]>
Commit | Line | Data |
---|---|---|
8a9562d2 | 1 | /* |
12d0d124 | 2 | * Copyright (c) 2014, 2015, 2016, 2017 Nicira, Inc. |
8a9562d2 PS |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
01961bbd | 18 | #include "netdev-dpdk.h" |
8a9562d2 | 19 | |
8a9562d2 PS |
20 | #include <string.h> |
21 | #include <signal.h> | |
22 | #include <stdlib.h> | |
8a9562d2 | 23 | #include <errno.h> |
8a9562d2 | 24 | #include <unistd.h> |
f3e7ec25 MW |
25 | #include <linux/virtio_net.h> |
26 | #include <sys/socket.h> | |
27 | #include <linux/if.h> | |
01961bbd | 28 | |
5e925ccc | 29 | #include <rte_bus_pci.h> |
01961bbd DDP |
30 | #include <rte_config.h> |
31 | #include <rte_cycles.h> | |
32 | #include <rte_errno.h> | |
33 | #include <rte_eth_ring.h> | |
34 | #include <rte_ethdev.h> | |
35 | #include <rte_malloc.h> | |
36 | #include <rte_mbuf.h> | |
37 | #include <rte_meter.h> | |
fc56f5e0 | 38 | #include <rte_pci.h> |
f3e7ec25 | 39 | #include <rte_vhost.h> |
8a9562d2 | 40 | |
7d1ced01 | 41 | #include "dirs.h" |
e14deea0 | 42 | #include "dp-packet.h" |
01961bbd | 43 | #include "dpdk.h" |
8a9562d2 | 44 | #include "dpif-netdev.h" |
e5c0f5a4 | 45 | #include "fatal-signal.h" |
8a9562d2 PS |
46 | #include "netdev-provider.h" |
47 | #include "netdev-vport.h" | |
48 | #include "odp-util.h" | |
eac84432 | 49 | #include "openvswitch/dynamic-string.h" |
25d436fb BW |
50 | #include "openvswitch/list.h" |
51 | #include "openvswitch/ofp-print.h" | |
52 | #include "openvswitch/vlog.h" | |
94143fc4 | 53 | #include "ovs-numa.h" |
8a9562d2 PS |
54 | #include "ovs-thread.h" |
55 | #include "ovs-rcu.h" | |
56 | #include "packets.h" | |
ee89ea7b | 57 | #include "openvswitch/shash.h" |
0bf765f7 | 58 | #include "smap.h" |
8a9562d2 PS |
59 | #include "sset.h" |
60 | #include "unaligned.h" | |
61 | #include "timeval.h" | |
62 | #include "unixctl.h" | |
8a9562d2 | 63 | |
f3e7ec25 MW |
64 | enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM}; |
65 | ||
05b49df6 | 66 | VLOG_DEFINE_THIS_MODULE(netdev_dpdk); |
8a9562d2 PS |
67 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
68 | ||
69 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
70 | ||
71 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
72 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
73 | ||
74 | /* | |
75 | * need to reserve tons of extra space in the mbufs so we can align the | |
76 | * DMA addresses to 4KB. | |
18f777b2 TP |
77 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
78 | * performance for standard Ethernet MTU. | |
8a9562d2 | 79 | */ |
58be5c0e MK |
80 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN \ |
81 | + (2 * VLAN_HEADER_LEN)) | |
4be4d22c MK |
82 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) |
83 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
58be5c0e MK |
84 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len) \ |
85 | - ETHER_HDR_LEN - ETHER_CRC_LEN) | |
31b88c97 SS |
86 | #define MBUF_SIZE(mtu) ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) \ |
87 | + sizeof(struct dp_packet) \ | |
88 | + RTE_PKTMBUF_HEADROOM), \ | |
89 | RTE_CACHE_LINE_SIZE) | |
4be4d22c | 90 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 91 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 92 | |
bc57ed90 IM |
93 | /* Min number of packets in the mempool. OVS tries to allocate a mempool with |
94 | * roughly estimated number of mbufs: if this fails (because the system doesn't | |
95 | * have enough hugepages) we keep halving the number until the allocation | |
96 | * succeeds or we reach MIN_NB_MBUF */ | |
da79ce2b DDP |
97 | #define MIN_NB_MBUF (4096 * 4) |
98 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
99 | ||
d6e3feb5 | 100 | /* |
101 | * DPDK XSTATS Counter names definition | |
102 | */ | |
103 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
104 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
105 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
106 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
107 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
108 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
109 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
110 | ||
111 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
112 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
113 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
114 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
115 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
116 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
117 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
118 | ||
d57f777f | 119 | #define XSTAT_RX_MULTICAST_PACKETS "rx_multicast_packets" |
d6e3feb5 | 120 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" |
121 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
122 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
123 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
124 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
125 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
126 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
127 | ||
8a9562d2 PS |
128 | #define SOCKET0 0 |
129 | ||
b685696b CL |
130 | /* Default size of Physical NIC RXQ */ |
131 | #define NIC_PORT_DEFAULT_RXQ_SIZE 2048 | |
132 | /* Default size of Physical NIC TXQ */ | |
133 | #define NIC_PORT_DEFAULT_TXQ_SIZE 2048 | |
134 | /* Maximum size of Physical NIC Queues */ | |
135 | #define NIC_PORT_MAX_Q_SIZE 4096 | |
79f5354c | 136 | |
585a5bea | 137 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
138 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
139 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
140 | * yet mapped to another queue. */ | |
585a5bea | 141 | |
bb37956a IM |
142 | #define DPDK_ETH_PORT_ID_INVALID RTE_MAX_ETHPORTS |
143 | ||
5e925ccc MK |
144 | /* DPDK library uses uint16_t for port_id. */ |
145 | typedef uint16_t dpdk_port_t; | |
bb37956a | 146 | |
31871ee3 | 147 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 148 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 149 | |
8a9562d2 | 150 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
151 | .rxmode = { |
152 | .mq_mode = ETH_MQ_RX_RSS, | |
153 | .split_hdr_size = 0, | |
154 | .header_split = 0, /* Header Split disabled */ | |
155 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
156 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
157 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
158 | .hw_strip_crc = 0, | |
159 | }, | |
160 | .rx_adv_conf = { | |
161 | .rss_conf = { | |
162 | .rss_key = NULL, | |
543342a4 | 163 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 164 | }, |
a28ddd11 DDP |
165 | }, |
166 | .txmode = { | |
167 | .mq_mode = ETH_MQ_TX_NONE, | |
168 | }, | |
8a9562d2 PS |
169 | }; |
170 | ||
f3e7ec25 MW |
171 | /* |
172 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
173 | * configuration has been fully completed. | |
174 | */ | |
175 | static int new_device(int vid); | |
176 | static void destroy_device(int vid); | |
177 | static int vring_state_changed(int vid, uint16_t queue_id, int enable); | |
178 | static const struct vhost_device_ops virtio_net_device_ops = | |
179 | { | |
180 | .new_device = new_device, | |
181 | .destroy_device = destroy_device, | |
182 | .vring_state_changed = vring_state_changed, | |
183 | .features_changed = NULL | |
184 | }; | |
185 | ||
58f7c37b DDP |
186 | enum { DPDK_RING_SIZE = 256 }; |
187 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
188 | enum { DRAIN_TSC = 200000ULL }; |
189 | ||
58397e6c KT |
190 | enum dpdk_dev_type { |
191 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 192 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
193 | }; |
194 | ||
0bf765f7 IS |
195 | /* Quality of Service */ |
196 | ||
197 | /* An instance of a QoS configuration. Always associated with a particular | |
198 | * network device. | |
199 | * | |
200 | * Each QoS implementation subclasses this with whatever additional data it | |
201 | * needs. | |
202 | */ | |
203 | struct qos_conf { | |
204 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 205 | rte_spinlock_t lock; |
0bf765f7 IS |
206 | }; |
207 | ||
208 | /* A particular implementation of dpdk QoS operations. | |
209 | * | |
210 | * The functions below return 0 if successful or a positive errno value on | |
211 | * failure, except where otherwise noted. All of them must be provided, except | |
212 | * where otherwise noted. | |
213 | */ | |
214 | struct dpdk_qos_ops { | |
215 | ||
216 | /* Name of the QoS type */ | |
217 | const char *qos_name; | |
218 | ||
78bd47cf DDP |
219 | /* Called to construct a qos_conf object. The implementation should make |
220 | * the appropriate calls to configure QoS according to 'details'. | |
0bf765f7 IS |
221 | * |
222 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
223 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
224 | * (which is built as ovs-vswitchd.conf.db(8)). | |
225 | * | |
78bd47cf DDP |
226 | * This function must return 0 if and only if it sets '*conf' to an |
227 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
228 | * |
229 | * For all QoS implementations it should always be non-null. | |
230 | */ | |
78bd47cf | 231 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
232 | |
233 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 234 | * 'qos_conf'. |
0bf765f7 IS |
235 | * |
236 | * For all QoS implementations it should always be non-null. | |
237 | */ | |
78bd47cf | 238 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 239 | |
78bd47cf | 240 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
241 | * |
242 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
243 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
244 | * (which is built as ovs-vswitchd.conf.db(8)). | |
245 | */ | |
78bd47cf | 246 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 247 | |
78bd47cf | 248 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
249 | * |
250 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
251 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
252 | * (which is built as ovs-vswitchd.conf.db(8)). | |
253 | * | |
78bd47cf | 254 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 255 | */ |
78bd47cf DDP |
256 | bool (*qos_is_equal)(const struct qos_conf *conf, |
257 | const struct smap *details); | |
0bf765f7 IS |
258 | |
259 | /* Modify an array of rte_mbufs. The modification is specific to | |
260 | * each qos implementation. | |
261 | * | |
262 | * The function should take and array of mbufs and an int representing | |
263 | * the current number of mbufs present in the array. | |
264 | * | |
265 | * After the function has performed a qos modification to the array of | |
266 | * mbufs it returns an int representing the number of mbufs now present in | |
267 | * the array. This value is can then be passed to the port send function | |
268 | * along with the modified array for transmission. | |
269 | * | |
270 | * For all QoS implementations it should always be non-null. | |
271 | */ | |
78bd47cf | 272 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
3e90f7d7 | 273 | int pkt_cnt, bool may_steal); |
0bf765f7 IS |
274 | }; |
275 | ||
276 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
277 | static const struct dpdk_qos_ops egress_policer_ops; | |
278 | ||
279 | /* | |
280 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
281 | * operations. | |
282 | */ | |
283 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
284 | &egress_policer_ops, | |
285 | NULL | |
286 | }; | |
287 | ||
c2adb102 IM |
288 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
289 | ||
8a9562d2 | 290 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 291 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 292 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 293 | |
c2adb102 IM |
294 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
295 | = OVS_MUTEX_INITIALIZER; | |
296 | ||
5a034064 AW |
297 | /* There should be one 'struct dpdk_tx_queue' created for |
298 | * each cpu core. */ | |
8a9562d2 | 299 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
300 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
301 | * from concurrent access. It is used only | |
302 | * if the queue is shared among different | |
324c8374 | 303 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
304 | int map; /* Mapping of configured vhost-user queues |
305 | * to enabled by guest. */ | |
8a9562d2 PS |
306 | }; |
307 | ||
95fb793a | 308 | /* dpdk has no way to remove dpdk ring ethernet devices |
309 | so we have to keep them around once they've been created | |
310 | */ | |
311 | ||
ca6ba700 | 312 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 313 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 314 | |
315 | struct dpdk_ring { | |
316 | /* For the client rings */ | |
317 | struct rte_ring *cring_tx; | |
318 | struct rte_ring *cring_rx; | |
b83a2df1 | 319 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
bb37956a | 320 | dpdk_port_t eth_port_id; /* ethernet device port id */ |
ca6ba700 | 321 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 322 | }; |
323 | ||
9509913a IS |
324 | struct ingress_policer { |
325 | struct rte_meter_srtcm_params app_srtcm_params; | |
326 | struct rte_meter_srtcm in_policer; | |
327 | rte_spinlock_t policer_lock; | |
328 | }; | |
329 | ||
1a2bb118 SC |
330 | enum dpdk_hw_ol_features { |
331 | NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0, | |
332 | }; | |
333 | ||
b2e72a9c IM |
334 | /* |
335 | * In order to avoid confusion in variables names, following naming convention | |
336 | * should be used, if possible: | |
337 | * | |
338 | * 'struct netdev' : 'netdev' | |
339 | * 'struct netdev_dpdk' : 'dev' | |
340 | * 'struct netdev_rxq' : 'rxq' | |
341 | * 'struct netdev_rxq_dpdk' : 'rx' | |
342 | * | |
343 | * Example: | |
344 | * struct netdev *netdev = netdev_from_name(name); | |
345 | * struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
346 | * | |
347 | * Also, 'netdev' should be used instead of 'dev->up', where 'netdev' was | |
348 | * already defined. | |
349 | */ | |
350 | ||
8a9562d2 | 351 | struct netdev_dpdk { |
23d4d53f BB |
352 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline0, |
353 | dpdk_port_t port_id; | |
354 | ||
355 | /* If true, device was attached by rte_eth_dev_attach(). */ | |
356 | bool attached; | |
357 | struct eth_addr hwaddr; | |
358 | int mtu; | |
359 | int socket_id; | |
360 | int buf_size; | |
361 | int max_packet_len; | |
362 | enum dpdk_dev_type type; | |
363 | enum netdev_flags flags; | |
364 | char *devargs; /* Device arguments for dpdk ports */ | |
365 | struct dpdk_tx_queue *tx_q; | |
366 | struct rte_eth_link link; | |
367 | int link_reset_cnt; | |
368 | /* 4 pad bytes here. */ | |
369 | ); | |
370 | ||
371 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline1, | |
372 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
24e78f93 | 373 | struct rte_mempool *mp; |
23d4d53f BB |
374 | |
375 | /* virtio identifier for vhost devices */ | |
376 | ovsrcu_index vid; | |
377 | ||
378 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
379 | bool vhost_reconfigured; | |
380 | /* 3 pad bytes here. */ | |
381 | ); | |
382 | ||
383 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
384 | /* Identifier used to distinguish vhost devices from each other. */ | |
385 | char vhost_id[PATH_MAX]; | |
386 | ); | |
387 | ||
388 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
389 | struct netdev up; | |
390 | /* In dpdk_list. */ | |
391 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); | |
392 | ||
393 | /* QoS configuration and lock for the device */ | |
394 | OVSRCU_TYPE(struct qos_conf *) qos_conf; | |
395 | ||
396 | /* Ingress Policer */ | |
397 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
398 | uint32_t policer_rate; | |
399 | uint32_t policer_burst; | |
400 | ); | |
401 | ||
402 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
403 | struct netdev_stats stats; | |
404 | /* Protects stats */ | |
405 | rte_spinlock_t stats_lock; | |
406 | /* 44 pad bytes here. */ | |
407 | ); | |
408 | ||
409 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
410 | /* The following properties cannot be changed when a device is running, | |
411 | * so we remember the request and update them next time | |
412 | * netdev_dpdk*_reconfigure() is called */ | |
413 | int requested_mtu; | |
414 | int requested_n_txq; | |
415 | int requested_n_rxq; | |
416 | int requested_rxq_size; | |
417 | int requested_txq_size; | |
418 | ||
419 | /* Number of rx/tx descriptors for physical devices */ | |
420 | int rxq_size; | |
421 | int txq_size; | |
422 | ||
423 | /* Socket ID detected when vHost device is brought up */ | |
424 | int requested_socket_id; | |
425 | ||
426 | /* Denotes whether vHost port is client/server mode */ | |
427 | uint64_t vhost_driver_flags; | |
428 | ||
429 | /* DPDK-ETH Flow control */ | |
430 | struct rte_eth_fc_conf fc_conf; | |
431 | ||
432 | /* DPDK-ETH hardware offload features, | |
433 | * from the enum set 'dpdk_hw_ol_features' */ | |
434 | uint32_t hw_ol_features; | |
435 | ); | |
8a9562d2 PS |
436 | }; |
437 | ||
438 | struct netdev_rxq_dpdk { | |
439 | struct netdev_rxq up; | |
bb37956a | 440 | dpdk_port_t port_id; |
8a9562d2 PS |
441 | }; |
442 | ||
f3e7ec25 MW |
443 | static void netdev_dpdk_destruct(struct netdev *netdev); |
444 | static void netdev_dpdk_vhost_destruct(struct netdev *netdev); | |
8a9562d2 | 445 | |
0a0f39df | 446 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 447 | |
9509913a IS |
448 | struct ingress_policer * |
449 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
450 | ||
8a9562d2 PS |
451 | static bool |
452 | is_dpdk_class(const struct netdev_class *class) | |
453 | { | |
f3e7ec25 MW |
454 | return class->destruct == netdev_dpdk_destruct |
455 | || class->destruct == netdev_dpdk_vhost_destruct; | |
8a9562d2 PS |
456 | } |
457 | ||
4be4d22c MK |
458 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
459 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
460 | * value, insufficient buffers are allocated to accomodate the packet in its | |
461 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
462 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
463 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
464 | * behaviour, which reduces performance. To prevent this, use a buffer size |
465 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
466 | */ |
467 | static uint32_t | |
468 | dpdk_buf_size(int mtu) | |
469 | { | |
470 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
471 | NETDEV_DPDK_MBUF_ALIGN); | |
472 | } | |
473 | ||
eff23640 DDP |
474 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
475 | * | |
476 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
477 | static void * |
478 | dpdk_rte_mzalloc(size_t sz) | |
479 | { | |
eff23640 | 480 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
481 | } |
482 | ||
483 | void | |
e14deea0 | 484 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 485 | { |
db73f716 | 486 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 487 | |
b00b4a81 | 488 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
489 | } |
490 | ||
b3cd9f9d | 491 | static void |
401b70d6 | 492 | ovs_rte_pktmbuf_init(struct rte_mempool *mp OVS_UNUSED, |
b3cd9f9d | 493 | void *opaque_arg OVS_UNUSED, |
2391135c | 494 | void *_p, |
b3cd9f9d PS |
495 | unsigned i OVS_UNUSED) |
496 | { | |
2391135c | 497 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 498 | |
2391135c | 499 | dp_packet_init_dpdk((struct dp_packet *) pkt, pkt->buf_len); |
b3cd9f9d PS |
500 | } |
501 | ||
24e78f93 IM |
502 | /* Returns a valid pointer when either of the following is true: |
503 | * - a new mempool was just created; | |
504 | * - a matching mempool already exists. */ | |
505 | static struct rte_mempool * | |
506 | dpdk_mp_create(struct netdev_dpdk *dev, int mtu) | |
8a9562d2 | 507 | { |
24e78f93 IM |
508 | char mp_name[RTE_MEMPOOL_NAMESIZE]; |
509 | const char *netdev_name = netdev_get_name(&dev->up); | |
510 | int socket_id = dev->requested_socket_id; | |
511 | uint32_t n_mbufs; | |
512 | uint32_t hash = hash_string(netdev_name, 0); | |
513 | struct rte_mempool *mp = NULL; | |
d555d9bd RW |
514 | |
515 | /* | |
ad9b5b9b | 516 | * XXX: rough estimation of number of mbufs required for this port: |
d555d9bd RW |
517 | * <packets required to fill the device rxqs> |
518 | * + <packets that could be stuck on other ports txqs> | |
519 | * + <packets in the pmd threads> | |
520 | * + <additional memory for corner cases> | |
0072e931 | 521 | */ |
24e78f93 IM |
522 | n_mbufs = dev->requested_n_rxq * dev->requested_rxq_size |
523 | + dev->requested_n_txq * dev->requested_txq_size | |
524 | + MIN(RTE_MAX_LCORE, dev->requested_n_rxq) * NETDEV_MAX_BURST | |
525 | + MIN_NB_MBUF; | |
d555d9bd | 526 | |
24e78f93 | 527 | ovs_mutex_lock(&dpdk_mp_mutex); |
da79ce2b | 528 | do { |
24e78f93 IM |
529 | /* Full DPDK memory pool name must be unique and cannot be |
530 | * longer than RTE_MEMPOOL_NAMESIZE. */ | |
af5b0dad IM |
531 | int ret = snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, |
532 | "ovs%08x%02d%05d%07u", | |
24e78f93 IM |
533 | hash, socket_id, mtu, n_mbufs); |
534 | if (ret < 0 || ret >= RTE_MEMPOOL_NAMESIZE) { | |
535 | VLOG_DBG("snprintf returned %d. " | |
536 | "Failed to generate a mempool name for \"%s\". " | |
537 | "Hash:0x%x, socket_id: %d, mtu:%d, mbufs:%u.", | |
538 | ret, netdev_name, hash, socket_id, mtu, n_mbufs); | |
539 | break; | |
65056fd7 | 540 | } |
95fb793a | 541 | |
f06546a5 FA |
542 | VLOG_DBG("Port %s: Requesting a mempool of %u mbufs " |
543 | "on socket %d for %d Rx and %d Tx queues.", | |
24e78f93 | 544 | netdev_name, n_mbufs, socket_id, |
f06546a5 | 545 | dev->requested_n_rxq, dev->requested_n_txq); |
d555d9bd | 546 | |
24e78f93 IM |
547 | mp = rte_pktmbuf_pool_create(mp_name, n_mbufs, MP_CACHE_SZ, |
548 | sizeof (struct dp_packet) - sizeof (struct rte_mbuf), | |
549 | MBUF_SIZE(mtu) - sizeof(struct dp_packet), socket_id); | |
550 | ||
551 | if (mp) { | |
552 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", | |
553 | mp_name, n_mbufs); | |
837c1761 FA |
554 | /* rte_pktmbuf_pool_create has done some initialization of the |
555 | * rte_mbuf part of each dp_packet. Some OvS specific fields | |
556 | * of the packet still need to be initialized by | |
557 | * ovs_rte_pktmbuf_init. */ | |
24e78f93 | 558 | rte_mempool_obj_iter(mp, ovs_rte_pktmbuf_init, NULL); |
d555d9bd RW |
559 | } else if (rte_errno == EEXIST) { |
560 | /* A mempool with the same name already exists. We just | |
561 | * retrieve its pointer to be returned to the caller. */ | |
24e78f93 | 562 | mp = rte_mempool_lookup(mp_name); |
d555d9bd RW |
563 | /* As the mempool create returned EEXIST we can expect the |
564 | * lookup has returned a valid pointer. If for some reason | |
565 | * that's not the case we keep track of it. */ | |
24e78f93 IM |
566 | VLOG_DBG("A mempool with name \"%s\" already exists at %p.", |
567 | mp_name, mp); | |
d555d9bd RW |
568 | } else { |
569 | VLOG_ERR("Failed mempool \"%s\" create request of %u mbufs", | |
24e78f93 | 570 | mp_name, n_mbufs); |
0c6f39e5 | 571 | } |
24e78f93 | 572 | } while (!mp && rte_errno == ENOMEM && (n_mbufs /= 2) >= MIN_NB_MBUF); |
2ae3d542 | 573 | |
c2adb102 | 574 | ovs_mutex_unlock(&dpdk_mp_mutex); |
24e78f93 | 575 | return mp; |
8a9562d2 PS |
576 | } |
577 | ||
a08a115d | 578 | /* Release an existing mempool. */ |
8a9562d2 | 579 | static void |
24e78f93 | 580 | dpdk_mp_free(struct rte_mempool *mp) |
8a9562d2 | 581 | { |
24e78f93 | 582 | if (!mp) { |
8a9562d2 PS |
583 | return; |
584 | } | |
585 | ||
c2adb102 | 586 | ovs_mutex_lock(&dpdk_mp_mutex); |
24e78f93 IM |
587 | VLOG_DBG("Releasing \"%s\" mempool", mp->name); |
588 | rte_mempool_free(mp); | |
c2adb102 | 589 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
590 | } |
591 | ||
b6b26021 FA |
592 | /* Tries to allocate a new mempool - or re-use an existing one where |
593 | * appropriate - on requested_socket_id with a size determined by | |
594 | * requested_mtu and requested Rx/Tx queues. | |
595 | * On success - or when re-using an existing mempool - the new configuration | |
596 | * will be applied. | |
0072e931 MK |
597 | * On error, device will be left unchanged. */ |
598 | static int | |
599 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
600 | OVS_REQUIRES(dev->mutex) |
601 | { | |
602 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
24e78f93 IM |
603 | struct rte_mempool *mp; |
604 | int ret = 0; | |
0072e931 | 605 | |
24e78f93 | 606 | mp = dpdk_mp_create(dev, FRAME_LEN_TO_MTU(buf_size)); |
0072e931 | 607 | if (!mp) { |
c67e46c0 MK |
608 | VLOG_ERR("Failed to create memory pool for netdev " |
609 | "%s, with MTU %d on socket %d: %s\n", | |
610 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
611 | rte_strerror(rte_errno)); | |
24e78f93 | 612 | ret = rte_errno; |
0072e931 | 613 | } else { |
24e78f93 IM |
614 | /* If a new MTU was requested and its rounded value equals the one |
615 | * that is currently used, then the existing mempool is returned. */ | |
616 | if (dev->mp != mp) { | |
617 | /* A new mempool was created, release the previous one. */ | |
618 | dpdk_mp_free(dev->mp); | |
619 | } else { | |
620 | ret = EEXIST; | |
621 | } | |
622 | dev->mp = mp; | |
0072e931 MK |
623 | dev->mtu = dev->requested_mtu; |
624 | dev->socket_id = dev->requested_socket_id; | |
625 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
626 | } | |
627 | ||
24e78f93 | 628 | return ret; |
0072e931 MK |
629 | } |
630 | ||
8a9562d2 PS |
631 | static void |
632 | check_link_status(struct netdev_dpdk *dev) | |
633 | { | |
634 | struct rte_eth_link link; | |
635 | ||
636 | rte_eth_link_get_nowait(dev->port_id, &link); | |
637 | ||
638 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 639 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
640 | |
641 | dev->link_reset_cnt++; | |
642 | dev->link = link; | |
643 | if (dev->link.link_status) { | |
bb37956a | 644 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Up - speed %u Mbps - %s", |
58be5c0e | 645 | dev->port_id, (unsigned) dev->link.link_speed, |
8a9562d2 PS |
646 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? |
647 | ("full-duplex") : ("half-duplex")); | |
648 | } else { | |
bb37956a | 649 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Down", dev->port_id); |
8a9562d2 PS |
650 | } |
651 | } | |
652 | } | |
653 | ||
654 | static void * | |
655 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
656 | { | |
657 | struct netdev_dpdk *dev; | |
658 | ||
659 | pthread_detach(pthread_self()); | |
660 | ||
661 | for (;;) { | |
662 | ovs_mutex_lock(&dpdk_mutex); | |
663 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
664 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
665 | if (dev->type == DPDK_DEV_ETH) { |
666 | check_link_status(dev); | |
667 | } | |
8a9562d2 PS |
668 | ovs_mutex_unlock(&dev->mutex); |
669 | } | |
670 | ovs_mutex_unlock(&dpdk_mutex); | |
671 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
672 | } | |
673 | ||
674 | return NULL; | |
675 | } | |
676 | ||
b98d7669 DDP |
677 | static int |
678 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
679 | { | |
680 | int diag = 0; | |
681 | int i; | |
0072e931 | 682 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 683 | |
67fe6d63 MK |
684 | /* For some NICs (e.g. Niantic), scatter_rx mode needs to be explicitly |
685 | * enabled. */ | |
0072e931 | 686 | if (dev->mtu > ETHER_MTU) { |
67fe6d63 | 687 | conf.rxmode.enable_scatter = 1; |
0072e931 | 688 | } |
67fe6d63 | 689 | |
1a2bb118 SC |
690 | conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & |
691 | NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
b98d7669 DDP |
692 | /* A device may report more queues than it makes available (this has |
693 | * been observed for Intel xl710, which reserves some of them for | |
694 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
695 | * available. When this happens we can retry the configuration | |
696 | * and request less queues */ | |
697 | while (n_rxq && n_txq) { | |
698 | if (diag) { | |
699 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
700 | } | |
701 | ||
0072e931 | 702 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 703 | if (diag) { |
0072e931 MK |
704 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
705 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
706 | break; |
707 | } | |
708 | ||
67fe6d63 MK |
709 | diag = rte_eth_dev_set_mtu(dev->port_id, dev->mtu); |
710 | if (diag) { | |
711 | VLOG_ERR("Interface %s MTU (%d) setup error: %s", | |
712 | dev->up.name, dev->mtu, rte_strerror(-diag)); | |
713 | break; | |
714 | } | |
715 | ||
b98d7669 | 716 | for (i = 0; i < n_txq; i++) { |
b685696b | 717 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
718 | dev->socket_id, NULL); |
719 | if (diag) { | |
720 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
721 | dev->up.name, i, rte_strerror(-diag)); | |
722 | break; | |
723 | } | |
724 | } | |
725 | ||
726 | if (i != n_txq) { | |
727 | /* Retry with less tx queues */ | |
728 | n_txq = i; | |
729 | continue; | |
730 | } | |
731 | ||
732 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 733 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
24e78f93 | 734 | dev->socket_id, NULL, dev->mp); |
b98d7669 DDP |
735 | if (diag) { |
736 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
737 | dev->up.name, i, rte_strerror(-diag)); | |
738 | break; | |
739 | } | |
740 | } | |
741 | ||
742 | if (i != n_rxq) { | |
743 | /* Retry with less rx queues */ | |
744 | n_rxq = i; | |
745 | continue; | |
746 | } | |
747 | ||
748 | dev->up.n_rxq = n_rxq; | |
81acebda | 749 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
750 | |
751 | return 0; | |
752 | } | |
753 | ||
754 | return diag; | |
755 | } | |
756 | ||
9fd39370 SC |
757 | static void |
758 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
759 | { | |
760 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
bb37956a IM |
761 | VLOG_WARN("Failed to enable flow control on device %"PRIu8, |
762 | dev->port_id); | |
9fd39370 SC |
763 | } |
764 | } | |
b98d7669 | 765 | |
8a9562d2 | 766 | static int |
c2adb102 IM |
767 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
768 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
769 | { |
770 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 771 | struct rte_eth_dev_info info; |
8a9562d2 PS |
772 | struct ether_addr eth_addr; |
773 | int diag; | |
b98d7669 | 774 | int n_rxq, n_txq; |
d4f5282c KT |
775 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | |
776 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
777 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
8a9562d2 | 778 | |
a0cb2d66 | 779 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 780 | |
d4f5282c KT |
781 | if ((info.rx_offload_capa & rx_chksm_offload_capa) != |
782 | rx_chksm_offload_capa) { | |
2cfe866f | 783 | VLOG_WARN("Rx checksum offload is not supported on port %"PRIu8, |
d4f5282c KT |
784 | dev->port_id); |
785 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; | |
786 | } else { | |
787 | dev->hw_ol_features |= NETDEV_RX_CHECKSUM_OFFLOAD; | |
788 | } | |
789 | ||
b98d7669 DDP |
790 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
791 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
792 | ||
793 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 794 | if (diag) { |
b98d7669 DDP |
795 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
796 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 797 | return -diag; |
8a9562d2 PS |
798 | } |
799 | ||
8a9562d2 PS |
800 | diag = rte_eth_dev_start(dev->port_id); |
801 | if (diag) { | |
b98d7669 DDP |
802 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
803 | rte_strerror(-diag)); | |
95fb793a | 804 | return -diag; |
8a9562d2 PS |
805 | } |
806 | ||
807 | rte_eth_promiscuous_enable(dev->port_id); | |
808 | rte_eth_allmulticast_enable(dev->port_id); | |
809 | ||
810 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
811 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
bb37956a | 812 | VLOG_INFO_RL(&rl, "Port %"PRIu8": "ETH_ADDR_FMT, |
ca92d173 | 813 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 814 | |
ca92d173 | 815 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
816 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
817 | ||
24e78f93 | 818 | mbp_priv = rte_mempool_get_priv(dev->mp); |
8a9562d2 PS |
819 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; |
820 | ||
9fd39370 SC |
821 | /* Get the Flow control configuration for DPDK-ETH */ |
822 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
823 | if (diag) { | |
bb37956a | 824 | VLOG_DBG("cannot get flow control parameters on port=%"PRIu8", err=%d", |
9fd39370 SC |
825 | dev->port_id, diag); |
826 | } | |
827 | ||
8a9562d2 PS |
828 | return 0; |
829 | } | |
830 | ||
831 | static struct netdev_dpdk * | |
832 | netdev_dpdk_cast(const struct netdev *netdev) | |
833 | { | |
834 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
835 | } | |
836 | ||
837 | static struct netdev * | |
838 | netdev_dpdk_alloc(void) | |
839 | { | |
bab69409 AC |
840 | struct netdev_dpdk *dev; |
841 | ||
65e19e70 DDP |
842 | dev = dpdk_rte_mzalloc(sizeof *dev); |
843 | if (dev) { | |
844 | return &dev->up; | |
bab69409 | 845 | } |
65e19e70 | 846 | |
bab69409 | 847 | return NULL; |
8a9562d2 PS |
848 | } |
849 | ||
eff23640 DDP |
850 | static struct dpdk_tx_queue * |
851 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 852 | { |
eff23640 | 853 | struct dpdk_tx_queue *txqs; |
bd5131ba | 854 | unsigned i; |
5a034064 | 855 | |
eff23640 DDP |
856 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
857 | if (txqs) { | |
858 | for (i = 0; i < n_txqs; i++) { | |
859 | /* Initialize map for vhost devices. */ | |
860 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
861 | rte_spinlock_init(&txqs[i].tx_lock); | |
862 | } | |
5a034064 | 863 | } |
eff23640 DDP |
864 | |
865 | return txqs; | |
5a034064 AW |
866 | } |
867 | ||
8a9562d2 | 868 | static int |
bb37956a | 869 | common_construct(struct netdev *netdev, dpdk_port_t port_no, |
1ce30dfd | 870 | enum dpdk_dev_type type, int socket_id) |
5a034064 | 871 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 872 | { |
d46285a2 | 873 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 874 | |
d46285a2 | 875 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 876 | |
d46285a2 | 877 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 878 | |
1b7a04e0 AW |
879 | /* If the 'sid' is negative, it means that the kernel fails |
880 | * to obtain the pci numa info. In that situation, always | |
881 | * use 'SOCKET0'. */ | |
1ce30dfd | 882 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 883 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
884 | dev->port_id = port_no; |
885 | dev->type = type; | |
886 | dev->flags = 0; | |
7f381c2e | 887 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 888 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
889 | ovsrcu_index_init(&dev->vid, -1); |
890 | dev->vhost_reconfigured = false; | |
5dcde09c | 891 | dev->attached = false; |
8a9562d2 | 892 | |
78bd47cf | 893 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 894 | |
9509913a IS |
895 | ovsrcu_init(&dev->ingress_policer, NULL); |
896 | dev->policer_rate = 0; | |
897 | dev->policer_burst = 0; | |
898 | ||
7f381c2e DDP |
899 | netdev->n_rxq = 0; |
900 | netdev->n_txq = 0; | |
901 | dev->requested_n_rxq = NR_QUEUE; | |
902 | dev->requested_n_txq = NR_QUEUE; | |
903 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
904 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 905 | |
9fd39370 SC |
906 | /* Initialize the flow control to NULL */ |
907 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
908 | |
909 | /* Initilize the hardware offload flags to 0 */ | |
910 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
911 | |
912 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
913 | ||
d46285a2 | 914 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 915 | |
7f381c2e DDP |
916 | netdev_request_reconfigure(netdev); |
917 | ||
1ce30dfd | 918 | return 0; |
95fb793a | 919 | } |
920 | ||
b83a2df1 MV |
921 | /* dev_name must be the prefix followed by a positive decimal number. |
922 | * (no leading + or - signs are allowed) */ | |
95fb793a | 923 | static int |
924 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
925 | unsigned int *port_no) | |
926 | { | |
927 | const char *cport; | |
928 | ||
929 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
930 | return ENODEV; | |
931 | } | |
932 | ||
933 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
934 | |
935 | if (str_to_uint(cport, 10, port_no)) { | |
936 | return 0; | |
937 | } else { | |
938 | return ENODEV; | |
939 | } | |
95fb793a | 940 | } |
941 | ||
1ce30dfd DDP |
942 | static int |
943 | vhost_common_construct(struct netdev *netdev) | |
944 | OVS_REQUIRES(dpdk_mutex) | |
945 | { | |
946 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
947 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
948 | ||
949 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); | |
950 | if (!dev->tx_q) { | |
951 | return ENOMEM; | |
952 | } | |
953 | ||
bb37956a IM |
954 | return common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
955 | DPDK_DEV_VHOST, socket_id); | |
1ce30dfd DDP |
956 | } |
957 | ||
7d1ced01 | 958 | static int |
53f50d24 | 959 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 960 | { |
d46285a2 DDP |
961 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
962 | const char *name = netdev->name; | |
7d1ced01 | 963 | int err; |
a0cb2d66 | 964 | |
1af27e8a DDP |
965 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
966 | * the file system. '/' or '\' would traverse directories, so they're not | |
967 | * acceptable in 'name'. */ | |
968 | if (strchr(name, '/') || strchr(name, '\\')) { | |
969 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
970 | "A valid name must not include '/' or '\\'", | |
971 | name); | |
972 | return EINVAL; | |
973 | } | |
974 | ||
7d1ced01 CL |
975 | ovs_mutex_lock(&dpdk_mutex); |
976 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 977 | * the socket is to be created, then register the socket. |
7d1ced01 | 978 | */ |
2d24d165 | 979 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 980 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 981 | |
2d24d165 CL |
982 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
983 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
984 | if (err) { |
985 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 986 | dev->vhost_id); |
f3e7ec25 | 987 | goto out; |
e5c0f5a4 | 988 | } else { |
2d24d165 CL |
989 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
990 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
991 | dev->vhost_id, name); | |
992 | } | |
f3e7ec25 MW |
993 | |
994 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
995 | &virtio_net_device_ops); | |
996 | if (err) { | |
997 | VLOG_ERR("rte_vhost_driver_callback_register failed for vhost user " | |
998 | "port: %s\n", name); | |
999 | goto out; | |
1000 | } | |
1001 | ||
1002 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
1003 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
1004 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
1005 | | 1ULL << VIRTIO_NET_F_CSUM); | |
1006 | if (err) { | |
1007 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
1008 | "port: %s\n", name); | |
1009 | goto out; | |
1010 | } | |
1011 | ||
1012 | err = rte_vhost_driver_start(dev->vhost_id); | |
1013 | if (err) { | |
1014 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
1015 | "port: %s\n", name); | |
1016 | goto out; | |
1017 | } | |
1018 | ||
1ce30dfd | 1019 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1020 | if (err) { |
1021 | VLOG_ERR("vhost_common_construct failed for vhost user " | |
1022 | "port: %s\n", name); | |
1023 | } | |
2d24d165 | 1024 | |
f3e7ec25 | 1025 | out: |
2d24d165 | 1026 | ovs_mutex_unlock(&dpdk_mutex); |
28ca969e AC |
1027 | VLOG_WARN_ONCE("dpdkvhostuser ports are considered deprecated; " |
1028 | "please migrate to dpdkvhostuserclient ports."); | |
2d24d165 CL |
1029 | return err; |
1030 | } | |
1031 | ||
1032 | static int | |
1033 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
1034 | { | |
1035 | int err; | |
1036 | ||
2d24d165 | 1037 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 1038 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1039 | if (err) { |
1040 | VLOG_ERR("vhost_common_construct failed for vhost user client" | |
1041 | "port: %s\n", netdev->name); | |
1042 | } | |
7d1ced01 | 1043 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
1044 | return err; |
1045 | } | |
1046 | ||
95fb793a | 1047 | static int |
1048 | netdev_dpdk_construct(struct netdev *netdev) | |
1049 | { | |
95fb793a | 1050 | int err; |
1051 | ||
95fb793a | 1052 | ovs_mutex_lock(&dpdk_mutex); |
bb37956a IM |
1053 | err = common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1054 | DPDK_DEV_ETH, SOCKET0); | |
8a9562d2 PS |
1055 | ovs_mutex_unlock(&dpdk_mutex); |
1056 | return err; | |
1057 | } | |
1058 | ||
1ce30dfd DDP |
1059 | static void |
1060 | common_destruct(struct netdev_dpdk *dev) | |
1061 | OVS_REQUIRES(dpdk_mutex) | |
1062 | OVS_EXCLUDED(dev->mutex) | |
1063 | { | |
1064 | rte_free(dev->tx_q); | |
24e78f93 | 1065 | dpdk_mp_free(dev->mp); |
1ce30dfd DDP |
1066 | |
1067 | ovs_list_remove(&dev->list_node); | |
1068 | free(ovsrcu_get_protected(struct ingress_policer *, | |
1069 | &dev->ingress_policer)); | |
1070 | ovs_mutex_destroy(&dev->mutex); | |
1071 | } | |
1072 | ||
8a9562d2 | 1073 | static void |
d46285a2 | 1074 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 1075 | { |
d46285a2 | 1076 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5dcde09c | 1077 | char devname[RTE_ETH_NAME_MAX_LEN]; |
8a9562d2 | 1078 | |
8d38823b | 1079 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1080 | |
8a9562d2 | 1081 | rte_eth_dev_stop(dev->port_id); |
5dcde09c IM |
1082 | |
1083 | if (dev->attached) { | |
1084 | rte_eth_dev_close(dev->port_id); | |
1085 | if (rte_eth_dev_detach(dev->port_id, devname) < 0) { | |
1086 | VLOG_ERR("Device '%s' can not be detached", dev->devargs); | |
1087 | } else { | |
0ee821c2 | 1088 | VLOG_INFO("Device '%s' has been detached", devname); |
5dcde09c IM |
1089 | } |
1090 | } | |
1091 | ||
55e075e6 | 1092 | free(dev->devargs); |
1ce30dfd | 1093 | common_destruct(dev); |
8d38823b | 1094 | |
8a9562d2 | 1095 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1096 | } |
8a9562d2 | 1097 | |
3f891bbe DDP |
1098 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1099 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1100 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1101 | static int | |
c1ff66ac CL |
1102 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1103 | char *vhost_id) | |
3f891bbe DDP |
1104 | OVS_EXCLUDED(dpdk_mutex) |
1105 | OVS_EXCLUDED(dev->mutex) | |
1106 | { | |
c1ff66ac | 1107 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1108 | } |
1109 | ||
58397e6c | 1110 | static void |
d46285a2 | 1111 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1112 | { |
d46285a2 | 1113 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1114 | char *vhost_id; |
58397e6c | 1115 | |
8d38823b | 1116 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1117 | |
c62da695 | 1118 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1119 | if (netdev_dpdk_get_vid(dev) >= 0 |
1120 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1121 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1122 | netdev->name); |
58be5c0e MK |
1123 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1124 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1125 | } |
1126 | ||
2d24d165 | 1127 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1128 | |
1ce30dfd DDP |
1129 | common_destruct(dev); |
1130 | ||
58397e6c | 1131 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1132 | |
569c26da | 1133 | if (!vhost_id[0]) { |
821b8664 IM |
1134 | goto out; |
1135 | } | |
1136 | ||
c1ff66ac | 1137 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1138 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1139 | netdev->name, vhost_id); | |
c1ff66ac CL |
1140 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1141 | /* OVS server mode - remove this socket from list for deletion */ | |
1142 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1143 | } |
821b8664 | 1144 | out: |
c1ff66ac | 1145 | free(vhost_id); |
8a9562d2 PS |
1146 | } |
1147 | ||
1148 | static void | |
d46285a2 | 1149 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1150 | { |
d46285a2 | 1151 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1152 | |
d46285a2 | 1153 | rte_free(dev); |
8a9562d2 PS |
1154 | } |
1155 | ||
1156 | static int | |
a14b8947 | 1157 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1158 | { |
a14b8947 | 1159 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1160 | |
1161 | ovs_mutex_lock(&dev->mutex); | |
1162 | ||
050c60bf | 1163 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1164 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1165 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1166 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1167 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1168 | |
1169 | if (dev->type == DPDK_DEV_ETH) { | |
1170 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1171 | dev->requested_rxq_size); | |
1172 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1173 | dev->rxq_size); | |
1174 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1175 | dev->requested_txq_size); | |
1176 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1177 | dev->txq_size); | |
1a2bb118 SC |
1178 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1179 | smap_add(args, "rx_csum_offload", "true"); | |
8155ab7e KT |
1180 | } else { |
1181 | smap_add(args, "rx_csum_offload", "false"); | |
1a2bb118 | 1182 | } |
451f26fd | 1183 | } |
8a9562d2 PS |
1184 | ovs_mutex_unlock(&dev->mutex); |
1185 | ||
1186 | return 0; | |
1187 | } | |
1188 | ||
55e075e6 | 1189 | static struct netdev_dpdk * |
bb37956a | 1190 | netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id) |
55e075e6 CL |
1191 | OVS_REQUIRES(dpdk_mutex) |
1192 | { | |
1193 | struct netdev_dpdk *dev; | |
1194 | ||
1195 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1196 | if (dev->port_id == port_id) { | |
1197 | return dev; | |
1198 | } | |
1199 | } | |
1200 | ||
1201 | return NULL; | |
1202 | } | |
1203 | ||
bb37956a | 1204 | static dpdk_port_t |
5dcde09c IM |
1205 | netdev_dpdk_process_devargs(struct netdev_dpdk *dev, |
1206 | const char *devargs, char **errp) | |
55e075e6 | 1207 | { |
03d6399e IM |
1208 | /* Get the name up to the first comma. */ |
1209 | char *name = xmemdup0(devargs, strcspn(devargs, ",")); | |
bb37956a | 1210 | dpdk_port_t new_port_id = DPDK_ETH_PORT_ID_INVALID; |
55e075e6 | 1211 | |
255b7bda | 1212 | if (rte_eth_dev_get_port_by_name(name, &new_port_id) |
69876ed7 CL |
1213 | || !rte_eth_dev_is_valid_port(new_port_id)) { |
1214 | /* Device not found in DPDK, attempt to attach it */ | |
1215 | if (!rte_eth_dev_attach(devargs, &new_port_id)) { | |
1216 | /* Attach successful */ | |
5dcde09c | 1217 | dev->attached = true; |
69876ed7 CL |
1218 | VLOG_INFO("Device '%s' attached to DPDK", devargs); |
1219 | } else { | |
1220 | /* Attach unsuccessful */ | |
bb37956a | 1221 | new_port_id = DPDK_ETH_PORT_ID_INVALID; |
47a45d86 KT |
1222 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", |
1223 | devargs); | |
55e075e6 CL |
1224 | } |
1225 | } | |
1226 | ||
03d6399e | 1227 | free(name); |
55e075e6 CL |
1228 | return new_port_id; |
1229 | } | |
1230 | ||
c3d062a7 CL |
1231 | static void |
1232 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1233 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1234 | { |
050c60bf | 1235 | int new_n_rxq; |
a14b8947 | 1236 | |
2a21e757 | 1237 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1238 | if (new_n_rxq != dev->requested_n_rxq) { |
1239 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1240 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1241 | } |
c3d062a7 CL |
1242 | } |
1243 | ||
b685696b CL |
1244 | static void |
1245 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1246 | const char *flag, int default_size, int *new_size) | |
1247 | { | |
1248 | int queue_size = smap_get_int(args, flag, default_size); | |
1249 | ||
1250 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1251 | || !is_pow2(queue_size)) { | |
1252 | queue_size = default_size; | |
1253 | } | |
1254 | ||
1255 | if (queue_size != *new_size) { | |
1256 | *new_size = queue_size; | |
1257 | netdev_request_reconfigure(netdev); | |
1258 | } | |
1259 | } | |
1260 | ||
c3d062a7 | 1261 | static int |
9fff138e DDP |
1262 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1263 | char **errp) | |
c3d062a7 CL |
1264 | { |
1265 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
b614c894 IM |
1266 | bool rx_fc_en, tx_fc_en, autoneg; |
1267 | enum rte_eth_fc_mode fc_mode; | |
1268 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1269 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1270 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1271 | }; | |
55e075e6 CL |
1272 | const char *new_devargs; |
1273 | int err = 0; | |
c3d062a7 | 1274 | |
55e075e6 | 1275 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1276 | ovs_mutex_lock(&dev->mutex); |
1277 | ||
1278 | dpdk_set_rxq_config(dev, args); | |
1279 | ||
b685696b CL |
1280 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1281 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1282 | &dev->requested_rxq_size); | |
1283 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1284 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1285 | &dev->requested_txq_size); | |
1286 | ||
55e075e6 CL |
1287 | new_devargs = smap_get(args, "dpdk-devargs"); |
1288 | ||
1289 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1290 | /* The user requested a new device. If we return error, the caller | |
1291 | * will delete this netdev and try to recreate it. */ | |
1292 | err = EAGAIN; | |
1293 | goto out; | |
1294 | } | |
1295 | ||
1296 | /* dpdk-devargs is required for device configuration */ | |
1297 | if (new_devargs && new_devargs[0]) { | |
1298 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1299 | * is valid */ | |
1300 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1301 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
bb37956a IM |
1302 | dpdk_port_t new_port_id = netdev_dpdk_process_devargs(dev, |
1303 | new_devargs, | |
1304 | errp); | |
55e075e6 CL |
1305 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1306 | err = EINVAL; | |
1307 | } else if (new_port_id == dev->port_id) { | |
1308 | /* Already configured, do not reconfigure again */ | |
1309 | err = 0; | |
1310 | } else { | |
1311 | struct netdev_dpdk *dup_dev; | |
bb37956a | 1312 | |
55e075e6 CL |
1313 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); |
1314 | if (dup_dev) { | |
9fff138e DDP |
1315 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
1316 | "which is already in use by '%s'", | |
1317 | netdev_get_name(netdev), new_devargs, | |
1318 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1319 | err = EADDRINUSE; |
1320 | } else { | |
bd4e172b | 1321 | int sid = rte_eth_dev_socket_id(new_port_id); |
bb37956a | 1322 | |
bd4e172b | 1323 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; |
55e075e6 CL |
1324 | dev->devargs = xstrdup(new_devargs); |
1325 | dev->port_id = new_port_id; | |
1326 | netdev_request_reconfigure(&dev->up); | |
1327 | err = 0; | |
1328 | } | |
1329 | } | |
1330 | } | |
1331 | } else { | |
9fff138e DDP |
1332 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1333 | "The old 'dpdk<port_id>' names are not supported", | |
1334 | netdev_get_name(netdev)); | |
55e075e6 CL |
1335 | err = EINVAL; |
1336 | } | |
1337 | ||
1338 | if (err) { | |
1339 | goto out; | |
1340 | } | |
1341 | ||
c3d062a7 CL |
1342 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1343 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1344 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1345 | |
b614c894 IM |
1346 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1347 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1348 | dev->fc_conf.mode = fc_mode; | |
1349 | dev->fc_conf.autoneg = autoneg; | |
1350 | dpdk_eth_flow_ctrl_setup(dev); | |
1351 | } | |
9fd39370 | 1352 | |
55e075e6 | 1353 | out: |
c3d062a7 | 1354 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1355 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1356 | |
55e075e6 | 1357 | return err; |
c3d062a7 CL |
1358 | } |
1359 | ||
1360 | static int | |
9fff138e DDP |
1361 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1362 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1363 | { |
1364 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1365 | ||
1366 | ovs_mutex_lock(&dev->mutex); | |
1367 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1368 | ovs_mutex_unlock(&dev->mutex); |
1369 | ||
1370 | return 0; | |
1371 | } | |
1372 | ||
c1ff66ac | 1373 | static int |
2d24d165 | 1374 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1375 | const struct smap *args, |
1376 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1377 | { |
1378 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1379 | const char *path; | |
1380 | ||
6881885a | 1381 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1382 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1383 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1384 | if (path && strcmp(path, dev->vhost_id)) { |
1385 | strcpy(dev->vhost_id, path); | |
c1ff66ac CL |
1386 | netdev_request_reconfigure(netdev); |
1387 | } | |
1388 | } | |
6881885a | 1389 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1390 | |
1391 | return 0; | |
1392 | } | |
1393 | ||
7dec44fe | 1394 | static int |
d46285a2 | 1395 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1396 | { |
d46285a2 | 1397 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1398 | |
d46285a2 | 1399 | return dev->socket_id; |
7dec44fe AW |
1400 | } |
1401 | ||
050c60bf | 1402 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1403 | static int |
050c60bf | 1404 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1405 | { |
d46285a2 | 1406 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1407 | |
d46285a2 | 1408 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1409 | |
050c60bf DDP |
1410 | if (dev->requested_n_txq == n_txq) { |
1411 | goto out; | |
4573fbd3 FL |
1412 | } |
1413 | ||
050c60bf DDP |
1414 | dev->requested_n_txq = n_txq; |
1415 | netdev_request_reconfigure(netdev); | |
58397e6c | 1416 | |
050c60bf | 1417 | out: |
d46285a2 | 1418 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1419 | return 0; |
58397e6c KT |
1420 | } |
1421 | ||
8a9562d2 PS |
1422 | static struct netdev_rxq * |
1423 | netdev_dpdk_rxq_alloc(void) | |
1424 | { | |
1425 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1426 | ||
eff23640 DDP |
1427 | if (rx) { |
1428 | return &rx->up; | |
1429 | } | |
1430 | ||
1431 | return NULL; | |
8a9562d2 PS |
1432 | } |
1433 | ||
1434 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1435 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1436 | { |
d46285a2 | 1437 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1438 | } |
1439 | ||
1440 | static int | |
d46285a2 | 1441 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1442 | { |
d46285a2 DDP |
1443 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1444 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1445 | |
d46285a2 DDP |
1446 | ovs_mutex_lock(&dev->mutex); |
1447 | rx->port_id = dev->port_id; | |
1448 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1449 | |
1450 | return 0; | |
1451 | } | |
1452 | ||
1453 | static void | |
d46285a2 | 1454 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1455 | { |
1456 | } | |
1457 | ||
1458 | static void | |
d46285a2 | 1459 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1460 | { |
d46285a2 | 1461 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1462 | |
1463 | rte_free(rx); | |
1464 | } | |
1465 | ||
819f13bd DDP |
1466 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1467 | * 'pkts', even in case of failure. | |
1468 | * | |
1469 | * Returns the number of packets that weren't transmitted. */ | |
1470 | static inline int | |
b59cc14e | 1471 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 1472 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 1473 | { |
1304f1f8 DDP |
1474 | uint32_t nb_tx = 0; |
1475 | ||
b59cc14e | 1476 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1477 | uint32_t ret; |
1478 | ||
b59cc14e | 1479 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1480 | if (!ret) { |
1481 | break; | |
1482 | } | |
1483 | ||
1484 | nb_tx += ret; | |
1485 | } | |
8a9562d2 | 1486 | |
b59cc14e | 1487 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 1488 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
1489 | * packet could come from a different mempool) */ |
1490 | int i; | |
1491 | ||
b59cc14e IM |
1492 | for (i = nb_tx; i < cnt; i++) { |
1493 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1494 | } |
8a9562d2 | 1495 | } |
819f13bd DDP |
1496 | |
1497 | return cnt - nb_tx; | |
8a9562d2 PS |
1498 | } |
1499 | ||
f3926f29 IS |
1500 | static inline bool |
1501 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1502 | struct rte_mbuf *pkt, uint64_t time) | |
1503 | { | |
1504 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1505 | ||
1506 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1507 | e_RTE_METER_GREEN; | |
1508 | } | |
1509 | ||
1510 | static int | |
1511 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
3e90f7d7 GZ |
1512 | struct rte_mbuf **pkts, int pkt_cnt, |
1513 | bool may_steal) | |
f3926f29 IS |
1514 | { |
1515 | int i = 0; | |
1516 | int cnt = 0; | |
1517 | struct rte_mbuf *pkt = NULL; | |
1518 | uint64_t current_time = rte_rdtsc(); | |
1519 | ||
1520 | for (i = 0; i < pkt_cnt; i++) { | |
1521 | pkt = pkts[i]; | |
1522 | /* Handle current packet */ | |
1523 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1524 | if (cnt != i) { | |
1525 | pkts[cnt] = pkt; | |
1526 | } | |
1527 | cnt++; | |
1528 | } else { | |
3e90f7d7 GZ |
1529 | if (may_steal) { |
1530 | rte_pktmbuf_free(pkt); | |
1531 | } | |
f3926f29 IS |
1532 | } |
1533 | } | |
1534 | ||
1535 | return cnt; | |
1536 | } | |
1537 | ||
9509913a IS |
1538 | static int |
1539 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
3e90f7d7 | 1540 | int pkt_cnt, bool may_steal) |
9509913a IS |
1541 | { |
1542 | int cnt = 0; | |
1543 | ||
1544 | rte_spinlock_lock(&policer->policer_lock); | |
3e90f7d7 GZ |
1545 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, |
1546 | pkt_cnt, may_steal); | |
9509913a IS |
1547 | rte_spinlock_unlock(&policer->policer_lock); |
1548 | ||
1549 | return cnt; | |
1550 | } | |
1551 | ||
58397e6c | 1552 | static bool |
0a0f39df | 1553 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1554 | { |
0a0f39df | 1555 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1556 | } |
1557 | ||
d6e3feb5 | 1558 | static inline void |
1559 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1560 | unsigned int packet_size) | |
1561 | { | |
1562 | /* Hard-coded search for the size bucket. */ | |
1563 | if (packet_size < 256) { | |
1564 | if (packet_size >= 128) { | |
1565 | stats->rx_128_to_255_packets++; | |
1566 | } else if (packet_size <= 64) { | |
1567 | stats->rx_1_to_64_packets++; | |
1568 | } else { | |
1569 | stats->rx_65_to_127_packets++; | |
1570 | } | |
1571 | } else { | |
1572 | if (packet_size >= 1523) { | |
1573 | stats->rx_1523_to_max_packets++; | |
1574 | } else if (packet_size >= 1024) { | |
1575 | stats->rx_1024_to_1522_packets++; | |
1576 | } else if (packet_size < 512) { | |
1577 | stats->rx_256_to_511_packets++; | |
1578 | } else { | |
1579 | stats->rx_512_to_1023_packets++; | |
1580 | } | |
1581 | } | |
1582 | } | |
1583 | ||
9e3ddd45 TP |
1584 | static inline void |
1585 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1586 | struct dp_packet **packets, int count, |
1587 | int dropped) | |
9e3ddd45 TP |
1588 | { |
1589 | int i; | |
d6e3feb5 | 1590 | unsigned int packet_size; |
9e3ddd45 TP |
1591 | struct dp_packet *packet; |
1592 | ||
1593 | stats->rx_packets += count; | |
9509913a | 1594 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1595 | for (i = 0; i < count; i++) { |
1596 | packet = packets[i]; | |
d6e3feb5 | 1597 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1598 | |
d6e3feb5 | 1599 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1600 | /* This only protects the following multicast counting from |
1601 | * too short packets, but it does not stop the packet from | |
1602 | * further processing. */ | |
1603 | stats->rx_errors++; | |
1604 | stats->rx_length_errors++; | |
1605 | continue; | |
1606 | } | |
1607 | ||
d6e3feb5 | 1608 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1609 | ||
9e3ddd45 TP |
1610 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1611 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1612 | stats->multicast++; | |
1613 | } | |
1614 | ||
d6e3feb5 | 1615 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1616 | } |
1617 | } | |
1618 | ||
58397e6c KT |
1619 | /* |
1620 | * The receive path for the vhost port is the TX path out from guest. | |
1621 | */ | |
1622 | static int | |
d46285a2 | 1623 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1624 | struct dp_packet_batch *batch) |
58397e6c | 1625 | { |
d46285a2 | 1626 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
9509913a | 1627 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1628 | uint16_t nb_rx = 0; |
9509913a | 1629 | uint16_t dropped = 0; |
daf22bf7 IM |
1630 | int qid = rxq->queue_id; |
1631 | int vid = netdev_dpdk_get_vid(dev); | |
58397e6c | 1632 | |
daf22bf7 | 1633 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured |
e543851d | 1634 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1635 | return EAGAIN; |
1636 | } | |
1637 | ||
daf22bf7 | 1638 | nb_rx = rte_vhost_dequeue_burst(vid, qid * VIRTIO_QNUM + VIRTIO_TXQ, |
24e78f93 | 1639 | dev->mp, |
64839cf4 | 1640 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1641 | NETDEV_MAX_BURST); |
58397e6c KT |
1642 | if (!nb_rx) { |
1643 | return EAGAIN; | |
1644 | } | |
1645 | ||
9509913a IS |
1646 | if (policer) { |
1647 | dropped = nb_rx; | |
64839cf4 WT |
1648 | nb_rx = ingress_policer_run(policer, |
1649 | (struct rte_mbuf **) batch->packets, | |
3e90f7d7 | 1650 | nb_rx, true); |
9509913a IS |
1651 | dropped -= nb_rx; |
1652 | } | |
1653 | ||
d46285a2 | 1654 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1655 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1656 | nb_rx, dropped); | |
d46285a2 | 1657 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1658 | |
75fb9148 ZB |
1659 | batch->count = nb_rx; |
1660 | dp_packet_batch_init_packet_fields(batch); | |
1661 | ||
58397e6c KT |
1662 | return 0; |
1663 | } | |
1664 | ||
8a9562d2 | 1665 | static int |
64839cf4 | 1666 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1667 | { |
d46285a2 DDP |
1668 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1669 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1670 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1671 | int nb_rx; |
9509913a | 1672 | int dropped = 0; |
8a9562d2 | 1673 | |
3b1fb077 DDP |
1674 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1675 | return EAGAIN; | |
1676 | } | |
1677 | ||
d46285a2 | 1678 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1679 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1680 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1681 | if (!nb_rx) { |
1682 | return EAGAIN; | |
1683 | } | |
1684 | ||
9509913a IS |
1685 | if (policer) { |
1686 | dropped = nb_rx; | |
64839cf4 | 1687 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 1688 | (struct rte_mbuf **) batch->packets, |
3e90f7d7 | 1689 | nb_rx, true); |
9509913a IS |
1690 | dropped -= nb_rx; |
1691 | } | |
1692 | ||
1693 | /* Update stats to reflect dropped packets */ | |
1694 | if (OVS_UNLIKELY(dropped)) { | |
1695 | rte_spinlock_lock(&dev->stats_lock); | |
1696 | dev->stats.rx_dropped += dropped; | |
1697 | rte_spinlock_unlock(&dev->stats_lock); | |
1698 | } | |
1699 | ||
64839cf4 | 1700 | batch->count = nb_rx; |
75fb9148 | 1701 | dp_packet_batch_init_packet_fields(batch); |
8a9562d2 PS |
1702 | |
1703 | return 0; | |
1704 | } | |
1705 | ||
0bf765f7 | 1706 | static inline int |
78bd47cf | 1707 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
3e90f7d7 | 1708 | int cnt, bool may_steal) |
0bf765f7 | 1709 | { |
78bd47cf | 1710 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 1711 | |
78bd47cf DDP |
1712 | if (qos_conf) { |
1713 | rte_spinlock_lock(&qos_conf->lock); | |
3e90f7d7 | 1714 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt, may_steal); |
78bd47cf | 1715 | rte_spinlock_unlock(&qos_conf->lock); |
0bf765f7 IS |
1716 | } |
1717 | ||
1718 | return cnt; | |
1719 | } | |
1720 | ||
c6ec9d17 IM |
1721 | static int |
1722 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1723 | int pkt_cnt) | |
1724 | { | |
1725 | int i = 0; | |
1726 | int cnt = 0; | |
1727 | struct rte_mbuf *pkt; | |
1728 | ||
1729 | for (i = 0; i < pkt_cnt; i++) { | |
1730 | pkt = pkts[i]; | |
1731 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
1732 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
1733 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
1734 | rte_pktmbuf_free(pkt); | |
1735 | continue; | |
1736 | } | |
1737 | ||
1738 | if (OVS_UNLIKELY(i != cnt)) { | |
1739 | pkts[cnt] = pkt; | |
1740 | } | |
1741 | cnt++; | |
1742 | } | |
1743 | ||
1744 | return cnt; | |
1745 | } | |
1746 | ||
9e3ddd45 TP |
1747 | static inline void |
1748 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1749 | struct dp_packet **packets, | |
1750 | int attempted, | |
1751 | int dropped) | |
1752 | { | |
1753 | int i; | |
1754 | int sent = attempted - dropped; | |
1755 | ||
1756 | stats->tx_packets += sent; | |
1757 | stats->tx_dropped += dropped; | |
1758 | ||
1759 | for (i = 0; i < sent; i++) { | |
1760 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1761 | } | |
1762 | } | |
1763 | ||
58397e6c | 1764 | static void |
4573fbd3 | 1765 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 1766 | struct dp_packet **pkts, int cnt) |
58397e6c | 1767 | { |
d46285a2 | 1768 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1769 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1770 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 1771 | unsigned int dropped = 0; |
dd52de45 | 1772 | int i, retries = 0; |
daf22bf7 | 1773 | int vid = netdev_dpdk_get_vid(dev); |
58397e6c | 1774 | |
81acebda | 1775 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1776 | |
daf22bf7 | 1777 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured || qid < 0 |
e543851d | 1778 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1779 | rte_spinlock_lock(&dev->stats_lock); |
1780 | dev->stats.tx_dropped+= cnt; | |
1781 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1782 | goto out; |
58397e6c KT |
1783 | } |
1784 | ||
d46285a2 | 1785 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1786 | |
c6ec9d17 | 1787 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 1788 | /* Check has QoS has been configured for the netdev */ |
3e90f7d7 | 1789 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt, true); |
c6ec9d17 | 1790 | dropped = total_pkts - cnt; |
0bf765f7 | 1791 | |
95e9881f | 1792 | do { |
4573fbd3 | 1793 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1794 | unsigned int tx_pkts; |
1795 | ||
daf22bf7 | 1796 | tx_pkts = rte_vhost_enqueue_burst(vid, vhost_qid, cur_pkts, cnt); |
95e9881f KT |
1797 | if (OVS_LIKELY(tx_pkts)) { |
1798 | /* Packets have been sent.*/ | |
1799 | cnt -= tx_pkts; | |
31871ee3 | 1800 | /* Prepare for possible retry.*/ |
95e9881f KT |
1801 | cur_pkts = &cur_pkts[tx_pkts]; |
1802 | } else { | |
31871ee3 KT |
1803 | /* No packets sent - do not retry.*/ |
1804 | break; | |
95e9881f | 1805 | } |
c6ec9d17 | 1806 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1807 | |
d46285a2 | 1808 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1809 | |
d46285a2 | 1810 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 1811 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 1812 | cnt + dropped); |
d46285a2 | 1813 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1814 | |
1815 | out: | |
c6ec9d17 | 1816 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 1817 | dp_packet_delete(pkts[i]); |
58397e6c KT |
1818 | } |
1819 | } | |
1820 | ||
8a9562d2 PS |
1821 | /* Tx function. Transmit packets indefinitely */ |
1822 | static void | |
64839cf4 | 1823 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1824 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1825 | { |
8a14bd7b | 1826 | const size_t batch_cnt = dp_packet_batch_size(batch); |
bce01e3a | 1827 | #if !defined(__CHECKER__) && !defined(_WIN32) |
8a14bd7b | 1828 | const size_t PKT_ARRAY_SIZE = batch_cnt; |
bce01e3a EJ |
1829 | #else |
1830 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1831 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1832 | #endif |
8a9562d2 | 1833 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 1834 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
8a14bd7b | 1835 | uint32_t cnt = batch_cnt; |
3e90f7d7 GZ |
1836 | uint32_t dropped = 0; |
1837 | ||
1838 | if (dev->type != DPDK_DEV_VHOST) { | |
1839 | /* Check if QoS has been configured for this netdev. */ | |
1840 | cnt = netdev_dpdk_qos_run(dev, (struct rte_mbuf **) batch->packets, | |
8a14bd7b BB |
1841 | batch_cnt, false); |
1842 | dropped += batch_cnt - cnt; | |
3e90f7d7 | 1843 | } |
8a9562d2 | 1844 | |
7d6d1a40 WT |
1845 | dp_packet_batch_apply_cutlen(batch); |
1846 | ||
3e90f7d7 GZ |
1847 | uint32_t txcnt = 0; |
1848 | ||
1849 | for (uint32_t i = 0; i < cnt; i++) { | |
8a14bd7b BB |
1850 | struct dp_packet *packet = batch->packets[i]; |
1851 | uint32_t size = dp_packet_size(packet); | |
95fb793a | 1852 | |
f98d7864 | 1853 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
3e90f7d7 GZ |
1854 | VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d", |
1855 | size, dev->max_packet_len); | |
f4fd623c | 1856 | |
175cf4de | 1857 | dropped++; |
f4fd623c DDP |
1858 | continue; |
1859 | } | |
8a9562d2 | 1860 | |
24e78f93 | 1861 | pkts[txcnt] = rte_pktmbuf_alloc(dev->mp); |
8a14bd7b | 1862 | if (OVS_UNLIKELY(!pkts[txcnt])) { |
3e90f7d7 | 1863 | dropped += cnt - i; |
175cf4de | 1864 | break; |
f4fd623c DDP |
1865 | } |
1866 | ||
1867 | /* We have to do a copy for now */ | |
3e90f7d7 | 1868 | memcpy(rte_pktmbuf_mtod(pkts[txcnt], void *), |
8a14bd7b BB |
1869 | dp_packet_data(packet), size); |
1870 | dp_packet_set_size((struct dp_packet *)pkts[txcnt], size); | |
f4fd623c | 1871 | |
3e90f7d7 | 1872 | txcnt++; |
f4fd623c | 1873 | } |
8a9562d2 | 1874 | |
3e90f7d7 GZ |
1875 | if (OVS_LIKELY(txcnt)) { |
1876 | if (dev->type == DPDK_DEV_VHOST) { | |
1877 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, | |
1878 | txcnt); | |
1879 | } else { | |
1880 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, txcnt); | |
1881 | } | |
58397e6c | 1882 | } |
db73f716 | 1883 | |
0bf765f7 IS |
1884 | if (OVS_UNLIKELY(dropped)) { |
1885 | rte_spinlock_lock(&dev->stats_lock); | |
1886 | dev->stats.tx_dropped += dropped; | |
1887 | rte_spinlock_unlock(&dev->stats_lock); | |
1888 | } | |
8a9562d2 PS |
1889 | } |
1890 | ||
58397e6c | 1891 | static int |
64839cf4 WT |
1892 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1893 | struct dp_packet_batch *batch, | |
324c8374 | 1894 | bool may_steal, bool concurrent_txq OVS_UNUSED) |
58397e6c | 1895 | { |
58397e6c | 1896 | |
dd52de45 | 1897 | if (OVS_UNLIKELY(!may_steal || batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 WT |
1898 | dpdk_do_tx_copy(netdev, qid, batch); |
1899 | dp_packet_delete_batch(batch, may_steal); | |
58397e6c | 1900 | } else { |
64839cf4 | 1901 | dp_packet_batch_apply_cutlen(batch); |
dd52de45 | 1902 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
1903 | } |
1904 | return 0; | |
1905 | } | |
1906 | ||
7251515e DV |
1907 | static inline void |
1908 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
324c8374 IM |
1909 | struct dp_packet_batch *batch, bool may_steal, |
1910 | bool concurrent_txq) | |
8a9562d2 | 1911 | { |
3b1fb077 DDP |
1912 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1913 | dp_packet_delete_batch(batch, may_steal); | |
1914 | return; | |
1915 | } | |
1916 | ||
324c8374 | 1917 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 1918 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
1919 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
1920 | } | |
1921 | ||
7251515e | 1922 | if (OVS_UNLIKELY(!may_steal || |
64839cf4 | 1923 | batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1924 | struct netdev *netdev = &dev->up; |
1925 | ||
64839cf4 WT |
1926 | dpdk_do_tx_copy(netdev, qid, batch); |
1927 | dp_packet_delete_batch(batch, may_steal); | |
8a9562d2 | 1928 | } else { |
fd57eeba BB |
1929 | int tx_cnt, dropped; |
1930 | int batch_cnt = dp_packet_batch_size(batch); | |
2391135c | 1931 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 1932 | |
7d6d1a40 WT |
1933 | dp_packet_batch_apply_cutlen(batch); |
1934 | ||
fd57eeba BB |
1935 | tx_cnt = netdev_dpdk_filter_packet_len(dev, pkts, batch_cnt); |
1936 | tx_cnt = netdev_dpdk_qos_run(dev, pkts, tx_cnt, true); | |
1937 | dropped = batch_cnt - tx_cnt; | |
1b99bb05 | 1938 | |
fd57eeba | 1939 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, tx_cnt); |
8a9562d2 | 1940 | |
f4fd623c | 1941 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1942 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1943 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1944 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1945 | } |
8a9562d2 | 1946 | } |
a0cb2d66 | 1947 | |
324c8374 | 1948 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
1949 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
1950 | } | |
7251515e DV |
1951 | } |
1952 | ||
1953 | static int | |
1954 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
324c8374 IM |
1955 | struct dp_packet_batch *batch, bool may_steal, |
1956 | bool concurrent_txq) | |
7251515e DV |
1957 | { |
1958 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1959 | |
324c8374 | 1960 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e | 1961 | return 0; |
8a9562d2 PS |
1962 | } |
1963 | ||
1964 | static int | |
74ff3298 | 1965 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1966 | { |
1967 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1968 | ||
1969 | ovs_mutex_lock(&dev->mutex); | |
1970 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1971 | dev->hwaddr = mac; |
045c0d1a | 1972 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1973 | } |
1974 | ovs_mutex_unlock(&dev->mutex); | |
1975 | ||
1976 | return 0; | |
1977 | } | |
1978 | ||
1979 | static int | |
74ff3298 | 1980 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1981 | { |
1982 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1983 | ||
1984 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1985 | *mac = dev->hwaddr; |
8a9562d2 PS |
1986 | ovs_mutex_unlock(&dev->mutex); |
1987 | ||
1988 | return 0; | |
1989 | } | |
1990 | ||
1991 | static int | |
1992 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1993 | { | |
1994 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1995 | ||
1996 | ovs_mutex_lock(&dev->mutex); | |
1997 | *mtup = dev->mtu; | |
1998 | ovs_mutex_unlock(&dev->mutex); | |
1999 | ||
2000 | return 0; | |
2001 | } | |
2002 | ||
0072e931 MK |
2003 | static int |
2004 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
2005 | { | |
2006 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2007 | ||
2008 | if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
2009 | || mtu < ETHER_MIN_MTU) { | |
2010 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
2011 | return EINVAL; | |
2012 | } | |
2013 | ||
2014 | ovs_mutex_lock(&dev->mutex); | |
2015 | if (dev->requested_mtu != mtu) { | |
2016 | dev->requested_mtu = mtu; | |
2017 | netdev_request_reconfigure(netdev); | |
2018 | } | |
2019 | ovs_mutex_unlock(&dev->mutex); | |
2020 | ||
2021 | return 0; | |
2022 | } | |
2023 | ||
8a9562d2 | 2024 | static int |
d46285a2 | 2025 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 2026 | |
58397e6c KT |
2027 | static int |
2028 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
2029 | struct netdev_stats *stats) | |
2030 | { | |
2031 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2032 | ||
2033 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 2034 | |
45d947c4 | 2035 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c | 2036 | /* Supported Stats */ |
50986e78 | 2037 | stats->rx_packets = dev->stats.rx_packets; |
2038 | stats->tx_packets = dev->stats.tx_packets; | |
9509913a | 2039 | stats->rx_dropped = dev->stats.rx_dropped; |
50986e78 | 2040 | stats->tx_dropped = dev->stats.tx_dropped; |
9e3ddd45 TP |
2041 | stats->multicast = dev->stats.multicast; |
2042 | stats->rx_bytes = dev->stats.rx_bytes; | |
2043 | stats->tx_bytes = dev->stats.tx_bytes; | |
2044 | stats->rx_errors = dev->stats.rx_errors; | |
2045 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 2046 | |
2047 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
2048 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
2049 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
2050 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
2051 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
2052 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
2053 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
2054 | ||
45d947c4 | 2055 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 2056 | |
58397e6c KT |
2057 | ovs_mutex_unlock(&dev->mutex); |
2058 | ||
2059 | return 0; | |
2060 | } | |
2061 | ||
d6e3feb5 | 2062 | static void |
2063 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
2064 | const struct rte_eth_xstat *xstats, |
2065 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 2066 | const unsigned int size) |
2067 | { | |
d6e3feb5 | 2068 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 2069 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2070 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2071 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2072 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2073 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2074 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2075 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2076 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2077 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2078 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2079 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2080 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2081 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2082 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 2083 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2084 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2085 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2086 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2087 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2088 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2089 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2090 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2091 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2092 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2093 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2094 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2095 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2096 | stats->tx_1523_to_max_packets = xstats[i].value; |
d57f777f PS |
2097 | } else if (strcmp(XSTAT_RX_MULTICAST_PACKETS, names[i].name) == 0) { |
2098 | stats->multicast = xstats[i].value; | |
0a0f39df | 2099 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2100 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 2101 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2102 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2103 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2104 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2105 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2106 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2107 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2108 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2109 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2110 | stats->rx_jabber_errors = xstats[i].value; |
2111 | } | |
2112 | } | |
2113 | } | |
2114 | ||
8a9562d2 PS |
2115 | static int |
2116 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2117 | { | |
2118 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2119 | struct rte_eth_stats rte_stats; | |
2120 | bool gg; | |
2121 | ||
2122 | netdev_dpdk_get_carrier(netdev, &gg); | |
2123 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2124 | |
0a0f39df CL |
2125 | struct rte_eth_xstat *rte_xstats = NULL; |
2126 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2127 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2128 | |
2129 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
bb37956a | 2130 | VLOG_ERR("Can't get ETH statistics for port: %"PRIu8, dev->port_id); |
f9256822 | 2131 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2132 | return EPROTO; |
2133 | } | |
2134 | ||
0a0f39df CL |
2135 | /* Get length of statistics */ |
2136 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2137 | if (rte_xstats_len < 0) { | |
bb37956a | 2138 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2139 | goto out; |
2140 | } | |
2141 | /* Reserve memory for xstats names and values */ | |
2142 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2143 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2144 | ||
2145 | /* Retreive xstats names */ | |
2146 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2147 | rte_xstats_names, | |
2148 | rte_xstats_len); | |
2149 | if (rte_xstats_new_len != rte_xstats_len) { | |
bb37956a | 2150 | VLOG_WARN("Cannot get XSTATS names for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2151 | goto out; |
2152 | } | |
2153 | /* Retreive xstats values */ | |
2154 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2155 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2156 | rte_xstats_len); | |
2157 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2158 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2159 | rte_xstats_len); | |
d6e3feb5 | 2160 | } else { |
bb37956a | 2161 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
d6e3feb5 | 2162 | } |
8a9562d2 | 2163 | |
0a0f39df CL |
2164 | out: |
2165 | free(rte_xstats); | |
2166 | free(rte_xstats_names); | |
2167 | ||
2f9dd77f PS |
2168 | stats->rx_packets = rte_stats.ipackets; |
2169 | stats->tx_packets = rte_stats.opackets; | |
2170 | stats->rx_bytes = rte_stats.ibytes; | |
2171 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2172 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2173 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2174 | |
45d947c4 | 2175 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2176 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2177 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2178 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2179 | |
2180 | /* These are the available DPDK counters for packets not received due to | |
2181 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2182 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2183 | stats->rx_missed_errors = rte_stats.imissed; |
2184 | ||
8a9562d2 PS |
2185 | ovs_mutex_unlock(&dev->mutex); |
2186 | ||
2187 | return 0; | |
2188 | } | |
2189 | ||
2190 | static int | |
d46285a2 | 2191 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2192 | enum netdev_features *current, |
ca3d4f55 BX |
2193 | enum netdev_features *advertised, |
2194 | enum netdev_features *supported, | |
2195 | enum netdev_features *peer) | |
8a9562d2 | 2196 | { |
d46285a2 | 2197 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2198 | struct rte_eth_link link; |
2199 | ||
2200 | ovs_mutex_lock(&dev->mutex); | |
2201 | link = dev->link; | |
2202 | ovs_mutex_unlock(&dev->mutex); | |
2203 | ||
362ca396 | 2204 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2205 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
2206 | *current = NETDEV_F_10MB_HD; |
2207 | } | |
362ca396 | 2208 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2209 | *current = NETDEV_F_100MB_HD; |
2210 | } | |
362ca396 | 2211 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2212 | *current = NETDEV_F_1GB_HD; |
2213 | } | |
2214 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 2215 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
2216 | *current = NETDEV_F_10MB_FD; |
2217 | } | |
362ca396 | 2218 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2219 | *current = NETDEV_F_100MB_FD; |
2220 | } | |
362ca396 | 2221 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2222 | *current = NETDEV_F_1GB_FD; |
2223 | } | |
362ca396 | 2224 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
2225 | *current = NETDEV_F_10GB_FD; |
2226 | } | |
2227 | } | |
2228 | ||
362ca396 | 2229 | if (link.link_autoneg) { |
2230 | *current |= NETDEV_F_AUTONEG; | |
2231 | } | |
2232 | ||
ca3d4f55 BX |
2233 | *advertised = *supported = *peer = 0; |
2234 | ||
8a9562d2 PS |
2235 | return 0; |
2236 | } | |
2237 | ||
9509913a IS |
2238 | static struct ingress_policer * |
2239 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2240 | { | |
2241 | struct ingress_policer *policer = NULL; | |
2242 | uint64_t rate_bytes; | |
2243 | uint64_t burst_bytes; | |
2244 | int err = 0; | |
2245 | ||
2246 | policer = xmalloc(sizeof *policer); | |
2247 | rte_spinlock_init(&policer->policer_lock); | |
2248 | ||
2249 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
602c8668 LR |
2250 | rate_bytes = rate * 1000ULL / 8; |
2251 | burst_bytes = burst * 1000ULL / 8; | |
9509913a IS |
2252 | |
2253 | policer->app_srtcm_params.cir = rate_bytes; | |
2254 | policer->app_srtcm_params.cbs = burst_bytes; | |
2255 | policer->app_srtcm_params.ebs = 0; | |
2256 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2257 | &policer->app_srtcm_params); | |
58be5c0e | 2258 | if (err) { |
9509913a IS |
2259 | VLOG_ERR("Could not create rte meter for ingress policer"); |
2260 | return NULL; | |
2261 | } | |
2262 | ||
2263 | return policer; | |
2264 | } | |
2265 | ||
2266 | static int | |
2267 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2268 | uint32_t policer_burst) | |
2269 | { | |
2270 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2271 | struct ingress_policer *policer; | |
2272 | ||
2273 | /* Force to 0 if no rate specified, | |
2274 | * default to 8000 kbits if burst is 0, | |
2275 | * else stick with user-specified value. | |
2276 | */ | |
2277 | policer_burst = (!policer_rate ? 0 | |
2278 | : !policer_burst ? 8000 | |
2279 | : policer_burst); | |
2280 | ||
2281 | ovs_mutex_lock(&dev->mutex); | |
2282 | ||
2283 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2284 | &dev->ingress_policer); | |
2285 | ||
2286 | if (dev->policer_rate == policer_rate && | |
2287 | dev->policer_burst == policer_burst) { | |
2288 | /* Assume that settings haven't changed since we last set them. */ | |
2289 | ovs_mutex_unlock(&dev->mutex); | |
2290 | return 0; | |
2291 | } | |
2292 | ||
2293 | /* Destroy any existing ingress policer for the device if one exists */ | |
2294 | if (policer) { | |
2295 | ovsrcu_postpone(free, policer); | |
2296 | } | |
2297 | ||
2298 | if (policer_rate != 0) { | |
2299 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2300 | } else { | |
2301 | policer = NULL; | |
2302 | } | |
2303 | ovsrcu_set(&dev->ingress_policer, policer); | |
2304 | dev->policer_rate = policer_rate; | |
2305 | dev->policer_burst = policer_burst; | |
2306 | ovs_mutex_unlock(&dev->mutex); | |
2307 | ||
2308 | return 0; | |
2309 | } | |
2310 | ||
8a9562d2 PS |
2311 | static int |
2312 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2313 | { | |
2314 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
2315 | |
2316 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
2317 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
2318 | * postive integer to meet RFC 2863 recommendations. | |
2319 | */ | |
2320 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
2321 | ovs_mutex_unlock(&dev->mutex); |
2322 | ||
2323 | return ifindex; | |
2324 | } | |
2325 | ||
2326 | static int | |
d46285a2 | 2327 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2328 | { |
d46285a2 | 2329 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2330 | |
2331 | ovs_mutex_lock(&dev->mutex); | |
2332 | check_link_status(dev); | |
2333 | *carrier = dev->link.link_status; | |
58397e6c KT |
2334 | |
2335 | ovs_mutex_unlock(&dev->mutex); | |
2336 | ||
2337 | return 0; | |
2338 | } | |
2339 | ||
2340 | static int | |
d46285a2 | 2341 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2342 | { |
d46285a2 | 2343 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2344 | |
2345 | ovs_mutex_lock(&dev->mutex); | |
2346 | ||
0a0f39df | 2347 | if (is_vhost_running(dev)) { |
58397e6c KT |
2348 | *carrier = 1; |
2349 | } else { | |
2350 | *carrier = 0; | |
2351 | } | |
2352 | ||
8a9562d2 PS |
2353 | ovs_mutex_unlock(&dev->mutex); |
2354 | ||
2355 | return 0; | |
2356 | } | |
2357 | ||
2358 | static long long int | |
d46285a2 | 2359 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2360 | { |
d46285a2 | 2361 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2362 | long long int carrier_resets; |
2363 | ||
2364 | ovs_mutex_lock(&dev->mutex); | |
2365 | carrier_resets = dev->link_reset_cnt; | |
2366 | ovs_mutex_unlock(&dev->mutex); | |
2367 | ||
2368 | return carrier_resets; | |
2369 | } | |
2370 | ||
2371 | static int | |
d46285a2 | 2372 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2373 | long long int interval OVS_UNUSED) |
2374 | { | |
ee32150e | 2375 | return EOPNOTSUPP; |
8a9562d2 PS |
2376 | } |
2377 | ||
2378 | static int | |
2379 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2380 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2381 | enum netdev_flags *old_flagsp) |
2382 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 2383 | { |
8a9562d2 PS |
2384 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
2385 | return EINVAL; | |
2386 | } | |
2387 | ||
2388 | *old_flagsp = dev->flags; | |
2389 | dev->flags |= on; | |
2390 | dev->flags &= ~off; | |
2391 | ||
2392 | if (dev->flags == *old_flagsp) { | |
2393 | return 0; | |
2394 | } | |
2395 | ||
58397e6c | 2396 | if (dev->type == DPDK_DEV_ETH) { |
58397e6c KT |
2397 | if (dev->flags & NETDEV_PROMISC) { |
2398 | rte_eth_promiscuous_enable(dev->port_id); | |
2399 | } | |
8a9562d2 | 2400 | |
314fb5ad | 2401 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
2402 | } else { |
2403 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2404 | * running then change netdev's change_seq to trigger link state | |
2405 | * update. */ | |
e543851d ZB |
2406 | |
2407 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2408 | && is_vhost_running(dev)) { |
e543851d ZB |
2409 | netdev_change_seq_changed(&dev->up); |
2410 | ||
2411 | /* Clear statistics if device is getting up. */ | |
2412 | if (NETDEV_UP & on) { | |
2413 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 2414 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
2415 | rte_spinlock_unlock(&dev->stats_lock); |
2416 | } | |
2417 | } | |
8a9562d2 PS |
2418 | } |
2419 | ||
2420 | return 0; | |
2421 | } | |
2422 | ||
2423 | static int | |
d46285a2 | 2424 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2425 | enum netdev_flags off, enum netdev_flags on, |
2426 | enum netdev_flags *old_flagsp) | |
2427 | { | |
d46285a2 | 2428 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2429 | int error; |
2430 | ||
d46285a2 DDP |
2431 | ovs_mutex_lock(&dev->mutex); |
2432 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2433 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2434 | |
2435 | return error; | |
2436 | } | |
2437 | ||
2438 | static int | |
d46285a2 | 2439 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2440 | { |
d46285a2 | 2441 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2442 | struct rte_eth_dev_info dev_info; |
2443 | ||
7cd1261d | 2444 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2445 | return ENODEV; |
7cd1261d | 2446 | } |
8a9562d2 PS |
2447 | |
2448 | ovs_mutex_lock(&dev->mutex); | |
2449 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2450 | ovs_mutex_unlock(&dev->mutex); | |
2451 | ||
95fb793a | 2452 | smap_add_format(args, "port_no", "%d", dev->port_id); |
58be5c0e MK |
2453 | smap_add_format(args, "numa_id", "%d", |
2454 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
2455 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
2456 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2457 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2458 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2459 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2460 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
2461 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
2462 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
2463 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
2464 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2465 | ||
39c2baa9 | 2466 | if (dev_info.pci_dev) { |
2467 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2468 | dev_info.pci_dev->id.vendor_id); | |
2469 | smap_add_format(args, "pci-device_id", "0x%x", | |
2470 | dev_info.pci_dev->id.device_id); | |
2471 | } | |
8a9562d2 PS |
2472 | |
2473 | return 0; | |
2474 | } | |
2475 | ||
2476 | static void | |
2477 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2478 | OVS_REQUIRES(dev->mutex) | |
2479 | { | |
2480 | enum netdev_flags old_flags; | |
2481 | ||
2482 | if (admin_state) { | |
2483 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2484 | } else { | |
2485 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2486 | } | |
2487 | } | |
2488 | ||
2489 | static void | |
2490 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2491 | const char *argv[], void *aux OVS_UNUSED) | |
2492 | { | |
2493 | bool up; | |
2494 | ||
2495 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2496 | up = true; | |
2497 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2498 | up = false; | |
2499 | } else { | |
2500 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2501 | return; | |
2502 | } | |
2503 | ||
2504 | if (argc > 2) { | |
2505 | struct netdev *netdev = netdev_from_name(argv[1]); | |
3d0d5ab1 | 2506 | |
8a9562d2 | 2507 | if (netdev && is_dpdk_class(netdev->netdev_class)) { |
3d0d5ab1 | 2508 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 2509 | |
3d0d5ab1 IM |
2510 | ovs_mutex_lock(&dev->mutex); |
2511 | netdev_dpdk_set_admin_state__(dev, up); | |
2512 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2513 | |
2514 | netdev_close(netdev); | |
2515 | } else { | |
2516 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2517 | netdev_close(netdev); | |
2518 | return; | |
2519 | } | |
2520 | } else { | |
3d0d5ab1 | 2521 | struct netdev_dpdk *dev; |
8a9562d2 PS |
2522 | |
2523 | ovs_mutex_lock(&dpdk_mutex); | |
3d0d5ab1 IM |
2524 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
2525 | ovs_mutex_lock(&dev->mutex); | |
2526 | netdev_dpdk_set_admin_state__(dev, up); | |
2527 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2528 | } |
2529 | ovs_mutex_unlock(&dpdk_mutex); | |
2530 | } | |
2531 | unixctl_command_reply(conn, "OK"); | |
2532 | } | |
2533 | ||
0ee821c2 DB |
2534 | static void |
2535 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2536 | const char *argv[], void *aux OVS_UNUSED) | |
2537 | { | |
2538 | int ret; | |
2539 | char *response; | |
7ee94cba | 2540 | dpdk_port_t port_id; |
0ee821c2 DB |
2541 | char devname[RTE_ETH_NAME_MAX_LEN]; |
2542 | struct netdev_dpdk *dev; | |
2543 | ||
2544 | ovs_mutex_lock(&dpdk_mutex); | |
2545 | ||
255b7bda | 2546 | if (rte_eth_dev_get_port_by_name(argv[1], &port_id)) { |
0ee821c2 DB |
2547 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); |
2548 | goto error; | |
2549 | } | |
2550 | ||
2551 | dev = netdev_dpdk_lookup_by_port_id(port_id); | |
2552 | if (dev) { | |
2553 | response = xasprintf("Device '%s' is being used by interface '%s'. " | |
2554 | "Remove it before detaching", | |
2555 | argv[1], netdev_get_name(&dev->up)); | |
2556 | goto error; | |
2557 | } | |
2558 | ||
2559 | rte_eth_dev_close(port_id); | |
2560 | ||
2561 | ret = rte_eth_dev_detach(port_id, devname); | |
2562 | if (ret < 0) { | |
2563 | response = xasprintf("Device '%s' can not be detached", argv[1]); | |
2564 | goto error; | |
2565 | } | |
2566 | ||
2567 | response = xasprintf("Device '%s' has been detached", argv[1]); | |
2568 | ||
2569 | ovs_mutex_unlock(&dpdk_mutex); | |
2570 | unixctl_command_reply(conn, response); | |
2571 | free(response); | |
2572 | return; | |
2573 | ||
2574 | error: | |
2575 | ovs_mutex_unlock(&dpdk_mutex); | |
2576 | unixctl_command_reply_error(conn, response); | |
2577 | free(response); | |
2578 | } | |
2579 | ||
58397e6c KT |
2580 | /* |
2581 | * Set virtqueue flags so that we do not receive interrupts. | |
2582 | */ | |
2583 | static void | |
0a0f39df | 2584 | set_irq_status(int vid) |
58397e6c | 2585 | { |
4573fbd3 | 2586 | uint32_t i; |
4573fbd3 | 2587 | |
f3e7ec25 MW |
2588 | for (i = 0; i < rte_vhost_get_vring_num(vid); i++) { |
2589 | rte_vhost_enable_guest_notification(vid, i, 0); | |
4573fbd3 FL |
2590 | } |
2591 | } | |
2592 | ||
585a5bea IM |
2593 | /* |
2594 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2595 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2596 | */ |
2597 | static void | |
d46285a2 DDP |
2598 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2599 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2600 | { |
2601 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2602 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 2603 | |
eff23640 | 2604 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
2605 | |
2606 | for (i = 0; i < total_txqs; i++) { | |
2607 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2608 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2609 | enabled_queues[n_enabled++] = i; |
2610 | } | |
2611 | } | |
2612 | ||
2613 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2614 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2615 | n_enabled = 1; |
2616 | } | |
2617 | ||
2618 | k = 0; | |
2619 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2620 | if (dev->tx_q[i].map != i) { |
2621 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2622 | k = (k + 1) % n_enabled; |
2623 | } | |
2624 | } | |
2625 | ||
2d24d165 | 2626 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2627 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2628 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2629 | } |
2630 | ||
eff23640 | 2631 | free(enabled_queues); |
585a5bea | 2632 | } |
4573fbd3 | 2633 | |
58397e6c KT |
2634 | /* |
2635 | * A new virtio-net device is added to a vhost port. | |
2636 | */ | |
2637 | static int | |
0a0f39df | 2638 | new_device(int vid) |
58397e6c | 2639 | { |
d46285a2 | 2640 | struct netdev_dpdk *dev; |
58397e6c | 2641 | bool exists = false; |
db8f13b0 | 2642 | int newnode = 0; |
0a0f39df CL |
2643 | char ifname[IF_NAME_SZ]; |
2644 | ||
58be5c0e | 2645 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2646 | |
2647 | ovs_mutex_lock(&dpdk_mutex); | |
2648 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2649 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 2650 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2651 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
f3e7ec25 | 2652 | uint32_t qp_num = rte_vhost_get_vring_num(vid)/VIRTIO_QNUM; |
db8f13b0 CL |
2653 | |
2654 | /* Get NUMA information */ | |
0a0f39df CL |
2655 | newnode = rte_vhost_get_numa_node(vid); |
2656 | if (newnode == -1) { | |
5b9bf9e0 | 2657 | #ifdef VHOST_NUMA |
db8f13b0 | 2658 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2659 | ifname); |
5b9bf9e0 | 2660 | #endif |
db8f13b0 | 2661 | newnode = dev->socket_id; |
db8f13b0 CL |
2662 | } |
2663 | ||
7f5f2bd0 IM |
2664 | if (dev->requested_n_txq != qp_num |
2665 | || dev->requested_n_rxq != qp_num | |
2666 | || dev->requested_socket_id != newnode) { | |
2667 | dev->requested_socket_id = newnode; | |
2668 | dev->requested_n_rxq = qp_num; | |
2669 | dev->requested_n_txq = qp_num; | |
2670 | netdev_request_reconfigure(&dev->up); | |
2671 | } else { | |
2672 | /* Reconfiguration not required. */ | |
2673 | dev->vhost_reconfigured = true; | |
2674 | } | |
81acebda | 2675 | |
0a0f39df | 2676 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2677 | exists = true; |
2678 | ||
58397e6c | 2679 | /* Disable notifications. */ |
0a0f39df | 2680 | set_irq_status(vid); |
e543851d | 2681 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2682 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2683 | break; |
2684 | } | |
c1ff66ac | 2685 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2686 | } |
2687 | ovs_mutex_unlock(&dpdk_mutex); | |
2688 | ||
2689 | if (!exists) { | |
0a0f39df | 2690 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2691 | |
2692 | return -1; | |
2693 | } | |
2694 | ||
0a0f39df CL |
2695 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2696 | ifname, newnode); | |
2697 | ||
58397e6c KT |
2698 | return 0; |
2699 | } | |
2700 | ||
f3ea2ad2 IM |
2701 | /* Clears mapping for all available queues of vhost interface. */ |
2702 | static void | |
2703 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2704 | OVS_REQUIRES(dev->mutex) | |
2705 | { | |
2706 | int i; | |
2707 | ||
81acebda | 2708 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
2709 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
2710 | } | |
2711 | } | |
2712 | ||
58397e6c KT |
2713 | /* |
2714 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2715 | * flag to stop any more packets from being sent or received to/from a VM and | |
2716 | * ensure all currently queued packets have been sent/received before removing | |
2717 | * the device. | |
2718 | */ | |
2719 | static void | |
0a0f39df | 2720 | destroy_device(int vid) |
58397e6c | 2721 | { |
d46285a2 | 2722 | struct netdev_dpdk *dev; |
afee281f | 2723 | bool exists = false; |
0a0f39df CL |
2724 | char ifname[IF_NAME_SZ]; |
2725 | ||
58be5c0e | 2726 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2727 | |
2728 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2729 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2730 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 2731 | |
d46285a2 | 2732 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
2733 | dev->vhost_reconfigured = false; |
2734 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 2735 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 2736 | |
e543851d | 2737 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2738 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 2739 | exists = true; |
afee281f | 2740 | break; |
58397e6c KT |
2741 | } |
2742 | } | |
afee281f | 2743 | |
58397e6c KT |
2744 | ovs_mutex_unlock(&dpdk_mutex); |
2745 | ||
0a0f39df | 2746 | if (exists) { |
afee281f KT |
2747 | /* |
2748 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2749 | * to NULL, before returning. | |
2750 | */ | |
2751 | ovsrcu_synchronize(); | |
2752 | /* | |
2753 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2754 | * put thread back into quiescent state before returning. | |
2755 | */ | |
2756 | ovsrcu_quiesce_start(); | |
0a0f39df | 2757 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 2758 | } else { |
0a0f39df | 2759 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 2760 | } |
58397e6c KT |
2761 | } |
2762 | ||
585a5bea | 2763 | static int |
0a0f39df | 2764 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 2765 | { |
d46285a2 | 2766 | struct netdev_dpdk *dev; |
585a5bea IM |
2767 | bool exists = false; |
2768 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
2769 | char ifname[IF_NAME_SZ]; |
2770 | ||
58be5c0e | 2771 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
2772 | |
2773 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2774 | return 0; | |
2775 | } | |
2776 | ||
2777 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2778 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 2779 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2780 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 2781 | if (enable) { |
d46285a2 | 2782 | dev->tx_q[qid].map = qid; |
585a5bea | 2783 | } else { |
d46285a2 | 2784 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2785 | } |
d46285a2 | 2786 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2787 | exists = true; |
d46285a2 | 2788 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2789 | break; |
2790 | } | |
c1ff66ac | 2791 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2792 | } |
2793 | ovs_mutex_unlock(&dpdk_mutex); | |
2794 | ||
2795 | if (exists) { | |
0a0f39df CL |
2796 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
2797 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 2798 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 2799 | } else { |
0a0f39df | 2800 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
2801 | return -1; |
2802 | } | |
2803 | ||
2804 | return 0; | |
2805 | } | |
2806 | ||
0a0f39df CL |
2807 | int |
2808 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 2809 | { |
0a0f39df | 2810 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
2811 | } |
2812 | ||
9509913a IS |
2813 | struct ingress_policer * |
2814 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
2815 | { | |
2816 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
2817 | } | |
2818 | ||
58397e6c | 2819 | static int |
ecc1a34e | 2820 | netdev_dpdk_class_init(void) |
7d1ced01 | 2821 | { |
ecc1a34e DDP |
2822 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2823 | ||
2824 | /* This function can be called for different classes. The initialization | |
2825 | * needs to be done only once */ | |
2826 | if (ovsthread_once_start(&once)) { | |
2827 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
2828 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2829 | "[netdev] up|down", 1, 2, | |
2830 | netdev_dpdk_set_admin_state, NULL); | |
2831 | ||
0ee821c2 DB |
2832 | unixctl_command_register("netdev-dpdk/detach", |
2833 | "pci address of device", 1, 1, | |
2834 | netdev_dpdk_detach, NULL); | |
2835 | ||
ecc1a34e DDP |
2836 | ovsthread_once_done(&once); |
2837 | } | |
362ca396 | 2838 | |
7d1ced01 CL |
2839 | return 0; |
2840 | } | |
2841 | ||
033e9df2 | 2842 | |
95fb793a | 2843 | /* Client Rings */ |
2844 | ||
95fb793a | 2845 | static int |
2846 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
bb37956a | 2847 | dpdk_port_t *eth_port_id) |
95fb793a | 2848 | { |
48fffdee | 2849 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 2850 | char *ring_name; |
b8374d0d | 2851 | int port_id; |
95fb793a | 2852 | |
48fffdee KT |
2853 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
2854 | if (!ring_pair) { | |
95fb793a | 2855 | return ENOMEM; |
2856 | } | |
2857 | ||
7251515e | 2858 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 2859 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 2860 | |
8f0a76c9 | 2861 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 2862 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2863 | RING_F_SP_ENQ); |
0c6f39e5 | 2864 | free(ring_name); |
48fffdee KT |
2865 | if (ring_pair->cring_tx == NULL) { |
2866 | rte_free(ring_pair); | |
95fb793a | 2867 | return ENOMEM; |
2868 | } | |
2869 | ||
0c6f39e5 | 2870 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 2871 | |
8f0a76c9 | 2872 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 2873 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2874 | RING_F_SC_DEQ); |
0c6f39e5 | 2875 | free(ring_name); |
48fffdee KT |
2876 | if (ring_pair->cring_rx == NULL) { |
2877 | rte_free(ring_pair); | |
95fb793a | 2878 | return ENOMEM; |
2879 | } | |
2880 | ||
b8374d0d MV |
2881 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
2882 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 2883 | |
b8374d0d | 2884 | if (port_id < 0) { |
48fffdee | 2885 | rte_free(ring_pair); |
95fb793a | 2886 | return ENODEV; |
2887 | } | |
2888 | ||
48fffdee | 2889 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
2890 | ring_pair->eth_port_id = port_id; |
2891 | *eth_port_id = port_id; | |
2892 | ||
48fffdee | 2893 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 2894 | |
95fb793a | 2895 | return 0; |
2896 | } | |
2897 | ||
2898 | static int | |
bb37956a | 2899 | dpdk_ring_open(const char dev_name[], dpdk_port_t *eth_port_id) |
64839cf4 | 2900 | OVS_REQUIRES(dpdk_mutex) |
95fb793a | 2901 | { |
48fffdee | 2902 | struct dpdk_ring *ring_pair; |
95fb793a | 2903 | unsigned int port_no; |
2904 | int err = 0; | |
2905 | ||
2906 | /* Names always start with "dpdkr" */ | |
2907 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2908 | if (err) { | |
2909 | return err; | |
2910 | } | |
2911 | ||
58be5c0e | 2912 | /* Look through our list to find the device */ |
48fffdee KT |
2913 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
2914 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 2915 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 2916 | /* Really all that is needed */ |
48fffdee | 2917 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 2918 | return 0; |
2919 | } | |
2920 | } | |
2921 | /* Need to create the device rings */ | |
2922 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2923 | } | |
2924 | ||
7251515e | 2925 | static int |
d46285a2 | 2926 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
324c8374 IM |
2927 | struct dp_packet_batch *batch, bool may_steal, |
2928 | bool concurrent_txq) | |
7251515e | 2929 | { |
d46285a2 | 2930 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a543eb0 | 2931 | struct dp_packet *packet; |
1b99bb05 | 2932 | |
58be5c0e MK |
2933 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
2934 | * the rss hash field is clear. This is because the same mbuf may be | |
2935 | * modified by the consumer of the ring and return into the datapath | |
2936 | * without recalculating the RSS hash. */ | |
8a543eb0 BB |
2937 | DP_PACKET_BATCH_FOR_EACH (packet, batch) { |
2938 | dp_packet_mbuf_rss_flag_reset(packet); | |
1b99bb05 | 2939 | } |
7251515e | 2940 | |
324c8374 | 2941 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e DV |
2942 | return 0; |
2943 | } | |
2944 | ||
95fb793a | 2945 | static int |
2946 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2947 | { | |
bb37956a | 2948 | dpdk_port_t port_no = 0; |
95fb793a | 2949 | int err = 0; |
2950 | ||
95fb793a | 2951 | ovs_mutex_lock(&dpdk_mutex); |
2952 | ||
2953 | err = dpdk_ring_open(netdev->name, &port_no); | |
2954 | if (err) { | |
2955 | goto unlock_dpdk; | |
2956 | } | |
2957 | ||
1ce30dfd DDP |
2958 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
2959 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 2960 | unlock_dpdk: |
2961 | ovs_mutex_unlock(&dpdk_mutex); | |
2962 | return err; | |
2963 | } | |
2964 | ||
0bf765f7 IS |
2965 | /* QoS Functions */ |
2966 | ||
2967 | /* | |
2968 | * Initialize QoS configuration operations. | |
2969 | */ | |
2970 | static void | |
2971 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
2972 | { | |
2973 | conf->ops = ops; | |
78bd47cf | 2974 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
2975 | } |
2976 | ||
2977 | /* | |
2978 | * Search existing QoS operations in qos_ops and compare each set of | |
2979 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
2980 | * else return NULL | |
2981 | */ | |
2982 | static const struct dpdk_qos_ops * | |
2983 | qos_lookup_name(const char *name) | |
2984 | { | |
2985 | const struct dpdk_qos_ops *const *opsp; | |
2986 | ||
2987 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2988 | const struct dpdk_qos_ops *ops = *opsp; | |
2989 | if (!strcmp(name, ops->qos_name)) { | |
2990 | return ops; | |
2991 | } | |
2992 | } | |
2993 | return NULL; | |
2994 | } | |
2995 | ||
0bf765f7 IS |
2996 | static int |
2997 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
2998 | struct sset *types) | |
2999 | { | |
3000 | const struct dpdk_qos_ops *const *opsp; | |
3001 | ||
3002 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3003 | const struct dpdk_qos_ops *ops = *opsp; | |
3004 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
3005 | sset_add(types, ops->qos_name); | |
3006 | } | |
3007 | } | |
3008 | return 0; | |
3009 | } | |
3010 | ||
3011 | static int | |
d46285a2 | 3012 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
3013 | const char **typep, struct smap *details) |
3014 | { | |
d46285a2 | 3015 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 3016 | struct qos_conf *qos_conf; |
0bf765f7 IS |
3017 | int error = 0; |
3018 | ||
d46285a2 | 3019 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
3020 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
3021 | if (qos_conf) { | |
3022 | *typep = qos_conf->ops->qos_name; | |
3023 | error = (qos_conf->ops->qos_get | |
3024 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
3025 | } else { |
3026 | /* No QoS configuration set, return an empty string */ | |
3027 | *typep = ""; | |
0bf765f7 | 3028 | } |
d46285a2 | 3029 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
3030 | |
3031 | return error; | |
3032 | } | |
3033 | ||
3034 | static int | |
78bd47cf DDP |
3035 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
3036 | const struct smap *details) | |
0bf765f7 | 3037 | { |
d46285a2 | 3038 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 3039 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 3040 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
3041 | int error = 0; |
3042 | ||
d46285a2 | 3043 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 3044 | |
78bd47cf | 3045 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 3046 | |
78bd47cf DDP |
3047 | new_ops = qos_lookup_name(type); |
3048 | ||
3049 | if (!new_ops || !new_ops->qos_construct) { | |
3050 | new_qos_conf = NULL; | |
3051 | if (type && type[0]) { | |
3052 | error = EOPNOTSUPP; | |
0bf765f7 | 3053 | } |
44975bb0 | 3054 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
3055 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
3056 | new_qos_conf = qos_conf; | |
0bf765f7 | 3057 | } else { |
78bd47cf | 3058 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
3059 | } |
3060 | ||
7ea266e9 | 3061 | if (error) { |
78bd47cf DDP |
3062 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
3063 | type, netdev->name, rte_strerror(error)); | |
3064 | } | |
3065 | ||
3066 | if (new_qos_conf != qos_conf) { | |
3067 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
3068 | if (qos_conf) { | |
3069 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
3070 | } | |
0bf765f7 IS |
3071 | } |
3072 | ||
d46285a2 | 3073 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 3074 | |
0bf765f7 IS |
3075 | return error; |
3076 | } | |
3077 | ||
3078 | /* egress-policer details */ | |
3079 | ||
3080 | struct egress_policer { | |
3081 | struct qos_conf qos_conf; | |
3082 | struct rte_meter_srtcm_params app_srtcm_params; | |
3083 | struct rte_meter_srtcm egress_meter; | |
3084 | }; | |
3085 | ||
78bd47cf DDP |
3086 | static void |
3087 | egress_policer_details_to_param(const struct smap *details, | |
3088 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 3089 | { |
78bd47cf DDP |
3090 | memset(params, 0, sizeof *params); |
3091 | params->cir = smap_get_ullong(details, "cir", 0); | |
3092 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
3093 | params->ebs = 0; | |
0bf765f7 IS |
3094 | } |
3095 | ||
3096 | static int | |
78bd47cf DDP |
3097 | egress_policer_qos_construct(const struct smap *details, |
3098 | struct qos_conf **conf) | |
0bf765f7 | 3099 | { |
0bf765f7 | 3100 | struct egress_policer *policer; |
0bf765f7 IS |
3101 | int err = 0; |
3102 | ||
0bf765f7 IS |
3103 | policer = xmalloc(sizeof *policer); |
3104 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 3105 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
0bf765f7 | 3106 | err = rte_meter_srtcm_config(&policer->egress_meter, |
78bd47cf DDP |
3107 | &policer->app_srtcm_params); |
3108 | if (!err) { | |
3109 | *conf = &policer->qos_conf; | |
3110 | } else { | |
7ea266e9 | 3111 | free(policer); |
78bd47cf | 3112 | *conf = NULL; |
7ea266e9 IS |
3113 | err = -err; |
3114 | } | |
0bf765f7 IS |
3115 | |
3116 | return err; | |
3117 | } | |
3118 | ||
3119 | static void | |
78bd47cf | 3120 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
3121 | { |
3122 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 3123 | qos_conf); |
0bf765f7 IS |
3124 | free(policer); |
3125 | } | |
3126 | ||
3127 | static int | |
78bd47cf | 3128 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 3129 | { |
78bd47cf DDP |
3130 | struct egress_policer *policer = |
3131 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3132 | ||
3133 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
3134 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 3135 | |
0bf765f7 IS |
3136 | return 0; |
3137 | } | |
3138 | ||
78bd47cf | 3139 | static bool |
47a45d86 KT |
3140 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
3141 | const struct smap *details) | |
0bf765f7 | 3142 | { |
78bd47cf DDP |
3143 | struct egress_policer *policer = |
3144 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3145 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 3146 | |
78bd47cf | 3147 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 3148 | |
78bd47cf | 3149 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
3150 | } |
3151 | ||
0bf765f7 | 3152 | static int |
3e90f7d7 GZ |
3153 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt, |
3154 | bool may_steal) | |
0bf765f7 | 3155 | { |
0bf765f7 | 3156 | int cnt = 0; |
78bd47cf DDP |
3157 | struct egress_policer *policer = |
3158 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 3159 | |
3e90f7d7 GZ |
3160 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, |
3161 | pkt_cnt, may_steal); | |
0bf765f7 IS |
3162 | |
3163 | return cnt; | |
3164 | } | |
3165 | ||
3166 | static const struct dpdk_qos_ops egress_policer_ops = { | |
3167 | "egress-policer", /* qos_name */ | |
3168 | egress_policer_qos_construct, | |
3169 | egress_policer_qos_destruct, | |
3170 | egress_policer_qos_get, | |
78bd47cf | 3171 | egress_policer_qos_is_equal, |
0bf765f7 IS |
3172 | egress_policer_run |
3173 | }; | |
3174 | ||
050c60bf DDP |
3175 | static int |
3176 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
3177 | { | |
3178 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3179 | int err = 0; | |
3180 | ||
050c60bf DDP |
3181 | ovs_mutex_lock(&dev->mutex); |
3182 | ||
3183 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 3184 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b CL |
3185 | && dev->mtu == dev->requested_mtu |
3186 | && dev->rxq_size == dev->requested_rxq_size | |
bd4e172b | 3187 | && dev->txq_size == dev->requested_txq_size |
3188 | && dev->socket_id == dev->requested_socket_id) { | |
050c60bf DDP |
3189 | /* Reconfiguration is unnecessary */ |
3190 | ||
3191 | goto out; | |
3192 | } | |
3193 | ||
3194 | rte_eth_dev_stop(dev->port_id); | |
3195 | ||
d555d9bd | 3196 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 | 3197 | if (err && err != EEXIST) { |
d555d9bd | 3198 | goto out; |
0072e931 MK |
3199 | } |
3200 | ||
050c60bf DDP |
3201 | netdev->n_txq = dev->requested_n_txq; |
3202 | netdev->n_rxq = dev->requested_n_rxq; | |
3203 | ||
b685696b CL |
3204 | dev->rxq_size = dev->requested_rxq_size; |
3205 | dev->txq_size = dev->requested_txq_size; | |
3206 | ||
050c60bf DDP |
3207 | rte_free(dev->tx_q); |
3208 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
3209 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
3210 | if (!dev->tx_q) { | |
3211 | err = ENOMEM; | |
3212 | } | |
050c60bf | 3213 | |
0072e931 MK |
3214 | netdev_change_seq_changed(netdev); |
3215 | ||
050c60bf | 3216 | out: |
050c60bf | 3217 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
3218 | return err; |
3219 | } | |
3220 | ||
7f381c2e | 3221 | static int |
2d24d165 | 3222 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 3223 | OVS_REQUIRES(dev->mutex) |
050c60bf | 3224 | { |
2d24d165 CL |
3225 | dev->up.n_txq = dev->requested_n_txq; |
3226 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 3227 | int err; |
050c60bf | 3228 | |
81acebda IM |
3229 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
3230 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
3231 | dev->tx_q[0].map = 0; | |
3232 | } | |
3233 | ||
3234 | netdev_dpdk_remap_txqs(dev); | |
3235 | ||
d555d9bd | 3236 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 FA |
3237 | if (!err) { |
3238 | /* A new mempool was created. */ | |
d555d9bd | 3239 | netdev_change_seq_changed(&dev->up); |
b6b26021 FA |
3240 | } else if (err != EEXIST){ |
3241 | return err; | |
db8f13b0 | 3242 | } |
0a0f39df | 3243 | if (netdev_dpdk_get_vid(dev) >= 0) { |
894af647 | 3244 | if (dev->vhost_reconfigured == false) { |
3245 | dev->vhost_reconfigured = true; | |
3246 | /* Carrier status may need updating. */ | |
3247 | netdev_change_seq_changed(&dev->up); | |
3248 | } | |
81acebda | 3249 | } |
7f381c2e DDP |
3250 | |
3251 | return 0; | |
2d24d165 CL |
3252 | } |
3253 | ||
3254 | static int | |
3255 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
3256 | { | |
3257 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3258 | int err; |
2d24d165 | 3259 | |
2d24d165 | 3260 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 3261 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 3262 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
3263 | |
3264 | return err; | |
2d24d165 CL |
3265 | } |
3266 | ||
3267 | static int | |
3268 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3269 | { | |
3270 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3271 | int err; |
2d24d165 | 3272 | |
2d24d165 CL |
3273 | ovs_mutex_lock(&dev->mutex); |
3274 | ||
c1ff66ac CL |
3275 | /* Configure vHost client mode if requested and if the following criteria |
3276 | * are met: | |
2d24d165 CL |
3277 | * 1. Device hasn't been registered yet. |
3278 | * 2. A path has been specified. | |
c1ff66ac CL |
3279 | */ |
3280 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 CL |
3281 | && strlen(dev->vhost_id)) { |
3282 | /* Register client-mode device */ | |
3283 | err = rte_vhost_driver_register(dev->vhost_id, | |
3284 | RTE_VHOST_USER_CLIENT); | |
c1ff66ac | 3285 | if (err) { |
2d24d165 CL |
3286 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3287 | dev->vhost_id); | |
7f381c2e | 3288 | goto unlock; |
c1ff66ac | 3289 | } else { |
2d24d165 CL |
3290 | /* Configuration successful */ |
3291 | dev->vhost_driver_flags |= RTE_VHOST_USER_CLIENT; | |
3292 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " | |
3293 | "using client socket '%s'", | |
3294 | dev->up.name, dev->vhost_id); | |
c1ff66ac | 3295 | } |
f3e7ec25 MW |
3296 | |
3297 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
3298 | &virtio_net_device_ops); | |
3299 | if (err) { | |
3300 | VLOG_ERR("rte_vhost_driver_callback_register failed for " | |
3301 | "vhost user client port: %s\n", dev->up.name); | |
3302 | goto unlock; | |
3303 | } | |
3304 | ||
3305 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
3306 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
3307 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
3308 | | 1ULL << VIRTIO_NET_F_CSUM); | |
3309 | if (err) { | |
3310 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
3311 | "client port: %s\n", dev->up.name); | |
3312 | goto unlock; | |
3313 | } | |
3314 | ||
3315 | err = rte_vhost_driver_start(dev->vhost_id); | |
3316 | if (err) { | |
3317 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
3318 | "client port: %s\n", dev->up.name); | |
3319 | goto unlock; | |
3320 | } | |
c1ff66ac CL |
3321 | } |
3322 | ||
7f381c2e DDP |
3323 | err = dpdk_vhost_reconfigure_helper(dev); |
3324 | ||
3325 | unlock: | |
050c60bf | 3326 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 3327 | |
7f381c2e | 3328 | return err; |
050c60bf DDP |
3329 | } |
3330 | ||
ecc1a34e | 3331 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
81acebda IM |
3332 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
3333 | GET_CARRIER, GET_STATS, \ | |
3334 | GET_FEATURES, GET_STATUS, \ | |
3335 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3336 | { \ |
3337 | NAME, \ | |
118c77b1 | 3338 | true, /* is_pmd */ \ |
ecc1a34e | 3339 | INIT, /* init */ \ |
95fb793a | 3340 | NULL, /* netdev_dpdk_run */ \ |
3341 | NULL, /* netdev_dpdk_wait */ \ | |
3342 | \ | |
3343 | netdev_dpdk_alloc, \ | |
3344 | CONSTRUCT, \ | |
58397e6c | 3345 | DESTRUCT, \ |
95fb793a | 3346 | netdev_dpdk_dealloc, \ |
3347 | netdev_dpdk_get_config, \ | |
81acebda | 3348 | SET_CONFIG, \ |
95fb793a | 3349 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3350 | NULL, /* build header */ \ |
3351 | NULL, /* push header */ \ | |
3352 | NULL, /* pop header */ \ | |
7dec44fe | 3353 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3354 | SET_TX_MULTIQ, \ |
95fb793a | 3355 | \ |
7251515e | 3356 | SEND, /* send */ \ |
95fb793a | 3357 | NULL, /* send_wait */ \ |
3358 | \ | |
3359 | netdev_dpdk_set_etheraddr, \ | |
3360 | netdev_dpdk_get_etheraddr, \ | |
3361 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3362 | netdev_dpdk_set_mtu, \ |
95fb793a | 3363 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3364 | GET_CARRIER, \ |
95fb793a | 3365 | netdev_dpdk_get_carrier_resets, \ |
3366 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
3367 | GET_STATS, \ |
3368 | GET_FEATURES, \ | |
95fb793a | 3369 | NULL, /* set_advertisements */ \ |
875ab130 | 3370 | NULL, /* get_pt_mode */ \ |
95fb793a | 3371 | \ |
9509913a | 3372 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3373 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3374 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3375 | netdev_dpdk_get_qos, \ |
3376 | netdev_dpdk_set_qos, \ | |
95fb793a | 3377 | NULL, /* get_queue */ \ |
3378 | NULL, /* set_queue */ \ | |
3379 | NULL, /* delete_queue */ \ | |
3380 | NULL, /* get_queue_stats */ \ | |
3381 | NULL, /* queue_dump_start */ \ | |
3382 | NULL, /* queue_dump_next */ \ | |
3383 | NULL, /* queue_dump_done */ \ | |
3384 | NULL, /* dump_queue_stats */ \ | |
3385 | \ | |
95fb793a | 3386 | NULL, /* set_in4 */ \ |
a8704b50 | 3387 | NULL, /* get_addr_list */ \ |
95fb793a | 3388 | NULL, /* add_router */ \ |
3389 | NULL, /* get_next_hop */ \ | |
58397e6c | 3390 | GET_STATUS, \ |
95fb793a | 3391 | NULL, /* arp_lookup */ \ |
3392 | \ | |
3393 | netdev_dpdk_update_flags, \ | |
050c60bf | 3394 | RECONFIGURE, \ |
95fb793a | 3395 | \ |
3396 | netdev_dpdk_rxq_alloc, \ | |
3397 | netdev_dpdk_rxq_construct, \ | |
3398 | netdev_dpdk_rxq_destruct, \ | |
3399 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3400 | RXQ_RECV, \ |
95fb793a | 3401 | NULL, /* rx_wait */ \ |
3402 | NULL, /* rxq_drain */ \ | |
18ebd48c | 3403 | NO_OFFLOAD_API \ |
95fb793a | 3404 | } |
8a9562d2 | 3405 | |
bce01e3a | 3406 | static const struct netdev_class dpdk_class = |
95fb793a | 3407 | NETDEV_DPDK_CLASS( |
3408 | "dpdk", | |
ecc1a34e | 3409 | netdev_dpdk_class_init, |
5496878c | 3410 | netdev_dpdk_construct, |
58397e6c | 3411 | netdev_dpdk_destruct, |
81acebda IM |
3412 | netdev_dpdk_set_config, |
3413 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3414 | netdev_dpdk_eth_send, |
3415 | netdev_dpdk_get_carrier, | |
3416 | netdev_dpdk_get_stats, | |
3417 | netdev_dpdk_get_features, | |
3418 | netdev_dpdk_get_status, | |
050c60bf | 3419 | netdev_dpdk_reconfigure, |
58397e6c | 3420 | netdev_dpdk_rxq_recv); |
95fb793a | 3421 | |
bce01e3a | 3422 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3423 | NETDEV_DPDK_CLASS( |
3424 | "dpdkr", | |
ecc1a34e | 3425 | netdev_dpdk_class_init, |
5496878c | 3426 | netdev_dpdk_ring_construct, |
58397e6c | 3427 | netdev_dpdk_destruct, |
c3d062a7 | 3428 | netdev_dpdk_ring_set_config, |
81acebda | 3429 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3430 | netdev_dpdk_ring_send, |
3431 | netdev_dpdk_get_carrier, | |
3432 | netdev_dpdk_get_stats, | |
3433 | netdev_dpdk_get_features, | |
3434 | netdev_dpdk_get_status, | |
050c60bf | 3435 | netdev_dpdk_reconfigure, |
58397e6c KT |
3436 | netdev_dpdk_rxq_recv); |
3437 | ||
53f50d24 | 3438 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3439 | NETDEV_DPDK_CLASS( |
3440 | "dpdkvhostuser", | |
f3e7ec25 | 3441 | NULL, |
53f50d24 | 3442 | netdev_dpdk_vhost_construct, |
58397e6c | 3443 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3444 | NULL, |
81acebda | 3445 | NULL, |
58397e6c KT |
3446 | netdev_dpdk_vhost_send, |
3447 | netdev_dpdk_vhost_get_carrier, | |
3448 | netdev_dpdk_vhost_get_stats, | |
3449 | NULL, | |
7251515e | 3450 | NULL, |
53f50d24 | 3451 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3452 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3453 | static const struct netdev_class dpdk_vhost_client_class = |
3454 | NETDEV_DPDK_CLASS( | |
3455 | "dpdkvhostuserclient", | |
f3e7ec25 | 3456 | NULL, |
2d24d165 CL |
3457 | netdev_dpdk_vhost_client_construct, |
3458 | netdev_dpdk_vhost_destruct, | |
3459 | netdev_dpdk_vhost_client_set_config, | |
3460 | NULL, | |
3461 | netdev_dpdk_vhost_send, | |
3462 | netdev_dpdk_vhost_get_carrier, | |
3463 | netdev_dpdk_vhost_get_stats, | |
3464 | NULL, | |
3465 | NULL, | |
3466 | netdev_dpdk_vhost_client_reconfigure, | |
3467 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 3468 | |
8a9562d2 PS |
3469 | void |
3470 | netdev_dpdk_register(void) | |
3471 | { | |
bab69409 AC |
3472 | netdev_register_provider(&dpdk_class); |
3473 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3474 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 3475 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 3476 | } |