]>
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> |
3eb8d4fa | 40 | #include <rte_version.h> |
8a9562d2 | 41 | |
7d1ced01 | 42 | #include "dirs.h" |
e14deea0 | 43 | #include "dp-packet.h" |
01961bbd | 44 | #include "dpdk.h" |
8a9562d2 | 45 | #include "dpif-netdev.h" |
e5c0f5a4 | 46 | #include "fatal-signal.h" |
8a9562d2 PS |
47 | #include "netdev-provider.h" |
48 | #include "netdev-vport.h" | |
49 | #include "odp-util.h" | |
eac84432 | 50 | #include "openvswitch/dynamic-string.h" |
25d436fb BW |
51 | #include "openvswitch/list.h" |
52 | #include "openvswitch/ofp-print.h" | |
53 | #include "openvswitch/vlog.h" | |
94143fc4 | 54 | #include "ovs-numa.h" |
8a9562d2 PS |
55 | #include "ovs-thread.h" |
56 | #include "ovs-rcu.h" | |
57 | #include "packets.h" | |
ee89ea7b | 58 | #include "openvswitch/shash.h" |
0bf765f7 | 59 | #include "smap.h" |
8a9562d2 PS |
60 | #include "sset.h" |
61 | #include "unaligned.h" | |
62 | #include "timeval.h" | |
63 | #include "unixctl.h" | |
8a9562d2 | 64 | |
f3e7ec25 MW |
65 | enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM}; |
66 | ||
05b49df6 | 67 | VLOG_DEFINE_THIS_MODULE(netdev_dpdk); |
8a9562d2 PS |
68 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
69 | ||
70 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
71 | ||
72 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
73 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
74 | ||
75 | /* | |
76 | * need to reserve tons of extra space in the mbufs so we can align the | |
77 | * DMA addresses to 4KB. | |
18f777b2 TP |
78 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
79 | * performance for standard Ethernet MTU. | |
8a9562d2 | 80 | */ |
58be5c0e MK |
81 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN \ |
82 | + (2 * VLAN_HEADER_LEN)) | |
4be4d22c MK |
83 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) |
84 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
58be5c0e MK |
85 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len) \ |
86 | - ETHER_HDR_LEN - ETHER_CRC_LEN) | |
31b88c97 SS |
87 | #define MBUF_SIZE(mtu) ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) \ |
88 | + sizeof(struct dp_packet) \ | |
89 | + RTE_PKTMBUF_HEADROOM), \ | |
90 | RTE_CACHE_LINE_SIZE) | |
4be4d22c | 91 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 92 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 93 | |
bc57ed90 IM |
94 | /* Min number of packets in the mempool. OVS tries to allocate a mempool with |
95 | * roughly estimated number of mbufs: if this fails (because the system doesn't | |
96 | * have enough hugepages) we keep halving the number until the allocation | |
97 | * succeeds or we reach MIN_NB_MBUF */ | |
da79ce2b DDP |
98 | #define MIN_NB_MBUF (4096 * 4) |
99 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
100 | ||
d6e3feb5 | 101 | /* |
102 | * DPDK XSTATS Counter names definition | |
103 | */ | |
104 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
105 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
106 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
107 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
108 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
109 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
110 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
111 | ||
112 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
113 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
114 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
115 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
116 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
117 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
118 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
119 | ||
d57f777f | 120 | #define XSTAT_RX_MULTICAST_PACKETS "rx_multicast_packets" |
d6e3feb5 | 121 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" |
122 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
123 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
124 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
125 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
126 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
127 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
128 | ||
8a9562d2 PS |
129 | #define SOCKET0 0 |
130 | ||
b685696b CL |
131 | /* Default size of Physical NIC RXQ */ |
132 | #define NIC_PORT_DEFAULT_RXQ_SIZE 2048 | |
133 | /* Default size of Physical NIC TXQ */ | |
134 | #define NIC_PORT_DEFAULT_TXQ_SIZE 2048 | |
135 | /* Maximum size of Physical NIC Queues */ | |
136 | #define NIC_PORT_MAX_Q_SIZE 4096 | |
79f5354c | 137 | |
585a5bea | 138 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
139 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
140 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
141 | * yet mapped to another queue. */ | |
585a5bea | 142 | |
bb37956a IM |
143 | #define DPDK_ETH_PORT_ID_INVALID RTE_MAX_ETHPORTS |
144 | ||
5e925ccc MK |
145 | /* DPDK library uses uint16_t for port_id. */ |
146 | typedef uint16_t dpdk_port_t; | |
fa9f4eeb | 147 | #define DPDK_PORT_ID_FMT "%"PRIu16 |
bb37956a | 148 | |
31871ee3 | 149 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 150 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 151 | |
8a9562d2 | 152 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
153 | .rxmode = { |
154 | .mq_mode = ETH_MQ_RX_RSS, | |
155 | .split_hdr_size = 0, | |
156 | .header_split = 0, /* Header Split disabled */ | |
157 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
158 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
159 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
160 | .hw_strip_crc = 0, | |
161 | }, | |
162 | .rx_adv_conf = { | |
163 | .rss_conf = { | |
164 | .rss_key = NULL, | |
543342a4 | 165 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 166 | }, |
a28ddd11 DDP |
167 | }, |
168 | .txmode = { | |
169 | .mq_mode = ETH_MQ_TX_NONE, | |
170 | }, | |
8a9562d2 PS |
171 | }; |
172 | ||
f3e7ec25 MW |
173 | /* |
174 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
175 | * configuration has been fully completed. | |
176 | */ | |
177 | static int new_device(int vid); | |
178 | static void destroy_device(int vid); | |
179 | static int vring_state_changed(int vid, uint16_t queue_id, int enable); | |
180 | static const struct vhost_device_ops virtio_net_device_ops = | |
181 | { | |
182 | .new_device = new_device, | |
183 | .destroy_device = destroy_device, | |
184 | .vring_state_changed = vring_state_changed, | |
185 | .features_changed = NULL | |
186 | }; | |
187 | ||
58f7c37b DDP |
188 | enum { DPDK_RING_SIZE = 256 }; |
189 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
190 | enum { DRAIN_TSC = 200000ULL }; |
191 | ||
58397e6c KT |
192 | enum dpdk_dev_type { |
193 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 194 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
195 | }; |
196 | ||
0bf765f7 IS |
197 | /* Quality of Service */ |
198 | ||
199 | /* An instance of a QoS configuration. Always associated with a particular | |
200 | * network device. | |
201 | * | |
202 | * Each QoS implementation subclasses this with whatever additional data it | |
203 | * needs. | |
204 | */ | |
205 | struct qos_conf { | |
206 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 207 | rte_spinlock_t lock; |
0bf765f7 IS |
208 | }; |
209 | ||
210 | /* A particular implementation of dpdk QoS operations. | |
211 | * | |
212 | * The functions below return 0 if successful or a positive errno value on | |
213 | * failure, except where otherwise noted. All of them must be provided, except | |
214 | * where otherwise noted. | |
215 | */ | |
216 | struct dpdk_qos_ops { | |
217 | ||
218 | /* Name of the QoS type */ | |
219 | const char *qos_name; | |
220 | ||
78bd47cf DDP |
221 | /* Called to construct a qos_conf object. The implementation should make |
222 | * the appropriate calls to configure QoS according to 'details'. | |
0bf765f7 IS |
223 | * |
224 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
225 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
226 | * (which is built as ovs-vswitchd.conf.db(8)). | |
227 | * | |
78bd47cf DDP |
228 | * This function must return 0 if and only if it sets '*conf' to an |
229 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
230 | * |
231 | * For all QoS implementations it should always be non-null. | |
232 | */ | |
78bd47cf | 233 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
234 | |
235 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 236 | * 'qos_conf'. |
0bf765f7 IS |
237 | * |
238 | * For all QoS implementations it should always be non-null. | |
239 | */ | |
78bd47cf | 240 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 241 | |
78bd47cf | 242 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
243 | * |
244 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
245 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
246 | * (which is built as ovs-vswitchd.conf.db(8)). | |
247 | */ | |
78bd47cf | 248 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 249 | |
78bd47cf | 250 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
251 | * |
252 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
253 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
254 | * (which is built as ovs-vswitchd.conf.db(8)). | |
255 | * | |
78bd47cf | 256 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 257 | */ |
78bd47cf DDP |
258 | bool (*qos_is_equal)(const struct qos_conf *conf, |
259 | const struct smap *details); | |
0bf765f7 IS |
260 | |
261 | /* Modify an array of rte_mbufs. The modification is specific to | |
262 | * each qos implementation. | |
263 | * | |
264 | * The function should take and array of mbufs and an int representing | |
265 | * the current number of mbufs present in the array. | |
266 | * | |
267 | * After the function has performed a qos modification to the array of | |
268 | * mbufs it returns an int representing the number of mbufs now present in | |
269 | * the array. This value is can then be passed to the port send function | |
270 | * along with the modified array for transmission. | |
271 | * | |
272 | * For all QoS implementations it should always be non-null. | |
273 | */ | |
78bd47cf | 274 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
3e90f7d7 | 275 | int pkt_cnt, bool may_steal); |
0bf765f7 IS |
276 | }; |
277 | ||
278 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
279 | static const struct dpdk_qos_ops egress_policer_ops; | |
280 | ||
281 | /* | |
282 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
283 | * operations. | |
284 | */ | |
285 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
286 | &egress_policer_ops, | |
287 | NULL | |
288 | }; | |
289 | ||
c2adb102 IM |
290 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
291 | ||
8a9562d2 | 292 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 293 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 294 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 295 | |
c2adb102 IM |
296 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
297 | = OVS_MUTEX_INITIALIZER; | |
298 | ||
91fccdad KT |
299 | /* Contains all 'struct dpdk_mp's. */ |
300 | static struct ovs_list dpdk_mp_free_list OVS_GUARDED_BY(dpdk_mp_mutex) | |
301 | = OVS_LIST_INITIALIZER(&dpdk_mp_free_list); | |
302 | ||
303 | /* Wrapper for a mempool released but not yet freed. */ | |
304 | struct dpdk_mp { | |
305 | struct rte_mempool *mp; | |
306 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mp_mutex); | |
307 | }; | |
308 | ||
5a034064 AW |
309 | /* There should be one 'struct dpdk_tx_queue' created for |
310 | * each cpu core. */ | |
8a9562d2 | 311 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
312 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
313 | * from concurrent access. It is used only | |
314 | * if the queue is shared among different | |
324c8374 | 315 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
316 | int map; /* Mapping of configured vhost-user queues |
317 | * to enabled by guest. */ | |
8a9562d2 PS |
318 | }; |
319 | ||
95fb793a | 320 | /* dpdk has no way to remove dpdk ring ethernet devices |
321 | so we have to keep them around once they've been created | |
322 | */ | |
323 | ||
ca6ba700 | 324 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 325 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 326 | |
327 | struct dpdk_ring { | |
328 | /* For the client rings */ | |
329 | struct rte_ring *cring_tx; | |
330 | struct rte_ring *cring_rx; | |
b83a2df1 | 331 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
bb37956a | 332 | dpdk_port_t eth_port_id; /* ethernet device port id */ |
ca6ba700 | 333 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 334 | }; |
335 | ||
9509913a IS |
336 | struct ingress_policer { |
337 | struct rte_meter_srtcm_params app_srtcm_params; | |
338 | struct rte_meter_srtcm in_policer; | |
339 | rte_spinlock_t policer_lock; | |
340 | }; | |
341 | ||
1a2bb118 SC |
342 | enum dpdk_hw_ol_features { |
343 | NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0, | |
344 | }; | |
345 | ||
b2e72a9c IM |
346 | /* |
347 | * In order to avoid confusion in variables names, following naming convention | |
348 | * should be used, if possible: | |
349 | * | |
350 | * 'struct netdev' : 'netdev' | |
351 | * 'struct netdev_dpdk' : 'dev' | |
352 | * 'struct netdev_rxq' : 'rxq' | |
353 | * 'struct netdev_rxq_dpdk' : 'rx' | |
354 | * | |
355 | * Example: | |
356 | * struct netdev *netdev = netdev_from_name(name); | |
357 | * struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
358 | * | |
359 | * Also, 'netdev' should be used instead of 'dev->up', where 'netdev' was | |
360 | * already defined. | |
361 | */ | |
362 | ||
8a9562d2 | 363 | struct netdev_dpdk { |
23d4d53f BB |
364 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline0, |
365 | dpdk_port_t port_id; | |
366 | ||
367 | /* If true, device was attached by rte_eth_dev_attach(). */ | |
368 | bool attached; | |
369 | struct eth_addr hwaddr; | |
370 | int mtu; | |
371 | int socket_id; | |
372 | int buf_size; | |
373 | int max_packet_len; | |
374 | enum dpdk_dev_type type; | |
375 | enum netdev_flags flags; | |
376 | char *devargs; /* Device arguments for dpdk ports */ | |
377 | struct dpdk_tx_queue *tx_q; | |
378 | struct rte_eth_link link; | |
379 | int link_reset_cnt; | |
380 | /* 4 pad bytes here. */ | |
381 | ); | |
382 | ||
383 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline1, | |
384 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
24e78f93 | 385 | struct rte_mempool *mp; |
23d4d53f BB |
386 | |
387 | /* virtio identifier for vhost devices */ | |
388 | ovsrcu_index vid; | |
389 | ||
390 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
391 | bool vhost_reconfigured; | |
392 | /* 3 pad bytes here. */ | |
393 | ); | |
394 | ||
395 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
396 | /* Identifier used to distinguish vhost devices from each other. */ | |
397 | char vhost_id[PATH_MAX]; | |
398 | ); | |
399 | ||
400 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
401 | struct netdev up; | |
402 | /* In dpdk_list. */ | |
403 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); | |
404 | ||
405 | /* QoS configuration and lock for the device */ | |
406 | OVSRCU_TYPE(struct qos_conf *) qos_conf; | |
407 | ||
408 | /* Ingress Policer */ | |
409 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
410 | uint32_t policer_rate; | |
411 | uint32_t policer_burst; | |
412 | ); | |
413 | ||
414 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
415 | struct netdev_stats stats; | |
416 | /* Protects stats */ | |
417 | rte_spinlock_t stats_lock; | |
418 | /* 44 pad bytes here. */ | |
419 | ); | |
420 | ||
421 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
422 | /* The following properties cannot be changed when a device is running, | |
423 | * so we remember the request and update them next time | |
424 | * netdev_dpdk*_reconfigure() is called */ | |
425 | int requested_mtu; | |
426 | int requested_n_txq; | |
427 | int requested_n_rxq; | |
428 | int requested_rxq_size; | |
429 | int requested_txq_size; | |
430 | ||
431 | /* Number of rx/tx descriptors for physical devices */ | |
432 | int rxq_size; | |
433 | int txq_size; | |
434 | ||
435 | /* Socket ID detected when vHost device is brought up */ | |
436 | int requested_socket_id; | |
437 | ||
438 | /* Denotes whether vHost port is client/server mode */ | |
439 | uint64_t vhost_driver_flags; | |
440 | ||
441 | /* DPDK-ETH Flow control */ | |
442 | struct rte_eth_fc_conf fc_conf; | |
443 | ||
444 | /* DPDK-ETH hardware offload features, | |
445 | * from the enum set 'dpdk_hw_ol_features' */ | |
446 | uint32_t hw_ol_features; | |
447 | ); | |
971f4b39 MW |
448 | |
449 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
450 | /* Names of all XSTATS counters */ | |
451 | struct rte_eth_xstat_name *rte_xstats_names; | |
452 | int rte_xstats_names_size; | |
453 | int rte_xstats_ids_size; | |
454 | uint64_t *rte_xstats_ids; | |
455 | ); | |
8a9562d2 PS |
456 | }; |
457 | ||
458 | struct netdev_rxq_dpdk { | |
459 | struct netdev_rxq up; | |
bb37956a | 460 | dpdk_port_t port_id; |
8a9562d2 PS |
461 | }; |
462 | ||
f3e7ec25 MW |
463 | static void netdev_dpdk_destruct(struct netdev *netdev); |
464 | static void netdev_dpdk_vhost_destruct(struct netdev *netdev); | |
8a9562d2 | 465 | |
ac1a9bb9 IM |
466 | static void netdev_dpdk_clear_xstats(struct netdev_dpdk *dev); |
467 | ||
0a0f39df | 468 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 469 | |
9509913a IS |
470 | struct ingress_policer * |
471 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
472 | ||
8a9562d2 PS |
473 | static bool |
474 | is_dpdk_class(const struct netdev_class *class) | |
475 | { | |
f3e7ec25 MW |
476 | return class->destruct == netdev_dpdk_destruct |
477 | || class->destruct == netdev_dpdk_vhost_destruct; | |
8a9562d2 PS |
478 | } |
479 | ||
4be4d22c MK |
480 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
481 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
482 | * value, insufficient buffers are allocated to accomodate the packet in its | |
483 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
484 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
485 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
486 | * behaviour, which reduces performance. To prevent this, use a buffer size |
487 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
488 | */ |
489 | static uint32_t | |
490 | dpdk_buf_size(int mtu) | |
491 | { | |
492 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
493 | NETDEV_DPDK_MBUF_ALIGN); | |
494 | } | |
495 | ||
eff23640 DDP |
496 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
497 | * | |
498 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
499 | static void * |
500 | dpdk_rte_mzalloc(size_t sz) | |
501 | { | |
eff23640 | 502 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
503 | } |
504 | ||
505 | void | |
e14deea0 | 506 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 507 | { |
db73f716 | 508 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 509 | |
b00b4a81 | 510 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
511 | } |
512 | ||
b3cd9f9d | 513 | static void |
401b70d6 | 514 | ovs_rte_pktmbuf_init(struct rte_mempool *mp OVS_UNUSED, |
b3cd9f9d | 515 | void *opaque_arg OVS_UNUSED, |
2391135c | 516 | void *_p, |
b3cd9f9d PS |
517 | unsigned i OVS_UNUSED) |
518 | { | |
2391135c | 519 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 520 | |
2391135c | 521 | dp_packet_init_dpdk((struct dp_packet *) pkt, pkt->buf_len); |
b3cd9f9d PS |
522 | } |
523 | ||
91fccdad KT |
524 | static int |
525 | dpdk_mp_full(const struct rte_mempool *mp) OVS_REQUIRES(dpdk_mp_mutex) | |
526 | { | |
527 | unsigned ring_count; | |
528 | /* This logic is needed because rte_mempool_full() is not guaranteed to | |
529 | * be atomic and mbufs could be moved from mempool cache --> mempool ring | |
530 | * during the call. However, as no mbufs will be taken from the mempool | |
531 | * at this time, we can work around it by also checking the ring entries | |
532 | * separately and ensuring that they have not changed. | |
533 | */ | |
534 | ring_count = rte_mempool_ops_get_count(mp); | |
535 | if (rte_mempool_full(mp) && rte_mempool_ops_get_count(mp) == ring_count) { | |
536 | return 1; | |
537 | } | |
538 | ||
539 | return 0; | |
540 | } | |
541 | ||
542 | /* Free unused mempools. */ | |
543 | static void | |
544 | dpdk_mp_sweep(void) | |
545 | { | |
546 | struct dpdk_mp *dmp, *next; | |
547 | ||
548 | ovs_mutex_lock(&dpdk_mp_mutex); | |
549 | LIST_FOR_EACH_SAFE (dmp, next, list_node, &dpdk_mp_free_list) { | |
550 | if (dpdk_mp_full(dmp->mp)) { | |
551 | VLOG_DBG("Freeing mempool \"%s\"", dmp->mp->name); | |
552 | ovs_list_remove(&dmp->list_node); | |
553 | rte_mempool_free(dmp->mp); | |
554 | rte_free(dmp); | |
555 | } | |
556 | } | |
557 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
558 | } | |
559 | ||
560 | /* Ensure a mempool will not be freed. */ | |
561 | static void | |
562 | dpdk_mp_do_not_free(struct rte_mempool *mp) OVS_REQUIRES(dpdk_mp_mutex) | |
563 | { | |
564 | struct dpdk_mp *dmp, *next; | |
565 | ||
566 | LIST_FOR_EACH_SAFE (dmp, next, list_node, &dpdk_mp_free_list) { | |
567 | if (dmp->mp == mp) { | |
568 | VLOG_DBG("Removing mempool \"%s\" from free list", dmp->mp->name); | |
569 | ovs_list_remove(&dmp->list_node); | |
570 | rte_free(dmp); | |
571 | break; | |
572 | } | |
573 | } | |
574 | } | |
575 | ||
24e78f93 IM |
576 | /* Returns a valid pointer when either of the following is true: |
577 | * - a new mempool was just created; | |
578 | * - a matching mempool already exists. */ | |
579 | static struct rte_mempool * | |
580 | dpdk_mp_create(struct netdev_dpdk *dev, int mtu) | |
8a9562d2 | 581 | { |
24e78f93 IM |
582 | char mp_name[RTE_MEMPOOL_NAMESIZE]; |
583 | const char *netdev_name = netdev_get_name(&dev->up); | |
584 | int socket_id = dev->requested_socket_id; | |
585 | uint32_t n_mbufs; | |
586 | uint32_t hash = hash_string(netdev_name, 0); | |
587 | struct rte_mempool *mp = NULL; | |
d555d9bd RW |
588 | |
589 | /* | |
ad9b5b9b | 590 | * XXX: rough estimation of number of mbufs required for this port: |
d555d9bd RW |
591 | * <packets required to fill the device rxqs> |
592 | * + <packets that could be stuck on other ports txqs> | |
593 | * + <packets in the pmd threads> | |
594 | * + <additional memory for corner cases> | |
0072e931 | 595 | */ |
24e78f93 IM |
596 | n_mbufs = dev->requested_n_rxq * dev->requested_rxq_size |
597 | + dev->requested_n_txq * dev->requested_txq_size | |
598 | + MIN(RTE_MAX_LCORE, dev->requested_n_rxq) * NETDEV_MAX_BURST | |
599 | + MIN_NB_MBUF; | |
d555d9bd | 600 | |
24e78f93 | 601 | ovs_mutex_lock(&dpdk_mp_mutex); |
da79ce2b | 602 | do { |
24e78f93 IM |
603 | /* Full DPDK memory pool name must be unique and cannot be |
604 | * longer than RTE_MEMPOOL_NAMESIZE. */ | |
af5b0dad IM |
605 | int ret = snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, |
606 | "ovs%08x%02d%05d%07u", | |
24e78f93 IM |
607 | hash, socket_id, mtu, n_mbufs); |
608 | if (ret < 0 || ret >= RTE_MEMPOOL_NAMESIZE) { | |
609 | VLOG_DBG("snprintf returned %d. " | |
610 | "Failed to generate a mempool name for \"%s\". " | |
611 | "Hash:0x%x, socket_id: %d, mtu:%d, mbufs:%u.", | |
612 | ret, netdev_name, hash, socket_id, mtu, n_mbufs); | |
613 | break; | |
65056fd7 | 614 | } |
95fb793a | 615 | |
f06546a5 FA |
616 | VLOG_DBG("Port %s: Requesting a mempool of %u mbufs " |
617 | "on socket %d for %d Rx and %d Tx queues.", | |
24e78f93 | 618 | netdev_name, n_mbufs, socket_id, |
f06546a5 | 619 | dev->requested_n_rxq, dev->requested_n_txq); |
d555d9bd | 620 | |
24e78f93 IM |
621 | mp = rte_pktmbuf_pool_create(mp_name, n_mbufs, MP_CACHE_SZ, |
622 | sizeof (struct dp_packet) - sizeof (struct rte_mbuf), | |
623 | MBUF_SIZE(mtu) - sizeof(struct dp_packet), socket_id); | |
624 | ||
625 | if (mp) { | |
626 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", | |
627 | mp_name, n_mbufs); | |
837c1761 FA |
628 | /* rte_pktmbuf_pool_create has done some initialization of the |
629 | * rte_mbuf part of each dp_packet. Some OvS specific fields | |
630 | * of the packet still need to be initialized by | |
631 | * ovs_rte_pktmbuf_init. */ | |
24e78f93 | 632 | rte_mempool_obj_iter(mp, ovs_rte_pktmbuf_init, NULL); |
d555d9bd RW |
633 | } else if (rte_errno == EEXIST) { |
634 | /* A mempool with the same name already exists. We just | |
635 | * retrieve its pointer to be returned to the caller. */ | |
24e78f93 | 636 | mp = rte_mempool_lookup(mp_name); |
d555d9bd RW |
637 | /* As the mempool create returned EEXIST we can expect the |
638 | * lookup has returned a valid pointer. If for some reason | |
639 | * that's not the case we keep track of it. */ | |
24e78f93 IM |
640 | VLOG_DBG("A mempool with name \"%s\" already exists at %p.", |
641 | mp_name, mp); | |
91fccdad KT |
642 | /* Ensure this reused mempool will not be freed. */ |
643 | dpdk_mp_do_not_free(mp); | |
d555d9bd RW |
644 | } else { |
645 | VLOG_ERR("Failed mempool \"%s\" create request of %u mbufs", | |
24e78f93 | 646 | mp_name, n_mbufs); |
0c6f39e5 | 647 | } |
24e78f93 | 648 | } while (!mp && rte_errno == ENOMEM && (n_mbufs /= 2) >= MIN_NB_MBUF); |
2ae3d542 | 649 | |
c2adb102 | 650 | ovs_mutex_unlock(&dpdk_mp_mutex); |
24e78f93 | 651 | return mp; |
8a9562d2 PS |
652 | } |
653 | ||
a08a115d | 654 | /* Release an existing mempool. */ |
8a9562d2 | 655 | static void |
91fccdad | 656 | dpdk_mp_release(struct rte_mempool *mp) |
8a9562d2 | 657 | { |
24e78f93 | 658 | if (!mp) { |
8a9562d2 PS |
659 | return; |
660 | } | |
661 | ||
c2adb102 | 662 | ovs_mutex_lock(&dpdk_mp_mutex); |
91fccdad KT |
663 | if (dpdk_mp_full(mp)) { |
664 | VLOG_DBG("Freeing mempool \"%s\"", mp->name); | |
665 | rte_mempool_free(mp); | |
666 | } else { | |
667 | struct dpdk_mp *dmp; | |
668 | ||
669 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
670 | if (dmp) { | |
671 | dmp->mp = mp; | |
672 | ovs_list_push_back(&dpdk_mp_free_list, &dmp->list_node); | |
673 | } | |
674 | } | |
c2adb102 | 675 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
676 | } |
677 | ||
b6b26021 FA |
678 | /* Tries to allocate a new mempool - or re-use an existing one where |
679 | * appropriate - on requested_socket_id with a size determined by | |
680 | * requested_mtu and requested Rx/Tx queues. | |
681 | * On success - or when re-using an existing mempool - the new configuration | |
682 | * will be applied. | |
0072e931 MK |
683 | * On error, device will be left unchanged. */ |
684 | static int | |
685 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
686 | OVS_REQUIRES(dev->mutex) |
687 | { | |
688 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
24e78f93 IM |
689 | struct rte_mempool *mp; |
690 | int ret = 0; | |
0072e931 | 691 | |
91fccdad KT |
692 | dpdk_mp_sweep(); |
693 | ||
24e78f93 | 694 | mp = dpdk_mp_create(dev, FRAME_LEN_TO_MTU(buf_size)); |
0072e931 | 695 | if (!mp) { |
c67e46c0 MK |
696 | VLOG_ERR("Failed to create memory pool for netdev " |
697 | "%s, with MTU %d on socket %d: %s\n", | |
698 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
699 | rte_strerror(rte_errno)); | |
24e78f93 | 700 | ret = rte_errno; |
0072e931 | 701 | } else { |
24e78f93 IM |
702 | /* If a new MTU was requested and its rounded value equals the one |
703 | * that is currently used, then the existing mempool is returned. */ | |
704 | if (dev->mp != mp) { | |
705 | /* A new mempool was created, release the previous one. */ | |
91fccdad | 706 | dpdk_mp_release(dev->mp); |
24e78f93 IM |
707 | } else { |
708 | ret = EEXIST; | |
709 | } | |
710 | dev->mp = mp; | |
0072e931 MK |
711 | dev->mtu = dev->requested_mtu; |
712 | dev->socket_id = dev->requested_socket_id; | |
713 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
714 | } | |
715 | ||
24e78f93 | 716 | return ret; |
0072e931 MK |
717 | } |
718 | ||
8a9562d2 PS |
719 | static void |
720 | check_link_status(struct netdev_dpdk *dev) | |
721 | { | |
722 | struct rte_eth_link link; | |
723 | ||
724 | rte_eth_link_get_nowait(dev->port_id, &link); | |
725 | ||
726 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 727 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
728 | |
729 | dev->link_reset_cnt++; | |
730 | dev->link = link; | |
731 | if (dev->link.link_status) { | |
fa9f4eeb IM |
732 | VLOG_DBG_RL(&rl, |
733 | "Port "DPDK_PORT_ID_FMT" Link Up - speed %u Mbps - %s", | |
58be5c0e | 734 | dev->port_id, (unsigned) dev->link.link_speed, |
fa9f4eeb IM |
735 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) |
736 | ? "full-duplex" : "half-duplex"); | |
8a9562d2 | 737 | } else { |
fa9f4eeb IM |
738 | VLOG_DBG_RL(&rl, "Port "DPDK_PORT_ID_FMT" Link Down", |
739 | dev->port_id); | |
8a9562d2 PS |
740 | } |
741 | } | |
742 | } | |
743 | ||
744 | static void * | |
745 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
746 | { | |
747 | struct netdev_dpdk *dev; | |
748 | ||
749 | pthread_detach(pthread_self()); | |
750 | ||
751 | for (;;) { | |
752 | ovs_mutex_lock(&dpdk_mutex); | |
753 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
754 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
755 | if (dev->type == DPDK_DEV_ETH) { |
756 | check_link_status(dev); | |
757 | } | |
8a9562d2 PS |
758 | ovs_mutex_unlock(&dev->mutex); |
759 | } | |
760 | ovs_mutex_unlock(&dpdk_mutex); | |
761 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
762 | } | |
763 | ||
764 | return NULL; | |
765 | } | |
766 | ||
b98d7669 DDP |
767 | static int |
768 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
769 | { | |
770 | int diag = 0; | |
771 | int i; | |
0072e931 | 772 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 773 | |
67fe6d63 MK |
774 | /* For some NICs (e.g. Niantic), scatter_rx mode needs to be explicitly |
775 | * enabled. */ | |
0072e931 | 776 | if (dev->mtu > ETHER_MTU) { |
67fe6d63 | 777 | conf.rxmode.enable_scatter = 1; |
0072e931 | 778 | } |
67fe6d63 | 779 | |
1a2bb118 SC |
780 | conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & |
781 | NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
b98d7669 DDP |
782 | /* A device may report more queues than it makes available (this has |
783 | * been observed for Intel xl710, which reserves some of them for | |
784 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
785 | * available. When this happens we can retry the configuration | |
786 | * and request less queues */ | |
787 | while (n_rxq && n_txq) { | |
788 | if (diag) { | |
789 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
790 | } | |
791 | ||
0072e931 | 792 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 793 | if (diag) { |
0072e931 MK |
794 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
795 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
796 | break; |
797 | } | |
798 | ||
67fe6d63 MK |
799 | diag = rte_eth_dev_set_mtu(dev->port_id, dev->mtu); |
800 | if (diag) { | |
801 | VLOG_ERR("Interface %s MTU (%d) setup error: %s", | |
802 | dev->up.name, dev->mtu, rte_strerror(-diag)); | |
803 | break; | |
804 | } | |
805 | ||
b98d7669 | 806 | for (i = 0; i < n_txq; i++) { |
b685696b | 807 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
808 | dev->socket_id, NULL); |
809 | if (diag) { | |
1dfebee9 | 810 | VLOG_INFO("Interface %s unable to setup txq(%d): %s", |
b98d7669 DDP |
811 | dev->up.name, i, rte_strerror(-diag)); |
812 | break; | |
813 | } | |
814 | } | |
815 | ||
816 | if (i != n_txq) { | |
817 | /* Retry with less tx queues */ | |
818 | n_txq = i; | |
819 | continue; | |
820 | } | |
821 | ||
822 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 823 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
24e78f93 | 824 | dev->socket_id, NULL, dev->mp); |
b98d7669 | 825 | if (diag) { |
1dfebee9 | 826 | VLOG_INFO("Interface %s unable to setup rxq(%d): %s", |
b98d7669 DDP |
827 | dev->up.name, i, rte_strerror(-diag)); |
828 | break; | |
829 | } | |
830 | } | |
831 | ||
832 | if (i != n_rxq) { | |
833 | /* Retry with less rx queues */ | |
834 | n_rxq = i; | |
835 | continue; | |
836 | } | |
837 | ||
838 | dev->up.n_rxq = n_rxq; | |
81acebda | 839 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
840 | |
841 | return 0; | |
842 | } | |
843 | ||
844 | return diag; | |
845 | } | |
846 | ||
9fd39370 SC |
847 | static void |
848 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
849 | { | |
850 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
fa9f4eeb | 851 | VLOG_WARN("Failed to enable flow control on device "DPDK_PORT_ID_FMT, |
bb37956a | 852 | dev->port_id); |
9fd39370 SC |
853 | } |
854 | } | |
b98d7669 | 855 | |
8a9562d2 | 856 | static int |
c2adb102 IM |
857 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
858 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
859 | { |
860 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 861 | struct rte_eth_dev_info info; |
8a9562d2 PS |
862 | struct ether_addr eth_addr; |
863 | int diag; | |
b98d7669 | 864 | int n_rxq, n_txq; |
d4f5282c KT |
865 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | |
866 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
867 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
8a9562d2 | 868 | |
a0cb2d66 | 869 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 870 | |
d4f5282c KT |
871 | if ((info.rx_offload_capa & rx_chksm_offload_capa) != |
872 | rx_chksm_offload_capa) { | |
fa9f4eeb IM |
873 | VLOG_WARN("Rx checksum offload is not supported on port " |
874 | DPDK_PORT_ID_FMT, dev->port_id); | |
d4f5282c KT |
875 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; |
876 | } else { | |
877 | dev->hw_ol_features |= NETDEV_RX_CHECKSUM_OFFLOAD; | |
878 | } | |
879 | ||
b98d7669 DDP |
880 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
881 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
882 | ||
883 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 884 | if (diag) { |
b98d7669 DDP |
885 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
886 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 887 | return -diag; |
8a9562d2 PS |
888 | } |
889 | ||
8a9562d2 PS |
890 | diag = rte_eth_dev_start(dev->port_id); |
891 | if (diag) { | |
b98d7669 DDP |
892 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
893 | rte_strerror(-diag)); | |
95fb793a | 894 | return -diag; |
8a9562d2 PS |
895 | } |
896 | ||
897 | rte_eth_promiscuous_enable(dev->port_id); | |
898 | rte_eth_allmulticast_enable(dev->port_id); | |
899 | ||
900 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
901 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
fa9f4eeb IM |
902 | VLOG_INFO_RL(&rl, "Port "DPDK_PORT_ID_FMT": "ETH_ADDR_FMT, |
903 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); | |
8a9562d2 | 904 | |
ca92d173 | 905 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
906 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
907 | ||
24e78f93 | 908 | mbp_priv = rte_mempool_get_priv(dev->mp); |
8a9562d2 PS |
909 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; |
910 | ||
9fd39370 SC |
911 | /* Get the Flow control configuration for DPDK-ETH */ |
912 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
913 | if (diag) { | |
fa9f4eeb IM |
914 | VLOG_DBG("cannot get flow control parameters on port "DPDK_PORT_ID_FMT |
915 | ", err=%d", dev->port_id, diag); | |
9fd39370 SC |
916 | } |
917 | ||
8a9562d2 PS |
918 | return 0; |
919 | } | |
920 | ||
921 | static struct netdev_dpdk * | |
922 | netdev_dpdk_cast(const struct netdev *netdev) | |
923 | { | |
924 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
925 | } | |
926 | ||
927 | static struct netdev * | |
928 | netdev_dpdk_alloc(void) | |
929 | { | |
bab69409 AC |
930 | struct netdev_dpdk *dev; |
931 | ||
65e19e70 DDP |
932 | dev = dpdk_rte_mzalloc(sizeof *dev); |
933 | if (dev) { | |
934 | return &dev->up; | |
bab69409 | 935 | } |
65e19e70 | 936 | |
bab69409 | 937 | return NULL; |
8a9562d2 PS |
938 | } |
939 | ||
eff23640 DDP |
940 | static struct dpdk_tx_queue * |
941 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 942 | { |
eff23640 | 943 | struct dpdk_tx_queue *txqs; |
bd5131ba | 944 | unsigned i; |
5a034064 | 945 | |
eff23640 DDP |
946 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
947 | if (txqs) { | |
948 | for (i = 0; i < n_txqs; i++) { | |
949 | /* Initialize map for vhost devices. */ | |
950 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
951 | rte_spinlock_init(&txqs[i].tx_lock); | |
952 | } | |
5a034064 | 953 | } |
eff23640 DDP |
954 | |
955 | return txqs; | |
5a034064 AW |
956 | } |
957 | ||
8a9562d2 | 958 | static int |
bb37956a | 959 | common_construct(struct netdev *netdev, dpdk_port_t port_no, |
1ce30dfd | 960 | enum dpdk_dev_type type, int socket_id) |
5a034064 | 961 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 962 | { |
d46285a2 | 963 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 964 | |
d46285a2 | 965 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 966 | |
d46285a2 | 967 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 968 | |
1b7a04e0 AW |
969 | /* If the 'sid' is negative, it means that the kernel fails |
970 | * to obtain the pci numa info. In that situation, always | |
971 | * use 'SOCKET0'. */ | |
1ce30dfd | 972 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 973 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
974 | dev->port_id = port_no; |
975 | dev->type = type; | |
976 | dev->flags = 0; | |
7f381c2e | 977 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 978 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
979 | ovsrcu_index_init(&dev->vid, -1); |
980 | dev->vhost_reconfigured = false; | |
5dcde09c | 981 | dev->attached = false; |
8a9562d2 | 982 | |
78bd47cf | 983 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 984 | |
9509913a IS |
985 | ovsrcu_init(&dev->ingress_policer, NULL); |
986 | dev->policer_rate = 0; | |
987 | dev->policer_burst = 0; | |
988 | ||
7f381c2e DDP |
989 | netdev->n_rxq = 0; |
990 | netdev->n_txq = 0; | |
991 | dev->requested_n_rxq = NR_QUEUE; | |
992 | dev->requested_n_txq = NR_QUEUE; | |
993 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
994 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 995 | |
9fd39370 SC |
996 | /* Initialize the flow control to NULL */ |
997 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
998 | |
999 | /* Initilize the hardware offload flags to 0 */ | |
1000 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
1001 | |
1002 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
1003 | ||
d46285a2 | 1004 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 1005 | |
7f381c2e DDP |
1006 | netdev_request_reconfigure(netdev); |
1007 | ||
971f4b39 MW |
1008 | dev->rte_xstats_names = NULL; |
1009 | dev->rte_xstats_names_size = 0; | |
1010 | ||
1011 | dev->rte_xstats_ids = NULL; | |
1012 | dev->rte_xstats_ids_size = 0; | |
1013 | ||
1ce30dfd | 1014 | return 0; |
95fb793a | 1015 | } |
1016 | ||
b83a2df1 MV |
1017 | /* dev_name must be the prefix followed by a positive decimal number. |
1018 | * (no leading + or - signs are allowed) */ | |
95fb793a | 1019 | static int |
1020 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
1021 | unsigned int *port_no) | |
1022 | { | |
1023 | const char *cport; | |
1024 | ||
1025 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
1026 | return ENODEV; | |
1027 | } | |
1028 | ||
1029 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
1030 | |
1031 | if (str_to_uint(cport, 10, port_no)) { | |
1032 | return 0; | |
1033 | } else { | |
1034 | return ENODEV; | |
1035 | } | |
95fb793a | 1036 | } |
1037 | ||
1ce30dfd DDP |
1038 | static int |
1039 | vhost_common_construct(struct netdev *netdev) | |
1040 | OVS_REQUIRES(dpdk_mutex) | |
1041 | { | |
1042 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
1043 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1044 | ||
1045 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); | |
1046 | if (!dev->tx_q) { | |
1047 | return ENOMEM; | |
1048 | } | |
1049 | ||
bb37956a IM |
1050 | return common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1051 | DPDK_DEV_VHOST, socket_id); | |
1ce30dfd DDP |
1052 | } |
1053 | ||
7d1ced01 | 1054 | static int |
53f50d24 | 1055 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 1056 | { |
d46285a2 DDP |
1057 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1058 | const char *name = netdev->name; | |
7d1ced01 | 1059 | int err; |
a0cb2d66 | 1060 | |
1af27e8a DDP |
1061 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
1062 | * the file system. '/' or '\' would traverse directories, so they're not | |
1063 | * acceptable in 'name'. */ | |
1064 | if (strchr(name, '/') || strchr(name, '\\')) { | |
1065 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
1066 | "A valid name must not include '/' or '\\'", | |
1067 | name); | |
1068 | return EINVAL; | |
1069 | } | |
1070 | ||
7d1ced01 CL |
1071 | ovs_mutex_lock(&dpdk_mutex); |
1072 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 1073 | * the socket is to be created, then register the socket. |
7d1ced01 | 1074 | */ |
2d24d165 | 1075 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 1076 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 1077 | |
2d24d165 CL |
1078 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
1079 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
1080 | if (err) { |
1081 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 1082 | dev->vhost_id); |
f3e7ec25 | 1083 | goto out; |
e5c0f5a4 | 1084 | } else { |
2d24d165 CL |
1085 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
1086 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
1087 | dev->vhost_id, name); | |
1088 | } | |
f3e7ec25 MW |
1089 | |
1090 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
1091 | &virtio_net_device_ops); | |
1092 | if (err) { | |
1093 | VLOG_ERR("rte_vhost_driver_callback_register failed for vhost user " | |
1094 | "port: %s\n", name); | |
1095 | goto out; | |
1096 | } | |
1097 | ||
1098 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
1099 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
1100 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
1101 | | 1ULL << VIRTIO_NET_F_CSUM); | |
1102 | if (err) { | |
1103 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
1104 | "port: %s\n", name); | |
1105 | goto out; | |
1106 | } | |
1107 | ||
1108 | err = rte_vhost_driver_start(dev->vhost_id); | |
1109 | if (err) { | |
1110 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
1111 | "port: %s\n", name); | |
1112 | goto out; | |
1113 | } | |
1114 | ||
1ce30dfd | 1115 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1116 | if (err) { |
1117 | VLOG_ERR("vhost_common_construct failed for vhost user " | |
1118 | "port: %s\n", name); | |
1119 | } | |
2d24d165 | 1120 | |
f3e7ec25 | 1121 | out: |
2d24d165 | 1122 | ovs_mutex_unlock(&dpdk_mutex); |
28ca969e AC |
1123 | VLOG_WARN_ONCE("dpdkvhostuser ports are considered deprecated; " |
1124 | "please migrate to dpdkvhostuserclient ports."); | |
2d24d165 CL |
1125 | return err; |
1126 | } | |
1127 | ||
1128 | static int | |
1129 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
1130 | { | |
1131 | int err; | |
1132 | ||
2d24d165 | 1133 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 1134 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1135 | if (err) { |
1136 | VLOG_ERR("vhost_common_construct failed for vhost user client" | |
1137 | "port: %s\n", netdev->name); | |
1138 | } | |
7d1ced01 | 1139 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
1140 | return err; |
1141 | } | |
1142 | ||
95fb793a | 1143 | static int |
1144 | netdev_dpdk_construct(struct netdev *netdev) | |
1145 | { | |
95fb793a | 1146 | int err; |
1147 | ||
95fb793a | 1148 | ovs_mutex_lock(&dpdk_mutex); |
bb37956a IM |
1149 | err = common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1150 | DPDK_DEV_ETH, SOCKET0); | |
8a9562d2 PS |
1151 | ovs_mutex_unlock(&dpdk_mutex); |
1152 | return err; | |
1153 | } | |
1154 | ||
1ce30dfd DDP |
1155 | static void |
1156 | common_destruct(struct netdev_dpdk *dev) | |
1157 | OVS_REQUIRES(dpdk_mutex) | |
1158 | OVS_EXCLUDED(dev->mutex) | |
1159 | { | |
1160 | rte_free(dev->tx_q); | |
91fccdad | 1161 | dpdk_mp_release(dev->mp); |
1ce30dfd DDP |
1162 | |
1163 | ovs_list_remove(&dev->list_node); | |
1164 | free(ovsrcu_get_protected(struct ingress_policer *, | |
1165 | &dev->ingress_policer)); | |
1166 | ovs_mutex_destroy(&dev->mutex); | |
1167 | } | |
1168 | ||
8a9562d2 | 1169 | static void |
d46285a2 | 1170 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 1171 | { |
d46285a2 | 1172 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5dcde09c | 1173 | char devname[RTE_ETH_NAME_MAX_LEN]; |
8a9562d2 | 1174 | |
8d38823b | 1175 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1176 | |
8a9562d2 | 1177 | rte_eth_dev_stop(dev->port_id); |
5dcde09c IM |
1178 | |
1179 | if (dev->attached) { | |
1180 | rte_eth_dev_close(dev->port_id); | |
1181 | if (rte_eth_dev_detach(dev->port_id, devname) < 0) { | |
1182 | VLOG_ERR("Device '%s' can not be detached", dev->devargs); | |
1183 | } else { | |
0ee821c2 | 1184 | VLOG_INFO("Device '%s' has been detached", devname); |
5dcde09c IM |
1185 | } |
1186 | } | |
1187 | ||
ac1a9bb9 | 1188 | netdev_dpdk_clear_xstats(dev); |
55e075e6 | 1189 | free(dev->devargs); |
1ce30dfd | 1190 | common_destruct(dev); |
8d38823b | 1191 | |
8a9562d2 | 1192 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1193 | } |
8a9562d2 | 1194 | |
3f891bbe DDP |
1195 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1196 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1197 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1198 | static int | |
c1ff66ac CL |
1199 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1200 | char *vhost_id) | |
3f891bbe DDP |
1201 | OVS_EXCLUDED(dpdk_mutex) |
1202 | OVS_EXCLUDED(dev->mutex) | |
1203 | { | |
c1ff66ac | 1204 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1205 | } |
1206 | ||
58397e6c | 1207 | static void |
d46285a2 | 1208 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1209 | { |
d46285a2 | 1210 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1211 | char *vhost_id; |
58397e6c | 1212 | |
8d38823b | 1213 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1214 | |
c62da695 | 1215 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1216 | if (netdev_dpdk_get_vid(dev) >= 0 |
1217 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1218 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1219 | netdev->name); |
58be5c0e MK |
1220 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1221 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1222 | } |
1223 | ||
2d24d165 | 1224 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1225 | |
1ce30dfd DDP |
1226 | common_destruct(dev); |
1227 | ||
58397e6c | 1228 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1229 | |
569c26da | 1230 | if (!vhost_id[0]) { |
821b8664 IM |
1231 | goto out; |
1232 | } | |
1233 | ||
c1ff66ac | 1234 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1235 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1236 | netdev->name, vhost_id); | |
c1ff66ac CL |
1237 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1238 | /* OVS server mode - remove this socket from list for deletion */ | |
1239 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1240 | } |
821b8664 | 1241 | out: |
c1ff66ac | 1242 | free(vhost_id); |
8a9562d2 PS |
1243 | } |
1244 | ||
1245 | static void | |
d46285a2 | 1246 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1247 | { |
d46285a2 | 1248 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1249 | |
d46285a2 | 1250 | rte_free(dev); |
8a9562d2 PS |
1251 | } |
1252 | ||
971f4b39 | 1253 | static void |
ac1a9bb9 | 1254 | netdev_dpdk_clear_xstats(struct netdev_dpdk *dev) |
971f4b39 MW |
1255 | { |
1256 | /* If statistics are already allocated, we have to | |
1257 | * reconfigure, as port_id could have been changed. */ | |
1258 | if (dev->rte_xstats_names) { | |
1259 | free(dev->rte_xstats_names); | |
1260 | dev->rte_xstats_names = NULL; | |
1261 | dev->rte_xstats_names_size = 0; | |
1262 | } | |
1263 | if (dev->rte_xstats_ids) { | |
1264 | free(dev->rte_xstats_ids); | |
1265 | dev->rte_xstats_ids = NULL; | |
1266 | dev->rte_xstats_ids_size = 0; | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | static const char* | |
1271 | netdev_dpdk_get_xstat_name(struct netdev_dpdk *dev, uint64_t id) | |
1272 | { | |
1273 | if (id >= dev->rte_xstats_names_size) { | |
1274 | return "UNKNOWN"; | |
1275 | } | |
1276 | return dev->rte_xstats_names[id].name; | |
1277 | } | |
1278 | ||
1279 | static bool | |
1280 | netdev_dpdk_configure_xstats(struct netdev_dpdk *dev) | |
1281 | OVS_REQUIRES(dev->mutex) | |
1282 | { | |
1283 | int rte_xstats_len; | |
1284 | bool ret; | |
1285 | struct rte_eth_xstat *rte_xstats; | |
1286 | uint64_t id; | |
1287 | int xstats_no; | |
1288 | const char *name; | |
1289 | ||
1290 | /* Retrieving all XSTATS names. If something will go wrong | |
1291 | * or amount of counters will be equal 0, rte_xstats_names | |
1292 | * buffer will be marked as NULL, and any further xstats | |
1293 | * query won't be performed (e.g. during netdev_dpdk_get_stats | |
1294 | * execution). */ | |
1295 | ||
1296 | ret = false; | |
1297 | rte_xstats = NULL; | |
1298 | ||
1299 | if (dev->rte_xstats_names == NULL || dev->rte_xstats_ids == NULL) { | |
1300 | dev->rte_xstats_names_size = | |
1301 | rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
1302 | ||
1303 | if (dev->rte_xstats_names_size < 0) { | |
fa9f4eeb IM |
1304 | VLOG_WARN("Cannot get XSTATS for port: "DPDK_PORT_ID_FMT, |
1305 | dev->port_id); | |
971f4b39 MW |
1306 | dev->rte_xstats_names_size = 0; |
1307 | } else { | |
1308 | /* Reserve memory for xstats names and values */ | |
1309 | dev->rte_xstats_names = xcalloc(dev->rte_xstats_names_size, | |
1310 | sizeof *dev->rte_xstats_names); | |
1311 | ||
1312 | if (dev->rte_xstats_names) { | |
1313 | /* Retreive xstats names */ | |
1314 | rte_xstats_len = | |
1315 | rte_eth_xstats_get_names(dev->port_id, | |
1316 | dev->rte_xstats_names, | |
1317 | dev->rte_xstats_names_size); | |
1318 | ||
1319 | if (rte_xstats_len < 0) { | |
fa9f4eeb IM |
1320 | VLOG_WARN("Cannot get XSTATS names for port: " |
1321 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 MW |
1322 | goto out; |
1323 | } else if (rte_xstats_len != dev->rte_xstats_names_size) { | |
fa9f4eeb IM |
1324 | VLOG_WARN("XSTATS size doesn't match for port: " |
1325 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 MW |
1326 | goto out; |
1327 | } | |
1328 | ||
1329 | dev->rte_xstats_ids = xcalloc(dev->rte_xstats_names_size, | |
1330 | sizeof(uint64_t)); | |
1331 | ||
1332 | /* We have to calculate number of counters */ | |
1333 | rte_xstats = xmalloc(rte_xstats_len * sizeof *rte_xstats); | |
1334 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
1335 | ||
1336 | /* Retreive xstats values */ | |
1337 | if (rte_eth_xstats_get(dev->port_id, rte_xstats, | |
1338 | rte_xstats_len) > 0) { | |
1339 | dev->rte_xstats_ids_size = 0; | |
1340 | xstats_no = 0; | |
1341 | for (uint32_t i = 0; i < rte_xstats_len; i++) { | |
1342 | id = rte_xstats[i].id; | |
1343 | name = netdev_dpdk_get_xstat_name(dev, id); | |
1344 | /* We need to filter out everything except | |
1345 | * dropped, error and management counters */ | |
1346 | if (string_ends_with(name, "_errors") || | |
1347 | strstr(name, "_management_") || | |
1348 | string_ends_with(name, "_dropped")) { | |
1349 | ||
1350 | dev->rte_xstats_ids[xstats_no] = id; | |
1351 | xstats_no++; | |
1352 | } | |
1353 | } | |
1354 | dev->rte_xstats_ids_size = xstats_no; | |
1355 | ret = true; | |
1356 | } else { | |
fa9f4eeb IM |
1357 | VLOG_WARN("Can't get XSTATS IDs for port: " |
1358 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 | 1359 | } |
34eb0863 IM |
1360 | |
1361 | free(rte_xstats); | |
971f4b39 MW |
1362 | } |
1363 | } | |
1364 | } else { | |
1365 | /* Already configured */ | |
1366 | ret = true; | |
1367 | } | |
1368 | ||
1369 | out: | |
1370 | if (!ret) { | |
1371 | netdev_dpdk_clear_xstats(dev); | |
1372 | } | |
1373 | return ret; | |
1374 | } | |
1375 | ||
8a9562d2 | 1376 | static int |
a14b8947 | 1377 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1378 | { |
a14b8947 | 1379 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1380 | |
1381 | ovs_mutex_lock(&dev->mutex); | |
1382 | ||
050c60bf | 1383 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1384 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1385 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1386 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1387 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1388 | |
1389 | if (dev->type == DPDK_DEV_ETH) { | |
1390 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1391 | dev->requested_rxq_size); | |
1392 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1393 | dev->rxq_size); | |
1394 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1395 | dev->requested_txq_size); | |
1396 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1397 | dev->txq_size); | |
1a2bb118 SC |
1398 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1399 | smap_add(args, "rx_csum_offload", "true"); | |
8155ab7e KT |
1400 | } else { |
1401 | smap_add(args, "rx_csum_offload", "false"); | |
1a2bb118 | 1402 | } |
451f26fd | 1403 | } |
8a9562d2 PS |
1404 | ovs_mutex_unlock(&dev->mutex); |
1405 | ||
1406 | return 0; | |
1407 | } | |
1408 | ||
55e075e6 | 1409 | static struct netdev_dpdk * |
bb37956a | 1410 | netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id) |
55e075e6 CL |
1411 | OVS_REQUIRES(dpdk_mutex) |
1412 | { | |
1413 | struct netdev_dpdk *dev; | |
1414 | ||
1415 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1416 | if (dev->port_id == port_id) { | |
1417 | return dev; | |
1418 | } | |
1419 | } | |
1420 | ||
1421 | return NULL; | |
1422 | } | |
1423 | ||
5e758818 YL |
1424 | static dpdk_port_t |
1425 | netdev_dpdk_get_port_by_mac(const char *mac_str) | |
1426 | { | |
1427 | dpdk_port_t port_id; | |
1428 | struct eth_addr mac, port_mac; | |
1429 | ||
1430 | if (!eth_addr_from_string(mac_str, &mac)) { | |
1431 | VLOG_ERR("invalid mac: %s", mac_str); | |
1432 | return DPDK_ETH_PORT_ID_INVALID; | |
1433 | } | |
1434 | ||
1435 | RTE_ETH_FOREACH_DEV (port_id) { | |
1436 | struct ether_addr ea; | |
1437 | ||
1438 | rte_eth_macaddr_get(port_id, &ea); | |
1439 | memcpy(port_mac.ea, ea.addr_bytes, ETH_ADDR_LEN); | |
1440 | if (eth_addr_equals(mac, port_mac)) { | |
1441 | return port_id; | |
1442 | } | |
1443 | } | |
1444 | ||
1445 | return DPDK_ETH_PORT_ID_INVALID; | |
1446 | } | |
1447 | ||
1448 | /* | |
1449 | * Normally, a PCI id is enough for identifying a specific DPDK port. | |
1450 | * However, for some NICs having multiple ports sharing the same PCI | |
1451 | * id, using PCI id won't work then. | |
1452 | * | |
1453 | * To fix that, here one more method is introduced: "class=eth,mac=$MAC". | |
1454 | * | |
1455 | * Note that the compatibility is fully kept: user can still use the | |
1456 | * PCI id for adding ports (when it's enough for them). | |
1457 | */ | |
bb37956a | 1458 | static dpdk_port_t |
5dcde09c IM |
1459 | netdev_dpdk_process_devargs(struct netdev_dpdk *dev, |
1460 | const char *devargs, char **errp) | |
55e075e6 | 1461 | { |
5e758818 | 1462 | char *name; |
bb37956a | 1463 | dpdk_port_t new_port_id = DPDK_ETH_PORT_ID_INVALID; |
55e075e6 | 1464 | |
5e758818 YL |
1465 | if (strncmp(devargs, "class=eth,mac=", 14) == 0) { |
1466 | new_port_id = netdev_dpdk_get_port_by_mac(&devargs[14]); | |
1467 | } else { | |
1468 | name = xmemdup0(devargs, strcspn(devargs, ",")); | |
1469 | if (rte_eth_dev_get_port_by_name(name, &new_port_id) | |
1470 | || !rte_eth_dev_is_valid_port(new_port_id)) { | |
1471 | /* Device not found in DPDK, attempt to attach it */ | |
1472 | if (!rte_eth_dev_attach(devargs, &new_port_id)) { | |
1473 | /* Attach successful */ | |
1474 | dev->attached = true; | |
1475 | VLOG_INFO("Device '%s' attached to DPDK", devargs); | |
1476 | } else { | |
1477 | /* Attach unsuccessful */ | |
1478 | new_port_id = DPDK_ETH_PORT_ID_INVALID; | |
1479 | } | |
55e075e6 | 1480 | } |
5e758818 YL |
1481 | free(name); |
1482 | } | |
1483 | ||
1484 | if (new_port_id == DPDK_ETH_PORT_ID_INVALID) { | |
1485 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", devargs); | |
55e075e6 CL |
1486 | } |
1487 | ||
1488 | return new_port_id; | |
1489 | } | |
1490 | ||
c3d062a7 CL |
1491 | static void |
1492 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1493 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1494 | { |
050c60bf | 1495 | int new_n_rxq; |
a14b8947 | 1496 | |
2a21e757 | 1497 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1498 | if (new_n_rxq != dev->requested_n_rxq) { |
1499 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1500 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1501 | } |
c3d062a7 CL |
1502 | } |
1503 | ||
b685696b CL |
1504 | static void |
1505 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1506 | const char *flag, int default_size, int *new_size) | |
1507 | { | |
1508 | int queue_size = smap_get_int(args, flag, default_size); | |
1509 | ||
1510 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1511 | || !is_pow2(queue_size)) { | |
1512 | queue_size = default_size; | |
1513 | } | |
1514 | ||
1515 | if (queue_size != *new_size) { | |
1516 | *new_size = queue_size; | |
1517 | netdev_request_reconfigure(netdev); | |
1518 | } | |
1519 | } | |
1520 | ||
c3d062a7 | 1521 | static int |
9fff138e DDP |
1522 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1523 | char **errp) | |
c3d062a7 CL |
1524 | { |
1525 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
b614c894 IM |
1526 | bool rx_fc_en, tx_fc_en, autoneg; |
1527 | enum rte_eth_fc_mode fc_mode; | |
1528 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1529 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1530 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1531 | }; | |
55e075e6 CL |
1532 | const char *new_devargs; |
1533 | int err = 0; | |
c3d062a7 | 1534 | |
55e075e6 | 1535 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1536 | ovs_mutex_lock(&dev->mutex); |
1537 | ||
1538 | dpdk_set_rxq_config(dev, args); | |
1539 | ||
b685696b CL |
1540 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1541 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1542 | &dev->requested_rxq_size); | |
1543 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1544 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1545 | &dev->requested_txq_size); | |
1546 | ||
55e075e6 CL |
1547 | new_devargs = smap_get(args, "dpdk-devargs"); |
1548 | ||
1549 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1550 | /* The user requested a new device. If we return error, the caller | |
1551 | * will delete this netdev and try to recreate it. */ | |
1552 | err = EAGAIN; | |
1553 | goto out; | |
1554 | } | |
1555 | ||
1556 | /* dpdk-devargs is required for device configuration */ | |
1557 | if (new_devargs && new_devargs[0]) { | |
1558 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1559 | * is valid */ | |
1560 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1561 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
bb37956a IM |
1562 | dpdk_port_t new_port_id = netdev_dpdk_process_devargs(dev, |
1563 | new_devargs, | |
1564 | errp); | |
55e075e6 CL |
1565 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1566 | err = EINVAL; | |
1567 | } else if (new_port_id == dev->port_id) { | |
1568 | /* Already configured, do not reconfigure again */ | |
1569 | err = 0; | |
1570 | } else { | |
1571 | struct netdev_dpdk *dup_dev; | |
bb37956a | 1572 | |
55e075e6 CL |
1573 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); |
1574 | if (dup_dev) { | |
9fff138e DDP |
1575 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
1576 | "which is already in use by '%s'", | |
1577 | netdev_get_name(netdev), new_devargs, | |
1578 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1579 | err = EADDRINUSE; |
1580 | } else { | |
bd4e172b | 1581 | int sid = rte_eth_dev_socket_id(new_port_id); |
bb37956a | 1582 | |
bd4e172b | 1583 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; |
55e075e6 CL |
1584 | dev->devargs = xstrdup(new_devargs); |
1585 | dev->port_id = new_port_id; | |
1586 | netdev_request_reconfigure(&dev->up); | |
971f4b39 | 1587 | netdev_dpdk_clear_xstats(dev); |
55e075e6 CL |
1588 | err = 0; |
1589 | } | |
1590 | } | |
1591 | } | |
1592 | } else { | |
9fff138e DDP |
1593 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1594 | "The old 'dpdk<port_id>' names are not supported", | |
1595 | netdev_get_name(netdev)); | |
55e075e6 CL |
1596 | err = EINVAL; |
1597 | } | |
1598 | ||
1599 | if (err) { | |
1600 | goto out; | |
1601 | } | |
1602 | ||
c3d062a7 CL |
1603 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1604 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1605 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1606 | |
b614c894 IM |
1607 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1608 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1609 | dev->fc_conf.mode = fc_mode; | |
1610 | dev->fc_conf.autoneg = autoneg; | |
1611 | dpdk_eth_flow_ctrl_setup(dev); | |
1612 | } | |
9fd39370 | 1613 | |
55e075e6 | 1614 | out: |
c3d062a7 | 1615 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1616 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1617 | |
55e075e6 | 1618 | return err; |
c3d062a7 CL |
1619 | } |
1620 | ||
1621 | static int | |
9fff138e DDP |
1622 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1623 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1624 | { |
1625 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1626 | ||
1627 | ovs_mutex_lock(&dev->mutex); | |
1628 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1629 | ovs_mutex_unlock(&dev->mutex); |
1630 | ||
1631 | return 0; | |
1632 | } | |
1633 | ||
c1ff66ac | 1634 | static int |
2d24d165 | 1635 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1636 | const struct smap *args, |
1637 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1638 | { |
1639 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1640 | const char *path; | |
1641 | ||
6881885a | 1642 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1643 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1644 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1645 | if (path && strcmp(path, dev->vhost_id)) { |
1646 | strcpy(dev->vhost_id, path); | |
10087cba CL |
1647 | /* check zero copy configuration */ |
1648 | if (smap_get_bool(args, "dq-zero-copy", false)) { | |
1649 | dev->vhost_driver_flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
1650 | } else { | |
1651 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
1652 | } | |
c1ff66ac CL |
1653 | netdev_request_reconfigure(netdev); |
1654 | } | |
1655 | } | |
6881885a | 1656 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1657 | |
1658 | return 0; | |
1659 | } | |
1660 | ||
7dec44fe | 1661 | static int |
d46285a2 | 1662 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1663 | { |
d46285a2 | 1664 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1665 | |
d46285a2 | 1666 | return dev->socket_id; |
7dec44fe AW |
1667 | } |
1668 | ||
050c60bf | 1669 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1670 | static int |
050c60bf | 1671 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1672 | { |
d46285a2 | 1673 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1674 | |
d46285a2 | 1675 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1676 | |
050c60bf DDP |
1677 | if (dev->requested_n_txq == n_txq) { |
1678 | goto out; | |
4573fbd3 FL |
1679 | } |
1680 | ||
050c60bf DDP |
1681 | dev->requested_n_txq = n_txq; |
1682 | netdev_request_reconfigure(netdev); | |
58397e6c | 1683 | |
050c60bf | 1684 | out: |
d46285a2 | 1685 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1686 | return 0; |
58397e6c KT |
1687 | } |
1688 | ||
8a9562d2 PS |
1689 | static struct netdev_rxq * |
1690 | netdev_dpdk_rxq_alloc(void) | |
1691 | { | |
1692 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1693 | ||
eff23640 DDP |
1694 | if (rx) { |
1695 | return &rx->up; | |
1696 | } | |
1697 | ||
1698 | return NULL; | |
8a9562d2 PS |
1699 | } |
1700 | ||
1701 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1702 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1703 | { |
d46285a2 | 1704 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1705 | } |
1706 | ||
1707 | static int | |
d46285a2 | 1708 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1709 | { |
d46285a2 DDP |
1710 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1711 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1712 | |
d46285a2 DDP |
1713 | ovs_mutex_lock(&dev->mutex); |
1714 | rx->port_id = dev->port_id; | |
1715 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1716 | |
1717 | return 0; | |
1718 | } | |
1719 | ||
1720 | static void | |
d46285a2 | 1721 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1722 | { |
1723 | } | |
1724 | ||
1725 | static void | |
d46285a2 | 1726 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1727 | { |
d46285a2 | 1728 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1729 | |
1730 | rte_free(rx); | |
1731 | } | |
1732 | ||
819f13bd DDP |
1733 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1734 | * 'pkts', even in case of failure. | |
1735 | * | |
1736 | * Returns the number of packets that weren't transmitted. */ | |
1737 | static inline int | |
b59cc14e | 1738 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 1739 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 1740 | { |
1304f1f8 DDP |
1741 | uint32_t nb_tx = 0; |
1742 | ||
b59cc14e | 1743 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1744 | uint32_t ret; |
1745 | ||
b59cc14e | 1746 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1747 | if (!ret) { |
1748 | break; | |
1749 | } | |
1750 | ||
1751 | nb_tx += ret; | |
1752 | } | |
8a9562d2 | 1753 | |
b59cc14e | 1754 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 1755 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
1756 | * packet could come from a different mempool) */ |
1757 | int i; | |
1758 | ||
b59cc14e IM |
1759 | for (i = nb_tx; i < cnt; i++) { |
1760 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1761 | } |
8a9562d2 | 1762 | } |
819f13bd DDP |
1763 | |
1764 | return cnt - nb_tx; | |
8a9562d2 PS |
1765 | } |
1766 | ||
f3926f29 IS |
1767 | static inline bool |
1768 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1769 | struct rte_mbuf *pkt, uint64_t time) | |
1770 | { | |
1771 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1772 | ||
1773 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1774 | e_RTE_METER_GREEN; | |
1775 | } | |
1776 | ||
1777 | static int | |
1778 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
3e90f7d7 GZ |
1779 | struct rte_mbuf **pkts, int pkt_cnt, |
1780 | bool may_steal) | |
f3926f29 IS |
1781 | { |
1782 | int i = 0; | |
1783 | int cnt = 0; | |
1784 | struct rte_mbuf *pkt = NULL; | |
1785 | uint64_t current_time = rte_rdtsc(); | |
1786 | ||
1787 | for (i = 0; i < pkt_cnt; i++) { | |
1788 | pkt = pkts[i]; | |
1789 | /* Handle current packet */ | |
1790 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1791 | if (cnt != i) { | |
1792 | pkts[cnt] = pkt; | |
1793 | } | |
1794 | cnt++; | |
1795 | } else { | |
3e90f7d7 GZ |
1796 | if (may_steal) { |
1797 | rte_pktmbuf_free(pkt); | |
1798 | } | |
f3926f29 IS |
1799 | } |
1800 | } | |
1801 | ||
1802 | return cnt; | |
1803 | } | |
1804 | ||
9509913a IS |
1805 | static int |
1806 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
3e90f7d7 | 1807 | int pkt_cnt, bool may_steal) |
9509913a IS |
1808 | { |
1809 | int cnt = 0; | |
1810 | ||
1811 | rte_spinlock_lock(&policer->policer_lock); | |
3e90f7d7 GZ |
1812 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, |
1813 | pkt_cnt, may_steal); | |
9509913a IS |
1814 | rte_spinlock_unlock(&policer->policer_lock); |
1815 | ||
1816 | return cnt; | |
1817 | } | |
1818 | ||
58397e6c | 1819 | static bool |
0a0f39df | 1820 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1821 | { |
0a0f39df | 1822 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1823 | } |
1824 | ||
d6e3feb5 | 1825 | static inline void |
1826 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1827 | unsigned int packet_size) | |
1828 | { | |
1829 | /* Hard-coded search for the size bucket. */ | |
1830 | if (packet_size < 256) { | |
1831 | if (packet_size >= 128) { | |
1832 | stats->rx_128_to_255_packets++; | |
1833 | } else if (packet_size <= 64) { | |
1834 | stats->rx_1_to_64_packets++; | |
1835 | } else { | |
1836 | stats->rx_65_to_127_packets++; | |
1837 | } | |
1838 | } else { | |
1839 | if (packet_size >= 1523) { | |
1840 | stats->rx_1523_to_max_packets++; | |
1841 | } else if (packet_size >= 1024) { | |
1842 | stats->rx_1024_to_1522_packets++; | |
1843 | } else if (packet_size < 512) { | |
1844 | stats->rx_256_to_511_packets++; | |
1845 | } else { | |
1846 | stats->rx_512_to_1023_packets++; | |
1847 | } | |
1848 | } | |
1849 | } | |
1850 | ||
9e3ddd45 TP |
1851 | static inline void |
1852 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1853 | struct dp_packet **packets, int count, |
1854 | int dropped) | |
9e3ddd45 TP |
1855 | { |
1856 | int i; | |
d6e3feb5 | 1857 | unsigned int packet_size; |
9e3ddd45 TP |
1858 | struct dp_packet *packet; |
1859 | ||
1860 | stats->rx_packets += count; | |
9509913a | 1861 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1862 | for (i = 0; i < count; i++) { |
1863 | packet = packets[i]; | |
d6e3feb5 | 1864 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1865 | |
d6e3feb5 | 1866 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1867 | /* This only protects the following multicast counting from |
1868 | * too short packets, but it does not stop the packet from | |
1869 | * further processing. */ | |
1870 | stats->rx_errors++; | |
1871 | stats->rx_length_errors++; | |
1872 | continue; | |
1873 | } | |
1874 | ||
d6e3feb5 | 1875 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1876 | ||
9e3ddd45 TP |
1877 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1878 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1879 | stats->multicast++; | |
1880 | } | |
1881 | ||
d6e3feb5 | 1882 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1883 | } |
1884 | } | |
1885 | ||
58397e6c KT |
1886 | /* |
1887 | * The receive path for the vhost port is the TX path out from guest. | |
1888 | */ | |
1889 | static int | |
d46285a2 | 1890 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1891 | struct dp_packet_batch *batch) |
58397e6c | 1892 | { |
d46285a2 | 1893 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
9509913a | 1894 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1895 | uint16_t nb_rx = 0; |
9509913a | 1896 | uint16_t dropped = 0; |
daf22bf7 IM |
1897 | int qid = rxq->queue_id; |
1898 | int vid = netdev_dpdk_get_vid(dev); | |
58397e6c | 1899 | |
daf22bf7 | 1900 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured |
e543851d | 1901 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1902 | return EAGAIN; |
1903 | } | |
1904 | ||
daf22bf7 | 1905 | nb_rx = rte_vhost_dequeue_burst(vid, qid * VIRTIO_QNUM + VIRTIO_TXQ, |
24e78f93 | 1906 | dev->mp, |
64839cf4 | 1907 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1908 | NETDEV_MAX_BURST); |
58397e6c KT |
1909 | if (!nb_rx) { |
1910 | return EAGAIN; | |
1911 | } | |
1912 | ||
9509913a IS |
1913 | if (policer) { |
1914 | dropped = nb_rx; | |
64839cf4 WT |
1915 | nb_rx = ingress_policer_run(policer, |
1916 | (struct rte_mbuf **) batch->packets, | |
3e90f7d7 | 1917 | nb_rx, true); |
9509913a IS |
1918 | dropped -= nb_rx; |
1919 | } | |
1920 | ||
d46285a2 | 1921 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1922 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1923 | nb_rx, dropped); | |
d46285a2 | 1924 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1925 | |
75fb9148 ZB |
1926 | batch->count = nb_rx; |
1927 | dp_packet_batch_init_packet_fields(batch); | |
1928 | ||
58397e6c KT |
1929 | return 0; |
1930 | } | |
1931 | ||
8a9562d2 | 1932 | static int |
64839cf4 | 1933 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1934 | { |
d46285a2 DDP |
1935 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1936 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1937 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1938 | int nb_rx; |
9509913a | 1939 | int dropped = 0; |
8a9562d2 | 1940 | |
3b1fb077 DDP |
1941 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1942 | return EAGAIN; | |
1943 | } | |
1944 | ||
d46285a2 | 1945 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1946 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1947 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1948 | if (!nb_rx) { |
1949 | return EAGAIN; | |
1950 | } | |
1951 | ||
9509913a IS |
1952 | if (policer) { |
1953 | dropped = nb_rx; | |
64839cf4 | 1954 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 1955 | (struct rte_mbuf **) batch->packets, |
3e90f7d7 | 1956 | nb_rx, true); |
9509913a IS |
1957 | dropped -= nb_rx; |
1958 | } | |
1959 | ||
1960 | /* Update stats to reflect dropped packets */ | |
1961 | if (OVS_UNLIKELY(dropped)) { | |
1962 | rte_spinlock_lock(&dev->stats_lock); | |
1963 | dev->stats.rx_dropped += dropped; | |
1964 | rte_spinlock_unlock(&dev->stats_lock); | |
1965 | } | |
1966 | ||
64839cf4 | 1967 | batch->count = nb_rx; |
75fb9148 | 1968 | dp_packet_batch_init_packet_fields(batch); |
8a9562d2 PS |
1969 | |
1970 | return 0; | |
1971 | } | |
1972 | ||
0bf765f7 | 1973 | static inline int |
78bd47cf | 1974 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
3e90f7d7 | 1975 | int cnt, bool may_steal) |
0bf765f7 | 1976 | { |
78bd47cf | 1977 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 1978 | |
78bd47cf DDP |
1979 | if (qos_conf) { |
1980 | rte_spinlock_lock(&qos_conf->lock); | |
3e90f7d7 | 1981 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt, may_steal); |
78bd47cf | 1982 | rte_spinlock_unlock(&qos_conf->lock); |
0bf765f7 IS |
1983 | } |
1984 | ||
1985 | return cnt; | |
1986 | } | |
1987 | ||
c6ec9d17 IM |
1988 | static int |
1989 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1990 | int pkt_cnt) | |
1991 | { | |
1992 | int i = 0; | |
1993 | int cnt = 0; | |
1994 | struct rte_mbuf *pkt; | |
1995 | ||
1996 | for (i = 0; i < pkt_cnt; i++) { | |
1997 | pkt = pkts[i]; | |
1998 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
1999 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
2000 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
2001 | rte_pktmbuf_free(pkt); | |
2002 | continue; | |
2003 | } | |
2004 | ||
2005 | if (OVS_UNLIKELY(i != cnt)) { | |
2006 | pkts[cnt] = pkt; | |
2007 | } | |
2008 | cnt++; | |
2009 | } | |
2010 | ||
2011 | return cnt; | |
2012 | } | |
2013 | ||
9e3ddd45 TP |
2014 | static inline void |
2015 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
2016 | struct dp_packet **packets, | |
2017 | int attempted, | |
2018 | int dropped) | |
2019 | { | |
2020 | int i; | |
2021 | int sent = attempted - dropped; | |
2022 | ||
2023 | stats->tx_packets += sent; | |
2024 | stats->tx_dropped += dropped; | |
2025 | ||
2026 | for (i = 0; i < sent; i++) { | |
2027 | stats->tx_bytes += dp_packet_size(packets[i]); | |
2028 | } | |
2029 | } | |
2030 | ||
58397e6c | 2031 | static void |
4573fbd3 | 2032 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 2033 | struct dp_packet **pkts, int cnt) |
58397e6c | 2034 | { |
d46285a2 | 2035 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
2036 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
2037 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 2038 | unsigned int dropped = 0; |
dd52de45 | 2039 | int i, retries = 0; |
daf22bf7 | 2040 | int vid = netdev_dpdk_get_vid(dev); |
58397e6c | 2041 | |
81acebda | 2042 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 2043 | |
daf22bf7 | 2044 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured || qid < 0 |
e543851d | 2045 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
2046 | rte_spinlock_lock(&dev->stats_lock); |
2047 | dev->stats.tx_dropped+= cnt; | |
2048 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 2049 | goto out; |
58397e6c KT |
2050 | } |
2051 | ||
d46285a2 | 2052 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 2053 | |
c6ec9d17 | 2054 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 2055 | /* Check has QoS has been configured for the netdev */ |
3e90f7d7 | 2056 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt, true); |
c6ec9d17 | 2057 | dropped = total_pkts - cnt; |
0bf765f7 | 2058 | |
95e9881f | 2059 | do { |
4573fbd3 | 2060 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
2061 | unsigned int tx_pkts; |
2062 | ||
daf22bf7 | 2063 | tx_pkts = rte_vhost_enqueue_burst(vid, vhost_qid, cur_pkts, cnt); |
95e9881f KT |
2064 | if (OVS_LIKELY(tx_pkts)) { |
2065 | /* Packets have been sent.*/ | |
2066 | cnt -= tx_pkts; | |
31871ee3 | 2067 | /* Prepare for possible retry.*/ |
95e9881f KT |
2068 | cur_pkts = &cur_pkts[tx_pkts]; |
2069 | } else { | |
31871ee3 KT |
2070 | /* No packets sent - do not retry.*/ |
2071 | break; | |
95e9881f | 2072 | } |
c6ec9d17 | 2073 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 2074 | |
d46285a2 | 2075 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 2076 | |
d46285a2 | 2077 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 2078 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 2079 | cnt + dropped); |
d46285a2 | 2080 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
2081 | |
2082 | out: | |
c6ec9d17 | 2083 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 2084 | dp_packet_delete(pkts[i]); |
58397e6c KT |
2085 | } |
2086 | } | |
2087 | ||
8a9562d2 PS |
2088 | /* Tx function. Transmit packets indefinitely */ |
2089 | static void | |
64839cf4 | 2090 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 2091 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 2092 | { |
8a14bd7b | 2093 | const size_t batch_cnt = dp_packet_batch_size(batch); |
bce01e3a | 2094 | #if !defined(__CHECKER__) && !defined(_WIN32) |
8a14bd7b | 2095 | const size_t PKT_ARRAY_SIZE = batch_cnt; |
bce01e3a EJ |
2096 | #else |
2097 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 2098 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 2099 | #endif |
8a9562d2 | 2100 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 2101 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
8a14bd7b | 2102 | uint32_t cnt = batch_cnt; |
3e90f7d7 GZ |
2103 | uint32_t dropped = 0; |
2104 | ||
2105 | if (dev->type != DPDK_DEV_VHOST) { | |
2106 | /* Check if QoS has been configured for this netdev. */ | |
2107 | cnt = netdev_dpdk_qos_run(dev, (struct rte_mbuf **) batch->packets, | |
8a14bd7b BB |
2108 | batch_cnt, false); |
2109 | dropped += batch_cnt - cnt; | |
3e90f7d7 | 2110 | } |
8a9562d2 | 2111 | |
3e90f7d7 GZ |
2112 | uint32_t txcnt = 0; |
2113 | ||
2114 | for (uint32_t i = 0; i < cnt; i++) { | |
8a14bd7b BB |
2115 | struct dp_packet *packet = batch->packets[i]; |
2116 | uint32_t size = dp_packet_size(packet); | |
95fb793a | 2117 | |
f98d7864 | 2118 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
3e90f7d7 GZ |
2119 | VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d", |
2120 | size, dev->max_packet_len); | |
f4fd623c | 2121 | |
175cf4de | 2122 | dropped++; |
f4fd623c DDP |
2123 | continue; |
2124 | } | |
8a9562d2 | 2125 | |
24e78f93 | 2126 | pkts[txcnt] = rte_pktmbuf_alloc(dev->mp); |
8a14bd7b | 2127 | if (OVS_UNLIKELY(!pkts[txcnt])) { |
3e90f7d7 | 2128 | dropped += cnt - i; |
175cf4de | 2129 | break; |
f4fd623c DDP |
2130 | } |
2131 | ||
2132 | /* We have to do a copy for now */ | |
3e90f7d7 | 2133 | memcpy(rte_pktmbuf_mtod(pkts[txcnt], void *), |
8a14bd7b BB |
2134 | dp_packet_data(packet), size); |
2135 | dp_packet_set_size((struct dp_packet *)pkts[txcnt], size); | |
f4fd623c | 2136 | |
3e90f7d7 | 2137 | txcnt++; |
f4fd623c | 2138 | } |
8a9562d2 | 2139 | |
3e90f7d7 GZ |
2140 | if (OVS_LIKELY(txcnt)) { |
2141 | if (dev->type == DPDK_DEV_VHOST) { | |
2142 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, | |
2143 | txcnt); | |
2144 | } else { | |
2145 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, txcnt); | |
2146 | } | |
58397e6c | 2147 | } |
db73f716 | 2148 | |
0bf765f7 IS |
2149 | if (OVS_UNLIKELY(dropped)) { |
2150 | rte_spinlock_lock(&dev->stats_lock); | |
2151 | dev->stats.tx_dropped += dropped; | |
2152 | rte_spinlock_unlock(&dev->stats_lock); | |
2153 | } | |
8a9562d2 PS |
2154 | } |
2155 | ||
58397e6c | 2156 | static int |
64839cf4 WT |
2157 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
2158 | struct dp_packet_batch *batch, | |
b30896c9 | 2159 | bool concurrent_txq OVS_UNUSED) |
58397e6c | 2160 | { |
58397e6c | 2161 | |
b30896c9 | 2162 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 | 2163 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2164 | dp_packet_delete_batch(batch, true); |
58397e6c | 2165 | } else { |
dd52de45 | 2166 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
2167 | } |
2168 | return 0; | |
2169 | } | |
2170 | ||
7251515e DV |
2171 | static inline void |
2172 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
b30896c9 | 2173 | struct dp_packet_batch *batch, |
324c8374 | 2174 | bool concurrent_txq) |
8a9562d2 | 2175 | { |
3b1fb077 | 2176 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
b30896c9 | 2177 | dp_packet_delete_batch(batch, true); |
3b1fb077 DDP |
2178 | return; |
2179 | } | |
2180 | ||
324c8374 | 2181 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 2182 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
2183 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
2184 | } | |
2185 | ||
b30896c9 | 2186 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
2187 | struct netdev *netdev = &dev->up; |
2188 | ||
64839cf4 | 2189 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2190 | dp_packet_delete_batch(batch, true); |
8a9562d2 | 2191 | } else { |
fd57eeba BB |
2192 | int tx_cnt, dropped; |
2193 | int batch_cnt = dp_packet_batch_size(batch); | |
2391135c | 2194 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 2195 | |
fd57eeba BB |
2196 | tx_cnt = netdev_dpdk_filter_packet_len(dev, pkts, batch_cnt); |
2197 | tx_cnt = netdev_dpdk_qos_run(dev, pkts, tx_cnt, true); | |
2198 | dropped = batch_cnt - tx_cnt; | |
1b99bb05 | 2199 | |
fd57eeba | 2200 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, tx_cnt); |
8a9562d2 | 2201 | |
f4fd623c | 2202 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 2203 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 2204 | dev->stats.tx_dropped += dropped; |
45d947c4 | 2205 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 2206 | } |
8a9562d2 | 2207 | } |
a0cb2d66 | 2208 | |
324c8374 | 2209 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
2210 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
2211 | } | |
7251515e DV |
2212 | } |
2213 | ||
2214 | static int | |
2215 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
b30896c9 | 2216 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e DV |
2217 | { |
2218 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 2219 | |
b30896c9 | 2220 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e | 2221 | return 0; |
8a9562d2 PS |
2222 | } |
2223 | ||
2224 | static int | |
74ff3298 | 2225 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
2226 | { |
2227 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2228 | ||
2229 | ovs_mutex_lock(&dev->mutex); | |
2230 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 2231 | dev->hwaddr = mac; |
045c0d1a | 2232 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
2233 | } |
2234 | ovs_mutex_unlock(&dev->mutex); | |
2235 | ||
2236 | return 0; | |
2237 | } | |
2238 | ||
2239 | static int | |
74ff3298 | 2240 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
2241 | { |
2242 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2243 | ||
2244 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 2245 | *mac = dev->hwaddr; |
8a9562d2 PS |
2246 | ovs_mutex_unlock(&dev->mutex); |
2247 | ||
2248 | return 0; | |
2249 | } | |
2250 | ||
2251 | static int | |
2252 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
2253 | { | |
2254 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2255 | ||
2256 | ovs_mutex_lock(&dev->mutex); | |
2257 | *mtup = dev->mtu; | |
2258 | ovs_mutex_unlock(&dev->mutex); | |
2259 | ||
2260 | return 0; | |
2261 | } | |
2262 | ||
0072e931 MK |
2263 | static int |
2264 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
2265 | { | |
2266 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2267 | ||
f6f50552 IS |
2268 | /* XXX: Ensure that the overall frame length of the requested MTU does not |
2269 | * surpass the NETDEV_DPDK_MAX_PKT_LEN. DPDK device drivers differ in how | |
2270 | * the L2 frame length is calculated for a given MTU when | |
2271 | * rte_eth_dev_set_mtu(mtu) is called e.g. i40e driver includes 2 x vlan | |
2272 | * headers, the em driver includes 1 x vlan header, the ixgbe driver does | |
2273 | * not include vlan headers. As such we should use | |
2274 | * MTU_TO_MAX_FRAME_LEN(mtu) which includes an additional 2 x vlan headers | |
2275 | * (8 bytes) for comparison. This avoids a failure later with | |
2276 | * rte_eth_dev_set_mtu(). This approach should be used until DPDK provides | |
2277 | * a method to retrieve the upper bound MTU for a given device. | |
2278 | */ | |
2279 | if (MTU_TO_MAX_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
0072e931 MK |
2280 | || mtu < ETHER_MIN_MTU) { |
2281 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
2282 | return EINVAL; | |
2283 | } | |
2284 | ||
2285 | ovs_mutex_lock(&dev->mutex); | |
2286 | if (dev->requested_mtu != mtu) { | |
2287 | dev->requested_mtu = mtu; | |
2288 | netdev_request_reconfigure(netdev); | |
2289 | } | |
2290 | ovs_mutex_unlock(&dev->mutex); | |
2291 | ||
2292 | return 0; | |
2293 | } | |
2294 | ||
8a9562d2 | 2295 | static int |
d46285a2 | 2296 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 2297 | |
58397e6c KT |
2298 | static int |
2299 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
2300 | struct netdev_stats *stats) | |
2301 | { | |
2302 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2303 | ||
2304 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 2305 | |
45d947c4 | 2306 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c | 2307 | /* Supported Stats */ |
50986e78 | 2308 | stats->rx_packets = dev->stats.rx_packets; |
2309 | stats->tx_packets = dev->stats.tx_packets; | |
9509913a | 2310 | stats->rx_dropped = dev->stats.rx_dropped; |
50986e78 | 2311 | stats->tx_dropped = dev->stats.tx_dropped; |
9e3ddd45 TP |
2312 | stats->multicast = dev->stats.multicast; |
2313 | stats->rx_bytes = dev->stats.rx_bytes; | |
2314 | stats->tx_bytes = dev->stats.tx_bytes; | |
2315 | stats->rx_errors = dev->stats.rx_errors; | |
2316 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 2317 | |
2318 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
2319 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
2320 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
2321 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
2322 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
2323 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
2324 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
2325 | ||
45d947c4 | 2326 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 2327 | |
58397e6c KT |
2328 | ovs_mutex_unlock(&dev->mutex); |
2329 | ||
2330 | return 0; | |
2331 | } | |
2332 | ||
d6e3feb5 | 2333 | static void |
2334 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
2335 | const struct rte_eth_xstat *xstats, |
2336 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 2337 | const unsigned int size) |
2338 | { | |
d6e3feb5 | 2339 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 2340 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2341 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2342 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2343 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2344 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2345 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2346 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2347 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2348 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2349 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2350 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2351 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2352 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2353 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 2354 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2355 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2356 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2357 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2358 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2359 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2360 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2361 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2362 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2363 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2364 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2365 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2366 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2367 | stats->tx_1523_to_max_packets = xstats[i].value; |
d57f777f PS |
2368 | } else if (strcmp(XSTAT_RX_MULTICAST_PACKETS, names[i].name) == 0) { |
2369 | stats->multicast = xstats[i].value; | |
0a0f39df | 2370 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2371 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 2372 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2373 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2374 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2375 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2376 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2377 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2378 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2379 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2380 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2381 | stats->rx_jabber_errors = xstats[i].value; |
2382 | } | |
2383 | } | |
2384 | } | |
2385 | ||
8a9562d2 PS |
2386 | static int |
2387 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2388 | { | |
2389 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2390 | struct rte_eth_stats rte_stats; | |
2391 | bool gg; | |
2392 | ||
2393 | netdev_dpdk_get_carrier(netdev, &gg); | |
2394 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2395 | |
0a0f39df CL |
2396 | struct rte_eth_xstat *rte_xstats = NULL; |
2397 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2398 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2399 | |
2400 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
fa9f4eeb IM |
2401 | VLOG_ERR("Can't get ETH statistics for port: "DPDK_PORT_ID_FMT, |
2402 | dev->port_id); | |
f9256822 | 2403 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2404 | return EPROTO; |
2405 | } | |
2406 | ||
0a0f39df CL |
2407 | /* Get length of statistics */ |
2408 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2409 | if (rte_xstats_len < 0) { | |
fa9f4eeb IM |
2410 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
2411 | dev->port_id); | |
0a0f39df CL |
2412 | goto out; |
2413 | } | |
2414 | /* Reserve memory for xstats names and values */ | |
2415 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2416 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2417 | ||
2418 | /* Retreive xstats names */ | |
2419 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2420 | rte_xstats_names, | |
2421 | rte_xstats_len); | |
2422 | if (rte_xstats_new_len != rte_xstats_len) { | |
fa9f4eeb IM |
2423 | VLOG_WARN("Cannot get XSTATS names for port: "DPDK_PORT_ID_FMT, |
2424 | dev->port_id); | |
0a0f39df CL |
2425 | goto out; |
2426 | } | |
2427 | /* Retreive xstats values */ | |
2428 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2429 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2430 | rte_xstats_len); | |
2431 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2432 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2433 | rte_xstats_len); | |
d6e3feb5 | 2434 | } else { |
fa9f4eeb IM |
2435 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
2436 | dev->port_id); | |
d6e3feb5 | 2437 | } |
8a9562d2 | 2438 | |
0a0f39df CL |
2439 | out: |
2440 | free(rte_xstats); | |
2441 | free(rte_xstats_names); | |
2442 | ||
2f9dd77f PS |
2443 | stats->rx_packets = rte_stats.ipackets; |
2444 | stats->tx_packets = rte_stats.opackets; | |
2445 | stats->rx_bytes = rte_stats.ibytes; | |
2446 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2447 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2448 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2449 | |
45d947c4 | 2450 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2451 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2452 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2453 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2454 | |
2455 | /* These are the available DPDK counters for packets not received due to | |
2456 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2457 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2458 | stats->rx_missed_errors = rte_stats.imissed; |
2459 | ||
8a9562d2 PS |
2460 | ovs_mutex_unlock(&dev->mutex); |
2461 | ||
2462 | return 0; | |
2463 | } | |
2464 | ||
971f4b39 MW |
2465 | static int |
2466 | netdev_dpdk_get_custom_stats(const struct netdev *netdev, | |
2467 | struct netdev_custom_stats *custom_stats) | |
2468 | { | |
2469 | ||
2470 | uint32_t i; | |
2471 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2472 | int rte_xstats_ret; | |
2473 | ||
2474 | ovs_mutex_lock(&dev->mutex); | |
2475 | ||
2476 | if (netdev_dpdk_configure_xstats(dev)) { | |
2477 | uint64_t *values = xcalloc(dev->rte_xstats_ids_size, | |
2478 | sizeof(uint64_t)); | |
2479 | ||
2480 | rte_xstats_ret = | |
2481 | rte_eth_xstats_get_by_id(dev->port_id, dev->rte_xstats_ids, | |
2482 | values, dev->rte_xstats_ids_size); | |
2483 | ||
2484 | if (rte_xstats_ret > 0 && | |
2485 | rte_xstats_ret <= dev->rte_xstats_ids_size) { | |
2486 | ||
2487 | custom_stats->size = rte_xstats_ret; | |
2488 | custom_stats->counters = | |
2489 | (struct netdev_custom_counter *) xcalloc(rte_xstats_ret, | |
2490 | sizeof(struct netdev_custom_counter)); | |
2491 | ||
2492 | for (i = 0; i < rte_xstats_ret; i++) { | |
2493 | ovs_strlcpy(custom_stats->counters[i].name, | |
2494 | netdev_dpdk_get_xstat_name(dev, | |
2495 | dev->rte_xstats_ids[i]), | |
2496 | NETDEV_CUSTOM_STATS_NAME_SIZE); | |
2497 | custom_stats->counters[i].value = values[i]; | |
2498 | } | |
2499 | } else { | |
fa9f4eeb | 2500 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
971f4b39 MW |
2501 | dev->port_id); |
2502 | custom_stats->counters = NULL; | |
2503 | custom_stats->size = 0; | |
2504 | /* Let's clear statistics cache, so it will be | |
2505 | * reconfigured */ | |
2506 | netdev_dpdk_clear_xstats(dev); | |
2507 | } | |
526259f2 IM |
2508 | |
2509 | free(values); | |
971f4b39 MW |
2510 | } |
2511 | ||
2512 | ovs_mutex_unlock(&dev->mutex); | |
2513 | ||
2514 | return 0; | |
2515 | } | |
2516 | ||
8a9562d2 | 2517 | static int |
d46285a2 | 2518 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2519 | enum netdev_features *current, |
ca3d4f55 BX |
2520 | enum netdev_features *advertised, |
2521 | enum netdev_features *supported, | |
2522 | enum netdev_features *peer) | |
8a9562d2 | 2523 | { |
d46285a2 | 2524 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2525 | struct rte_eth_link link; |
2526 | ||
2527 | ovs_mutex_lock(&dev->mutex); | |
2528 | link = dev->link; | |
2529 | ovs_mutex_unlock(&dev->mutex); | |
2530 | ||
362ca396 | 2531 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2532 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
2533 | *current = NETDEV_F_10MB_HD; |
2534 | } | |
362ca396 | 2535 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2536 | *current = NETDEV_F_100MB_HD; |
2537 | } | |
362ca396 | 2538 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2539 | *current = NETDEV_F_1GB_HD; |
2540 | } | |
2541 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 2542 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
2543 | *current = NETDEV_F_10MB_FD; |
2544 | } | |
362ca396 | 2545 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2546 | *current = NETDEV_F_100MB_FD; |
2547 | } | |
362ca396 | 2548 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2549 | *current = NETDEV_F_1GB_FD; |
2550 | } | |
362ca396 | 2551 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
2552 | *current = NETDEV_F_10GB_FD; |
2553 | } | |
2554 | } | |
2555 | ||
362ca396 | 2556 | if (link.link_autoneg) { |
2557 | *current |= NETDEV_F_AUTONEG; | |
2558 | } | |
2559 | ||
ca3d4f55 BX |
2560 | *advertised = *supported = *peer = 0; |
2561 | ||
8a9562d2 PS |
2562 | return 0; |
2563 | } | |
2564 | ||
9509913a IS |
2565 | static struct ingress_policer * |
2566 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2567 | { | |
2568 | struct ingress_policer *policer = NULL; | |
2569 | uint64_t rate_bytes; | |
2570 | uint64_t burst_bytes; | |
2571 | int err = 0; | |
2572 | ||
2573 | policer = xmalloc(sizeof *policer); | |
2574 | rte_spinlock_init(&policer->policer_lock); | |
2575 | ||
2576 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
602c8668 LR |
2577 | rate_bytes = rate * 1000ULL / 8; |
2578 | burst_bytes = burst * 1000ULL / 8; | |
9509913a IS |
2579 | |
2580 | policer->app_srtcm_params.cir = rate_bytes; | |
2581 | policer->app_srtcm_params.cbs = burst_bytes; | |
2582 | policer->app_srtcm_params.ebs = 0; | |
2583 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2584 | &policer->app_srtcm_params); | |
58be5c0e | 2585 | if (err) { |
9509913a | 2586 | VLOG_ERR("Could not create rte meter for ingress policer"); |
4c47ddde | 2587 | free(policer); |
9509913a IS |
2588 | return NULL; |
2589 | } | |
2590 | ||
2591 | return policer; | |
2592 | } | |
2593 | ||
2594 | static int | |
2595 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2596 | uint32_t policer_burst) | |
2597 | { | |
2598 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2599 | struct ingress_policer *policer; | |
2600 | ||
2601 | /* Force to 0 if no rate specified, | |
2602 | * default to 8000 kbits if burst is 0, | |
2603 | * else stick with user-specified value. | |
2604 | */ | |
2605 | policer_burst = (!policer_rate ? 0 | |
2606 | : !policer_burst ? 8000 | |
2607 | : policer_burst); | |
2608 | ||
2609 | ovs_mutex_lock(&dev->mutex); | |
2610 | ||
2611 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2612 | &dev->ingress_policer); | |
2613 | ||
2614 | if (dev->policer_rate == policer_rate && | |
2615 | dev->policer_burst == policer_burst) { | |
2616 | /* Assume that settings haven't changed since we last set them. */ | |
2617 | ovs_mutex_unlock(&dev->mutex); | |
2618 | return 0; | |
2619 | } | |
2620 | ||
2621 | /* Destroy any existing ingress policer for the device if one exists */ | |
2622 | if (policer) { | |
2623 | ovsrcu_postpone(free, policer); | |
2624 | } | |
2625 | ||
2626 | if (policer_rate != 0) { | |
2627 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2628 | } else { | |
2629 | policer = NULL; | |
2630 | } | |
2631 | ovsrcu_set(&dev->ingress_policer, policer); | |
2632 | dev->policer_rate = policer_rate; | |
2633 | dev->policer_burst = policer_burst; | |
2634 | ovs_mutex_unlock(&dev->mutex); | |
2635 | ||
2636 | return 0; | |
2637 | } | |
2638 | ||
8a9562d2 PS |
2639 | static int |
2640 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2641 | { | |
2642 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
2643 | |
2644 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
2645 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
2646 | * postive integer to meet RFC 2863 recommendations. | |
2647 | */ | |
2648 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
2649 | ovs_mutex_unlock(&dev->mutex); |
2650 | ||
2651 | return ifindex; | |
2652 | } | |
2653 | ||
2654 | static int | |
d46285a2 | 2655 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2656 | { |
d46285a2 | 2657 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2658 | |
2659 | ovs_mutex_lock(&dev->mutex); | |
2660 | check_link_status(dev); | |
2661 | *carrier = dev->link.link_status; | |
58397e6c KT |
2662 | |
2663 | ovs_mutex_unlock(&dev->mutex); | |
2664 | ||
2665 | return 0; | |
2666 | } | |
2667 | ||
2668 | static int | |
d46285a2 | 2669 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2670 | { |
d46285a2 | 2671 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2672 | |
2673 | ovs_mutex_lock(&dev->mutex); | |
2674 | ||
0a0f39df | 2675 | if (is_vhost_running(dev)) { |
58397e6c KT |
2676 | *carrier = 1; |
2677 | } else { | |
2678 | *carrier = 0; | |
2679 | } | |
2680 | ||
8a9562d2 PS |
2681 | ovs_mutex_unlock(&dev->mutex); |
2682 | ||
2683 | return 0; | |
2684 | } | |
2685 | ||
2686 | static long long int | |
d46285a2 | 2687 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2688 | { |
d46285a2 | 2689 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2690 | long long int carrier_resets; |
2691 | ||
2692 | ovs_mutex_lock(&dev->mutex); | |
2693 | carrier_resets = dev->link_reset_cnt; | |
2694 | ovs_mutex_unlock(&dev->mutex); | |
2695 | ||
2696 | return carrier_resets; | |
2697 | } | |
2698 | ||
2699 | static int | |
d46285a2 | 2700 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2701 | long long int interval OVS_UNUSED) |
2702 | { | |
ee32150e | 2703 | return EOPNOTSUPP; |
8a9562d2 PS |
2704 | } |
2705 | ||
2706 | static int | |
2707 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2708 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2709 | enum netdev_flags *old_flagsp) |
2710 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 2711 | { |
8a9562d2 PS |
2712 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
2713 | return EINVAL; | |
2714 | } | |
2715 | ||
2716 | *old_flagsp = dev->flags; | |
2717 | dev->flags |= on; | |
2718 | dev->flags &= ~off; | |
2719 | ||
2720 | if (dev->flags == *old_flagsp) { | |
2721 | return 0; | |
2722 | } | |
2723 | ||
58397e6c | 2724 | if (dev->type == DPDK_DEV_ETH) { |
58397e6c KT |
2725 | if (dev->flags & NETDEV_PROMISC) { |
2726 | rte_eth_promiscuous_enable(dev->port_id); | |
2727 | } | |
8a9562d2 | 2728 | |
314fb5ad | 2729 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
2730 | } else { |
2731 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2732 | * running then change netdev's change_seq to trigger link state | |
2733 | * update. */ | |
e543851d ZB |
2734 | |
2735 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2736 | && is_vhost_running(dev)) { |
e543851d ZB |
2737 | netdev_change_seq_changed(&dev->up); |
2738 | ||
2739 | /* Clear statistics if device is getting up. */ | |
2740 | if (NETDEV_UP & on) { | |
2741 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 2742 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
2743 | rte_spinlock_unlock(&dev->stats_lock); |
2744 | } | |
2745 | } | |
8a9562d2 PS |
2746 | } |
2747 | ||
2748 | return 0; | |
2749 | } | |
2750 | ||
2751 | static int | |
d46285a2 | 2752 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2753 | enum netdev_flags off, enum netdev_flags on, |
2754 | enum netdev_flags *old_flagsp) | |
2755 | { | |
d46285a2 | 2756 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2757 | int error; |
2758 | ||
d46285a2 DDP |
2759 | ovs_mutex_lock(&dev->mutex); |
2760 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2761 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2762 | |
2763 | return error; | |
2764 | } | |
2765 | ||
b2e8b12f FL |
2766 | static int |
2767 | netdev_dpdk_vhost_user_get_status(const struct netdev *netdev, | |
2768 | struct smap *args) | |
2769 | { | |
2770 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2771 | ||
2772 | ovs_mutex_lock(&dev->mutex); | |
2773 | ||
2774 | bool client_mode = dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT; | |
2775 | smap_add_format(args, "mode", "%s", client_mode ? "client" : "server"); | |
2776 | ||
2777 | int vid = netdev_dpdk_get_vid(dev); | |
2778 | if (vid < 0) { | |
2779 | smap_add_format(args, "status", "disconnected"); | |
2780 | ovs_mutex_unlock(&dev->mutex); | |
2781 | return 0; | |
2782 | } else { | |
2783 | smap_add_format(args, "status", "connected"); | |
2784 | } | |
2785 | ||
2786 | char socket_name[PATH_MAX]; | |
2787 | if (!rte_vhost_get_ifname(vid, socket_name, PATH_MAX)) { | |
2788 | smap_add_format(args, "socket", "%s", socket_name); | |
2789 | } | |
2790 | ||
2791 | uint64_t features; | |
2792 | if (!rte_vhost_get_negotiated_features(vid, &features)) { | |
2793 | smap_add_format(args, "features", "0x%016"PRIx64, features); | |
2794 | } | |
2795 | ||
2796 | uint16_t mtu; | |
2797 | if (!rte_vhost_get_mtu(vid, &mtu)) { | |
2798 | smap_add_format(args, "mtu", "%d", mtu); | |
2799 | } | |
2800 | ||
2801 | int numa = rte_vhost_get_numa_node(vid); | |
2802 | if (numa >= 0) { | |
2803 | smap_add_format(args, "numa", "%d", numa); | |
2804 | } | |
2805 | ||
2806 | uint16_t vring_num = rte_vhost_get_vring_num(vid); | |
2807 | if (vring_num) { | |
2808 | smap_add_format(args, "num_of_vrings", "%d", vring_num); | |
2809 | } | |
2810 | ||
2811 | for (int i = 0; i < vring_num; i++) { | |
2812 | struct rte_vhost_vring vring; | |
2813 | char vhost_vring[16]; | |
2814 | ||
2815 | rte_vhost_get_vhost_vring(vid, i, &vring); | |
2816 | snprintf(vhost_vring, 16, "vring_%d_size", i); | |
2817 | smap_add_format(args, vhost_vring, "%d", vring.size); | |
2818 | } | |
2819 | ||
2820 | ovs_mutex_unlock(&dev->mutex); | |
2821 | return 0; | |
2822 | } | |
2823 | ||
8a9562d2 | 2824 | static int |
d46285a2 | 2825 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2826 | { |
d46285a2 | 2827 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2828 | struct rte_eth_dev_info dev_info; |
2829 | ||
7cd1261d | 2830 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2831 | return ENODEV; |
7cd1261d | 2832 | } |
8a9562d2 PS |
2833 | |
2834 | ovs_mutex_lock(&dev->mutex); | |
2835 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2836 | ovs_mutex_unlock(&dev->mutex); | |
2837 | ||
fa9f4eeb | 2838 | smap_add_format(args, "port_no", DPDK_PORT_ID_FMT, dev->port_id); |
58be5c0e MK |
2839 | smap_add_format(args, "numa_id", "%d", |
2840 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
2841 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
2842 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2843 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2844 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2845 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2846 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
2847 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
2848 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
2849 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
2850 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2851 | ||
3eb8d4fa MW |
2852 | /* Querying the DPDK library for iftype may be done in future, pending |
2853 | * support; cf. RFC 3635 Section 3.2.4. */ | |
2854 | enum { IF_TYPE_ETHERNETCSMACD = 6 }; | |
2855 | ||
2856 | smap_add_format(args, "if_type", "%"PRIu32, IF_TYPE_ETHERNETCSMACD); | |
2857 | smap_add_format(args, "if_descr", "%s %s", rte_version(), | |
2858 | dev_info.driver_name); | |
2859 | ||
39c2baa9 | 2860 | if (dev_info.pci_dev) { |
2861 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2862 | dev_info.pci_dev->id.vendor_id); | |
2863 | smap_add_format(args, "pci-device_id", "0x%x", | |
2864 | dev_info.pci_dev->id.device_id); | |
2865 | } | |
8a9562d2 PS |
2866 | |
2867 | return 0; | |
2868 | } | |
2869 | ||
2870 | static void | |
2871 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2872 | OVS_REQUIRES(dev->mutex) | |
2873 | { | |
2874 | enum netdev_flags old_flags; | |
2875 | ||
2876 | if (admin_state) { | |
2877 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2878 | } else { | |
2879 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2880 | } | |
2881 | } | |
2882 | ||
2883 | static void | |
2884 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2885 | const char *argv[], void *aux OVS_UNUSED) | |
2886 | { | |
2887 | bool up; | |
2888 | ||
2889 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2890 | up = true; | |
2891 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2892 | up = false; | |
2893 | } else { | |
2894 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2895 | return; | |
2896 | } | |
2897 | ||
2898 | if (argc > 2) { | |
2899 | struct netdev *netdev = netdev_from_name(argv[1]); | |
3d0d5ab1 | 2900 | |
8a9562d2 | 2901 | if (netdev && is_dpdk_class(netdev->netdev_class)) { |
3d0d5ab1 | 2902 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 2903 | |
3d0d5ab1 IM |
2904 | ovs_mutex_lock(&dev->mutex); |
2905 | netdev_dpdk_set_admin_state__(dev, up); | |
2906 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2907 | |
2908 | netdev_close(netdev); | |
2909 | } else { | |
2910 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2911 | netdev_close(netdev); | |
2912 | return; | |
2913 | } | |
2914 | } else { | |
3d0d5ab1 | 2915 | struct netdev_dpdk *dev; |
8a9562d2 PS |
2916 | |
2917 | ovs_mutex_lock(&dpdk_mutex); | |
3d0d5ab1 IM |
2918 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
2919 | ovs_mutex_lock(&dev->mutex); | |
2920 | netdev_dpdk_set_admin_state__(dev, up); | |
2921 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2922 | } |
2923 | ovs_mutex_unlock(&dpdk_mutex); | |
2924 | } | |
2925 | unixctl_command_reply(conn, "OK"); | |
2926 | } | |
2927 | ||
0ee821c2 DB |
2928 | static void |
2929 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2930 | const char *argv[], void *aux OVS_UNUSED) | |
2931 | { | |
2932 | int ret; | |
2933 | char *response; | |
7ee94cba | 2934 | dpdk_port_t port_id; |
0ee821c2 DB |
2935 | char devname[RTE_ETH_NAME_MAX_LEN]; |
2936 | struct netdev_dpdk *dev; | |
2937 | ||
2938 | ovs_mutex_lock(&dpdk_mutex); | |
2939 | ||
255b7bda | 2940 | if (rte_eth_dev_get_port_by_name(argv[1], &port_id)) { |
0ee821c2 DB |
2941 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); |
2942 | goto error; | |
2943 | } | |
2944 | ||
2945 | dev = netdev_dpdk_lookup_by_port_id(port_id); | |
2946 | if (dev) { | |
2947 | response = xasprintf("Device '%s' is being used by interface '%s'. " | |
2948 | "Remove it before detaching", | |
2949 | argv[1], netdev_get_name(&dev->up)); | |
2950 | goto error; | |
2951 | } | |
2952 | ||
2953 | rte_eth_dev_close(port_id); | |
2954 | ||
2955 | ret = rte_eth_dev_detach(port_id, devname); | |
2956 | if (ret < 0) { | |
2957 | response = xasprintf("Device '%s' can not be detached", argv[1]); | |
2958 | goto error; | |
2959 | } | |
2960 | ||
2961 | response = xasprintf("Device '%s' has been detached", argv[1]); | |
2962 | ||
2963 | ovs_mutex_unlock(&dpdk_mutex); | |
2964 | unixctl_command_reply(conn, response); | |
2965 | free(response); | |
2966 | return; | |
2967 | ||
2968 | error: | |
2969 | ovs_mutex_unlock(&dpdk_mutex); | |
2970 | unixctl_command_reply_error(conn, response); | |
2971 | free(response); | |
2972 | } | |
2973 | ||
be481733 IM |
2974 | static void |
2975 | netdev_dpdk_get_mempool_info(struct unixctl_conn *conn, | |
2976 | int argc, const char *argv[], | |
2977 | void *aux OVS_UNUSED) | |
2978 | { | |
2979 | size_t size; | |
2980 | FILE *stream; | |
2981 | char *response = NULL; | |
2982 | struct netdev *netdev = NULL; | |
2983 | ||
2984 | if (argc == 2) { | |
2985 | netdev = netdev_from_name(argv[1]); | |
2986 | if (!netdev || !is_dpdk_class(netdev->netdev_class)) { | |
2987 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2988 | goto out; | |
2989 | } | |
2990 | } | |
2991 | ||
2992 | stream = open_memstream(&response, &size); | |
2993 | if (!stream) { | |
2994 | response = xasprintf("Unable to open memstream: %s.", | |
2995 | ovs_strerror(errno)); | |
2996 | unixctl_command_reply_error(conn, response); | |
2997 | goto out; | |
2998 | } | |
2999 | ||
3000 | if (netdev) { | |
3001 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3002 | ||
3003 | ovs_mutex_lock(&dev->mutex); | |
3004 | ovs_mutex_lock(&dpdk_mp_mutex); | |
3005 | ||
3006 | rte_mempool_dump(stream, dev->mp); | |
3007 | ||
3008 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
3009 | ovs_mutex_unlock(&dev->mutex); | |
3010 | } else { | |
3011 | ovs_mutex_lock(&dpdk_mp_mutex); | |
3012 | rte_mempool_list_dump(stream); | |
3013 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
3014 | } | |
3015 | ||
3016 | fclose(stream); | |
3017 | ||
3018 | unixctl_command_reply(conn, response); | |
3019 | out: | |
3020 | free(response); | |
3021 | netdev_close(netdev); | |
3022 | } | |
3023 | ||
58397e6c KT |
3024 | /* |
3025 | * Set virtqueue flags so that we do not receive interrupts. | |
3026 | */ | |
3027 | static void | |
0a0f39df | 3028 | set_irq_status(int vid) |
58397e6c | 3029 | { |
4573fbd3 | 3030 | uint32_t i; |
4573fbd3 | 3031 | |
f3e7ec25 MW |
3032 | for (i = 0; i < rte_vhost_get_vring_num(vid); i++) { |
3033 | rte_vhost_enable_guest_notification(vid, i, 0); | |
4573fbd3 FL |
3034 | } |
3035 | } | |
3036 | ||
585a5bea IM |
3037 | /* |
3038 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 3039 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
3040 | */ |
3041 | static void | |
d46285a2 DDP |
3042 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
3043 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
3044 | { |
3045 | int *enabled_queues, n_enabled = 0; | |
81acebda | 3046 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 3047 | |
eff23640 | 3048 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
3049 | |
3050 | for (i = 0; i < total_txqs; i++) { | |
3051 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 3052 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
3053 | enabled_queues[n_enabled++] = i; |
3054 | } | |
3055 | } | |
3056 | ||
3057 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 3058 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
3059 | n_enabled = 1; |
3060 | } | |
3061 | ||
3062 | k = 0; | |
3063 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
3064 | if (dev->tx_q[i].map != i) { |
3065 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
3066 | k = (k + 1) % n_enabled; |
3067 | } | |
3068 | } | |
3069 | ||
2d24d165 | 3070 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 3071 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 3072 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
3073 | } |
3074 | ||
eff23640 | 3075 | free(enabled_queues); |
585a5bea | 3076 | } |
4573fbd3 | 3077 | |
58397e6c KT |
3078 | /* |
3079 | * A new virtio-net device is added to a vhost port. | |
3080 | */ | |
3081 | static int | |
0a0f39df | 3082 | new_device(int vid) |
58397e6c | 3083 | { |
d46285a2 | 3084 | struct netdev_dpdk *dev; |
58397e6c | 3085 | bool exists = false; |
db8f13b0 | 3086 | int newnode = 0; |
0a0f39df CL |
3087 | char ifname[IF_NAME_SZ]; |
3088 | ||
58be5c0e | 3089 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
3090 | |
3091 | ovs_mutex_lock(&dpdk_mutex); | |
3092 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 3093 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 3094 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 3095 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
f3e7ec25 | 3096 | uint32_t qp_num = rte_vhost_get_vring_num(vid)/VIRTIO_QNUM; |
db8f13b0 CL |
3097 | |
3098 | /* Get NUMA information */ | |
0a0f39df CL |
3099 | newnode = rte_vhost_get_numa_node(vid); |
3100 | if (newnode == -1) { | |
5b9bf9e0 | 3101 | #ifdef VHOST_NUMA |
db8f13b0 | 3102 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 3103 | ifname); |
5b9bf9e0 | 3104 | #endif |
db8f13b0 | 3105 | newnode = dev->socket_id; |
db8f13b0 CL |
3106 | } |
3107 | ||
7f5f2bd0 IM |
3108 | if (dev->requested_n_txq != qp_num |
3109 | || dev->requested_n_rxq != qp_num | |
3110 | || dev->requested_socket_id != newnode) { | |
3111 | dev->requested_socket_id = newnode; | |
3112 | dev->requested_n_rxq = qp_num; | |
3113 | dev->requested_n_txq = qp_num; | |
3114 | netdev_request_reconfigure(&dev->up); | |
3115 | } else { | |
3116 | /* Reconfiguration not required. */ | |
3117 | dev->vhost_reconfigured = true; | |
3118 | } | |
81acebda | 3119 | |
0a0f39df | 3120 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
3121 | exists = true; |
3122 | ||
58397e6c | 3123 | /* Disable notifications. */ |
0a0f39df | 3124 | set_irq_status(vid); |
e543851d | 3125 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 3126 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
3127 | break; |
3128 | } | |
c1ff66ac | 3129 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
3130 | } |
3131 | ovs_mutex_unlock(&dpdk_mutex); | |
3132 | ||
3133 | if (!exists) { | |
0a0f39df | 3134 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
3135 | |
3136 | return -1; | |
3137 | } | |
3138 | ||
0a0f39df CL |
3139 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
3140 | ifname, newnode); | |
3141 | ||
58397e6c KT |
3142 | return 0; |
3143 | } | |
3144 | ||
f3ea2ad2 IM |
3145 | /* Clears mapping for all available queues of vhost interface. */ |
3146 | static void | |
3147 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
3148 | OVS_REQUIRES(dev->mutex) | |
3149 | { | |
3150 | int i; | |
3151 | ||
81acebda | 3152 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
3153 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
3154 | } | |
3155 | } | |
3156 | ||
58397e6c KT |
3157 | /* |
3158 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
3159 | * flag to stop any more packets from being sent or received to/from a VM and | |
3160 | * ensure all currently queued packets have been sent/received before removing | |
3161 | * the device. | |
3162 | */ | |
3163 | static void | |
0a0f39df | 3164 | destroy_device(int vid) |
58397e6c | 3165 | { |
d46285a2 | 3166 | struct netdev_dpdk *dev; |
afee281f | 3167 | bool exists = false; |
0a0f39df CL |
3168 | char ifname[IF_NAME_SZ]; |
3169 | ||
58be5c0e | 3170 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
3171 | |
3172 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 3173 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 3174 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 3175 | |
d46285a2 | 3176 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
3177 | dev->vhost_reconfigured = false; |
3178 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 3179 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 3180 | |
e543851d | 3181 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 3182 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 3183 | exists = true; |
afee281f | 3184 | break; |
58397e6c KT |
3185 | } |
3186 | } | |
afee281f | 3187 | |
58397e6c KT |
3188 | ovs_mutex_unlock(&dpdk_mutex); |
3189 | ||
0a0f39df | 3190 | if (exists) { |
afee281f KT |
3191 | /* |
3192 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
3193 | * to NULL, before returning. | |
3194 | */ | |
3195 | ovsrcu_synchronize(); | |
3196 | /* | |
3197 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
3198 | * put thread back into quiescent state before returning. | |
3199 | */ | |
3200 | ovsrcu_quiesce_start(); | |
0a0f39df | 3201 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 3202 | } else { |
0a0f39df | 3203 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 3204 | } |
58397e6c KT |
3205 | } |
3206 | ||
585a5bea | 3207 | static int |
0a0f39df | 3208 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 3209 | { |
d46285a2 | 3210 | struct netdev_dpdk *dev; |
585a5bea IM |
3211 | bool exists = false; |
3212 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
3213 | char ifname[IF_NAME_SZ]; |
3214 | ||
58be5c0e | 3215 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
3216 | |
3217 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
3218 | return 0; | |
3219 | } | |
3220 | ||
3221 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 3222 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 3223 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 3224 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 3225 | if (enable) { |
d46285a2 | 3226 | dev->tx_q[qid].map = qid; |
585a5bea | 3227 | } else { |
d46285a2 | 3228 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 3229 | } |
d46285a2 | 3230 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 3231 | exists = true; |
d46285a2 | 3232 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3233 | break; |
3234 | } | |
c1ff66ac | 3235 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3236 | } |
3237 | ovs_mutex_unlock(&dpdk_mutex); | |
3238 | ||
3239 | if (exists) { | |
0a0f39df CL |
3240 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
3241 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 3242 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 3243 | } else { |
0a0f39df | 3244 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
3245 | return -1; |
3246 | } | |
3247 | ||
3248 | return 0; | |
3249 | } | |
3250 | ||
0a0f39df CL |
3251 | int |
3252 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 3253 | { |
0a0f39df | 3254 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
3255 | } |
3256 | ||
9509913a IS |
3257 | struct ingress_policer * |
3258 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
3259 | { | |
3260 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
3261 | } | |
3262 | ||
58397e6c | 3263 | static int |
ecc1a34e | 3264 | netdev_dpdk_class_init(void) |
7d1ced01 | 3265 | { |
ecc1a34e DDP |
3266 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
3267 | ||
3268 | /* This function can be called for different classes. The initialization | |
3269 | * needs to be done only once */ | |
3270 | if (ovsthread_once_start(&once)) { | |
3271 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
3272 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
3273 | "[netdev] up|down", 1, 2, | |
3274 | netdev_dpdk_set_admin_state, NULL); | |
3275 | ||
0ee821c2 DB |
3276 | unixctl_command_register("netdev-dpdk/detach", |
3277 | "pci address of device", 1, 1, | |
3278 | netdev_dpdk_detach, NULL); | |
3279 | ||
be481733 IM |
3280 | unixctl_command_register("netdev-dpdk/get-mempool-info", |
3281 | "[netdev]", 0, 1, | |
3282 | netdev_dpdk_get_mempool_info, NULL); | |
3283 | ||
ecc1a34e DDP |
3284 | ovsthread_once_done(&once); |
3285 | } | |
362ca396 | 3286 | |
7d1ced01 CL |
3287 | return 0; |
3288 | } | |
3289 | ||
033e9df2 | 3290 | |
95fb793a | 3291 | /* Client Rings */ |
3292 | ||
95fb793a | 3293 | static int |
3294 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
bb37956a | 3295 | dpdk_port_t *eth_port_id) |
95fb793a | 3296 | { |
48fffdee | 3297 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 3298 | char *ring_name; |
b8374d0d | 3299 | int port_id; |
95fb793a | 3300 | |
48fffdee KT |
3301 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
3302 | if (!ring_pair) { | |
95fb793a | 3303 | return ENOMEM; |
3304 | } | |
3305 | ||
7251515e | 3306 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 3307 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 3308 | |
8f0a76c9 | 3309 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 3310 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3311 | RING_F_SP_ENQ); |
0c6f39e5 | 3312 | free(ring_name); |
48fffdee KT |
3313 | if (ring_pair->cring_tx == NULL) { |
3314 | rte_free(ring_pair); | |
95fb793a | 3315 | return ENOMEM; |
3316 | } | |
3317 | ||
0c6f39e5 | 3318 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 3319 | |
8f0a76c9 | 3320 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 3321 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3322 | RING_F_SC_DEQ); |
0c6f39e5 | 3323 | free(ring_name); |
48fffdee KT |
3324 | if (ring_pair->cring_rx == NULL) { |
3325 | rte_free(ring_pair); | |
95fb793a | 3326 | return ENOMEM; |
3327 | } | |
3328 | ||
b8374d0d MV |
3329 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
3330 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 3331 | |
b8374d0d | 3332 | if (port_id < 0) { |
48fffdee | 3333 | rte_free(ring_pair); |
95fb793a | 3334 | return ENODEV; |
3335 | } | |
3336 | ||
48fffdee | 3337 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
3338 | ring_pair->eth_port_id = port_id; |
3339 | *eth_port_id = port_id; | |
3340 | ||
48fffdee | 3341 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 3342 | |
95fb793a | 3343 | return 0; |
3344 | } | |
3345 | ||
3346 | static int | |
bb37956a | 3347 | dpdk_ring_open(const char dev_name[], dpdk_port_t *eth_port_id) |
64839cf4 | 3348 | OVS_REQUIRES(dpdk_mutex) |
95fb793a | 3349 | { |
48fffdee | 3350 | struct dpdk_ring *ring_pair; |
95fb793a | 3351 | unsigned int port_no; |
3352 | int err = 0; | |
3353 | ||
3354 | /* Names always start with "dpdkr" */ | |
3355 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
3356 | if (err) { | |
3357 | return err; | |
3358 | } | |
3359 | ||
58be5c0e | 3360 | /* Look through our list to find the device */ |
48fffdee KT |
3361 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
3362 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 3363 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 3364 | /* Really all that is needed */ |
48fffdee | 3365 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 3366 | return 0; |
3367 | } | |
3368 | } | |
3369 | /* Need to create the device rings */ | |
3370 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
3371 | } | |
3372 | ||
7251515e | 3373 | static int |
d46285a2 | 3374 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
b30896c9 | 3375 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e | 3376 | { |
d46285a2 | 3377 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a543eb0 | 3378 | struct dp_packet *packet; |
1b99bb05 | 3379 | |
58be5c0e MK |
3380 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
3381 | * the rss hash field is clear. This is because the same mbuf may be | |
3382 | * modified by the consumer of the ring and return into the datapath | |
3383 | * without recalculating the RSS hash. */ | |
e883448e | 3384 | DP_PACKET_BATCH_FOR_EACH (i, packet, batch) { |
8a543eb0 | 3385 | dp_packet_mbuf_rss_flag_reset(packet); |
1b99bb05 | 3386 | } |
7251515e | 3387 | |
b30896c9 | 3388 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e DV |
3389 | return 0; |
3390 | } | |
3391 | ||
95fb793a | 3392 | static int |
3393 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
3394 | { | |
bb37956a | 3395 | dpdk_port_t port_no = 0; |
95fb793a | 3396 | int err = 0; |
3397 | ||
95fb793a | 3398 | ovs_mutex_lock(&dpdk_mutex); |
3399 | ||
3400 | err = dpdk_ring_open(netdev->name, &port_no); | |
3401 | if (err) { | |
3402 | goto unlock_dpdk; | |
3403 | } | |
3404 | ||
1ce30dfd DDP |
3405 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
3406 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 3407 | unlock_dpdk: |
3408 | ovs_mutex_unlock(&dpdk_mutex); | |
3409 | return err; | |
3410 | } | |
3411 | ||
0bf765f7 IS |
3412 | /* QoS Functions */ |
3413 | ||
3414 | /* | |
3415 | * Initialize QoS configuration operations. | |
3416 | */ | |
3417 | static void | |
3418 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
3419 | { | |
3420 | conf->ops = ops; | |
78bd47cf | 3421 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
3422 | } |
3423 | ||
3424 | /* | |
3425 | * Search existing QoS operations in qos_ops and compare each set of | |
3426 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
3427 | * else return NULL | |
3428 | */ | |
3429 | static const struct dpdk_qos_ops * | |
3430 | qos_lookup_name(const char *name) | |
3431 | { | |
3432 | const struct dpdk_qos_ops *const *opsp; | |
3433 | ||
3434 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3435 | const struct dpdk_qos_ops *ops = *opsp; | |
3436 | if (!strcmp(name, ops->qos_name)) { | |
3437 | return ops; | |
3438 | } | |
3439 | } | |
3440 | return NULL; | |
3441 | } | |
3442 | ||
0bf765f7 IS |
3443 | static int |
3444 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
3445 | struct sset *types) | |
3446 | { | |
3447 | const struct dpdk_qos_ops *const *opsp; | |
3448 | ||
3449 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3450 | const struct dpdk_qos_ops *ops = *opsp; | |
3451 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
3452 | sset_add(types, ops->qos_name); | |
3453 | } | |
3454 | } | |
3455 | return 0; | |
3456 | } | |
3457 | ||
3458 | static int | |
d46285a2 | 3459 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
3460 | const char **typep, struct smap *details) |
3461 | { | |
d46285a2 | 3462 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 3463 | struct qos_conf *qos_conf; |
0bf765f7 IS |
3464 | int error = 0; |
3465 | ||
d46285a2 | 3466 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
3467 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
3468 | if (qos_conf) { | |
3469 | *typep = qos_conf->ops->qos_name; | |
3470 | error = (qos_conf->ops->qos_get | |
3471 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
3472 | } else { |
3473 | /* No QoS configuration set, return an empty string */ | |
3474 | *typep = ""; | |
0bf765f7 | 3475 | } |
d46285a2 | 3476 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
3477 | |
3478 | return error; | |
3479 | } | |
3480 | ||
3481 | static int | |
78bd47cf DDP |
3482 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
3483 | const struct smap *details) | |
0bf765f7 | 3484 | { |
d46285a2 | 3485 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 3486 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 3487 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
3488 | int error = 0; |
3489 | ||
d46285a2 | 3490 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 3491 | |
78bd47cf | 3492 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 3493 | |
78bd47cf DDP |
3494 | new_ops = qos_lookup_name(type); |
3495 | ||
3496 | if (!new_ops || !new_ops->qos_construct) { | |
3497 | new_qos_conf = NULL; | |
3498 | if (type && type[0]) { | |
3499 | error = EOPNOTSUPP; | |
0bf765f7 | 3500 | } |
44975bb0 | 3501 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
3502 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
3503 | new_qos_conf = qos_conf; | |
0bf765f7 | 3504 | } else { |
78bd47cf | 3505 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
3506 | } |
3507 | ||
7ea266e9 | 3508 | if (error) { |
78bd47cf DDP |
3509 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
3510 | type, netdev->name, rte_strerror(error)); | |
3511 | } | |
3512 | ||
3513 | if (new_qos_conf != qos_conf) { | |
3514 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
3515 | if (qos_conf) { | |
3516 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
3517 | } | |
0bf765f7 IS |
3518 | } |
3519 | ||
d46285a2 | 3520 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 3521 | |
0bf765f7 IS |
3522 | return error; |
3523 | } | |
3524 | ||
3525 | /* egress-policer details */ | |
3526 | ||
3527 | struct egress_policer { | |
3528 | struct qos_conf qos_conf; | |
3529 | struct rte_meter_srtcm_params app_srtcm_params; | |
3530 | struct rte_meter_srtcm egress_meter; | |
3531 | }; | |
3532 | ||
78bd47cf DDP |
3533 | static void |
3534 | egress_policer_details_to_param(const struct smap *details, | |
3535 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 3536 | { |
78bd47cf DDP |
3537 | memset(params, 0, sizeof *params); |
3538 | params->cir = smap_get_ullong(details, "cir", 0); | |
3539 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
3540 | params->ebs = 0; | |
0bf765f7 IS |
3541 | } |
3542 | ||
3543 | static int | |
78bd47cf DDP |
3544 | egress_policer_qos_construct(const struct smap *details, |
3545 | struct qos_conf **conf) | |
0bf765f7 | 3546 | { |
0bf765f7 | 3547 | struct egress_policer *policer; |
0bf765f7 IS |
3548 | int err = 0; |
3549 | ||
0bf765f7 IS |
3550 | policer = xmalloc(sizeof *policer); |
3551 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 3552 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
0bf765f7 | 3553 | err = rte_meter_srtcm_config(&policer->egress_meter, |
78bd47cf DDP |
3554 | &policer->app_srtcm_params); |
3555 | if (!err) { | |
3556 | *conf = &policer->qos_conf; | |
3557 | } else { | |
7ea266e9 | 3558 | free(policer); |
78bd47cf | 3559 | *conf = NULL; |
7ea266e9 IS |
3560 | err = -err; |
3561 | } | |
0bf765f7 IS |
3562 | |
3563 | return err; | |
3564 | } | |
3565 | ||
3566 | static void | |
78bd47cf | 3567 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
3568 | { |
3569 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 3570 | qos_conf); |
0bf765f7 IS |
3571 | free(policer); |
3572 | } | |
3573 | ||
3574 | static int | |
78bd47cf | 3575 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 3576 | { |
78bd47cf DDP |
3577 | struct egress_policer *policer = |
3578 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3579 | ||
3580 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
3581 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 3582 | |
0bf765f7 IS |
3583 | return 0; |
3584 | } | |
3585 | ||
78bd47cf | 3586 | static bool |
47a45d86 KT |
3587 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
3588 | const struct smap *details) | |
0bf765f7 | 3589 | { |
78bd47cf DDP |
3590 | struct egress_policer *policer = |
3591 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3592 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 3593 | |
78bd47cf | 3594 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 3595 | |
78bd47cf | 3596 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
3597 | } |
3598 | ||
0bf765f7 | 3599 | static int |
3e90f7d7 GZ |
3600 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt, |
3601 | bool may_steal) | |
0bf765f7 | 3602 | { |
0bf765f7 | 3603 | int cnt = 0; |
78bd47cf DDP |
3604 | struct egress_policer *policer = |
3605 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 3606 | |
3e90f7d7 GZ |
3607 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, |
3608 | pkt_cnt, may_steal); | |
0bf765f7 IS |
3609 | |
3610 | return cnt; | |
3611 | } | |
3612 | ||
3613 | static const struct dpdk_qos_ops egress_policer_ops = { | |
3614 | "egress-policer", /* qos_name */ | |
3615 | egress_policer_qos_construct, | |
3616 | egress_policer_qos_destruct, | |
3617 | egress_policer_qos_get, | |
78bd47cf | 3618 | egress_policer_qos_is_equal, |
0bf765f7 IS |
3619 | egress_policer_run |
3620 | }; | |
3621 | ||
050c60bf DDP |
3622 | static int |
3623 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
3624 | { | |
3625 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3626 | int err = 0; | |
3627 | ||
050c60bf DDP |
3628 | ovs_mutex_lock(&dev->mutex); |
3629 | ||
3630 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 3631 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b CL |
3632 | && dev->mtu == dev->requested_mtu |
3633 | && dev->rxq_size == dev->requested_rxq_size | |
bd4e172b | 3634 | && dev->txq_size == dev->requested_txq_size |
3635 | && dev->socket_id == dev->requested_socket_id) { | |
050c60bf DDP |
3636 | /* Reconfiguration is unnecessary */ |
3637 | ||
3638 | goto out; | |
3639 | } | |
3640 | ||
3641 | rte_eth_dev_stop(dev->port_id); | |
3642 | ||
d555d9bd | 3643 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 | 3644 | if (err && err != EEXIST) { |
d555d9bd | 3645 | goto out; |
0072e931 MK |
3646 | } |
3647 | ||
050c60bf DDP |
3648 | netdev->n_txq = dev->requested_n_txq; |
3649 | netdev->n_rxq = dev->requested_n_rxq; | |
3650 | ||
b685696b CL |
3651 | dev->rxq_size = dev->requested_rxq_size; |
3652 | dev->txq_size = dev->requested_txq_size; | |
3653 | ||
050c60bf DDP |
3654 | rte_free(dev->tx_q); |
3655 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
3656 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
3657 | if (!dev->tx_q) { | |
3658 | err = ENOMEM; | |
3659 | } | |
050c60bf | 3660 | |
0072e931 MK |
3661 | netdev_change_seq_changed(netdev); |
3662 | ||
050c60bf | 3663 | out: |
050c60bf | 3664 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
3665 | return err; |
3666 | } | |
3667 | ||
7f381c2e | 3668 | static int |
2d24d165 | 3669 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 3670 | OVS_REQUIRES(dev->mutex) |
050c60bf | 3671 | { |
2d24d165 CL |
3672 | dev->up.n_txq = dev->requested_n_txq; |
3673 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 3674 | int err; |
050c60bf | 3675 | |
81acebda IM |
3676 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
3677 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
3678 | dev->tx_q[0].map = 0; | |
3679 | } | |
3680 | ||
3681 | netdev_dpdk_remap_txqs(dev); | |
3682 | ||
d555d9bd | 3683 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 FA |
3684 | if (!err) { |
3685 | /* A new mempool was created. */ | |
d555d9bd | 3686 | netdev_change_seq_changed(&dev->up); |
b6b26021 FA |
3687 | } else if (err != EEXIST){ |
3688 | return err; | |
db8f13b0 | 3689 | } |
0a0f39df | 3690 | if (netdev_dpdk_get_vid(dev) >= 0) { |
894af647 | 3691 | if (dev->vhost_reconfigured == false) { |
3692 | dev->vhost_reconfigured = true; | |
3693 | /* Carrier status may need updating. */ | |
3694 | netdev_change_seq_changed(&dev->up); | |
3695 | } | |
81acebda | 3696 | } |
7f381c2e DDP |
3697 | |
3698 | return 0; | |
2d24d165 CL |
3699 | } |
3700 | ||
3701 | static int | |
3702 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
3703 | { | |
3704 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3705 | int err; |
2d24d165 | 3706 | |
2d24d165 | 3707 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 3708 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 3709 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
3710 | |
3711 | return err; | |
2d24d165 CL |
3712 | } |
3713 | ||
3714 | static int | |
3715 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3716 | { | |
3717 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3718 | int err; |
a14d1cc8 | 3719 | uint64_t vhost_flags = 0; |
10087cba | 3720 | bool zc_enabled; |
2d24d165 | 3721 | |
2d24d165 CL |
3722 | ovs_mutex_lock(&dev->mutex); |
3723 | ||
c1ff66ac CL |
3724 | /* Configure vHost client mode if requested and if the following criteria |
3725 | * are met: | |
2d24d165 CL |
3726 | * 1. Device hasn't been registered yet. |
3727 | * 2. A path has been specified. | |
c1ff66ac CL |
3728 | */ |
3729 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 | 3730 | && strlen(dev->vhost_id)) { |
a14d1cc8 MK |
3731 | /* Register client-mode device. */ |
3732 | vhost_flags |= RTE_VHOST_USER_CLIENT; | |
3733 | ||
3734 | /* Enable IOMMU support, if explicitly requested. */ | |
3735 | if (dpdk_vhost_iommu_enabled()) { | |
3736 | vhost_flags |= RTE_VHOST_USER_IOMMU_SUPPORT; | |
3737 | } | |
10087cba CL |
3738 | |
3739 | zc_enabled = dev->vhost_driver_flags | |
3740 | & RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
3741 | /* Enable zero copy flag, if requested */ | |
3742 | if (zc_enabled) { | |
3743 | vhost_flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
3744 | } | |
3745 | ||
a14d1cc8 | 3746 | err = rte_vhost_driver_register(dev->vhost_id, vhost_flags); |
c1ff66ac | 3747 | if (err) { |
2d24d165 CL |
3748 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3749 | dev->vhost_id); | |
7f381c2e | 3750 | goto unlock; |
c1ff66ac | 3751 | } else { |
2d24d165 | 3752 | /* Configuration successful */ |
a14d1cc8 | 3753 | dev->vhost_driver_flags |= vhost_flags; |
2d24d165 CL |
3754 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " |
3755 | "using client socket '%s'", | |
3756 | dev->up.name, dev->vhost_id); | |
10087cba CL |
3757 | if (zc_enabled) { |
3758 | VLOG_INFO("Zero copy enabled for vHost port %s", dev->up.name); | |
3759 | } | |
c1ff66ac | 3760 | } |
f3e7ec25 MW |
3761 | |
3762 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
3763 | &virtio_net_device_ops); | |
3764 | if (err) { | |
3765 | VLOG_ERR("rte_vhost_driver_callback_register failed for " | |
3766 | "vhost user client port: %s\n", dev->up.name); | |
3767 | goto unlock; | |
3768 | } | |
3769 | ||
3770 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
3771 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
3772 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
3773 | | 1ULL << VIRTIO_NET_F_CSUM); | |
3774 | if (err) { | |
3775 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
3776 | "client port: %s\n", dev->up.name); | |
3777 | goto unlock; | |
3778 | } | |
3779 | ||
3780 | err = rte_vhost_driver_start(dev->vhost_id); | |
3781 | if (err) { | |
3782 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
3783 | "client port: %s\n", dev->up.name); | |
3784 | goto unlock; | |
3785 | } | |
c1ff66ac CL |
3786 | } |
3787 | ||
7f381c2e DDP |
3788 | err = dpdk_vhost_reconfigure_helper(dev); |
3789 | ||
3790 | unlock: | |
050c60bf | 3791 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 3792 | |
7f381c2e | 3793 | return err; |
050c60bf DDP |
3794 | } |
3795 | ||
ecc1a34e | 3796 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
81acebda | 3797 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
971f4b39 MW |
3798 | GET_CARRIER, GET_STATS, \ |
3799 | GET_CUSTOM_STATS, \ | |
81acebda IM |
3800 | GET_FEATURES, GET_STATUS, \ |
3801 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3802 | { \ |
3803 | NAME, \ | |
118c77b1 | 3804 | true, /* is_pmd */ \ |
ecc1a34e | 3805 | INIT, /* init */ \ |
95fb793a | 3806 | NULL, /* netdev_dpdk_run */ \ |
3807 | NULL, /* netdev_dpdk_wait */ \ | |
3808 | \ | |
3809 | netdev_dpdk_alloc, \ | |
3810 | CONSTRUCT, \ | |
58397e6c | 3811 | DESTRUCT, \ |
95fb793a | 3812 | netdev_dpdk_dealloc, \ |
3813 | netdev_dpdk_get_config, \ | |
81acebda | 3814 | SET_CONFIG, \ |
95fb793a | 3815 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3816 | NULL, /* build header */ \ |
3817 | NULL, /* push header */ \ | |
3818 | NULL, /* pop header */ \ | |
7dec44fe | 3819 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3820 | SET_TX_MULTIQ, \ |
95fb793a | 3821 | \ |
7251515e | 3822 | SEND, /* send */ \ |
95fb793a | 3823 | NULL, /* send_wait */ \ |
3824 | \ | |
3825 | netdev_dpdk_set_etheraddr, \ | |
3826 | netdev_dpdk_get_etheraddr, \ | |
3827 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3828 | netdev_dpdk_set_mtu, \ |
95fb793a | 3829 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3830 | GET_CARRIER, \ |
95fb793a | 3831 | netdev_dpdk_get_carrier_resets, \ |
3832 | netdev_dpdk_set_miimon, \ | |
58397e6c | 3833 | GET_STATS, \ |
971f4b39 | 3834 | GET_CUSTOM_STATS, \ |
58397e6c | 3835 | GET_FEATURES, \ |
95fb793a | 3836 | NULL, /* set_advertisements */ \ |
875ab130 | 3837 | NULL, /* get_pt_mode */ \ |
95fb793a | 3838 | \ |
9509913a | 3839 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3840 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3841 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3842 | netdev_dpdk_get_qos, \ |
3843 | netdev_dpdk_set_qos, \ | |
95fb793a | 3844 | NULL, /* get_queue */ \ |
3845 | NULL, /* set_queue */ \ | |
3846 | NULL, /* delete_queue */ \ | |
3847 | NULL, /* get_queue_stats */ \ | |
3848 | NULL, /* queue_dump_start */ \ | |
3849 | NULL, /* queue_dump_next */ \ | |
3850 | NULL, /* queue_dump_done */ \ | |
3851 | NULL, /* dump_queue_stats */ \ | |
3852 | \ | |
95fb793a | 3853 | NULL, /* set_in4 */ \ |
a8704b50 | 3854 | NULL, /* get_addr_list */ \ |
95fb793a | 3855 | NULL, /* add_router */ \ |
3856 | NULL, /* get_next_hop */ \ | |
58397e6c | 3857 | GET_STATUS, \ |
95fb793a | 3858 | NULL, /* arp_lookup */ \ |
3859 | \ | |
3860 | netdev_dpdk_update_flags, \ | |
050c60bf | 3861 | RECONFIGURE, \ |
95fb793a | 3862 | \ |
3863 | netdev_dpdk_rxq_alloc, \ | |
3864 | netdev_dpdk_rxq_construct, \ | |
3865 | netdev_dpdk_rxq_destruct, \ | |
3866 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3867 | RXQ_RECV, \ |
95fb793a | 3868 | NULL, /* rx_wait */ \ |
3869 | NULL, /* rxq_drain */ \ | |
18ebd48c | 3870 | NO_OFFLOAD_API \ |
95fb793a | 3871 | } |
8a9562d2 | 3872 | |
bce01e3a | 3873 | static const struct netdev_class dpdk_class = |
95fb793a | 3874 | NETDEV_DPDK_CLASS( |
3875 | "dpdk", | |
ecc1a34e | 3876 | netdev_dpdk_class_init, |
5496878c | 3877 | netdev_dpdk_construct, |
58397e6c | 3878 | netdev_dpdk_destruct, |
81acebda IM |
3879 | netdev_dpdk_set_config, |
3880 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3881 | netdev_dpdk_eth_send, |
3882 | netdev_dpdk_get_carrier, | |
3883 | netdev_dpdk_get_stats, | |
971f4b39 | 3884 | netdev_dpdk_get_custom_stats, |
58397e6c KT |
3885 | netdev_dpdk_get_features, |
3886 | netdev_dpdk_get_status, | |
050c60bf | 3887 | netdev_dpdk_reconfigure, |
58397e6c | 3888 | netdev_dpdk_rxq_recv); |
95fb793a | 3889 | |
bce01e3a | 3890 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3891 | NETDEV_DPDK_CLASS( |
3892 | "dpdkr", | |
ecc1a34e | 3893 | netdev_dpdk_class_init, |
5496878c | 3894 | netdev_dpdk_ring_construct, |
58397e6c | 3895 | netdev_dpdk_destruct, |
c3d062a7 | 3896 | netdev_dpdk_ring_set_config, |
81acebda | 3897 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3898 | netdev_dpdk_ring_send, |
3899 | netdev_dpdk_get_carrier, | |
3900 | netdev_dpdk_get_stats, | |
971f4b39 | 3901 | netdev_dpdk_get_custom_stats, |
58397e6c KT |
3902 | netdev_dpdk_get_features, |
3903 | netdev_dpdk_get_status, | |
050c60bf | 3904 | netdev_dpdk_reconfigure, |
58397e6c KT |
3905 | netdev_dpdk_rxq_recv); |
3906 | ||
53f50d24 | 3907 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3908 | NETDEV_DPDK_CLASS( |
3909 | "dpdkvhostuser", | |
f3e7ec25 | 3910 | NULL, |
53f50d24 | 3911 | netdev_dpdk_vhost_construct, |
58397e6c | 3912 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3913 | NULL, |
81acebda | 3914 | NULL, |
58397e6c KT |
3915 | netdev_dpdk_vhost_send, |
3916 | netdev_dpdk_vhost_get_carrier, | |
3917 | netdev_dpdk_vhost_get_stats, | |
3918 | NULL, | |
7251515e | 3919 | NULL, |
b2e8b12f | 3920 | netdev_dpdk_vhost_user_get_status, |
53f50d24 | 3921 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3922 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3923 | static const struct netdev_class dpdk_vhost_client_class = |
3924 | NETDEV_DPDK_CLASS( | |
3925 | "dpdkvhostuserclient", | |
f3e7ec25 | 3926 | NULL, |
2d24d165 CL |
3927 | netdev_dpdk_vhost_client_construct, |
3928 | netdev_dpdk_vhost_destruct, | |
3929 | netdev_dpdk_vhost_client_set_config, | |
3930 | NULL, | |
3931 | netdev_dpdk_vhost_send, | |
3932 | netdev_dpdk_vhost_get_carrier, | |
3933 | netdev_dpdk_vhost_get_stats, | |
3934 | NULL, | |
3935 | NULL, | |
b2e8b12f | 3936 | netdev_dpdk_vhost_user_get_status, |
2d24d165 CL |
3937 | netdev_dpdk_vhost_client_reconfigure, |
3938 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 3939 | |
8a9562d2 PS |
3940 | void |
3941 | netdev_dpdk_register(void) | |
3942 | { | |
bab69409 AC |
3943 | netdev_register_provider(&dpdk_class); |
3944 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3945 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 3946 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 3947 | } |