } | |
2533 | ||
2d24d165 | 2534 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2535 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2536 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2537 | } |
2538 | ||
eff23640 | 2539 | free(enabled_queues); |
585a5bea | 2540 | } |
4573fbd3 | 2541 | |
58397e6c KT |
2542 | /* |
2543 | * A new virtio-net device is added to a vhost port. | |
2544 | */ | |
2545 | static int | |
0a0f39df | 2546 | new_device(int vid) |
58397e6c | 2547 | { |
d46285a2 | 2548 | struct netdev_dpdk *dev; |
58397e6c | 2549 | bool exists = false; |
db8f13b0 | 2550 | int newnode = 0; |
0a0f39df CL |
2551 | char ifname[IF_NAME_SZ]; |
2552 | ||
58be5c0e | 2553 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2554 | |
2555 | ovs_mutex_lock(&dpdk_mutex); | |
2556 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2557 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 2558 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2559 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
0a0f39df | 2560 | uint32_t qp_num = rte_vhost_get_queue_num(vid); |
db8f13b0 CL |
2561 | |
2562 | /* Get NUMA information */ | |
0a0f39df CL |
2563 | newnode = rte_vhost_get_numa_node(vid); |
2564 | if (newnode == -1) { | |
5b9bf9e0 | 2565 | #ifdef VHOST_NUMA |
db8f13b0 | 2566 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2567 | ifname); |
5b9bf9e0 | 2568 | #endif |
db8f13b0 | 2569 | newnode = dev->socket_id; |
db8f13b0 CL |
2570 | } |
2571 | ||
7f5f2bd0 IM |
2572 | if (dev->requested_n_txq != qp_num |
2573 | || dev->requested_n_rxq != qp_num | |
2574 | || dev->requested_socket_id != newnode) { | |
2575 | dev->requested_socket_id = newnode; | |
2576 | dev->requested_n_rxq = qp_num; | |
2577 | dev->requested_n_txq = qp_num; | |
2578 | netdev_request_reconfigure(&dev->up); | |
2579 | } else { | |
2580 | /* Reconfiguration not required. */ | |
2581 | dev->vhost_reconfigured = true; | |
2582 | } | |
81acebda | 2583 | |
0a0f39df | 2584 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2585 | exists = true; |
2586 | ||
58397e6c | 2587 | /* Disable notifications. */ |
0a0f39df | 2588 | set_irq_status(vid); |
e543851d | 2589 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2590 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2591 | break; |
2592 | } | |
c1ff66ac | 2593 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2594 | } |
2595 | ovs_mutex_unlock(&dpdk_mutex); | |
2596 | ||
2597 | if (!exists) { | |
0a0f39df | 2598 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2599 | |
2600 | return -1; | |
2601 | } | |
2602 | ||
0a0f39df CL |
2603 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2604 | ifname, newnode); | |
2605 | ||
58397e6c KT |
2606 | return 0; |
2607 | } | |
2608 | ||
f3ea2ad2 IM |
2609 | /* Clears mapping for all available queues of vhost interface. */ |
2610 | static void | |
2611 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2612 | OVS_REQUIRES(dev->mutex) | |
2613 | { | |
2614 | int i; | |
2615 | ||
81acebda | 2616 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
2617 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
2618 | } | |
2619 | } | |
2620 | ||
58397e6c KT |
2621 | /* |
2622 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2623 | * flag to stop any more packets from being sent or received to/from a VM and | |
2624 | * ensure all currently queued packets have been sent/received before removing | |
2625 | * the device. | |
2626 | */ | |
2627 | static void | |
0a0f39df | 2628 | destroy_device(int vid) |
58397e6c | 2629 | { |
d46285a2 | 2630 | struct netdev_dpdk *dev; |
afee281f | 2631 | bool exists = false; |
0a0f39df CL |
2632 | char ifname[IF_NAME_SZ]; |
2633 | ||
58be5c0e | 2634 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2635 | |
2636 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2637 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2638 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 2639 | |
d46285a2 | 2640 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
2641 | dev->vhost_reconfigured = false; |
2642 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 2643 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 2644 | |
e543851d | 2645 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2646 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 2647 | exists = true; |
afee281f | 2648 | break; |
58397e6c KT |
2649 | } |
2650 | } | |
afee281f | 2651 | |
58397e6c KT |
2652 | ovs_mutex_unlock(&dpdk_mutex); |
2653 | ||
0a0f39df | 2654 | if (exists) { |
afee281f KT |
2655 | /* |
2656 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2657 | * to NULL, before returning. | |
2658 | */ | |
2659 | ovsrcu_synchronize(); | |
2660 | /* | |
2661 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2662 | * put thread back into quiescent state before returning. | |
2663 | */ | |
2664 | ovsrcu_quiesce_start(); | |
0a0f39df | 2665 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 2666 | } else { |
0a0f39df | 2667 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 2668 | } |
58397e6c KT |
2669 | } |
2670 | ||
585a5bea | 2671 | static int |
0a0f39df | 2672 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 2673 | { |
d46285a2 | 2674 | struct netdev_dpdk *dev; |
585a5bea IM |
2675 | bool exists = false; |
2676 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
2677 | char ifname[IF_NAME_SZ]; |
2678 | ||
58be5c0e | 2679 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
2680 | |
2681 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2682 | return 0; | |
2683 | } | |
2684 | ||
2685 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2686 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 2687 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2688 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 2689 | if (enable) { |
d46285a2 | 2690 | dev->tx_q[qid].map = qid; |
585a5bea | 2691 | } else { |
d46285a2 | 2692 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2693 | } |
d46285a2 | 2694 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2695 | exists = true; |
d46285a2 | 2696 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2697 | break; |
2698 | } | |
c1ff66ac | 2699 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2700 | } |
2701 | ovs_mutex_unlock(&dpdk_mutex); | |
2702 | ||
2703 | if (exists) { | |
0a0f39df CL |
2704 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
2705 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 2706 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 2707 | } else { |
0a0f39df | 2708 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
2709 | return -1; |
2710 | } | |
2711 | ||
2712 | return 0; | |
2713 | } | |
2714 | ||
0a0f39df CL |
2715 | int |
2716 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 2717 | { |
0a0f39df | 2718 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
2719 | } |
2720 | ||
9509913a IS |
2721 | struct ingress_policer * |
2722 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
2723 | { | |
2724 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
2725 | } | |
2726 | ||
58397e6c KT |
2727 | /* |
2728 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
2729 | * configuration has been fully complete. | |
2730 | */ | |
bce01e3a | 2731 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
2732 | { |
2733 | .new_device = new_device, | |
2734 | .destroy_device = destroy_device, | |
585a5bea | 2735 | .vring_state_changed = vring_state_changed |
58397e6c KT |
2736 | }; |
2737 | ||
2738 | static void * | |
7d1ced01 | 2739 | start_vhost_loop(void *dummy OVS_UNUSED) |
58397e6c KT |
2740 | { |
2741 | pthread_detach(pthread_self()); | |
41987644 | 2742 | /* Put the vhost thread into quiescent state. */ |
618f44f7 | 2743 | ovsrcu_quiesce_start(); |
58397e6c KT |
2744 | rte_vhost_driver_session_start(); |
2745 | return NULL; | |
2746 | } | |
2747 | ||
2748 | static int | |
ecc1a34e | 2749 | netdev_dpdk_class_init(void) |
7d1ced01 | 2750 | { |
ecc1a34e DDP |
2751 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2752 | ||
2753 | /* This function can be called for different classes. The initialization | |
2754 | * needs to be done only once */ | |
2755 | if (ovsthread_once_start(&once)) { | |
2756 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
2757 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2758 | "[netdev] up|down", 1, 2, | |
2759 | netdev_dpdk_set_admin_state, NULL); | |
b8374d0d | 2760 | unixctl_command_register("netdev-dpdk/detach", |
55e075e6 | 2761 | "pci address of device", 1, 1, |
b8374d0d | 2762 | netdev_dpdk_detach, NULL); |
ecc1a34e DDP |
2763 | |
2764 | ovsthread_once_done(&once); | |
2765 | } | |
362ca396 | 2766 | |
7d1ced01 CL |
2767 | return 0; |
2768 | } | |
2769 | ||
ecc1a34e DDP |
2770 | static int |
2771 | netdev_dpdk_vhost_class_init(void) | |
033e9df2 | 2772 | { |
ecc1a34e | 2773 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
033e9df2 | 2774 | |
ecc1a34e DDP |
2775 | /* This function can be called for different classes. The initialization |
2776 | * needs to be done only once */ | |
2777 | if (ovsthread_once_start(&once)) { | |
2778 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
2779 | rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4 | |
2780 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
86bd9ed9 | 2781 | | 1ULL << VIRTIO_NET_F_CSUM); |
ecc1a34e DDP |
2782 | ovs_thread_create("vhost_thread", start_vhost_loop, NULL); |
2783 | ||
2784 | ovsthread_once_done(&once); | |
2785 | } | |
2786 | ||
2787 | return 0; | |
033e9df2 DDP |
2788 | } |
2789 | ||
95fb793a | 2790 | /* Client Rings */ |
2791 | ||
95fb793a | 2792 | static int |
2793 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
2794 | unsigned int *eth_port_id) | |
2795 | { | |
48fffdee | 2796 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 2797 | char *ring_name; |
b8374d0d | 2798 | int port_id; |
95fb793a | 2799 | |
48fffdee KT |
2800 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
2801 | if (!ring_pair) { | |
95fb793a | 2802 | return ENOMEM; |
2803 | } | |
2804 | ||
7251515e | 2805 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 2806 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 2807 | |
8f0a76c9 | 2808 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 2809 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2810 | RING_F_SP_ENQ); |
0c6f39e5 | 2811 | free(ring_name); |
48fffdee KT |
2812 | if (ring_pair->cring_tx == NULL) { |
2813 | rte_free(ring_pair); | |
95fb793a | 2814 | return ENOMEM; |
2815 | } | |
2816 | ||
0c6f39e5 | 2817 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 2818 | |
8f0a76c9 | 2819 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 2820 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2821 | RING_F_SC_DEQ); |
0c6f39e5 | 2822 | free(ring_name); |
48fffdee KT |
2823 | if (ring_pair->cring_rx == NULL) { |
2824 | rte_free(ring_pair); | |
95fb793a | 2825 | return ENOMEM; |
2826 | } | |
2827 | ||
b8374d0d MV |
2828 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
2829 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 2830 | |
b8374d0d | 2831 | if (port_id < 0) { |
48fffdee | 2832 | rte_free(ring_pair); |
95fb793a | 2833 | return ENODEV; |
2834 | } | |
2835 | ||
48fffdee | 2836 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
2837 | ring_pair->eth_port_id = port_id; |
2838 | *eth_port_id = port_id; | |
2839 | ||
48fffdee | 2840 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 2841 | |
95fb793a | 2842 | return 0; |
2843 | } | |
2844 | ||
2845 | static int | |
64839cf4 WT |
2846 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) |
2847 | OVS_REQUIRES(dpdk_mutex) | |
95fb793a | 2848 | { |
48fffdee | 2849 | struct dpdk_ring *ring_pair; |
95fb793a | 2850 | unsigned int port_no; |
2851 | int err = 0; | |
2852 | ||
2853 | /* Names always start with "dpdkr" */ | |
2854 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2855 | if (err) { | |
2856 | return err; | |
2857 | } | |
2858 | ||
58be5c0e | 2859 | /* Look through our list to find the device */ |
48fffdee KT |
2860 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
2861 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 2862 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 2863 | /* Really all that is needed */ |
48fffdee | 2864 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 2865 | return 0; |
2866 | } | |
2867 | } | |
2868 | /* Need to create the device rings */ | |
2869 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2870 | } | |
2871 | ||
7251515e | 2872 | static int |
d46285a2 | 2873 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
324c8374 IM |
2874 | struct dp_packet_batch *batch, bool may_steal, |
2875 | bool concurrent_txq) | |
7251515e | 2876 | { |
d46285a2 | 2877 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b99bb05 MG |
2878 | unsigned i; |
2879 | ||
58be5c0e MK |
2880 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
2881 | * the rss hash field is clear. This is because the same mbuf may be | |
2882 | * modified by the consumer of the ring and return into the datapath | |
2883 | * without recalculating the RSS hash. */ | |
64839cf4 WT |
2884 | for (i = 0; i < batch->count; i++) { |
2885 | dp_packet_rss_invalidate(batch->packets[i]); | |
1b99bb05 | 2886 | } |
7251515e | 2887 | |
324c8374 | 2888 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e DV |
2889 | return 0; |
2890 | } | |
2891 | ||
95fb793a | 2892 | static int |
2893 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2894 | { | |
2895 | unsigned int port_no = 0; | |
2896 | int err = 0; | |
2897 | ||
95fb793a | 2898 | ovs_mutex_lock(&dpdk_mutex); |
2899 | ||
2900 | err = dpdk_ring_open(netdev->name, &port_no); | |
2901 | if (err) { | |
2902 | goto unlock_dpdk; | |
2903 | } | |
2904 | ||
1ce30dfd DDP |
2905 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
2906 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 2907 | unlock_dpdk: |
2908 | ovs_mutex_unlock(&dpdk_mutex); | |
2909 | return err; | |
2910 | } | |
2911 | ||
0bf765f7 IS |
2912 | /* QoS Functions */ |
2913 | ||
2914 | /* | |
2915 | * Initialize QoS configuration operations. | |
2916 | */ | |
2917 | static void | |
2918 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
2919 | { | |
2920 | conf->ops = ops; | |
78bd47cf | 2921 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
2922 | } |
2923 | ||
2924 | /* | |
2925 | * Search existing QoS operations in qos_ops and compare each set of | |
2926 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
2927 | * else return NULL | |
2928 | */ | |
2929 | static const struct dpdk_qos_ops * | |
2930 | qos_lookup_name(const char *name) | |
2931 | { | |
2932 | const struct dpdk_qos_ops *const *opsp; | |
2933 | ||
2934 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2935 | const struct dpdk_qos_ops *ops = *opsp; | |
2936 | if (!strcmp(name, ops->qos_name)) { | |
2937 | return ops; | |
2938 | } | |
2939 | } | |
2940 | return NULL; | |
2941 | } | |
2942 | ||
0bf765f7 IS |
2943 | static int |
2944 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
2945 | struct sset *types) | |
2946 | { | |
2947 | const struct dpdk_qos_ops *const *opsp; | |
2948 | ||
2949 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2950 | const struct dpdk_qos_ops *ops = *opsp; | |
2951 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
2952 | sset_add(types, ops->qos_name); | |
2953 | } | |
2954 | } | |
2955 | return 0; | |
2956 | } | |
2957 | ||
2958 | static int | |
d46285a2 | 2959 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
2960 | const char **typep, struct smap *details) |
2961 | { | |
d46285a2 | 2962 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 2963 | struct qos_conf *qos_conf; |
0bf765f7 IS |
2964 | int error = 0; |
2965 | ||
d46285a2 | 2966 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
2967 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
2968 | if (qos_conf) { | |
2969 | *typep = qos_conf->ops->qos_name; | |
2970 | error = (qos_conf->ops->qos_get | |
2971 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
2972 | } else { |
2973 | /* No QoS configuration set, return an empty string */ | |
2974 | *typep = ""; | |
0bf765f7 | 2975 | } |
d46285a2 | 2976 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2977 | |
2978 | return error; | |
2979 | } | |
2980 | ||
2981 | static int | |
78bd47cf DDP |
2982 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
2983 | const struct smap *details) | |
0bf765f7 | 2984 | { |
d46285a2 | 2985 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 2986 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 2987 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
2988 | int error = 0; |
2989 | ||
d46285a2 | 2990 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 2991 | |
78bd47cf | 2992 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 2993 | |
78bd47cf DDP |
2994 | new_ops = qos_lookup_name(type); |
2995 | ||
2996 | if (!new_ops || !new_ops->qos_construct) { | |
2997 | new_qos_conf = NULL; | |
2998 | if (type && type[0]) { | |
2999 | error = EOPNOTSUPP; | |
0bf765f7 | 3000 | } |
44975bb0 | 3001 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
3002 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
3003 | new_qos_conf = qos_conf; | |
0bf765f7 | 3004 | } else { |
78bd47cf | 3005 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
3006 | } |
3007 | ||
7ea266e9 | 3008 | if (error) { |
78bd47cf DDP |
3009 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
3010 | type, netdev->name, rte_strerror(error)); | |
3011 | } | |
3012 | ||
3013 | if (new_qos_conf != qos_conf) { | |
3014 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
3015 | if (qos_conf) { | |
3016 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
3017 | } | |
0bf765f7 IS |
3018 | } |
3019 | ||
d46285a2 | 3020 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 3021 | |
0bf765f7 IS |
3022 | return error; |
3023 | } | |
3024 | ||
3025 | /* egress-policer details */ | |
3026 | ||
3027 | struct egress_policer { | |
3028 | struct qos_conf qos_conf; | |
3029 | struct rte_meter_srtcm_params app_srtcm_params; | |
3030 | struct rte_meter_srtcm egress_meter; | |
3031 | }; | |
3032 | ||
78bd47cf DDP |
3033 | static void |
3034 | egress_policer_details_to_param(const struct smap *details, | |
3035 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 3036 | { |
78bd47cf DDP |
3037 | memset(params, 0, sizeof *params); |
3038 | params->cir = smap_get_ullong(details, "cir", 0); | |
3039 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
3040 | params->ebs = 0; | |
0bf765f7 IS |
3041 | } |
3042 | ||
3043 | static int | |
78bd47cf DDP |
3044 | egress_policer_qos_construct(const struct smap *details, |
3045 | struct qos_conf **conf) | |
0bf765f7 | 3046 | { |
0bf765f7 | 3047 | struct egress_policer *policer; |
0bf765f7 IS |
3048 | int err = 0; |
3049 | ||
0bf765f7 IS |
3050 | policer = xmalloc(sizeof *policer); |
3051 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 3052 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
0bf765f7 | 3053 | err = rte_meter_srtcm_config(&policer->egress_meter, |
78bd47cf DDP |
3054 | &policer->app_srtcm_params); |
3055 | if (!err) { | |
3056 | *conf = &policer->qos_conf; | |
3057 | } else { | |
7ea266e9 | 3058 | free(policer); |
78bd47cf | 3059 | *conf = NULL; |
7ea266e9 IS |
3060 | err = -err; |
3061 | } | |
0bf765f7 IS |
3062 | |
3063 | return err; | |
3064 | } | |
3065 | ||
3066 | static void | |
78bd47cf | 3067 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
3068 | { |
3069 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 3070 | qos_conf); |
0bf765f7 IS |
3071 | free(policer); |
3072 | } | |
3073 | ||
3074 | static int | |
78bd47cf | 3075 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 3076 | { |
78bd47cf DDP |
3077 | struct egress_policer *policer = |
3078 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3079 | ||
3080 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
3081 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 3082 | |
0bf765f7 IS |
3083 | return 0; |
3084 | } | |
3085 | ||
78bd47cf | 3086 | static bool |
47a45d86 KT |
3087 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
3088 | const struct smap *details) | |
0bf765f7 | 3089 | { |
78bd47cf DDP |
3090 | struct egress_policer *policer = |
3091 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3092 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 3093 | |
78bd47cf | 3094 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 3095 | |
78bd47cf | 3096 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
3097 | } |
3098 | ||
0bf765f7 | 3099 | static int |
78bd47cf | 3100 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt) |
0bf765f7 | 3101 | { |
0bf765f7 | 3102 | int cnt = 0; |
78bd47cf DDP |
3103 | struct egress_policer *policer = |
3104 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 3105 | |
f3926f29 | 3106 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, pkt_cnt); |
0bf765f7 IS |
3107 | |
3108 | return cnt; | |
3109 | } | |
3110 | ||
3111 | static const struct dpdk_qos_ops egress_policer_ops = { | |
3112 | "egress-policer", /* qos_name */ | |
3113 | egress_policer_qos_construct, | |
3114 | egress_policer_qos_destruct, | |
3115 | egress_policer_qos_get, | |
78bd47cf | 3116 | egress_policer_qos_is_equal, |
0bf765f7 IS |
3117 | egress_policer_run |
3118 | }; | |
3119 | ||
050c60bf DDP |
3120 | static int |
3121 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
3122 | { | |
3123 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3124 | int err = 0; | |
3125 | ||
050c60bf DDP |
3126 | ovs_mutex_lock(&dev->mutex); |
3127 | ||
3128 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 3129 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b CL |
3130 | && dev->mtu == dev->requested_mtu |
3131 | && dev->rxq_size == dev->requested_rxq_size | |
bd4e172b | 3132 | && dev->txq_size == dev->requested_txq_size |
3133 | && dev->socket_id == dev->requested_socket_id) { | |
050c60bf DDP |
3134 | /* Reconfiguration is unnecessary */ |
3135 | ||
3136 | goto out; | |
3137 | } | |
3138 | ||
3139 | rte_eth_dev_stop(dev->port_id); | |
3140 | ||
bd4e172b | 3141 | if (dev->mtu != dev->requested_mtu |
3142 | || dev->socket_id != dev->requested_socket_id) { | |
96e9b168 IS |
3143 | err = netdev_dpdk_mempool_configure(dev); |
3144 | if (err) { | |
3145 | goto out; | |
3146 | } | |
0072e931 MK |
3147 | } |
3148 | ||
050c60bf DDP |
3149 | netdev->n_txq = dev->requested_n_txq; |
3150 | netdev->n_rxq = dev->requested_n_rxq; | |
3151 | ||
b685696b CL |
3152 | dev->rxq_size = dev->requested_rxq_size; |
3153 | dev->txq_size = dev->requested_txq_size; | |
3154 | ||
050c60bf DDP |
3155 | rte_free(dev->tx_q); |
3156 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
3157 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
3158 | if (!dev->tx_q) { | |
3159 | err = ENOMEM; | |
3160 | } | |
050c60bf | 3161 | |
0072e931 MK |
3162 | netdev_change_seq_changed(netdev); |
3163 | ||
050c60bf | 3164 | out: |
050c60bf | 3165 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
3166 | return err; |
3167 | } | |
3168 | ||
7f381c2e | 3169 | static int |
2d24d165 | 3170 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 3171 | OVS_REQUIRES(dev->mutex) |
050c60bf | 3172 | { |
2d24d165 CL |
3173 | dev->up.n_txq = dev->requested_n_txq; |
3174 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 3175 | int err; |
050c60bf | 3176 | |
81acebda IM |
3177 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
3178 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
3179 | dev->tx_q[0].map = 0; | |
3180 | } | |
3181 | ||
3182 | netdev_dpdk_remap_txqs(dev); | |
3183 | ||
0072e931 MK |
3184 | if (dev->requested_socket_id != dev->socket_id |
3185 | || dev->requested_mtu != dev->mtu) { | |
96e9b168 IS |
3186 | err = netdev_dpdk_mempool_configure(dev); |
3187 | if (err) { | |
3188 | return err; | |
3189 | } else { | |
2d24d165 | 3190 | netdev_change_seq_changed(&dev->up); |
db8f13b0 CL |
3191 | } |
3192 | } | |
3193 | ||
0a0f39df CL |
3194 | if (netdev_dpdk_get_vid(dev) >= 0) { |
3195 | dev->vhost_reconfigured = true; | |
81acebda | 3196 | } |
7f381c2e DDP |
3197 | |
3198 | return 0; | |
2d24d165 CL |
3199 | } |
3200 | ||
3201 | static int | |
3202 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
3203 | { | |
3204 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3205 | int err; |
2d24d165 | 3206 | |
2d24d165 | 3207 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 3208 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 3209 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
3210 | |
3211 | return err; | |
2d24d165 CL |
3212 | } |
3213 | ||
3214 | static int | |
3215 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3216 | { | |
3217 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3218 | int err; |
2d24d165 | 3219 | |
2d24d165 CL |
3220 | ovs_mutex_lock(&dev->mutex); |
3221 | ||
c1ff66ac CL |
3222 | /* Configure vHost client mode if requested and if the following criteria |
3223 | * are met: | |
2d24d165 CL |
3224 | * 1. Device hasn't been registered yet. |
3225 | * 2. A path has been specified. | |
c1ff66ac CL |
3226 | */ |
3227 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 CL |
3228 | && strlen(dev->vhost_id)) { |
3229 | /* Register client-mode device */ | |
3230 | err = rte_vhost_driver_register(dev->vhost_id, | |
3231 | RTE_VHOST_USER_CLIENT); | |
c1ff66ac | 3232 | if (err) { |
2d24d165 CL |
3233 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3234 | dev->vhost_id); | |
7f381c2e | 3235 | goto unlock; |
c1ff66ac | 3236 | } else { |
2d24d165 CL |
3237 | /* Configuration successful */ |
3238 | dev->vhost_driver_flags |= RTE_VHOST_USER_CLIENT; | |
3239 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " | |
3240 | "using client socket '%s'", | |
3241 | dev->up.name, dev->vhost_id); | |
c1ff66ac CL |
3242 | } |
3243 | } | |
3244 | ||
7f381c2e DDP |
3245 | err = dpdk_vhost_reconfigure_helper(dev); |
3246 | ||
3247 | unlock: | |
050c60bf | 3248 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 3249 | |
7f381c2e | 3250 | return err; |
050c60bf DDP |
3251 | } |
3252 | ||
ecc1a34e | 3253 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
81acebda IM |
3254 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
3255 | GET_CARRIER, GET_STATS, \ | |
3256 | GET_FEATURES, GET_STATUS, \ | |
3257 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3258 | { \ |
3259 | NAME, \ | |
118c77b1 | 3260 | true, /* is_pmd */ \ |
ecc1a34e | 3261 | INIT, /* init */ \ |
95fb793a | 3262 | NULL, /* netdev_dpdk_run */ \ |
3263 | NULL, /* netdev_dpdk_wait */ \ | |
3264 | \ | |
3265 | netdev_dpdk_alloc, \ | |
3266 | CONSTRUCT, \ | |
58397e6c | 3267 | DESTRUCT, \ |
95fb793a | 3268 | netdev_dpdk_dealloc, \ |
3269 | netdev_dpdk_get_config, \ | |
81acebda | 3270 | SET_CONFIG, \ |
95fb793a | 3271 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3272 | NULL, /* build header */ \ |
3273 | NULL, /* push header */ \ | |
3274 | NULL, /* pop header */ \ | |
7dec44fe | 3275 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3276 | SET_TX_MULTIQ, \ |
95fb793a | 3277 | \ |
7251515e | 3278 | SEND, /* send */ \ |
95fb793a | 3279 | NULL, /* send_wait */ \ |
3280 | \ | |
3281 | netdev_dpdk_set_etheraddr, \ | |
3282 | netdev_dpdk_get_etheraddr, \ | |
3283 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3284 | netdev_dpdk_set_mtu, \ |
95fb793a | 3285 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3286 | GET_CARRIER, \ |
95fb793a | 3287 | netdev_dpdk_get_carrier_resets, \ |
3288 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
3289 | GET_STATS, \ |
3290 | GET_FEATURES, \ | |
95fb793a | 3291 | NULL, /* set_advertisements */ \ |
3292 | \ | |
9509913a | 3293 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3294 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3295 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3296 | netdev_dpdk_get_qos, \ |
3297 | netdev_dpdk_set_qos, \ | |
95fb793a | 3298 | NULL, /* get_queue */ \ |
3299 | NULL, /* set_queue */ \ | |
3300 | NULL, /* delete_queue */ \ | |
3301 | NULL, /* get_queue_stats */ \ | |
3302 | NULL, /* queue_dump_start */ \ | |
3303 | NULL, /* queue_dump_next */ \ | |
3304 | NULL, /* queue_dump_done */ \ | |
3305 | NULL, /* dump_queue_stats */ \ | |
3306 | \ | |
95fb793a | 3307 | NULL, /* set_in4 */ \ |
a8704b50 | 3308 | NULL, /* get_addr_list */ \ |
95fb793a | 3309 | NULL, /* add_router */ \ |
3310 | NULL, /* get_next_hop */ \ | |
58397e6c | 3311 | GET_STATUS, \ |
95fb793a | 3312 | NULL, /* arp_lookup */ \ |
3313 | \ | |
3314 | netdev_dpdk_update_flags, \ | |
050c60bf | 3315 | RECONFIGURE, \ |
95fb793a | 3316 | \ |
3317 | netdev_dpdk_rxq_alloc, \ | |
3318 | netdev_dpdk_rxq_construct, \ | |
3319 | netdev_dpdk_rxq_destruct, \ | |
3320 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3321 | RXQ_RECV, \ |
95fb793a | 3322 | NULL, /* rx_wait */ \ |
3323 | NULL, /* rxq_drain */ \ | |
3324 | } | |
8a9562d2 | 3325 | |
bce01e3a | 3326 | static const struct netdev_class dpdk_class = |
95fb793a | 3327 | NETDEV_DPDK_CLASS( |
3328 | "dpdk", | |
ecc1a34e | 3329 | netdev_dpdk_class_init, |
5496878c | 3330 | netdev_dpdk_construct, |
58397e6c | 3331 | netdev_dpdk_destruct, |
81acebda IM |
3332 | netdev_dpdk_set_config, |
3333 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3334 | netdev_dpdk_eth_send, |
3335 | netdev_dpdk_get_carrier, | |
3336 | netdev_dpdk_get_stats, | |
3337 | netdev_dpdk_get_features, | |
3338 | netdev_dpdk_get_status, | |
050c60bf | 3339 | netdev_dpdk_reconfigure, |
58397e6c | 3340 | netdev_dpdk_rxq_recv); |
95fb793a | 3341 | |
bce01e3a | 3342 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3343 | NETDEV_DPDK_CLASS( |
3344 | "dpdkr", | |
ecc1a34e | 3345 | netdev_dpdk_class_init, |
5496878c | 3346 | netdev_dpdk_ring_construct, |
58397e6c | 3347 | netdev_dpdk_destruct, |
c3d062a7 | 3348 | netdev_dpdk_ring_set_config, |
81acebda | 3349 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3350 | netdev_dpdk_ring_send, |
3351 | netdev_dpdk_get_carrier, | |
3352 | netdev_dpdk_get_stats, | |
3353 | netdev_dpdk_get_features, | |
3354 | netdev_dpdk_get_status, | |
050c60bf | 3355 | netdev_dpdk_reconfigure, |
58397e6c KT |
3356 | netdev_dpdk_rxq_recv); |
3357 | ||
53f50d24 | 3358 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3359 | NETDEV_DPDK_CLASS( |
3360 | "dpdkvhostuser", | |
ecc1a34e | 3361 | netdev_dpdk_vhost_class_init, |
53f50d24 | 3362 | netdev_dpdk_vhost_construct, |
58397e6c | 3363 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3364 | NULL, |
81acebda | 3365 | NULL, |
58397e6c KT |
3366 | netdev_dpdk_vhost_send, |
3367 | netdev_dpdk_vhost_get_carrier, | |
3368 | netdev_dpdk_vhost_get_stats, | |
3369 | NULL, | |
7251515e | 3370 | NULL, |
53f50d24 | 3371 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3372 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3373 | static const struct netdev_class dpdk_vhost_client_class = |
3374 | NETDEV_DPDK_CLASS( | |
3375 | "dpdkvhostuserclient", | |
ecc1a34e | 3376 | netdev_dpdk_vhost_class_init, |
2d24d165 CL |
3377 | netdev_dpdk_vhost_client_construct, |
3378 | netdev_dpdk_vhost_destruct, | |
3379 | netdev_dpdk_vhost_client_set_config, | |
3380 | NULL, | |
3381 | netdev_dpdk_vhost_send, | |
3382 | netdev_dpdk_vhost_get_carrier, | |
3383 | netdev_dpdk_vhost_get_stats, | |
3384 | NULL, | |
3385 | NULL, | |
3386 | netdev_dpdk_vhost_client_reconfigure, | |
3387 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 3388 | |
8a9562d2 PS |
3389 | void |
3390 | netdev_dpdk_register(void) | |
3391 | { | |
bab69409 AC |
3392 | netdev_register_provider(&dpdk_class); |
3393 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3394 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 3395 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 3396 | } |