]>
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, |
7d7ded7a | 275 | int pkt_cnt, bool should_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; | |
606f6650 EC |
369 | /* If true, rte_eth_dev_start() was successfully called */ |
370 | bool started; | |
23d4d53f BB |
371 | struct eth_addr hwaddr; |
372 | int mtu; | |
373 | int socket_id; | |
374 | int buf_size; | |
375 | int max_packet_len; | |
376 | enum dpdk_dev_type type; | |
377 | enum netdev_flags flags; | |
eaa43581 | 378 | int link_reset_cnt; |
23d4d53f BB |
379 | char *devargs; /* Device arguments for dpdk ports */ |
380 | struct dpdk_tx_queue *tx_q; | |
381 | struct rte_eth_link link; | |
23d4d53f BB |
382 | ); |
383 | ||
384 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline1, | |
385 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
24e78f93 | 386 | struct rte_mempool *mp; |
23d4d53f BB |
387 | |
388 | /* virtio identifier for vhost devices */ | |
389 | ovsrcu_index vid; | |
390 | ||
391 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
392 | bool vhost_reconfigured; | |
393 | /* 3 pad bytes here. */ | |
394 | ); | |
395 | ||
396 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
397 | /* Identifier used to distinguish vhost devices from each other. */ | |
398 | char vhost_id[PATH_MAX]; | |
399 | ); | |
400 | ||
401 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
402 | struct netdev up; | |
403 | /* In dpdk_list. */ | |
404 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); | |
405 | ||
406 | /* QoS configuration and lock for the device */ | |
407 | OVSRCU_TYPE(struct qos_conf *) qos_conf; | |
408 | ||
409 | /* Ingress Policer */ | |
410 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
411 | uint32_t policer_rate; | |
412 | uint32_t policer_burst; | |
413 | ); | |
414 | ||
415 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
416 | struct netdev_stats stats; | |
417 | /* Protects stats */ | |
418 | rte_spinlock_t stats_lock; | |
419 | /* 44 pad bytes here. */ | |
420 | ); | |
421 | ||
422 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
423 | /* The following properties cannot be changed when a device is running, | |
424 | * so we remember the request and update them next time | |
425 | * netdev_dpdk*_reconfigure() is called */ | |
426 | int requested_mtu; | |
427 | int requested_n_txq; | |
428 | int requested_n_rxq; | |
429 | int requested_rxq_size; | |
430 | int requested_txq_size; | |
431 | ||
432 | /* Number of rx/tx descriptors for physical devices */ | |
433 | int rxq_size; | |
434 | int txq_size; | |
435 | ||
436 | /* Socket ID detected when vHost device is brought up */ | |
437 | int requested_socket_id; | |
438 | ||
439 | /* Denotes whether vHost port is client/server mode */ | |
440 | uint64_t vhost_driver_flags; | |
441 | ||
442 | /* DPDK-ETH Flow control */ | |
443 | struct rte_eth_fc_conf fc_conf; | |
444 | ||
445 | /* DPDK-ETH hardware offload features, | |
446 | * from the enum set 'dpdk_hw_ol_features' */ | |
447 | uint32_t hw_ol_features; | |
f8b64a61 RM |
448 | |
449 | /* Properties for link state change detection mode. | |
450 | * If lsc_interrupt_mode is set to false, poll mode is used, | |
451 | * otherwise interrupt mode is used. */ | |
452 | bool requested_lsc_interrupt_mode; | |
453 | bool lsc_interrupt_mode; | |
23d4d53f | 454 | ); |
971f4b39 MW |
455 | |
456 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
457 | /* Names of all XSTATS counters */ | |
458 | struct rte_eth_xstat_name *rte_xstats_names; | |
459 | int rte_xstats_names_size; | |
460 | int rte_xstats_ids_size; | |
461 | uint64_t *rte_xstats_ids; | |
462 | ); | |
8a9562d2 PS |
463 | }; |
464 | ||
465 | struct netdev_rxq_dpdk { | |
466 | struct netdev_rxq up; | |
bb37956a | 467 | dpdk_port_t port_id; |
8a9562d2 PS |
468 | }; |
469 | ||
f3e7ec25 MW |
470 | static void netdev_dpdk_destruct(struct netdev *netdev); |
471 | static void netdev_dpdk_vhost_destruct(struct netdev *netdev); | |
8a9562d2 | 472 | |
ac1a9bb9 IM |
473 | static void netdev_dpdk_clear_xstats(struct netdev_dpdk *dev); |
474 | ||
0a0f39df | 475 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 476 | |
9509913a IS |
477 | struct ingress_policer * |
478 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
479 | ||
8a9562d2 PS |
480 | static bool |
481 | is_dpdk_class(const struct netdev_class *class) | |
482 | { | |
f3e7ec25 MW |
483 | return class->destruct == netdev_dpdk_destruct |
484 | || class->destruct == netdev_dpdk_vhost_destruct; | |
8a9562d2 PS |
485 | } |
486 | ||
4be4d22c MK |
487 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
488 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
489 | * value, insufficient buffers are allocated to accomodate the packet in its | |
490 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
491 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
492 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
493 | * behaviour, which reduces performance. To prevent this, use a buffer size |
494 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
495 | */ |
496 | static uint32_t | |
497 | dpdk_buf_size(int mtu) | |
498 | { | |
499 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
500 | NETDEV_DPDK_MBUF_ALIGN); | |
501 | } | |
502 | ||
eff23640 DDP |
503 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
504 | * | |
505 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
506 | static void * |
507 | dpdk_rte_mzalloc(size_t sz) | |
508 | { | |
eff23640 | 509 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
510 | } |
511 | ||
512 | void | |
e14deea0 | 513 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 514 | { |
db73f716 | 515 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 516 | |
b00b4a81 | 517 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
518 | } |
519 | ||
b3cd9f9d | 520 | static void |
401b70d6 | 521 | ovs_rte_pktmbuf_init(struct rte_mempool *mp OVS_UNUSED, |
b3cd9f9d | 522 | void *opaque_arg OVS_UNUSED, |
2391135c | 523 | void *_p, |
b3cd9f9d PS |
524 | unsigned i OVS_UNUSED) |
525 | { | |
2391135c | 526 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 527 | |
2391135c | 528 | dp_packet_init_dpdk((struct dp_packet *) pkt, pkt->buf_len); |
b3cd9f9d PS |
529 | } |
530 | ||
91fccdad KT |
531 | static int |
532 | dpdk_mp_full(const struct rte_mempool *mp) OVS_REQUIRES(dpdk_mp_mutex) | |
533 | { | |
1f84a2d5 KT |
534 | /* At this point we want to know if all the mbufs are back |
535 | * in the mempool. rte_mempool_full() is not atomic but it's | |
536 | * the best available and as we are no longer requesting mbufs | |
537 | * from the mempool, it means mbufs will not move from | |
538 | * 'mempool ring' --> 'mempool cache'. In rte_mempool_full() | |
539 | * the ring is counted before caches, so we won't get false | |
540 | * positives in this use case and we handle false negatives. | |
541 | * | |
542 | * If future implementations of rte_mempool_full() were to change | |
543 | * it could be possible for a false positive. Even that would | |
544 | * likely be ok, as there are additional checks during mempool | |
545 | * freeing but it would make things racey. | |
91fccdad | 546 | */ |
1f84a2d5 | 547 | return rte_mempool_full(mp); |
91fccdad KT |
548 | } |
549 | ||
550 | /* Free unused mempools. */ | |
551 | static void | |
552 | dpdk_mp_sweep(void) | |
553 | { | |
554 | struct dpdk_mp *dmp, *next; | |
555 | ||
556 | ovs_mutex_lock(&dpdk_mp_mutex); | |
557 | LIST_FOR_EACH_SAFE (dmp, next, list_node, &dpdk_mp_free_list) { | |
558 | if (dpdk_mp_full(dmp->mp)) { | |
559 | VLOG_DBG("Freeing mempool \"%s\"", dmp->mp->name); | |
560 | ovs_list_remove(&dmp->list_node); | |
561 | rte_mempool_free(dmp->mp); | |
562 | rte_free(dmp); | |
563 | } | |
564 | } | |
565 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
566 | } | |
567 | ||
568 | /* Ensure a mempool will not be freed. */ | |
569 | static void | |
570 | dpdk_mp_do_not_free(struct rte_mempool *mp) OVS_REQUIRES(dpdk_mp_mutex) | |
571 | { | |
572 | struct dpdk_mp *dmp, *next; | |
573 | ||
574 | LIST_FOR_EACH_SAFE (dmp, next, list_node, &dpdk_mp_free_list) { | |
575 | if (dmp->mp == mp) { | |
576 | VLOG_DBG("Removing mempool \"%s\" from free list", dmp->mp->name); | |
577 | ovs_list_remove(&dmp->list_node); | |
578 | rte_free(dmp); | |
579 | break; | |
580 | } | |
581 | } | |
582 | } | |
583 | ||
24e78f93 IM |
584 | /* Returns a valid pointer when either of the following is true: |
585 | * - a new mempool was just created; | |
586 | * - a matching mempool already exists. */ | |
587 | static struct rte_mempool * | |
588 | dpdk_mp_create(struct netdev_dpdk *dev, int mtu) | |
8a9562d2 | 589 | { |
24e78f93 IM |
590 | char mp_name[RTE_MEMPOOL_NAMESIZE]; |
591 | const char *netdev_name = netdev_get_name(&dev->up); | |
592 | int socket_id = dev->requested_socket_id; | |
593 | uint32_t n_mbufs; | |
594 | uint32_t hash = hash_string(netdev_name, 0); | |
595 | struct rte_mempool *mp = NULL; | |
d555d9bd RW |
596 | |
597 | /* | |
ad9b5b9b | 598 | * XXX: rough estimation of number of mbufs required for this port: |
d555d9bd RW |
599 | * <packets required to fill the device rxqs> |
600 | * + <packets that could be stuck on other ports txqs> | |
601 | * + <packets in the pmd threads> | |
602 | * + <additional memory for corner cases> | |
0072e931 | 603 | */ |
24e78f93 IM |
604 | n_mbufs = dev->requested_n_rxq * dev->requested_rxq_size |
605 | + dev->requested_n_txq * dev->requested_txq_size | |
606 | + MIN(RTE_MAX_LCORE, dev->requested_n_rxq) * NETDEV_MAX_BURST | |
607 | + MIN_NB_MBUF; | |
d555d9bd | 608 | |
24e78f93 | 609 | ovs_mutex_lock(&dpdk_mp_mutex); |
da79ce2b | 610 | do { |
24e78f93 IM |
611 | /* Full DPDK memory pool name must be unique and cannot be |
612 | * longer than RTE_MEMPOOL_NAMESIZE. */ | |
af5b0dad IM |
613 | int ret = snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, |
614 | "ovs%08x%02d%05d%07u", | |
24e78f93 IM |
615 | hash, socket_id, mtu, n_mbufs); |
616 | if (ret < 0 || ret >= RTE_MEMPOOL_NAMESIZE) { | |
617 | VLOG_DBG("snprintf returned %d. " | |
618 | "Failed to generate a mempool name for \"%s\". " | |
619 | "Hash:0x%x, socket_id: %d, mtu:%d, mbufs:%u.", | |
620 | ret, netdev_name, hash, socket_id, mtu, n_mbufs); | |
621 | break; | |
65056fd7 | 622 | } |
95fb793a | 623 | |
f06546a5 FA |
624 | VLOG_DBG("Port %s: Requesting a mempool of %u mbufs " |
625 | "on socket %d for %d Rx and %d Tx queues.", | |
24e78f93 | 626 | netdev_name, n_mbufs, socket_id, |
f06546a5 | 627 | dev->requested_n_rxq, dev->requested_n_txq); |
d555d9bd | 628 | |
24e78f93 IM |
629 | mp = rte_pktmbuf_pool_create(mp_name, n_mbufs, MP_CACHE_SZ, |
630 | sizeof (struct dp_packet) - sizeof (struct rte_mbuf), | |
631 | MBUF_SIZE(mtu) - sizeof(struct dp_packet), socket_id); | |
632 | ||
633 | if (mp) { | |
634 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", | |
635 | mp_name, n_mbufs); | |
837c1761 FA |
636 | /* rte_pktmbuf_pool_create has done some initialization of the |
637 | * rte_mbuf part of each dp_packet. Some OvS specific fields | |
638 | * of the packet still need to be initialized by | |
639 | * ovs_rte_pktmbuf_init. */ | |
24e78f93 | 640 | rte_mempool_obj_iter(mp, ovs_rte_pktmbuf_init, NULL); |
d555d9bd RW |
641 | } else if (rte_errno == EEXIST) { |
642 | /* A mempool with the same name already exists. We just | |
643 | * retrieve its pointer to be returned to the caller. */ | |
24e78f93 | 644 | mp = rte_mempool_lookup(mp_name); |
d555d9bd RW |
645 | /* As the mempool create returned EEXIST we can expect the |
646 | * lookup has returned a valid pointer. If for some reason | |
647 | * that's not the case we keep track of it. */ | |
24e78f93 IM |
648 | VLOG_DBG("A mempool with name \"%s\" already exists at %p.", |
649 | mp_name, mp); | |
91fccdad KT |
650 | /* Ensure this reused mempool will not be freed. */ |
651 | dpdk_mp_do_not_free(mp); | |
d555d9bd RW |
652 | } else { |
653 | VLOG_ERR("Failed mempool \"%s\" create request of %u mbufs", | |
24e78f93 | 654 | mp_name, n_mbufs); |
0c6f39e5 | 655 | } |
24e78f93 | 656 | } while (!mp && rte_errno == ENOMEM && (n_mbufs /= 2) >= MIN_NB_MBUF); |
2ae3d542 | 657 | |
c2adb102 | 658 | ovs_mutex_unlock(&dpdk_mp_mutex); |
24e78f93 | 659 | return mp; |
8a9562d2 PS |
660 | } |
661 | ||
a08a115d | 662 | /* Release an existing mempool. */ |
8a9562d2 | 663 | static void |
91fccdad | 664 | dpdk_mp_release(struct rte_mempool *mp) |
8a9562d2 | 665 | { |
24e78f93 | 666 | if (!mp) { |
8a9562d2 PS |
667 | return; |
668 | } | |
669 | ||
c2adb102 | 670 | ovs_mutex_lock(&dpdk_mp_mutex); |
91fccdad KT |
671 | if (dpdk_mp_full(mp)) { |
672 | VLOG_DBG("Freeing mempool \"%s\"", mp->name); | |
673 | rte_mempool_free(mp); | |
674 | } else { | |
675 | struct dpdk_mp *dmp; | |
676 | ||
677 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
678 | if (dmp) { | |
679 | dmp->mp = mp; | |
680 | ovs_list_push_back(&dpdk_mp_free_list, &dmp->list_node); | |
681 | } | |
682 | } | |
c2adb102 | 683 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
684 | } |
685 | ||
b6b26021 FA |
686 | /* Tries to allocate a new mempool - or re-use an existing one where |
687 | * appropriate - on requested_socket_id with a size determined by | |
688 | * requested_mtu and requested Rx/Tx queues. | |
689 | * On success - or when re-using an existing mempool - the new configuration | |
690 | * will be applied. | |
0072e931 MK |
691 | * On error, device will be left unchanged. */ |
692 | static int | |
693 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
694 | OVS_REQUIRES(dev->mutex) |
695 | { | |
696 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
24e78f93 IM |
697 | struct rte_mempool *mp; |
698 | int ret = 0; | |
0072e931 | 699 | |
91fccdad KT |
700 | dpdk_mp_sweep(); |
701 | ||
24e78f93 | 702 | mp = dpdk_mp_create(dev, FRAME_LEN_TO_MTU(buf_size)); |
0072e931 | 703 | if (!mp) { |
c67e46c0 MK |
704 | VLOG_ERR("Failed to create memory pool for netdev " |
705 | "%s, with MTU %d on socket %d: %s\n", | |
706 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
707 | rte_strerror(rte_errno)); | |
24e78f93 | 708 | ret = rte_errno; |
0072e931 | 709 | } else { |
24e78f93 IM |
710 | /* If a new MTU was requested and its rounded value equals the one |
711 | * that is currently used, then the existing mempool is returned. */ | |
712 | if (dev->mp != mp) { | |
713 | /* A new mempool was created, release the previous one. */ | |
91fccdad | 714 | dpdk_mp_release(dev->mp); |
24e78f93 IM |
715 | } else { |
716 | ret = EEXIST; | |
717 | } | |
718 | dev->mp = mp; | |
0072e931 MK |
719 | dev->mtu = dev->requested_mtu; |
720 | dev->socket_id = dev->requested_socket_id; | |
721 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
722 | } | |
723 | ||
24e78f93 | 724 | return ret; |
0072e931 MK |
725 | } |
726 | ||
8a9562d2 PS |
727 | static void |
728 | check_link_status(struct netdev_dpdk *dev) | |
729 | { | |
730 | struct rte_eth_link link; | |
731 | ||
732 | rte_eth_link_get_nowait(dev->port_id, &link); | |
733 | ||
734 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 735 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
736 | |
737 | dev->link_reset_cnt++; | |
738 | dev->link = link; | |
739 | if (dev->link.link_status) { | |
fa9f4eeb IM |
740 | VLOG_DBG_RL(&rl, |
741 | "Port "DPDK_PORT_ID_FMT" Link Up - speed %u Mbps - %s", | |
58be5c0e | 742 | dev->port_id, (unsigned) dev->link.link_speed, |
fa9f4eeb IM |
743 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) |
744 | ? "full-duplex" : "half-duplex"); | |
8a9562d2 | 745 | } else { |
fa9f4eeb IM |
746 | VLOG_DBG_RL(&rl, "Port "DPDK_PORT_ID_FMT" Link Down", |
747 | dev->port_id); | |
8a9562d2 PS |
748 | } |
749 | } | |
750 | } | |
751 | ||
752 | static void * | |
753 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
754 | { | |
755 | struct netdev_dpdk *dev; | |
756 | ||
757 | pthread_detach(pthread_self()); | |
758 | ||
759 | for (;;) { | |
760 | ovs_mutex_lock(&dpdk_mutex); | |
761 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
762 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
763 | if (dev->type == DPDK_DEV_ETH) { |
764 | check_link_status(dev); | |
765 | } | |
8a9562d2 PS |
766 | ovs_mutex_unlock(&dev->mutex); |
767 | } | |
768 | ovs_mutex_unlock(&dpdk_mutex); | |
769 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
770 | } | |
771 | ||
772 | return NULL; | |
773 | } | |
774 | ||
b98d7669 | 775 | static int |
f8b64a61 | 776 | dpdk_eth_dev_port_config(struct netdev_dpdk *dev, int n_rxq, int n_txq) |
b98d7669 DDP |
777 | { |
778 | int diag = 0; | |
779 | int i; | |
0072e931 | 780 | struct rte_eth_conf conf = port_conf; |
65a87968 PC |
781 | struct rte_eth_dev_info info; |
782 | ||
783 | /* As of DPDK 17.11.1 a few PMDs require to explicitly enable | |
784 | * scatter to support jumbo RX. Checking the offload capabilities | |
785 | * is not an option as PMDs are not required yet to report | |
786 | * them. The only reliable info is the driver name and knowledge | |
787 | * (testing or code review). Listing all such PMDs feels harder | |
788 | * than highlighting the one known not to need scatter */ | |
0072e931 | 789 | if (dev->mtu > ETHER_MTU) { |
65a87968 | 790 | rte_eth_dev_info_get(dev->port_id, &info); |
7bbc2e1d | 791 | if (strncmp(info.driver_name, "net_nfp", 7)) { |
65a87968 PC |
792 | conf.rxmode.enable_scatter = 1; |
793 | } | |
0072e931 | 794 | } |
67fe6d63 | 795 | |
f8b64a61 | 796 | conf.intr_conf.lsc = dev->lsc_interrupt_mode; |
1a2bb118 SC |
797 | conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & |
798 | NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
b98d7669 DDP |
799 | /* A device may report more queues than it makes available (this has |
800 | * been observed for Intel xl710, which reserves some of them for | |
801 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
802 | * available. When this happens we can retry the configuration | |
803 | * and request less queues */ | |
804 | while (n_rxq && n_txq) { | |
805 | if (diag) { | |
806 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
807 | } | |
808 | ||
0072e931 | 809 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 810 | if (diag) { |
0072e931 MK |
811 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
812 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
813 | break; |
814 | } | |
815 | ||
67fe6d63 MK |
816 | diag = rte_eth_dev_set_mtu(dev->port_id, dev->mtu); |
817 | if (diag) { | |
818 | VLOG_ERR("Interface %s MTU (%d) setup error: %s", | |
819 | dev->up.name, dev->mtu, rte_strerror(-diag)); | |
820 | break; | |
821 | } | |
822 | ||
b98d7669 | 823 | for (i = 0; i < n_txq; i++) { |
b685696b | 824 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
825 | dev->socket_id, NULL); |
826 | if (diag) { | |
1dfebee9 | 827 | VLOG_INFO("Interface %s unable to setup txq(%d): %s", |
b98d7669 DDP |
828 | dev->up.name, i, rte_strerror(-diag)); |
829 | break; | |
830 | } | |
831 | } | |
832 | ||
833 | if (i != n_txq) { | |
834 | /* Retry with less tx queues */ | |
835 | n_txq = i; | |
836 | continue; | |
837 | } | |
838 | ||
839 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 840 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
24e78f93 | 841 | dev->socket_id, NULL, dev->mp); |
b98d7669 | 842 | if (diag) { |
1dfebee9 | 843 | VLOG_INFO("Interface %s unable to setup rxq(%d): %s", |
b98d7669 DDP |
844 | dev->up.name, i, rte_strerror(-diag)); |
845 | break; | |
846 | } | |
847 | } | |
848 | ||
849 | if (i != n_rxq) { | |
850 | /* Retry with less rx queues */ | |
851 | n_rxq = i; | |
852 | continue; | |
853 | } | |
854 | ||
855 | dev->up.n_rxq = n_rxq; | |
81acebda | 856 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
857 | |
858 | return 0; | |
859 | } | |
860 | ||
861 | return diag; | |
862 | } | |
863 | ||
9fd39370 SC |
864 | static void |
865 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
866 | { | |
867 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
fa9f4eeb | 868 | VLOG_WARN("Failed to enable flow control on device "DPDK_PORT_ID_FMT, |
bb37956a | 869 | dev->port_id); |
9fd39370 SC |
870 | } |
871 | } | |
b98d7669 | 872 | |
8a9562d2 | 873 | static int |
c2adb102 IM |
874 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
875 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
876 | { |
877 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 878 | struct rte_eth_dev_info info; |
8a9562d2 PS |
879 | struct ether_addr eth_addr; |
880 | int diag; | |
b98d7669 | 881 | int n_rxq, n_txq; |
d4f5282c KT |
882 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | |
883 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
884 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
8a9562d2 | 885 | |
a0cb2d66 | 886 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 887 | |
d4f5282c KT |
888 | if ((info.rx_offload_capa & rx_chksm_offload_capa) != |
889 | rx_chksm_offload_capa) { | |
fa9f4eeb IM |
890 | VLOG_WARN("Rx checksum offload is not supported on port " |
891 | DPDK_PORT_ID_FMT, dev->port_id); | |
d4f5282c KT |
892 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; |
893 | } else { | |
894 | dev->hw_ol_features |= NETDEV_RX_CHECKSUM_OFFLOAD; | |
895 | } | |
896 | ||
b98d7669 DDP |
897 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
898 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
899 | ||
f8b64a61 | 900 | diag = dpdk_eth_dev_port_config(dev, n_rxq, n_txq); |
8a9562d2 | 901 | if (diag) { |
f8b64a61 RM |
902 | VLOG_ERR("Interface %s(rxq:%d txq:%d lsc interrupt mode:%s) " |
903 | "configure error: %s", | |
904 | dev->up.name, n_rxq, n_txq, | |
905 | dev->lsc_interrupt_mode ? "true" : "false", | |
906 | rte_strerror(-diag)); | |
95fb793a | 907 | return -diag; |
8a9562d2 PS |
908 | } |
909 | ||
8a9562d2 PS |
910 | diag = rte_eth_dev_start(dev->port_id); |
911 | if (diag) { | |
b98d7669 DDP |
912 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
913 | rte_strerror(-diag)); | |
95fb793a | 914 | return -diag; |
8a9562d2 | 915 | } |
606f6650 | 916 | dev->started = true; |
8a9562d2 PS |
917 | |
918 | rte_eth_promiscuous_enable(dev->port_id); | |
919 | rte_eth_allmulticast_enable(dev->port_id); | |
920 | ||
921 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
922 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
fa9f4eeb IM |
923 | VLOG_INFO_RL(&rl, "Port "DPDK_PORT_ID_FMT": "ETH_ADDR_FMT, |
924 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); | |
8a9562d2 | 925 | |
ca92d173 | 926 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
927 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
928 | ||
24e78f93 | 929 | mbp_priv = rte_mempool_get_priv(dev->mp); |
8a9562d2 PS |
930 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; |
931 | ||
9fd39370 SC |
932 | /* Get the Flow control configuration for DPDK-ETH */ |
933 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
934 | if (diag) { | |
fa9f4eeb IM |
935 | VLOG_DBG("cannot get flow control parameters on port "DPDK_PORT_ID_FMT |
936 | ", err=%d", dev->port_id, diag); | |
9fd39370 SC |
937 | } |
938 | ||
8a9562d2 PS |
939 | return 0; |
940 | } | |
941 | ||
942 | static struct netdev_dpdk * | |
943 | netdev_dpdk_cast(const struct netdev *netdev) | |
944 | { | |
945 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
946 | } | |
947 | ||
948 | static struct netdev * | |
949 | netdev_dpdk_alloc(void) | |
950 | { | |
bab69409 AC |
951 | struct netdev_dpdk *dev; |
952 | ||
65e19e70 DDP |
953 | dev = dpdk_rte_mzalloc(sizeof *dev); |
954 | if (dev) { | |
955 | return &dev->up; | |
bab69409 | 956 | } |
65e19e70 | 957 | |
bab69409 | 958 | return NULL; |
8a9562d2 PS |
959 | } |
960 | ||
eff23640 DDP |
961 | static struct dpdk_tx_queue * |
962 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 963 | { |
eff23640 | 964 | struct dpdk_tx_queue *txqs; |
bd5131ba | 965 | unsigned i; |
5a034064 | 966 | |
eff23640 DDP |
967 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
968 | if (txqs) { | |
969 | for (i = 0; i < n_txqs; i++) { | |
970 | /* Initialize map for vhost devices. */ | |
971 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
972 | rte_spinlock_init(&txqs[i].tx_lock); | |
973 | } | |
5a034064 | 974 | } |
eff23640 DDP |
975 | |
976 | return txqs; | |
5a034064 AW |
977 | } |
978 | ||
8a9562d2 | 979 | static int |
bb37956a | 980 | common_construct(struct netdev *netdev, dpdk_port_t port_no, |
1ce30dfd | 981 | enum dpdk_dev_type type, int socket_id) |
5a034064 | 982 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 983 | { |
d46285a2 | 984 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 985 | |
d46285a2 | 986 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 987 | |
d46285a2 | 988 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 989 | |
1b7a04e0 AW |
990 | /* If the 'sid' is negative, it means that the kernel fails |
991 | * to obtain the pci numa info. In that situation, always | |
992 | * use 'SOCKET0'. */ | |
1ce30dfd | 993 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 994 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
995 | dev->port_id = port_no; |
996 | dev->type = type; | |
997 | dev->flags = 0; | |
7f381c2e | 998 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 999 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
f8b64a61 | 1000 | dev->requested_lsc_interrupt_mode = 0; |
0a0f39df CL |
1001 | ovsrcu_index_init(&dev->vid, -1); |
1002 | dev->vhost_reconfigured = false; | |
5dcde09c | 1003 | dev->attached = false; |
8a9562d2 | 1004 | |
78bd47cf | 1005 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 1006 | |
9509913a IS |
1007 | ovsrcu_init(&dev->ingress_policer, NULL); |
1008 | dev->policer_rate = 0; | |
1009 | dev->policer_burst = 0; | |
1010 | ||
7f381c2e DDP |
1011 | netdev->n_rxq = 0; |
1012 | netdev->n_txq = 0; | |
1013 | dev->requested_n_rxq = NR_QUEUE; | |
1014 | dev->requested_n_txq = NR_QUEUE; | |
1015 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
1016 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 1017 | |
9fd39370 SC |
1018 | /* Initialize the flow control to NULL */ |
1019 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
1020 | |
1021 | /* Initilize the hardware offload flags to 0 */ | |
1022 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
1023 | |
1024 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
1025 | ||
d46285a2 | 1026 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 1027 | |
7f381c2e DDP |
1028 | netdev_request_reconfigure(netdev); |
1029 | ||
971f4b39 MW |
1030 | dev->rte_xstats_names = NULL; |
1031 | dev->rte_xstats_names_size = 0; | |
1032 | ||
1033 | dev->rte_xstats_ids = NULL; | |
1034 | dev->rte_xstats_ids_size = 0; | |
1035 | ||
1ce30dfd | 1036 | return 0; |
95fb793a | 1037 | } |
1038 | ||
b83a2df1 MV |
1039 | /* dev_name must be the prefix followed by a positive decimal number. |
1040 | * (no leading + or - signs are allowed) */ | |
95fb793a | 1041 | static int |
1042 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
1043 | unsigned int *port_no) | |
1044 | { | |
1045 | const char *cport; | |
1046 | ||
1047 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
1048 | return ENODEV; | |
1049 | } | |
1050 | ||
1051 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
1052 | |
1053 | if (str_to_uint(cport, 10, port_no)) { | |
1054 | return 0; | |
1055 | } else { | |
1056 | return ENODEV; | |
1057 | } | |
95fb793a | 1058 | } |
1059 | ||
1ce30dfd DDP |
1060 | static int |
1061 | vhost_common_construct(struct netdev *netdev) | |
1062 | OVS_REQUIRES(dpdk_mutex) | |
1063 | { | |
1064 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
1065 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1066 | ||
1067 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); | |
1068 | if (!dev->tx_q) { | |
1069 | return ENOMEM; | |
1070 | } | |
1071 | ||
bb37956a IM |
1072 | return common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1073 | DPDK_DEV_VHOST, socket_id); | |
1ce30dfd DDP |
1074 | } |
1075 | ||
7d1ced01 | 1076 | static int |
53f50d24 | 1077 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 1078 | { |
d46285a2 DDP |
1079 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1080 | const char *name = netdev->name; | |
7d1ced01 | 1081 | int err; |
a0cb2d66 | 1082 | |
1af27e8a DDP |
1083 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
1084 | * the file system. '/' or '\' would traverse directories, so they're not | |
1085 | * acceptable in 'name'. */ | |
1086 | if (strchr(name, '/') || strchr(name, '\\')) { | |
1087 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
1088 | "A valid name must not include '/' or '\\'", | |
1089 | name); | |
1090 | return EINVAL; | |
1091 | } | |
1092 | ||
7d1ced01 CL |
1093 | ovs_mutex_lock(&dpdk_mutex); |
1094 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 1095 | * the socket is to be created, then register the socket. |
7d1ced01 | 1096 | */ |
2d24d165 | 1097 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 1098 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 1099 | |
2d24d165 CL |
1100 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
1101 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
1102 | if (err) { |
1103 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 1104 | dev->vhost_id); |
f3e7ec25 | 1105 | goto out; |
e5c0f5a4 | 1106 | } else { |
2d24d165 CL |
1107 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
1108 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
1109 | dev->vhost_id, name); | |
1110 | } | |
f3e7ec25 MW |
1111 | |
1112 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
1113 | &virtio_net_device_ops); | |
1114 | if (err) { | |
1115 | VLOG_ERR("rte_vhost_driver_callback_register failed for vhost user " | |
1116 | "port: %s\n", name); | |
1117 | goto out; | |
1118 | } | |
1119 | ||
1120 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
1121 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
1122 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
1123 | | 1ULL << VIRTIO_NET_F_CSUM); | |
1124 | if (err) { | |
1125 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
1126 | "port: %s\n", name); | |
1127 | goto out; | |
1128 | } | |
1129 | ||
1130 | err = rte_vhost_driver_start(dev->vhost_id); | |
1131 | if (err) { | |
1132 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
1133 | "port: %s\n", name); | |
1134 | goto out; | |
1135 | } | |
1136 | ||
1ce30dfd | 1137 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1138 | if (err) { |
1139 | VLOG_ERR("vhost_common_construct failed for vhost user " | |
1140 | "port: %s\n", name); | |
1141 | } | |
2d24d165 | 1142 | |
f3e7ec25 | 1143 | out: |
2d24d165 | 1144 | ovs_mutex_unlock(&dpdk_mutex); |
28ca969e AC |
1145 | VLOG_WARN_ONCE("dpdkvhostuser ports are considered deprecated; " |
1146 | "please migrate to dpdkvhostuserclient ports."); | |
2d24d165 CL |
1147 | return err; |
1148 | } | |
1149 | ||
1150 | static int | |
1151 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
1152 | { | |
1153 | int err; | |
1154 | ||
2d24d165 | 1155 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 1156 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1157 | if (err) { |
1158 | VLOG_ERR("vhost_common_construct failed for vhost user client" | |
1159 | "port: %s\n", netdev->name); | |
1160 | } | |
7d1ced01 | 1161 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
1162 | return err; |
1163 | } | |
1164 | ||
95fb793a | 1165 | static int |
1166 | netdev_dpdk_construct(struct netdev *netdev) | |
1167 | { | |
95fb793a | 1168 | int err; |
1169 | ||
95fb793a | 1170 | ovs_mutex_lock(&dpdk_mutex); |
bb37956a IM |
1171 | err = common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1172 | DPDK_DEV_ETH, SOCKET0); | |
8a9562d2 PS |
1173 | ovs_mutex_unlock(&dpdk_mutex); |
1174 | return err; | |
1175 | } | |
1176 | ||
1ce30dfd DDP |
1177 | static void |
1178 | common_destruct(struct netdev_dpdk *dev) | |
1179 | OVS_REQUIRES(dpdk_mutex) | |
1180 | OVS_EXCLUDED(dev->mutex) | |
1181 | { | |
1182 | rte_free(dev->tx_q); | |
91fccdad | 1183 | dpdk_mp_release(dev->mp); |
1ce30dfd DDP |
1184 | |
1185 | ovs_list_remove(&dev->list_node); | |
1186 | free(ovsrcu_get_protected(struct ingress_policer *, | |
1187 | &dev->ingress_policer)); | |
1188 | ovs_mutex_destroy(&dev->mutex); | |
1189 | } | |
1190 | ||
8a9562d2 | 1191 | static void |
d46285a2 | 1192 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 1193 | { |
d46285a2 | 1194 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5dcde09c | 1195 | char devname[RTE_ETH_NAME_MAX_LEN]; |
8a9562d2 | 1196 | |
8d38823b | 1197 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1198 | |
8a9562d2 | 1199 | rte_eth_dev_stop(dev->port_id); |
606f6650 | 1200 | dev->started = false; |
5dcde09c IM |
1201 | |
1202 | if (dev->attached) { | |
1203 | rte_eth_dev_close(dev->port_id); | |
1204 | if (rte_eth_dev_detach(dev->port_id, devname) < 0) { | |
1205 | VLOG_ERR("Device '%s' can not be detached", dev->devargs); | |
1206 | } else { | |
0ee821c2 | 1207 | VLOG_INFO("Device '%s' has been detached", devname); |
5dcde09c IM |
1208 | } |
1209 | } | |
1210 | ||
ac1a9bb9 | 1211 | netdev_dpdk_clear_xstats(dev); |
55e075e6 | 1212 | free(dev->devargs); |
1ce30dfd | 1213 | common_destruct(dev); |
8d38823b | 1214 | |
8a9562d2 | 1215 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1216 | } |
8a9562d2 | 1217 | |
3f891bbe DDP |
1218 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1219 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1220 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1221 | static int | |
c1ff66ac CL |
1222 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1223 | char *vhost_id) | |
3f891bbe DDP |
1224 | OVS_EXCLUDED(dpdk_mutex) |
1225 | OVS_EXCLUDED(dev->mutex) | |
1226 | { | |
c1ff66ac | 1227 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1228 | } |
1229 | ||
58397e6c | 1230 | static void |
d46285a2 | 1231 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1232 | { |
d46285a2 | 1233 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1234 | char *vhost_id; |
58397e6c | 1235 | |
8d38823b | 1236 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1237 | |
c62da695 | 1238 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1239 | if (netdev_dpdk_get_vid(dev) >= 0 |
1240 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1241 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1242 | netdev->name); |
58be5c0e MK |
1243 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1244 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1245 | } |
1246 | ||
2d24d165 | 1247 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1248 | |
1ce30dfd DDP |
1249 | common_destruct(dev); |
1250 | ||
58397e6c | 1251 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1252 | |
569c26da | 1253 | if (!vhost_id[0]) { |
821b8664 IM |
1254 | goto out; |
1255 | } | |
1256 | ||
c1ff66ac | 1257 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1258 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1259 | netdev->name, vhost_id); | |
c1ff66ac CL |
1260 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1261 | /* OVS server mode - remove this socket from list for deletion */ | |
1262 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1263 | } |
821b8664 | 1264 | out: |
c1ff66ac | 1265 | free(vhost_id); |
8a9562d2 PS |
1266 | } |
1267 | ||
1268 | static void | |
d46285a2 | 1269 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1270 | { |
d46285a2 | 1271 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1272 | |
d46285a2 | 1273 | rte_free(dev); |
8a9562d2 PS |
1274 | } |
1275 | ||
971f4b39 | 1276 | static void |
ac1a9bb9 | 1277 | netdev_dpdk_clear_xstats(struct netdev_dpdk *dev) |
971f4b39 MW |
1278 | { |
1279 | /* If statistics are already allocated, we have to | |
1280 | * reconfigure, as port_id could have been changed. */ | |
1281 | if (dev->rte_xstats_names) { | |
1282 | free(dev->rte_xstats_names); | |
1283 | dev->rte_xstats_names = NULL; | |
1284 | dev->rte_xstats_names_size = 0; | |
1285 | } | |
1286 | if (dev->rte_xstats_ids) { | |
1287 | free(dev->rte_xstats_ids); | |
1288 | dev->rte_xstats_ids = NULL; | |
1289 | dev->rte_xstats_ids_size = 0; | |
1290 | } | |
1291 | } | |
1292 | ||
1293 | static const char* | |
1294 | netdev_dpdk_get_xstat_name(struct netdev_dpdk *dev, uint64_t id) | |
1295 | { | |
1296 | if (id >= dev->rte_xstats_names_size) { | |
1297 | return "UNKNOWN"; | |
1298 | } | |
1299 | return dev->rte_xstats_names[id].name; | |
1300 | } | |
1301 | ||
1302 | static bool | |
1303 | netdev_dpdk_configure_xstats(struct netdev_dpdk *dev) | |
1304 | OVS_REQUIRES(dev->mutex) | |
1305 | { | |
1306 | int rte_xstats_len; | |
1307 | bool ret; | |
1308 | struct rte_eth_xstat *rte_xstats; | |
1309 | uint64_t id; | |
1310 | int xstats_no; | |
1311 | const char *name; | |
1312 | ||
1313 | /* Retrieving all XSTATS names. If something will go wrong | |
1314 | * or amount of counters will be equal 0, rte_xstats_names | |
1315 | * buffer will be marked as NULL, and any further xstats | |
1316 | * query won't be performed (e.g. during netdev_dpdk_get_stats | |
1317 | * execution). */ | |
1318 | ||
1319 | ret = false; | |
1320 | rte_xstats = NULL; | |
1321 | ||
1322 | if (dev->rte_xstats_names == NULL || dev->rte_xstats_ids == NULL) { | |
1323 | dev->rte_xstats_names_size = | |
1324 | rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
1325 | ||
1326 | if (dev->rte_xstats_names_size < 0) { | |
fa9f4eeb IM |
1327 | VLOG_WARN("Cannot get XSTATS for port: "DPDK_PORT_ID_FMT, |
1328 | dev->port_id); | |
971f4b39 MW |
1329 | dev->rte_xstats_names_size = 0; |
1330 | } else { | |
1331 | /* Reserve memory for xstats names and values */ | |
1332 | dev->rte_xstats_names = xcalloc(dev->rte_xstats_names_size, | |
1333 | sizeof *dev->rte_xstats_names); | |
1334 | ||
1335 | if (dev->rte_xstats_names) { | |
1336 | /* Retreive xstats names */ | |
1337 | rte_xstats_len = | |
1338 | rte_eth_xstats_get_names(dev->port_id, | |
1339 | dev->rte_xstats_names, | |
1340 | dev->rte_xstats_names_size); | |
1341 | ||
1342 | if (rte_xstats_len < 0) { | |
fa9f4eeb IM |
1343 | VLOG_WARN("Cannot get XSTATS names for port: " |
1344 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 MW |
1345 | goto out; |
1346 | } else if (rte_xstats_len != dev->rte_xstats_names_size) { | |
fa9f4eeb IM |
1347 | VLOG_WARN("XSTATS size doesn't match for port: " |
1348 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 MW |
1349 | goto out; |
1350 | } | |
1351 | ||
1352 | dev->rte_xstats_ids = xcalloc(dev->rte_xstats_names_size, | |
1353 | sizeof(uint64_t)); | |
1354 | ||
1355 | /* We have to calculate number of counters */ | |
1356 | rte_xstats = xmalloc(rte_xstats_len * sizeof *rte_xstats); | |
1357 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
1358 | ||
1359 | /* Retreive xstats values */ | |
1360 | if (rte_eth_xstats_get(dev->port_id, rte_xstats, | |
1361 | rte_xstats_len) > 0) { | |
1362 | dev->rte_xstats_ids_size = 0; | |
1363 | xstats_no = 0; | |
1364 | for (uint32_t i = 0; i < rte_xstats_len; i++) { | |
1365 | id = rte_xstats[i].id; | |
1366 | name = netdev_dpdk_get_xstat_name(dev, id); | |
1367 | /* We need to filter out everything except | |
1368 | * dropped, error and management counters */ | |
1369 | if (string_ends_with(name, "_errors") || | |
1370 | strstr(name, "_management_") || | |
1371 | string_ends_with(name, "_dropped")) { | |
1372 | ||
1373 | dev->rte_xstats_ids[xstats_no] = id; | |
1374 | xstats_no++; | |
1375 | } | |
1376 | } | |
1377 | dev->rte_xstats_ids_size = xstats_no; | |
1378 | ret = true; | |
1379 | } else { | |
fa9f4eeb IM |
1380 | VLOG_WARN("Can't get XSTATS IDs for port: " |
1381 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 | 1382 | } |
34eb0863 IM |
1383 | |
1384 | free(rte_xstats); | |
971f4b39 MW |
1385 | } |
1386 | } | |
1387 | } else { | |
1388 | /* Already configured */ | |
1389 | ret = true; | |
1390 | } | |
1391 | ||
1392 | out: | |
1393 | if (!ret) { | |
1394 | netdev_dpdk_clear_xstats(dev); | |
1395 | } | |
1396 | return ret; | |
1397 | } | |
1398 | ||
8a9562d2 | 1399 | static int |
a14b8947 | 1400 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1401 | { |
a14b8947 | 1402 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1403 | |
1404 | ovs_mutex_lock(&dev->mutex); | |
1405 | ||
050c60bf | 1406 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1407 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1408 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1409 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1410 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1411 | |
1412 | if (dev->type == DPDK_DEV_ETH) { | |
1413 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1414 | dev->requested_rxq_size); | |
1415 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1416 | dev->rxq_size); | |
1417 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1418 | dev->requested_txq_size); | |
1419 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1420 | dev->txq_size); | |
1a2bb118 SC |
1421 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1422 | smap_add(args, "rx_csum_offload", "true"); | |
8155ab7e KT |
1423 | } else { |
1424 | smap_add(args, "rx_csum_offload", "false"); | |
1a2bb118 | 1425 | } |
f8b64a61 RM |
1426 | smap_add(args, "lsc_interrupt_mode", |
1427 | dev->lsc_interrupt_mode ? "true" : "false"); | |
451f26fd | 1428 | } |
8a9562d2 PS |
1429 | ovs_mutex_unlock(&dev->mutex); |
1430 | ||
1431 | return 0; | |
1432 | } | |
1433 | ||
55e075e6 | 1434 | static struct netdev_dpdk * |
bb37956a | 1435 | netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id) |
55e075e6 CL |
1436 | OVS_REQUIRES(dpdk_mutex) |
1437 | { | |
1438 | struct netdev_dpdk *dev; | |
1439 | ||
1440 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1441 | if (dev->port_id == port_id) { | |
1442 | return dev; | |
1443 | } | |
1444 | } | |
1445 | ||
1446 | return NULL; | |
1447 | } | |
1448 | ||
5e758818 YL |
1449 | static dpdk_port_t |
1450 | netdev_dpdk_get_port_by_mac(const char *mac_str) | |
1451 | { | |
1452 | dpdk_port_t port_id; | |
1453 | struct eth_addr mac, port_mac; | |
1454 | ||
1455 | if (!eth_addr_from_string(mac_str, &mac)) { | |
1456 | VLOG_ERR("invalid mac: %s", mac_str); | |
1457 | return DPDK_ETH_PORT_ID_INVALID; | |
1458 | } | |
1459 | ||
1460 | RTE_ETH_FOREACH_DEV (port_id) { | |
1461 | struct ether_addr ea; | |
1462 | ||
1463 | rte_eth_macaddr_get(port_id, &ea); | |
1464 | memcpy(port_mac.ea, ea.addr_bytes, ETH_ADDR_LEN); | |
1465 | if (eth_addr_equals(mac, port_mac)) { | |
1466 | return port_id; | |
1467 | } | |
1468 | } | |
1469 | ||
1470 | return DPDK_ETH_PORT_ID_INVALID; | |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * Normally, a PCI id is enough for identifying a specific DPDK port. | |
1475 | * However, for some NICs having multiple ports sharing the same PCI | |
1476 | * id, using PCI id won't work then. | |
1477 | * | |
1478 | * To fix that, here one more method is introduced: "class=eth,mac=$MAC". | |
1479 | * | |
1480 | * Note that the compatibility is fully kept: user can still use the | |
1481 | * PCI id for adding ports (when it's enough for them). | |
1482 | */ | |
bb37956a | 1483 | static dpdk_port_t |
5dcde09c IM |
1484 | netdev_dpdk_process_devargs(struct netdev_dpdk *dev, |
1485 | const char *devargs, char **errp) | |
55e075e6 | 1486 | { |
5e758818 | 1487 | char *name; |
bb37956a | 1488 | dpdk_port_t new_port_id = DPDK_ETH_PORT_ID_INVALID; |
55e075e6 | 1489 | |
5e758818 YL |
1490 | if (strncmp(devargs, "class=eth,mac=", 14) == 0) { |
1491 | new_port_id = netdev_dpdk_get_port_by_mac(&devargs[14]); | |
1492 | } else { | |
1493 | name = xmemdup0(devargs, strcspn(devargs, ",")); | |
1494 | if (rte_eth_dev_get_port_by_name(name, &new_port_id) | |
1495 | || !rte_eth_dev_is_valid_port(new_port_id)) { | |
1496 | /* Device not found in DPDK, attempt to attach it */ | |
1497 | if (!rte_eth_dev_attach(devargs, &new_port_id)) { | |
1498 | /* Attach successful */ | |
1499 | dev->attached = true; | |
1500 | VLOG_INFO("Device '%s' attached to DPDK", devargs); | |
1501 | } else { | |
1502 | /* Attach unsuccessful */ | |
1503 | new_port_id = DPDK_ETH_PORT_ID_INVALID; | |
1504 | } | |
55e075e6 | 1505 | } |
5e758818 YL |
1506 | free(name); |
1507 | } | |
1508 | ||
1509 | if (new_port_id == DPDK_ETH_PORT_ID_INVALID) { | |
1510 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", devargs); | |
55e075e6 CL |
1511 | } |
1512 | ||
1513 | return new_port_id; | |
1514 | } | |
1515 | ||
c3d062a7 CL |
1516 | static void |
1517 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1518 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1519 | { |
050c60bf | 1520 | int new_n_rxq; |
a14b8947 | 1521 | |
2a21e757 | 1522 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1523 | if (new_n_rxq != dev->requested_n_rxq) { |
1524 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1525 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1526 | } |
c3d062a7 CL |
1527 | } |
1528 | ||
b685696b CL |
1529 | static void |
1530 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1531 | const char *flag, int default_size, int *new_size) | |
1532 | { | |
1533 | int queue_size = smap_get_int(args, flag, default_size); | |
1534 | ||
1535 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1536 | || !is_pow2(queue_size)) { | |
1537 | queue_size = default_size; | |
1538 | } | |
1539 | ||
1540 | if (queue_size != *new_size) { | |
1541 | *new_size = queue_size; | |
1542 | netdev_request_reconfigure(netdev); | |
1543 | } | |
1544 | } | |
1545 | ||
c3d062a7 | 1546 | static int |
9fff138e DDP |
1547 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1548 | char **errp) | |
c3d062a7 CL |
1549 | { |
1550 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
f8b64a61 | 1551 | bool rx_fc_en, tx_fc_en, autoneg, lsc_interrupt_mode; |
b614c894 IM |
1552 | enum rte_eth_fc_mode fc_mode; |
1553 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1554 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1555 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1556 | }; | |
55e075e6 CL |
1557 | const char *new_devargs; |
1558 | int err = 0; | |
c3d062a7 | 1559 | |
55e075e6 | 1560 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1561 | ovs_mutex_lock(&dev->mutex); |
1562 | ||
1563 | dpdk_set_rxq_config(dev, args); | |
1564 | ||
b685696b CL |
1565 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1566 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1567 | &dev->requested_rxq_size); | |
1568 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1569 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1570 | &dev->requested_txq_size); | |
1571 | ||
55e075e6 CL |
1572 | new_devargs = smap_get(args, "dpdk-devargs"); |
1573 | ||
1574 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1575 | /* The user requested a new device. If we return error, the caller | |
1576 | * will delete this netdev and try to recreate it. */ | |
1577 | err = EAGAIN; | |
1578 | goto out; | |
1579 | } | |
1580 | ||
1581 | /* dpdk-devargs is required for device configuration */ | |
1582 | if (new_devargs && new_devargs[0]) { | |
1583 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1584 | * is valid */ | |
1585 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1586 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
bb37956a IM |
1587 | dpdk_port_t new_port_id = netdev_dpdk_process_devargs(dev, |
1588 | new_devargs, | |
1589 | errp); | |
55e075e6 CL |
1590 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1591 | err = EINVAL; | |
1592 | } else if (new_port_id == dev->port_id) { | |
1593 | /* Already configured, do not reconfigure again */ | |
1594 | err = 0; | |
1595 | } else { | |
1596 | struct netdev_dpdk *dup_dev; | |
bb37956a | 1597 | |
55e075e6 CL |
1598 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); |
1599 | if (dup_dev) { | |
9fff138e DDP |
1600 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
1601 | "which is already in use by '%s'", | |
1602 | netdev_get_name(netdev), new_devargs, | |
1603 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1604 | err = EADDRINUSE; |
1605 | } else { | |
bd4e172b | 1606 | int sid = rte_eth_dev_socket_id(new_port_id); |
bb37956a | 1607 | |
bd4e172b | 1608 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; |
55e075e6 CL |
1609 | dev->devargs = xstrdup(new_devargs); |
1610 | dev->port_id = new_port_id; | |
1611 | netdev_request_reconfigure(&dev->up); | |
971f4b39 | 1612 | netdev_dpdk_clear_xstats(dev); |
55e075e6 CL |
1613 | err = 0; |
1614 | } | |
1615 | } | |
1616 | } | |
1617 | } else { | |
9fff138e DDP |
1618 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1619 | "The old 'dpdk<port_id>' names are not supported", | |
1620 | netdev_get_name(netdev)); | |
55e075e6 CL |
1621 | err = EINVAL; |
1622 | } | |
1623 | ||
1624 | if (err) { | |
1625 | goto out; | |
1626 | } | |
1627 | ||
f8b64a61 RM |
1628 | lsc_interrupt_mode = smap_get_bool(args, "dpdk-lsc-interrupt", false); |
1629 | if (dev->requested_lsc_interrupt_mode != lsc_interrupt_mode) { | |
1630 | dev->requested_lsc_interrupt_mode = lsc_interrupt_mode; | |
1631 | netdev_request_reconfigure(netdev); | |
1632 | } | |
1633 | ||
c3d062a7 CL |
1634 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1635 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1636 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1637 | |
b614c894 IM |
1638 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1639 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1640 | dev->fc_conf.mode = fc_mode; | |
1641 | dev->fc_conf.autoneg = autoneg; | |
1642 | dpdk_eth_flow_ctrl_setup(dev); | |
1643 | } | |
9fd39370 | 1644 | |
55e075e6 | 1645 | out: |
c3d062a7 | 1646 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1647 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1648 | |
55e075e6 | 1649 | return err; |
c3d062a7 CL |
1650 | } |
1651 | ||
1652 | static int | |
9fff138e DDP |
1653 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1654 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1655 | { |
1656 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1657 | ||
1658 | ovs_mutex_lock(&dev->mutex); | |
1659 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1660 | ovs_mutex_unlock(&dev->mutex); |
1661 | ||
1662 | return 0; | |
1663 | } | |
1664 | ||
c1ff66ac | 1665 | static int |
2d24d165 | 1666 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1667 | const struct smap *args, |
1668 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1669 | { |
1670 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1671 | const char *path; | |
1672 | ||
6881885a | 1673 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1674 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1675 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1676 | if (path && strcmp(path, dev->vhost_id)) { |
1677 | strcpy(dev->vhost_id, path); | |
10087cba CL |
1678 | /* check zero copy configuration */ |
1679 | if (smap_get_bool(args, "dq-zero-copy", false)) { | |
1680 | dev->vhost_driver_flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
1681 | } else { | |
1682 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
1683 | } | |
c1ff66ac CL |
1684 | netdev_request_reconfigure(netdev); |
1685 | } | |
1686 | } | |
6881885a | 1687 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1688 | |
1689 | return 0; | |
1690 | } | |
1691 | ||
7dec44fe | 1692 | static int |
d46285a2 | 1693 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1694 | { |
d46285a2 | 1695 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1696 | |
d46285a2 | 1697 | return dev->socket_id; |
7dec44fe AW |
1698 | } |
1699 | ||
050c60bf | 1700 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1701 | static int |
050c60bf | 1702 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1703 | { |
d46285a2 | 1704 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1705 | |
d46285a2 | 1706 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1707 | |
050c60bf DDP |
1708 | if (dev->requested_n_txq == n_txq) { |
1709 | goto out; | |
4573fbd3 FL |
1710 | } |
1711 | ||
050c60bf DDP |
1712 | dev->requested_n_txq = n_txq; |
1713 | netdev_request_reconfigure(netdev); | |
58397e6c | 1714 | |
050c60bf | 1715 | out: |
d46285a2 | 1716 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1717 | return 0; |
58397e6c KT |
1718 | } |
1719 | ||
8a9562d2 PS |
1720 | static struct netdev_rxq * |
1721 | netdev_dpdk_rxq_alloc(void) | |
1722 | { | |
1723 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1724 | ||
eff23640 DDP |
1725 | if (rx) { |
1726 | return &rx->up; | |
1727 | } | |
1728 | ||
1729 | return NULL; | |
8a9562d2 PS |
1730 | } |
1731 | ||
1732 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1733 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1734 | { |
d46285a2 | 1735 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1736 | } |
1737 | ||
1738 | static int | |
d46285a2 | 1739 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1740 | { |
d46285a2 DDP |
1741 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1742 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1743 | |
d46285a2 DDP |
1744 | ovs_mutex_lock(&dev->mutex); |
1745 | rx->port_id = dev->port_id; | |
1746 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1747 | |
1748 | return 0; | |
1749 | } | |
1750 | ||
1751 | static void | |
d46285a2 | 1752 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1753 | { |
1754 | } | |
1755 | ||
1756 | static void | |
d46285a2 | 1757 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1758 | { |
d46285a2 | 1759 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1760 | |
1761 | rte_free(rx); | |
1762 | } | |
1763 | ||
819f13bd DDP |
1764 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1765 | * 'pkts', even in case of failure. | |
1766 | * | |
1767 | * Returns the number of packets that weren't transmitted. */ | |
1768 | static inline int | |
b59cc14e | 1769 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 1770 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 1771 | { |
1304f1f8 DDP |
1772 | uint32_t nb_tx = 0; |
1773 | ||
b59cc14e | 1774 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1775 | uint32_t ret; |
1776 | ||
b59cc14e | 1777 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1778 | if (!ret) { |
1779 | break; | |
1780 | } | |
1781 | ||
1782 | nb_tx += ret; | |
1783 | } | |
8a9562d2 | 1784 | |
b59cc14e | 1785 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 1786 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
1787 | * packet could come from a different mempool) */ |
1788 | int i; | |
1789 | ||
b59cc14e IM |
1790 | for (i = nb_tx; i < cnt; i++) { |
1791 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1792 | } |
8a9562d2 | 1793 | } |
819f13bd DDP |
1794 | |
1795 | return cnt - nb_tx; | |
8a9562d2 PS |
1796 | } |
1797 | ||
f3926f29 IS |
1798 | static inline bool |
1799 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1800 | struct rte_mbuf *pkt, uint64_t time) | |
1801 | { | |
1802 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1803 | ||
1804 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1805 | e_RTE_METER_GREEN; | |
1806 | } | |
1807 | ||
1808 | static int | |
1809 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
3e90f7d7 | 1810 | struct rte_mbuf **pkts, int pkt_cnt, |
7d7ded7a | 1811 | bool should_steal) |
f3926f29 IS |
1812 | { |
1813 | int i = 0; | |
1814 | int cnt = 0; | |
1815 | struct rte_mbuf *pkt = NULL; | |
1816 | uint64_t current_time = rte_rdtsc(); | |
1817 | ||
1818 | for (i = 0; i < pkt_cnt; i++) { | |
1819 | pkt = pkts[i]; | |
1820 | /* Handle current packet */ | |
1821 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1822 | if (cnt != i) { | |
1823 | pkts[cnt] = pkt; | |
1824 | } | |
1825 | cnt++; | |
1826 | } else { | |
7d7ded7a | 1827 | if (should_steal) { |
3e90f7d7 GZ |
1828 | rte_pktmbuf_free(pkt); |
1829 | } | |
f3926f29 IS |
1830 | } |
1831 | } | |
1832 | ||
1833 | return cnt; | |
1834 | } | |
1835 | ||
9509913a IS |
1836 | static int |
1837 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
7d7ded7a | 1838 | int pkt_cnt, bool should_steal) |
9509913a IS |
1839 | { |
1840 | int cnt = 0; | |
1841 | ||
1842 | rte_spinlock_lock(&policer->policer_lock); | |
3e90f7d7 | 1843 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, |
7d7ded7a | 1844 | pkt_cnt, should_steal); |
9509913a IS |
1845 | rte_spinlock_unlock(&policer->policer_lock); |
1846 | ||
1847 | return cnt; | |
1848 | } | |
1849 | ||
58397e6c | 1850 | static bool |
0a0f39df | 1851 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1852 | { |
0a0f39df | 1853 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1854 | } |
1855 | ||
d6e3feb5 | 1856 | static inline void |
1857 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1858 | unsigned int packet_size) | |
1859 | { | |
1860 | /* Hard-coded search for the size bucket. */ | |
1861 | if (packet_size < 256) { | |
1862 | if (packet_size >= 128) { | |
1863 | stats->rx_128_to_255_packets++; | |
1864 | } else if (packet_size <= 64) { | |
1865 | stats->rx_1_to_64_packets++; | |
1866 | } else { | |
1867 | stats->rx_65_to_127_packets++; | |
1868 | } | |
1869 | } else { | |
1870 | if (packet_size >= 1523) { | |
1871 | stats->rx_1523_to_max_packets++; | |
1872 | } else if (packet_size >= 1024) { | |
1873 | stats->rx_1024_to_1522_packets++; | |
1874 | } else if (packet_size < 512) { | |
1875 | stats->rx_256_to_511_packets++; | |
1876 | } else { | |
1877 | stats->rx_512_to_1023_packets++; | |
1878 | } | |
1879 | } | |
1880 | } | |
1881 | ||
9e3ddd45 TP |
1882 | static inline void |
1883 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1884 | struct dp_packet **packets, int count, |
1885 | int dropped) | |
9e3ddd45 TP |
1886 | { |
1887 | int i; | |
d6e3feb5 | 1888 | unsigned int packet_size; |
9e3ddd45 TP |
1889 | struct dp_packet *packet; |
1890 | ||
1891 | stats->rx_packets += count; | |
9509913a | 1892 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1893 | for (i = 0; i < count; i++) { |
1894 | packet = packets[i]; | |
d6e3feb5 | 1895 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1896 | |
d6e3feb5 | 1897 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1898 | /* This only protects the following multicast counting from |
1899 | * too short packets, but it does not stop the packet from | |
1900 | * further processing. */ | |
1901 | stats->rx_errors++; | |
1902 | stats->rx_length_errors++; | |
1903 | continue; | |
1904 | } | |
1905 | ||
d6e3feb5 | 1906 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1907 | ||
9e3ddd45 TP |
1908 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1909 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1910 | stats->multicast++; | |
1911 | } | |
1912 | ||
d6e3feb5 | 1913 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1914 | } |
1915 | } | |
1916 | ||
58397e6c KT |
1917 | /* |
1918 | * The receive path for the vhost port is the TX path out from guest. | |
1919 | */ | |
1920 | static int | |
d46285a2 | 1921 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
8492adc2 | 1922 | struct dp_packet_batch *batch, int *qfill) |
58397e6c | 1923 | { |
d46285a2 | 1924 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
9509913a | 1925 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1926 | uint16_t nb_rx = 0; |
9509913a | 1927 | uint16_t dropped = 0; |
8492adc2 | 1928 | int qid = rxq->queue_id * VIRTIO_QNUM + VIRTIO_TXQ; |
daf22bf7 | 1929 | int vid = netdev_dpdk_get_vid(dev); |
58397e6c | 1930 | |
daf22bf7 | 1931 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured |
e543851d | 1932 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1933 | return EAGAIN; |
1934 | } | |
1935 | ||
8492adc2 | 1936 | nb_rx = rte_vhost_dequeue_burst(vid, qid, dev->mp, |
64839cf4 | 1937 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1938 | NETDEV_MAX_BURST); |
58397e6c KT |
1939 | if (!nb_rx) { |
1940 | return EAGAIN; | |
1941 | } | |
1942 | ||
8492adc2 JS |
1943 | if (qfill) { |
1944 | if (nb_rx == NETDEV_MAX_BURST) { | |
1945 | /* The DPDK API returns a uint32_t which often has invalid bits in | |
1946 | * the upper 16-bits. Need to restrict the value to uint16_t. */ | |
1947 | *qfill = rte_vhost_rx_queue_count(vid, qid) & UINT16_MAX; | |
1948 | } else { | |
1949 | *qfill = 0; | |
1950 | } | |
1951 | } | |
1952 | ||
9509913a IS |
1953 | if (policer) { |
1954 | dropped = nb_rx; | |
64839cf4 WT |
1955 | nb_rx = ingress_policer_run(policer, |
1956 | (struct rte_mbuf **) batch->packets, | |
3e90f7d7 | 1957 | nb_rx, true); |
9509913a IS |
1958 | dropped -= nb_rx; |
1959 | } | |
1960 | ||
d46285a2 | 1961 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1962 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1963 | nb_rx, dropped); | |
d46285a2 | 1964 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1965 | |
75fb9148 ZB |
1966 | batch->count = nb_rx; |
1967 | dp_packet_batch_init_packet_fields(batch); | |
1968 | ||
58397e6c KT |
1969 | return 0; |
1970 | } | |
1971 | ||
8a9562d2 | 1972 | static int |
8492adc2 JS |
1973 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch, |
1974 | int *qfill) | |
8a9562d2 | 1975 | { |
d46285a2 DDP |
1976 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1977 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1978 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1979 | int nb_rx; |
9509913a | 1980 | int dropped = 0; |
8a9562d2 | 1981 | |
3b1fb077 DDP |
1982 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1983 | return EAGAIN; | |
1984 | } | |
1985 | ||
d46285a2 | 1986 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1987 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1988 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1989 | if (!nb_rx) { |
1990 | return EAGAIN; | |
1991 | } | |
1992 | ||
9509913a IS |
1993 | if (policer) { |
1994 | dropped = nb_rx; | |
64839cf4 | 1995 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 1996 | (struct rte_mbuf **) batch->packets, |
3e90f7d7 | 1997 | nb_rx, true); |
9509913a IS |
1998 | dropped -= nb_rx; |
1999 | } | |
2000 | ||
2001 | /* Update stats to reflect dropped packets */ | |
2002 | if (OVS_UNLIKELY(dropped)) { | |
2003 | rte_spinlock_lock(&dev->stats_lock); | |
2004 | dev->stats.rx_dropped += dropped; | |
2005 | rte_spinlock_unlock(&dev->stats_lock); | |
2006 | } | |
2007 | ||
64839cf4 | 2008 | batch->count = nb_rx; |
75fb9148 | 2009 | dp_packet_batch_init_packet_fields(batch); |
8a9562d2 | 2010 | |
8492adc2 JS |
2011 | if (qfill) { |
2012 | if (nb_rx == NETDEV_MAX_BURST) { | |
2013 | *qfill = rte_eth_rx_queue_count(rx->port_id, rxq->queue_id); | |
2014 | } else { | |
2015 | *qfill = 0; | |
2016 | } | |
2017 | } | |
2018 | ||
8a9562d2 PS |
2019 | return 0; |
2020 | } | |
2021 | ||
0bf765f7 | 2022 | static inline int |
78bd47cf | 2023 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
7d7ded7a | 2024 | int cnt, bool should_steal) |
0bf765f7 | 2025 | { |
78bd47cf | 2026 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 2027 | |
78bd47cf DDP |
2028 | if (qos_conf) { |
2029 | rte_spinlock_lock(&qos_conf->lock); | |
7d7ded7a | 2030 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt, should_steal); |
78bd47cf | 2031 | rte_spinlock_unlock(&qos_conf->lock); |
0bf765f7 IS |
2032 | } |
2033 | ||
2034 | return cnt; | |
2035 | } | |
2036 | ||
c6ec9d17 IM |
2037 | static int |
2038 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
2039 | int pkt_cnt) | |
2040 | { | |
2041 | int i = 0; | |
2042 | int cnt = 0; | |
2043 | struct rte_mbuf *pkt; | |
2044 | ||
2045 | for (i = 0; i < pkt_cnt; i++) { | |
2046 | pkt = pkts[i]; | |
2047 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
2048 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
2049 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
2050 | rte_pktmbuf_free(pkt); | |
2051 | continue; | |
2052 | } | |
2053 | ||
2054 | if (OVS_UNLIKELY(i != cnt)) { | |
2055 | pkts[cnt] = pkt; | |
2056 | } | |
2057 | cnt++; | |
2058 | } | |
2059 | ||
2060 | return cnt; | |
2061 | } | |
2062 | ||
9e3ddd45 TP |
2063 | static inline void |
2064 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
2065 | struct dp_packet **packets, | |
2066 | int attempted, | |
2067 | int dropped) | |
2068 | { | |
2069 | int i; | |
2070 | int sent = attempted - dropped; | |
2071 | ||
2072 | stats->tx_packets += sent; | |
2073 | stats->tx_dropped += dropped; | |
2074 | ||
2075 | for (i = 0; i < sent; i++) { | |
2076 | stats->tx_bytes += dp_packet_size(packets[i]); | |
2077 | } | |
2078 | } | |
2079 | ||
58397e6c | 2080 | static void |
4573fbd3 | 2081 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 2082 | struct dp_packet **pkts, int cnt) |
58397e6c | 2083 | { |
d46285a2 | 2084 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
2085 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
2086 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 2087 | unsigned int dropped = 0; |
dd52de45 | 2088 | int i, retries = 0; |
daf22bf7 | 2089 | int vid = netdev_dpdk_get_vid(dev); |
58397e6c | 2090 | |
81acebda | 2091 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 2092 | |
daf22bf7 | 2093 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured || qid < 0 |
e543851d | 2094 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
2095 | rte_spinlock_lock(&dev->stats_lock); |
2096 | dev->stats.tx_dropped+= cnt; | |
2097 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 2098 | goto out; |
58397e6c KT |
2099 | } |
2100 | ||
d46285a2 | 2101 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 2102 | |
c6ec9d17 | 2103 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 2104 | /* Check has QoS has been configured for the netdev */ |
3e90f7d7 | 2105 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt, true); |
c6ec9d17 | 2106 | dropped = total_pkts - cnt; |
0bf765f7 | 2107 | |
95e9881f | 2108 | do { |
4573fbd3 | 2109 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
2110 | unsigned int tx_pkts; |
2111 | ||
daf22bf7 | 2112 | tx_pkts = rte_vhost_enqueue_burst(vid, vhost_qid, cur_pkts, cnt); |
95e9881f KT |
2113 | if (OVS_LIKELY(tx_pkts)) { |
2114 | /* Packets have been sent.*/ | |
2115 | cnt -= tx_pkts; | |
31871ee3 | 2116 | /* Prepare for possible retry.*/ |
95e9881f KT |
2117 | cur_pkts = &cur_pkts[tx_pkts]; |
2118 | } else { | |
31871ee3 KT |
2119 | /* No packets sent - do not retry.*/ |
2120 | break; | |
95e9881f | 2121 | } |
c6ec9d17 | 2122 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 2123 | |
d46285a2 | 2124 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 2125 | |
d46285a2 | 2126 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 2127 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 2128 | cnt + dropped); |
d46285a2 | 2129 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
2130 | |
2131 | out: | |
c6ec9d17 | 2132 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 2133 | dp_packet_delete(pkts[i]); |
58397e6c KT |
2134 | } |
2135 | } | |
2136 | ||
8a9562d2 PS |
2137 | /* Tx function. Transmit packets indefinitely */ |
2138 | static void | |
64839cf4 | 2139 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 2140 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 2141 | { |
8a14bd7b | 2142 | const size_t batch_cnt = dp_packet_batch_size(batch); |
bce01e3a | 2143 | #if !defined(__CHECKER__) && !defined(_WIN32) |
8a14bd7b | 2144 | const size_t PKT_ARRAY_SIZE = batch_cnt; |
bce01e3a EJ |
2145 | #else |
2146 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 2147 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 2148 | #endif |
8a9562d2 | 2149 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 2150 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
8a14bd7b | 2151 | uint32_t cnt = batch_cnt; |
3e90f7d7 GZ |
2152 | uint32_t dropped = 0; |
2153 | ||
2154 | if (dev->type != DPDK_DEV_VHOST) { | |
2155 | /* Check if QoS has been configured for this netdev. */ | |
2156 | cnt = netdev_dpdk_qos_run(dev, (struct rte_mbuf **) batch->packets, | |
8a14bd7b BB |
2157 | batch_cnt, false); |
2158 | dropped += batch_cnt - cnt; | |
3e90f7d7 | 2159 | } |
8a9562d2 | 2160 | |
3e90f7d7 GZ |
2161 | uint32_t txcnt = 0; |
2162 | ||
2163 | for (uint32_t i = 0; i < cnt; i++) { | |
8a14bd7b BB |
2164 | struct dp_packet *packet = batch->packets[i]; |
2165 | uint32_t size = dp_packet_size(packet); | |
95fb793a | 2166 | |
f98d7864 | 2167 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
3e90f7d7 GZ |
2168 | VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d", |
2169 | size, dev->max_packet_len); | |
f4fd623c | 2170 | |
175cf4de | 2171 | dropped++; |
f4fd623c DDP |
2172 | continue; |
2173 | } | |
8a9562d2 | 2174 | |
24e78f93 | 2175 | pkts[txcnt] = rte_pktmbuf_alloc(dev->mp); |
8a14bd7b | 2176 | if (OVS_UNLIKELY(!pkts[txcnt])) { |
3e90f7d7 | 2177 | dropped += cnt - i; |
175cf4de | 2178 | break; |
f4fd623c DDP |
2179 | } |
2180 | ||
2181 | /* We have to do a copy for now */ | |
3e90f7d7 | 2182 | memcpy(rte_pktmbuf_mtod(pkts[txcnt], void *), |
8a14bd7b BB |
2183 | dp_packet_data(packet), size); |
2184 | dp_packet_set_size((struct dp_packet *)pkts[txcnt], size); | |
f4fd623c | 2185 | |
3e90f7d7 | 2186 | txcnt++; |
f4fd623c | 2187 | } |
8a9562d2 | 2188 | |
3e90f7d7 GZ |
2189 | if (OVS_LIKELY(txcnt)) { |
2190 | if (dev->type == DPDK_DEV_VHOST) { | |
2191 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, | |
2192 | txcnt); | |
2193 | } else { | |
2194 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, txcnt); | |
2195 | } | |
58397e6c | 2196 | } |
db73f716 | 2197 | |
0bf765f7 IS |
2198 | if (OVS_UNLIKELY(dropped)) { |
2199 | rte_spinlock_lock(&dev->stats_lock); | |
2200 | dev->stats.tx_dropped += dropped; | |
2201 | rte_spinlock_unlock(&dev->stats_lock); | |
2202 | } | |
8a9562d2 PS |
2203 | } |
2204 | ||
58397e6c | 2205 | static int |
64839cf4 WT |
2206 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
2207 | struct dp_packet_batch *batch, | |
b30896c9 | 2208 | bool concurrent_txq OVS_UNUSED) |
58397e6c | 2209 | { |
58397e6c | 2210 | |
b30896c9 | 2211 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 | 2212 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2213 | dp_packet_delete_batch(batch, true); |
58397e6c | 2214 | } else { |
dd52de45 | 2215 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
2216 | } |
2217 | return 0; | |
2218 | } | |
2219 | ||
7251515e DV |
2220 | static inline void |
2221 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
b30896c9 | 2222 | struct dp_packet_batch *batch, |
324c8374 | 2223 | bool concurrent_txq) |
8a9562d2 | 2224 | { |
3b1fb077 | 2225 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
b30896c9 | 2226 | dp_packet_delete_batch(batch, true); |
3b1fb077 DDP |
2227 | return; |
2228 | } | |
2229 | ||
324c8374 | 2230 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 2231 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
2232 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
2233 | } | |
2234 | ||
b30896c9 | 2235 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
2236 | struct netdev *netdev = &dev->up; |
2237 | ||
64839cf4 | 2238 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2239 | dp_packet_delete_batch(batch, true); |
8a9562d2 | 2240 | } else { |
fd57eeba BB |
2241 | int tx_cnt, dropped; |
2242 | int batch_cnt = dp_packet_batch_size(batch); | |
2391135c | 2243 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 2244 | |
fd57eeba BB |
2245 | tx_cnt = netdev_dpdk_filter_packet_len(dev, pkts, batch_cnt); |
2246 | tx_cnt = netdev_dpdk_qos_run(dev, pkts, tx_cnt, true); | |
2247 | dropped = batch_cnt - tx_cnt; | |
1b99bb05 | 2248 | |
fd57eeba | 2249 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, tx_cnt); |
8a9562d2 | 2250 | |
f4fd623c | 2251 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 2252 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 2253 | dev->stats.tx_dropped += dropped; |
45d947c4 | 2254 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 2255 | } |
8a9562d2 | 2256 | } |
a0cb2d66 | 2257 | |
324c8374 | 2258 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
2259 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
2260 | } | |
7251515e DV |
2261 | } |
2262 | ||
2263 | static int | |
2264 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
b30896c9 | 2265 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e DV |
2266 | { |
2267 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 2268 | |
b30896c9 | 2269 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e | 2270 | return 0; |
8a9562d2 PS |
2271 | } |
2272 | ||
2273 | static int | |
74ff3298 | 2274 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
2275 | { |
2276 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2277 | ||
2278 | ovs_mutex_lock(&dev->mutex); | |
2279 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 2280 | dev->hwaddr = mac; |
045c0d1a | 2281 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
2282 | } |
2283 | ovs_mutex_unlock(&dev->mutex); | |
2284 | ||
2285 | return 0; | |
2286 | } | |
2287 | ||
2288 | static int | |
74ff3298 | 2289 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
2290 | { |
2291 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2292 | ||
2293 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 2294 | *mac = dev->hwaddr; |
8a9562d2 PS |
2295 | ovs_mutex_unlock(&dev->mutex); |
2296 | ||
2297 | return 0; | |
2298 | } | |
2299 | ||
2300 | static int | |
2301 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
2302 | { | |
2303 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2304 | ||
2305 | ovs_mutex_lock(&dev->mutex); | |
2306 | *mtup = dev->mtu; | |
2307 | ovs_mutex_unlock(&dev->mutex); | |
2308 | ||
2309 | return 0; | |
2310 | } | |
2311 | ||
0072e931 MK |
2312 | static int |
2313 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
2314 | { | |
2315 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2316 | ||
f6f50552 IS |
2317 | /* XXX: Ensure that the overall frame length of the requested MTU does not |
2318 | * surpass the NETDEV_DPDK_MAX_PKT_LEN. DPDK device drivers differ in how | |
2319 | * the L2 frame length is calculated for a given MTU when | |
2320 | * rte_eth_dev_set_mtu(mtu) is called e.g. i40e driver includes 2 x vlan | |
2321 | * headers, the em driver includes 1 x vlan header, the ixgbe driver does | |
2322 | * not include vlan headers. As such we should use | |
2323 | * MTU_TO_MAX_FRAME_LEN(mtu) which includes an additional 2 x vlan headers | |
2324 | * (8 bytes) for comparison. This avoids a failure later with | |
2325 | * rte_eth_dev_set_mtu(). This approach should be used until DPDK provides | |
2326 | * a method to retrieve the upper bound MTU for a given device. | |
2327 | */ | |
2328 | if (MTU_TO_MAX_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
0072e931 MK |
2329 | || mtu < ETHER_MIN_MTU) { |
2330 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
2331 | return EINVAL; | |
2332 | } | |
2333 | ||
2334 | ovs_mutex_lock(&dev->mutex); | |
2335 | if (dev->requested_mtu != mtu) { | |
2336 | dev->requested_mtu = mtu; | |
2337 | netdev_request_reconfigure(netdev); | |
2338 | } | |
2339 | ovs_mutex_unlock(&dev->mutex); | |
2340 | ||
2341 | return 0; | |
2342 | } | |
2343 | ||
8a9562d2 | 2344 | static int |
d46285a2 | 2345 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 2346 | |
58397e6c KT |
2347 | static int |
2348 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
2349 | struct netdev_stats *stats) | |
2350 | { | |
2351 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2352 | ||
2353 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 2354 | |
45d947c4 | 2355 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c | 2356 | /* Supported Stats */ |
50986e78 | 2357 | stats->rx_packets = dev->stats.rx_packets; |
2358 | stats->tx_packets = dev->stats.tx_packets; | |
9509913a | 2359 | stats->rx_dropped = dev->stats.rx_dropped; |
50986e78 | 2360 | stats->tx_dropped = dev->stats.tx_dropped; |
9e3ddd45 TP |
2361 | stats->multicast = dev->stats.multicast; |
2362 | stats->rx_bytes = dev->stats.rx_bytes; | |
2363 | stats->tx_bytes = dev->stats.tx_bytes; | |
2364 | stats->rx_errors = dev->stats.rx_errors; | |
2365 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 2366 | |
2367 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
2368 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
2369 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
2370 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
2371 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
2372 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
2373 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
2374 | ||
45d947c4 | 2375 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 2376 | |
58397e6c KT |
2377 | ovs_mutex_unlock(&dev->mutex); |
2378 | ||
2379 | return 0; | |
2380 | } | |
2381 | ||
d6e3feb5 | 2382 | static void |
2383 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
2384 | const struct rte_eth_xstat *xstats, |
2385 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 2386 | const unsigned int size) |
2387 | { | |
d6e3feb5 | 2388 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 2389 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2390 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2391 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2392 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2393 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2394 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2395 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2396 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2397 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2398 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2399 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2400 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2401 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2402 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 2403 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2404 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2405 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2406 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2407 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2408 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2409 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2410 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2411 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2412 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2413 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2414 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2415 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2416 | stats->tx_1523_to_max_packets = xstats[i].value; |
d57f777f PS |
2417 | } else if (strcmp(XSTAT_RX_MULTICAST_PACKETS, names[i].name) == 0) { |
2418 | stats->multicast = xstats[i].value; | |
0a0f39df | 2419 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2420 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 2421 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2422 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2423 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2424 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2425 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2426 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2427 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2428 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2429 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2430 | stats->rx_jabber_errors = xstats[i].value; |
2431 | } | |
2432 | } | |
2433 | } | |
2434 | ||
8a9562d2 PS |
2435 | static int |
2436 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2437 | { | |
2438 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2439 | struct rte_eth_stats rte_stats; | |
2440 | bool gg; | |
2441 | ||
2442 | netdev_dpdk_get_carrier(netdev, &gg); | |
2443 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2444 | |
0a0f39df CL |
2445 | struct rte_eth_xstat *rte_xstats = NULL; |
2446 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2447 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2448 | |
2449 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
fa9f4eeb IM |
2450 | VLOG_ERR("Can't get ETH statistics for port: "DPDK_PORT_ID_FMT, |
2451 | dev->port_id); | |
f9256822 | 2452 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2453 | return EPROTO; |
2454 | } | |
2455 | ||
0a0f39df CL |
2456 | /* Get length of statistics */ |
2457 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2458 | if (rte_xstats_len < 0) { | |
fa9f4eeb IM |
2459 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
2460 | dev->port_id); | |
0a0f39df CL |
2461 | goto out; |
2462 | } | |
2463 | /* Reserve memory for xstats names and values */ | |
2464 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2465 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2466 | ||
2467 | /* Retreive xstats names */ | |
2468 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2469 | rte_xstats_names, | |
2470 | rte_xstats_len); | |
2471 | if (rte_xstats_new_len != rte_xstats_len) { | |
fa9f4eeb IM |
2472 | VLOG_WARN("Cannot get XSTATS names for port: "DPDK_PORT_ID_FMT, |
2473 | dev->port_id); | |
0a0f39df CL |
2474 | goto out; |
2475 | } | |
2476 | /* Retreive xstats values */ | |
2477 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2478 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2479 | rte_xstats_len); | |
2480 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2481 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2482 | rte_xstats_len); | |
d6e3feb5 | 2483 | } else { |
fa9f4eeb IM |
2484 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
2485 | dev->port_id); | |
d6e3feb5 | 2486 | } |
8a9562d2 | 2487 | |
0a0f39df CL |
2488 | out: |
2489 | free(rte_xstats); | |
2490 | free(rte_xstats_names); | |
2491 | ||
2f9dd77f PS |
2492 | stats->rx_packets = rte_stats.ipackets; |
2493 | stats->tx_packets = rte_stats.opackets; | |
2494 | stats->rx_bytes = rte_stats.ibytes; | |
2495 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2496 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2497 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2498 | |
45d947c4 | 2499 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2500 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2501 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2502 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2503 | |
2504 | /* These are the available DPDK counters for packets not received due to | |
2505 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2506 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2507 | stats->rx_missed_errors = rte_stats.imissed; |
2508 | ||
8a9562d2 PS |
2509 | ovs_mutex_unlock(&dev->mutex); |
2510 | ||
2511 | return 0; | |
2512 | } | |
2513 | ||
971f4b39 MW |
2514 | static int |
2515 | netdev_dpdk_get_custom_stats(const struct netdev *netdev, | |
2516 | struct netdev_custom_stats *custom_stats) | |
2517 | { | |
2518 | ||
2519 | uint32_t i; | |
2520 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2521 | int rte_xstats_ret; | |
2522 | ||
2523 | ovs_mutex_lock(&dev->mutex); | |
2524 | ||
2525 | if (netdev_dpdk_configure_xstats(dev)) { | |
2526 | uint64_t *values = xcalloc(dev->rte_xstats_ids_size, | |
2527 | sizeof(uint64_t)); | |
2528 | ||
2529 | rte_xstats_ret = | |
2530 | rte_eth_xstats_get_by_id(dev->port_id, dev->rte_xstats_ids, | |
2531 | values, dev->rte_xstats_ids_size); | |
2532 | ||
2533 | if (rte_xstats_ret > 0 && | |
2534 | rte_xstats_ret <= dev->rte_xstats_ids_size) { | |
2535 | ||
2536 | custom_stats->size = rte_xstats_ret; | |
2537 | custom_stats->counters = | |
2538 | (struct netdev_custom_counter *) xcalloc(rte_xstats_ret, | |
2539 | sizeof(struct netdev_custom_counter)); | |
2540 | ||
2541 | for (i = 0; i < rte_xstats_ret; i++) { | |
2542 | ovs_strlcpy(custom_stats->counters[i].name, | |
2543 | netdev_dpdk_get_xstat_name(dev, | |
2544 | dev->rte_xstats_ids[i]), | |
2545 | NETDEV_CUSTOM_STATS_NAME_SIZE); | |
2546 | custom_stats->counters[i].value = values[i]; | |
2547 | } | |
2548 | } else { | |
fa9f4eeb | 2549 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
971f4b39 MW |
2550 | dev->port_id); |
2551 | custom_stats->counters = NULL; | |
2552 | custom_stats->size = 0; | |
2553 | /* Let's clear statistics cache, so it will be | |
2554 | * reconfigured */ | |
2555 | netdev_dpdk_clear_xstats(dev); | |
2556 | } | |
526259f2 IM |
2557 | |
2558 | free(values); | |
971f4b39 MW |
2559 | } |
2560 | ||
2561 | ovs_mutex_unlock(&dev->mutex); | |
2562 | ||
2563 | return 0; | |
2564 | } | |
2565 | ||
8a9562d2 | 2566 | static int |
d46285a2 | 2567 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2568 | enum netdev_features *current, |
ca3d4f55 BX |
2569 | enum netdev_features *advertised, |
2570 | enum netdev_features *supported, | |
2571 | enum netdev_features *peer) | |
8a9562d2 | 2572 | { |
d46285a2 | 2573 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2574 | struct rte_eth_link link; |
2575 | ||
2576 | ovs_mutex_lock(&dev->mutex); | |
2577 | link = dev->link; | |
2578 | ovs_mutex_unlock(&dev->mutex); | |
2579 | ||
362ca396 | 2580 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2581 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
2582 | *current = NETDEV_F_10MB_HD; |
2583 | } | |
362ca396 | 2584 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2585 | *current = NETDEV_F_100MB_HD; |
2586 | } | |
362ca396 | 2587 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2588 | *current = NETDEV_F_1GB_HD; |
2589 | } | |
2590 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 2591 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
2592 | *current = NETDEV_F_10MB_FD; |
2593 | } | |
362ca396 | 2594 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2595 | *current = NETDEV_F_100MB_FD; |
2596 | } | |
362ca396 | 2597 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2598 | *current = NETDEV_F_1GB_FD; |
2599 | } | |
362ca396 | 2600 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
2601 | *current = NETDEV_F_10GB_FD; |
2602 | } | |
2603 | } | |
2604 | ||
362ca396 | 2605 | if (link.link_autoneg) { |
2606 | *current |= NETDEV_F_AUTONEG; | |
2607 | } | |
2608 | ||
ca3d4f55 BX |
2609 | *advertised = *supported = *peer = 0; |
2610 | ||
8a9562d2 PS |
2611 | return 0; |
2612 | } | |
2613 | ||
9509913a IS |
2614 | static struct ingress_policer * |
2615 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2616 | { | |
2617 | struct ingress_policer *policer = NULL; | |
2618 | uint64_t rate_bytes; | |
2619 | uint64_t burst_bytes; | |
2620 | int err = 0; | |
2621 | ||
2622 | policer = xmalloc(sizeof *policer); | |
2623 | rte_spinlock_init(&policer->policer_lock); | |
2624 | ||
2625 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
602c8668 LR |
2626 | rate_bytes = rate * 1000ULL / 8; |
2627 | burst_bytes = burst * 1000ULL / 8; | |
9509913a IS |
2628 | |
2629 | policer->app_srtcm_params.cir = rate_bytes; | |
2630 | policer->app_srtcm_params.cbs = burst_bytes; | |
2631 | policer->app_srtcm_params.ebs = 0; | |
2632 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2633 | &policer->app_srtcm_params); | |
58be5c0e | 2634 | if (err) { |
9509913a | 2635 | VLOG_ERR("Could not create rte meter for ingress policer"); |
4c47ddde | 2636 | free(policer); |
9509913a IS |
2637 | return NULL; |
2638 | } | |
2639 | ||
2640 | return policer; | |
2641 | } | |
2642 | ||
2643 | static int | |
2644 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2645 | uint32_t policer_burst) | |
2646 | { | |
2647 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2648 | struct ingress_policer *policer; | |
2649 | ||
2650 | /* Force to 0 if no rate specified, | |
2651 | * default to 8000 kbits if burst is 0, | |
2652 | * else stick with user-specified value. | |
2653 | */ | |
2654 | policer_burst = (!policer_rate ? 0 | |
2655 | : !policer_burst ? 8000 | |
2656 | : policer_burst); | |
2657 | ||
2658 | ovs_mutex_lock(&dev->mutex); | |
2659 | ||
2660 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2661 | &dev->ingress_policer); | |
2662 | ||
2663 | if (dev->policer_rate == policer_rate && | |
2664 | dev->policer_burst == policer_burst) { | |
2665 | /* Assume that settings haven't changed since we last set them. */ | |
2666 | ovs_mutex_unlock(&dev->mutex); | |
2667 | return 0; | |
2668 | } | |
2669 | ||
2670 | /* Destroy any existing ingress policer for the device if one exists */ | |
2671 | if (policer) { | |
2672 | ovsrcu_postpone(free, policer); | |
2673 | } | |
2674 | ||
2675 | if (policer_rate != 0) { | |
2676 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2677 | } else { | |
2678 | policer = NULL; | |
2679 | } | |
2680 | ovsrcu_set(&dev->ingress_policer, policer); | |
2681 | dev->policer_rate = policer_rate; | |
2682 | dev->policer_burst = policer_burst; | |
2683 | ovs_mutex_unlock(&dev->mutex); | |
2684 | ||
2685 | return 0; | |
2686 | } | |
2687 | ||
8a9562d2 PS |
2688 | static int |
2689 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2690 | { | |
2691 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
2692 | |
2693 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
2694 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
2695 | * postive integer to meet RFC 2863 recommendations. | |
2696 | */ | |
2697 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
2698 | ovs_mutex_unlock(&dev->mutex); |
2699 | ||
2700 | return ifindex; | |
2701 | } | |
2702 | ||
2703 | static int | |
d46285a2 | 2704 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2705 | { |
d46285a2 | 2706 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2707 | |
2708 | ovs_mutex_lock(&dev->mutex); | |
2709 | check_link_status(dev); | |
2710 | *carrier = dev->link.link_status; | |
58397e6c KT |
2711 | |
2712 | ovs_mutex_unlock(&dev->mutex); | |
2713 | ||
2714 | return 0; | |
2715 | } | |
2716 | ||
2717 | static int | |
d46285a2 | 2718 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2719 | { |
d46285a2 | 2720 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2721 | |
2722 | ovs_mutex_lock(&dev->mutex); | |
2723 | ||
0a0f39df | 2724 | if (is_vhost_running(dev)) { |
58397e6c KT |
2725 | *carrier = 1; |
2726 | } else { | |
2727 | *carrier = 0; | |
2728 | } | |
2729 | ||
8a9562d2 PS |
2730 | ovs_mutex_unlock(&dev->mutex); |
2731 | ||
2732 | return 0; | |
2733 | } | |
2734 | ||
2735 | static long long int | |
d46285a2 | 2736 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2737 | { |
d46285a2 | 2738 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2739 | long long int carrier_resets; |
2740 | ||
2741 | ovs_mutex_lock(&dev->mutex); | |
2742 | carrier_resets = dev->link_reset_cnt; | |
2743 | ovs_mutex_unlock(&dev->mutex); | |
2744 | ||
2745 | return carrier_resets; | |
2746 | } | |
2747 | ||
2748 | static int | |
d46285a2 | 2749 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2750 | long long int interval OVS_UNUSED) |
2751 | { | |
ee32150e | 2752 | return EOPNOTSUPP; |
8a9562d2 PS |
2753 | } |
2754 | ||
2755 | static int | |
2756 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2757 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2758 | enum netdev_flags *old_flagsp) |
2759 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 2760 | { |
8a9562d2 PS |
2761 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
2762 | return EINVAL; | |
2763 | } | |
2764 | ||
2765 | *old_flagsp = dev->flags; | |
2766 | dev->flags |= on; | |
2767 | dev->flags &= ~off; | |
2768 | ||
2769 | if (dev->flags == *old_flagsp) { | |
2770 | return 0; | |
2771 | } | |
2772 | ||
58397e6c | 2773 | if (dev->type == DPDK_DEV_ETH) { |
58397e6c KT |
2774 | if (dev->flags & NETDEV_PROMISC) { |
2775 | rte_eth_promiscuous_enable(dev->port_id); | |
2776 | } | |
8a9562d2 | 2777 | |
314fb5ad | 2778 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
2779 | } else { |
2780 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2781 | * running then change netdev's change_seq to trigger link state | |
2782 | * update. */ | |
e543851d ZB |
2783 | |
2784 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2785 | && is_vhost_running(dev)) { |
e543851d ZB |
2786 | netdev_change_seq_changed(&dev->up); |
2787 | ||
2788 | /* Clear statistics if device is getting up. */ | |
2789 | if (NETDEV_UP & on) { | |
2790 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 2791 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
2792 | rte_spinlock_unlock(&dev->stats_lock); |
2793 | } | |
2794 | } | |
8a9562d2 PS |
2795 | } |
2796 | ||
2797 | return 0; | |
2798 | } | |
2799 | ||
2800 | static int | |
d46285a2 | 2801 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2802 | enum netdev_flags off, enum netdev_flags on, |
2803 | enum netdev_flags *old_flagsp) | |
2804 | { | |
d46285a2 | 2805 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2806 | int error; |
2807 | ||
d46285a2 DDP |
2808 | ovs_mutex_lock(&dev->mutex); |
2809 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2810 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2811 | |
2812 | return error; | |
2813 | } | |
2814 | ||
b2e8b12f FL |
2815 | static int |
2816 | netdev_dpdk_vhost_user_get_status(const struct netdev *netdev, | |
2817 | struct smap *args) | |
2818 | { | |
2819 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2820 | ||
2821 | ovs_mutex_lock(&dev->mutex); | |
2822 | ||
2823 | bool client_mode = dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT; | |
2824 | smap_add_format(args, "mode", "%s", client_mode ? "client" : "server"); | |
2825 | ||
2826 | int vid = netdev_dpdk_get_vid(dev); | |
2827 | if (vid < 0) { | |
2828 | smap_add_format(args, "status", "disconnected"); | |
2829 | ovs_mutex_unlock(&dev->mutex); | |
2830 | return 0; | |
2831 | } else { | |
2832 | smap_add_format(args, "status", "connected"); | |
2833 | } | |
2834 | ||
2835 | char socket_name[PATH_MAX]; | |
2836 | if (!rte_vhost_get_ifname(vid, socket_name, PATH_MAX)) { | |
2837 | smap_add_format(args, "socket", "%s", socket_name); | |
2838 | } | |
2839 | ||
2840 | uint64_t features; | |
2841 | if (!rte_vhost_get_negotiated_features(vid, &features)) { | |
2842 | smap_add_format(args, "features", "0x%016"PRIx64, features); | |
2843 | } | |
2844 | ||
2845 | uint16_t mtu; | |
2846 | if (!rte_vhost_get_mtu(vid, &mtu)) { | |
2847 | smap_add_format(args, "mtu", "%d", mtu); | |
2848 | } | |
2849 | ||
2850 | int numa = rte_vhost_get_numa_node(vid); | |
2851 | if (numa >= 0) { | |
2852 | smap_add_format(args, "numa", "%d", numa); | |
2853 | } | |
2854 | ||
2855 | uint16_t vring_num = rte_vhost_get_vring_num(vid); | |
2856 | if (vring_num) { | |
2857 | smap_add_format(args, "num_of_vrings", "%d", vring_num); | |
2858 | } | |
2859 | ||
2860 | for (int i = 0; i < vring_num; i++) { | |
2861 | struct rte_vhost_vring vring; | |
2862 | char vhost_vring[16]; | |
2863 | ||
2864 | rte_vhost_get_vhost_vring(vid, i, &vring); | |
2865 | snprintf(vhost_vring, 16, "vring_%d_size", i); | |
2866 | smap_add_format(args, vhost_vring, "%d", vring.size); | |
2867 | } | |
2868 | ||
2869 | ovs_mutex_unlock(&dev->mutex); | |
2870 | return 0; | |
2871 | } | |
2872 | ||
8a9562d2 | 2873 | static int |
d46285a2 | 2874 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2875 | { |
d46285a2 | 2876 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2877 | struct rte_eth_dev_info dev_info; |
2878 | ||
7cd1261d | 2879 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2880 | return ENODEV; |
7cd1261d | 2881 | } |
8a9562d2 PS |
2882 | |
2883 | ovs_mutex_lock(&dev->mutex); | |
2884 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2885 | ovs_mutex_unlock(&dev->mutex); | |
2886 | ||
fa9f4eeb | 2887 | smap_add_format(args, "port_no", DPDK_PORT_ID_FMT, dev->port_id); |
58be5c0e MK |
2888 | smap_add_format(args, "numa_id", "%d", |
2889 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
2890 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
2891 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2892 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2893 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2894 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2895 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
2896 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
2897 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
2898 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
2899 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2900 | ||
3eb8d4fa MW |
2901 | /* Querying the DPDK library for iftype may be done in future, pending |
2902 | * support; cf. RFC 3635 Section 3.2.4. */ | |
2903 | enum { IF_TYPE_ETHERNETCSMACD = 6 }; | |
2904 | ||
2905 | smap_add_format(args, "if_type", "%"PRIu32, IF_TYPE_ETHERNETCSMACD); | |
2906 | smap_add_format(args, "if_descr", "%s %s", rte_version(), | |
2907 | dev_info.driver_name); | |
2908 | ||
39c2baa9 | 2909 | if (dev_info.pci_dev) { |
2910 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2911 | dev_info.pci_dev->id.vendor_id); | |
2912 | smap_add_format(args, "pci-device_id", "0x%x", | |
2913 | dev_info.pci_dev->id.device_id); | |
2914 | } | |
8a9562d2 PS |
2915 | |
2916 | return 0; | |
2917 | } | |
2918 | ||
2919 | static void | |
2920 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2921 | OVS_REQUIRES(dev->mutex) | |
2922 | { | |
2923 | enum netdev_flags old_flags; | |
2924 | ||
2925 | if (admin_state) { | |
2926 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2927 | } else { | |
2928 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2929 | } | |
2930 | } | |
2931 | ||
2932 | static void | |
2933 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2934 | const char *argv[], void *aux OVS_UNUSED) | |
2935 | { | |
2936 | bool up; | |
2937 | ||
2938 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2939 | up = true; | |
2940 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2941 | up = false; | |
2942 | } else { | |
2943 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2944 | return; | |
2945 | } | |
2946 | ||
2947 | if (argc > 2) { | |
2948 | struct netdev *netdev = netdev_from_name(argv[1]); | |
3d0d5ab1 | 2949 | |
8a9562d2 | 2950 | if (netdev && is_dpdk_class(netdev->netdev_class)) { |
3d0d5ab1 | 2951 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 2952 | |
3d0d5ab1 IM |
2953 | ovs_mutex_lock(&dev->mutex); |
2954 | netdev_dpdk_set_admin_state__(dev, up); | |
2955 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2956 | |
2957 | netdev_close(netdev); | |
2958 | } else { | |
2959 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2960 | netdev_close(netdev); | |
2961 | return; | |
2962 | } | |
2963 | } else { | |
3d0d5ab1 | 2964 | struct netdev_dpdk *dev; |
8a9562d2 PS |
2965 | |
2966 | ovs_mutex_lock(&dpdk_mutex); | |
3d0d5ab1 IM |
2967 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
2968 | ovs_mutex_lock(&dev->mutex); | |
2969 | netdev_dpdk_set_admin_state__(dev, up); | |
2970 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2971 | } |
2972 | ovs_mutex_unlock(&dpdk_mutex); | |
2973 | } | |
2974 | unixctl_command_reply(conn, "OK"); | |
2975 | } | |
2976 | ||
0ee821c2 DB |
2977 | static void |
2978 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2979 | const char *argv[], void *aux OVS_UNUSED) | |
2980 | { | |
2981 | int ret; | |
2982 | char *response; | |
7ee94cba | 2983 | dpdk_port_t port_id; |
0ee821c2 DB |
2984 | char devname[RTE_ETH_NAME_MAX_LEN]; |
2985 | struct netdev_dpdk *dev; | |
2986 | ||
2987 | ovs_mutex_lock(&dpdk_mutex); | |
2988 | ||
255b7bda | 2989 | if (rte_eth_dev_get_port_by_name(argv[1], &port_id)) { |
0ee821c2 DB |
2990 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); |
2991 | goto error; | |
2992 | } | |
2993 | ||
2994 | dev = netdev_dpdk_lookup_by_port_id(port_id); | |
2995 | if (dev) { | |
2996 | response = xasprintf("Device '%s' is being used by interface '%s'. " | |
2997 | "Remove it before detaching", | |
2998 | argv[1], netdev_get_name(&dev->up)); | |
2999 | goto error; | |
3000 | } | |
3001 | ||
3002 | rte_eth_dev_close(port_id); | |
3003 | ||
3004 | ret = rte_eth_dev_detach(port_id, devname); | |
3005 | if (ret < 0) { | |
3006 | response = xasprintf("Device '%s' can not be detached", argv[1]); | |
3007 | goto error; | |
3008 | } | |
3009 | ||
3010 | response = xasprintf("Device '%s' has been detached", argv[1]); | |
3011 | ||
3012 | ovs_mutex_unlock(&dpdk_mutex); | |
3013 | unixctl_command_reply(conn, response); | |
3014 | free(response); | |
3015 | return; | |
3016 | ||
3017 | error: | |
3018 | ovs_mutex_unlock(&dpdk_mutex); | |
3019 | unixctl_command_reply_error(conn, response); | |
3020 | free(response); | |
3021 | } | |
3022 | ||
be481733 IM |
3023 | static void |
3024 | netdev_dpdk_get_mempool_info(struct unixctl_conn *conn, | |
3025 | int argc, const char *argv[], | |
3026 | void *aux OVS_UNUSED) | |
3027 | { | |
3028 | size_t size; | |
3029 | FILE *stream; | |
3030 | char *response = NULL; | |
3031 | struct netdev *netdev = NULL; | |
3032 | ||
3033 | if (argc == 2) { | |
3034 | netdev = netdev_from_name(argv[1]); | |
3035 | if (!netdev || !is_dpdk_class(netdev->netdev_class)) { | |
3036 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
3037 | goto out; | |
3038 | } | |
3039 | } | |
3040 | ||
3041 | stream = open_memstream(&response, &size); | |
3042 | if (!stream) { | |
3043 | response = xasprintf("Unable to open memstream: %s.", | |
3044 | ovs_strerror(errno)); | |
3045 | unixctl_command_reply_error(conn, response); | |
3046 | goto out; | |
3047 | } | |
3048 | ||
3049 | if (netdev) { | |
3050 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3051 | ||
3052 | ovs_mutex_lock(&dev->mutex); | |
3053 | ovs_mutex_lock(&dpdk_mp_mutex); | |
3054 | ||
3055 | rte_mempool_dump(stream, dev->mp); | |
3056 | ||
3057 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
3058 | ovs_mutex_unlock(&dev->mutex); | |
3059 | } else { | |
3060 | ovs_mutex_lock(&dpdk_mp_mutex); | |
3061 | rte_mempool_list_dump(stream); | |
3062 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
3063 | } | |
3064 | ||
3065 | fclose(stream); | |
3066 | ||
3067 | unixctl_command_reply(conn, response); | |
3068 | out: | |
3069 | free(response); | |
3070 | netdev_close(netdev); | |
3071 | } | |
3072 | ||
58397e6c KT |
3073 | /* |
3074 | * Set virtqueue flags so that we do not receive interrupts. | |
3075 | */ | |
3076 | static void | |
0a0f39df | 3077 | set_irq_status(int vid) |
58397e6c | 3078 | { |
4573fbd3 | 3079 | uint32_t i; |
4573fbd3 | 3080 | |
f3e7ec25 MW |
3081 | for (i = 0; i < rte_vhost_get_vring_num(vid); i++) { |
3082 | rte_vhost_enable_guest_notification(vid, i, 0); | |
4573fbd3 FL |
3083 | } |
3084 | } | |
3085 | ||
585a5bea IM |
3086 | /* |
3087 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 3088 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
3089 | */ |
3090 | static void | |
d46285a2 DDP |
3091 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
3092 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
3093 | { |
3094 | int *enabled_queues, n_enabled = 0; | |
81acebda | 3095 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 3096 | |
eff23640 | 3097 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
3098 | |
3099 | for (i = 0; i < total_txqs; i++) { | |
3100 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 3101 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
3102 | enabled_queues[n_enabled++] = i; |
3103 | } | |
3104 | } | |
3105 | ||
3106 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 3107 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
3108 | n_enabled = 1; |
3109 | } | |
3110 | ||
3111 | k = 0; | |
3112 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
3113 | if (dev->tx_q[i].map != i) { |
3114 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
3115 | k = (k + 1) % n_enabled; |
3116 | } | |
3117 | } | |
3118 | ||
2d24d165 | 3119 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 3120 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 3121 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
3122 | } |
3123 | ||
eff23640 | 3124 | free(enabled_queues); |
585a5bea | 3125 | } |
4573fbd3 | 3126 | |
58397e6c KT |
3127 | /* |
3128 | * A new virtio-net device is added to a vhost port. | |
3129 | */ | |
3130 | static int | |
0a0f39df | 3131 | new_device(int vid) |
58397e6c | 3132 | { |
d46285a2 | 3133 | struct netdev_dpdk *dev; |
58397e6c | 3134 | bool exists = false; |
db8f13b0 | 3135 | int newnode = 0; |
0a0f39df CL |
3136 | char ifname[IF_NAME_SZ]; |
3137 | ||
58be5c0e | 3138 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
3139 | |
3140 | ovs_mutex_lock(&dpdk_mutex); | |
3141 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 3142 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 3143 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 3144 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
f3e7ec25 | 3145 | uint32_t qp_num = rte_vhost_get_vring_num(vid)/VIRTIO_QNUM; |
db8f13b0 CL |
3146 | |
3147 | /* Get NUMA information */ | |
0a0f39df CL |
3148 | newnode = rte_vhost_get_numa_node(vid); |
3149 | if (newnode == -1) { | |
5b9bf9e0 | 3150 | #ifdef VHOST_NUMA |
db8f13b0 | 3151 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 3152 | ifname); |
5b9bf9e0 | 3153 | #endif |
db8f13b0 | 3154 | newnode = dev->socket_id; |
db8f13b0 CL |
3155 | } |
3156 | ||
7f5f2bd0 IM |
3157 | if (dev->requested_n_txq != qp_num |
3158 | || dev->requested_n_rxq != qp_num | |
3159 | || dev->requested_socket_id != newnode) { | |
3160 | dev->requested_socket_id = newnode; | |
3161 | dev->requested_n_rxq = qp_num; | |
3162 | dev->requested_n_txq = qp_num; | |
3163 | netdev_request_reconfigure(&dev->up); | |
3164 | } else { | |
3165 | /* Reconfiguration not required. */ | |
3166 | dev->vhost_reconfigured = true; | |
3167 | } | |
81acebda | 3168 | |
0a0f39df | 3169 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
3170 | exists = true; |
3171 | ||
58397e6c | 3172 | /* Disable notifications. */ |
0a0f39df | 3173 | set_irq_status(vid); |
e543851d | 3174 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 3175 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
3176 | break; |
3177 | } | |
c1ff66ac | 3178 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
3179 | } |
3180 | ovs_mutex_unlock(&dpdk_mutex); | |
3181 | ||
3182 | if (!exists) { | |
0a0f39df | 3183 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
3184 | |
3185 | return -1; | |
3186 | } | |
3187 | ||
0a0f39df CL |
3188 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
3189 | ifname, newnode); | |
3190 | ||
58397e6c KT |
3191 | return 0; |
3192 | } | |
3193 | ||
f3ea2ad2 IM |
3194 | /* Clears mapping for all available queues of vhost interface. */ |
3195 | static void | |
3196 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
3197 | OVS_REQUIRES(dev->mutex) | |
3198 | { | |
3199 | int i; | |
3200 | ||
81acebda | 3201 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
3202 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
3203 | } | |
3204 | } | |
3205 | ||
58397e6c KT |
3206 | /* |
3207 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
3208 | * flag to stop any more packets from being sent or received to/from a VM and | |
3209 | * ensure all currently queued packets have been sent/received before removing | |
3210 | * the device. | |
3211 | */ | |
3212 | static void | |
0a0f39df | 3213 | destroy_device(int vid) |
58397e6c | 3214 | { |
d46285a2 | 3215 | struct netdev_dpdk *dev; |
afee281f | 3216 | bool exists = false; |
0a0f39df CL |
3217 | char ifname[IF_NAME_SZ]; |
3218 | ||
58be5c0e | 3219 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
3220 | |
3221 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 3222 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 3223 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 3224 | |
d46285a2 | 3225 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
3226 | dev->vhost_reconfigured = false; |
3227 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 3228 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 3229 | |
e543851d | 3230 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 3231 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 3232 | exists = true; |
afee281f | 3233 | break; |
58397e6c KT |
3234 | } |
3235 | } | |
afee281f | 3236 | |
58397e6c KT |
3237 | ovs_mutex_unlock(&dpdk_mutex); |
3238 | ||
0a0f39df | 3239 | if (exists) { |
afee281f KT |
3240 | /* |
3241 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
3242 | * to NULL, before returning. | |
3243 | */ | |
3244 | ovsrcu_synchronize(); | |
3245 | /* | |
3246 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
3247 | * put thread back into quiescent state before returning. | |
3248 | */ | |
3249 | ovsrcu_quiesce_start(); | |
0a0f39df | 3250 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 3251 | } else { |
0a0f39df | 3252 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 3253 | } |
58397e6c KT |
3254 | } |
3255 | ||
585a5bea | 3256 | static int |
0a0f39df | 3257 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 3258 | { |
d46285a2 | 3259 | struct netdev_dpdk *dev; |
585a5bea IM |
3260 | bool exists = false; |
3261 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
3262 | char ifname[IF_NAME_SZ]; |
3263 | ||
58be5c0e | 3264 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
3265 | |
3266 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
3267 | return 0; | |
3268 | } | |
3269 | ||
3270 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 3271 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 3272 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 3273 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 3274 | if (enable) { |
d46285a2 | 3275 | dev->tx_q[qid].map = qid; |
585a5bea | 3276 | } else { |
d46285a2 | 3277 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 3278 | } |
d46285a2 | 3279 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 3280 | exists = true; |
d46285a2 | 3281 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3282 | break; |
3283 | } | |
c1ff66ac | 3284 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3285 | } |
3286 | ovs_mutex_unlock(&dpdk_mutex); | |
3287 | ||
3288 | if (exists) { | |
0a0f39df CL |
3289 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
3290 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 3291 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 3292 | } else { |
0a0f39df | 3293 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
3294 | return -1; |
3295 | } | |
3296 | ||
3297 | return 0; | |
3298 | } | |
3299 | ||
8492adc2 JS |
3300 | /* |
3301 | * Retrieve the DPDK virtio device ID (vid) associated with a vhostuser | |
3302 | * or vhostuserclient netdev. | |
3303 | * | |
3304 | * Returns a value greater or equal to zero for a valid vid or '-1' if | |
3305 | * there is no valid vid associated. A vid of '-1' must not be used in | |
3306 | * rte_vhost_ APi calls. | |
3307 | * | |
3308 | * Once obtained and validated, a vid can be used by a PMD for multiple | |
3309 | * subsequent rte_vhost API calls until the PMD quiesces. A PMD should | |
3310 | * not fetch the vid again for each of a series of API calls. | |
3311 | */ | |
3312 | ||
0a0f39df CL |
3313 | int |
3314 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 3315 | { |
0a0f39df | 3316 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
3317 | } |
3318 | ||
9509913a IS |
3319 | struct ingress_policer * |
3320 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
3321 | { | |
3322 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
3323 | } | |
3324 | ||
58397e6c | 3325 | static int |
ecc1a34e | 3326 | netdev_dpdk_class_init(void) |
7d1ced01 | 3327 | { |
ecc1a34e DDP |
3328 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
3329 | ||
3330 | /* This function can be called for different classes. The initialization | |
3331 | * needs to be done only once */ | |
3332 | if (ovsthread_once_start(&once)) { | |
3333 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
3334 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
3335 | "[netdev] up|down", 1, 2, | |
3336 | netdev_dpdk_set_admin_state, NULL); | |
3337 | ||
0ee821c2 DB |
3338 | unixctl_command_register("netdev-dpdk/detach", |
3339 | "pci address of device", 1, 1, | |
3340 | netdev_dpdk_detach, NULL); | |
3341 | ||
be481733 IM |
3342 | unixctl_command_register("netdev-dpdk/get-mempool-info", |
3343 | "[netdev]", 0, 1, | |
3344 | netdev_dpdk_get_mempool_info, NULL); | |
3345 | ||
ecc1a34e DDP |
3346 | ovsthread_once_done(&once); |
3347 | } | |
362ca396 | 3348 | |
7d1ced01 CL |
3349 | return 0; |
3350 | } | |
3351 | ||
033e9df2 | 3352 | |
95fb793a | 3353 | /* Client Rings */ |
3354 | ||
95fb793a | 3355 | static int |
3356 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
bb37956a | 3357 | dpdk_port_t *eth_port_id) |
95fb793a | 3358 | { |
48fffdee | 3359 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 3360 | char *ring_name; |
b8374d0d | 3361 | int port_id; |
95fb793a | 3362 | |
48fffdee KT |
3363 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
3364 | if (!ring_pair) { | |
95fb793a | 3365 | return ENOMEM; |
3366 | } | |
3367 | ||
7251515e | 3368 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 3369 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 3370 | |
8f0a76c9 | 3371 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 3372 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3373 | RING_F_SP_ENQ); |
0c6f39e5 | 3374 | free(ring_name); |
48fffdee KT |
3375 | if (ring_pair->cring_tx == NULL) { |
3376 | rte_free(ring_pair); | |
95fb793a | 3377 | return ENOMEM; |
3378 | } | |
3379 | ||
0c6f39e5 | 3380 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 3381 | |
8f0a76c9 | 3382 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 3383 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3384 | RING_F_SC_DEQ); |
0c6f39e5 | 3385 | free(ring_name); |
48fffdee KT |
3386 | if (ring_pair->cring_rx == NULL) { |
3387 | rte_free(ring_pair); | |
95fb793a | 3388 | return ENOMEM; |
3389 | } | |
3390 | ||
b8374d0d MV |
3391 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
3392 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 3393 | |
b8374d0d | 3394 | if (port_id < 0) { |
48fffdee | 3395 | rte_free(ring_pair); |
95fb793a | 3396 | return ENODEV; |
3397 | } | |
3398 | ||
48fffdee | 3399 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
3400 | ring_pair->eth_port_id = port_id; |
3401 | *eth_port_id = port_id; | |
3402 | ||
48fffdee | 3403 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 3404 | |
95fb793a | 3405 | return 0; |
3406 | } | |
3407 | ||
3408 | static int | |
bb37956a | 3409 | dpdk_ring_open(const char dev_name[], dpdk_port_t *eth_port_id) |
64839cf4 | 3410 | OVS_REQUIRES(dpdk_mutex) |
95fb793a | 3411 | { |
48fffdee | 3412 | struct dpdk_ring *ring_pair; |
95fb793a | 3413 | unsigned int port_no; |
3414 | int err = 0; | |
3415 | ||
3416 | /* Names always start with "dpdkr" */ | |
3417 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
3418 | if (err) { | |
3419 | return err; | |
3420 | } | |
3421 | ||
58be5c0e | 3422 | /* Look through our list to find the device */ |
48fffdee KT |
3423 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
3424 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 3425 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 3426 | /* Really all that is needed */ |
48fffdee | 3427 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 3428 | return 0; |
3429 | } | |
3430 | } | |
3431 | /* Need to create the device rings */ | |
3432 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
3433 | } | |
3434 | ||
7251515e | 3435 | static int |
d46285a2 | 3436 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
b30896c9 | 3437 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e | 3438 | { |
d46285a2 | 3439 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a543eb0 | 3440 | struct dp_packet *packet; |
1b99bb05 | 3441 | |
58be5c0e MK |
3442 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
3443 | * the rss hash field is clear. This is because the same mbuf may be | |
3444 | * modified by the consumer of the ring and return into the datapath | |
3445 | * without recalculating the RSS hash. */ | |
e883448e | 3446 | DP_PACKET_BATCH_FOR_EACH (i, packet, batch) { |
8a543eb0 | 3447 | dp_packet_mbuf_rss_flag_reset(packet); |
1b99bb05 | 3448 | } |
7251515e | 3449 | |
b30896c9 | 3450 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e DV |
3451 | return 0; |
3452 | } | |
3453 | ||
95fb793a | 3454 | static int |
3455 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
3456 | { | |
bb37956a | 3457 | dpdk_port_t port_no = 0; |
95fb793a | 3458 | int err = 0; |
3459 | ||
95fb793a | 3460 | ovs_mutex_lock(&dpdk_mutex); |
3461 | ||
3462 | err = dpdk_ring_open(netdev->name, &port_no); | |
3463 | if (err) { | |
3464 | goto unlock_dpdk; | |
3465 | } | |
3466 | ||
1ce30dfd DDP |
3467 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
3468 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 3469 | unlock_dpdk: |
3470 | ovs_mutex_unlock(&dpdk_mutex); | |
3471 | return err; | |
3472 | } | |
3473 | ||
0bf765f7 IS |
3474 | /* QoS Functions */ |
3475 | ||
3476 | /* | |
3477 | * Initialize QoS configuration operations. | |
3478 | */ | |
3479 | static void | |
3480 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
3481 | { | |
3482 | conf->ops = ops; | |
78bd47cf | 3483 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
3484 | } |
3485 | ||
3486 | /* | |
3487 | * Search existing QoS operations in qos_ops and compare each set of | |
3488 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
3489 | * else return NULL | |
3490 | */ | |
3491 | static const struct dpdk_qos_ops * | |
3492 | qos_lookup_name(const char *name) | |
3493 | { | |
3494 | const struct dpdk_qos_ops *const *opsp; | |
3495 | ||
3496 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3497 | const struct dpdk_qos_ops *ops = *opsp; | |
3498 | if (!strcmp(name, ops->qos_name)) { | |
3499 | return ops; | |
3500 | } | |
3501 | } | |
3502 | return NULL; | |
3503 | } | |
3504 | ||
0bf765f7 IS |
3505 | static int |
3506 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
3507 | struct sset *types) | |
3508 | { | |
3509 | const struct dpdk_qos_ops *const *opsp; | |
3510 | ||
3511 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3512 | const struct dpdk_qos_ops *ops = *opsp; | |
3513 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
3514 | sset_add(types, ops->qos_name); | |
3515 | } | |
3516 | } | |
3517 | return 0; | |
3518 | } | |
3519 | ||
3520 | static int | |
d46285a2 | 3521 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
3522 | const char **typep, struct smap *details) |
3523 | { | |
d46285a2 | 3524 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 3525 | struct qos_conf *qos_conf; |
0bf765f7 IS |
3526 | int error = 0; |
3527 | ||
d46285a2 | 3528 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
3529 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
3530 | if (qos_conf) { | |
3531 | *typep = qos_conf->ops->qos_name; | |
3532 | error = (qos_conf->ops->qos_get | |
3533 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
3534 | } else { |
3535 | /* No QoS configuration set, return an empty string */ | |
3536 | *typep = ""; | |
0bf765f7 | 3537 | } |
d46285a2 | 3538 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
3539 | |
3540 | return error; | |
3541 | } | |
3542 | ||
3543 | static int | |
78bd47cf DDP |
3544 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
3545 | const struct smap *details) | |
0bf765f7 | 3546 | { |
d46285a2 | 3547 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 3548 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 3549 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
3550 | int error = 0; |
3551 | ||
d46285a2 | 3552 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 3553 | |
78bd47cf | 3554 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 3555 | |
78bd47cf DDP |
3556 | new_ops = qos_lookup_name(type); |
3557 | ||
3558 | if (!new_ops || !new_ops->qos_construct) { | |
3559 | new_qos_conf = NULL; | |
3560 | if (type && type[0]) { | |
3561 | error = EOPNOTSUPP; | |
0bf765f7 | 3562 | } |
44975bb0 | 3563 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
3564 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
3565 | new_qos_conf = qos_conf; | |
0bf765f7 | 3566 | } else { |
78bd47cf | 3567 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
3568 | } |
3569 | ||
7ea266e9 | 3570 | if (error) { |
78bd47cf DDP |
3571 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
3572 | type, netdev->name, rte_strerror(error)); | |
3573 | } | |
3574 | ||
3575 | if (new_qos_conf != qos_conf) { | |
3576 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
3577 | if (qos_conf) { | |
3578 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
3579 | } | |
0bf765f7 IS |
3580 | } |
3581 | ||
d46285a2 | 3582 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 3583 | |
0bf765f7 IS |
3584 | return error; |
3585 | } | |
3586 | ||
3587 | /* egress-policer details */ | |
3588 | ||
3589 | struct egress_policer { | |
3590 | struct qos_conf qos_conf; | |
3591 | struct rte_meter_srtcm_params app_srtcm_params; | |
3592 | struct rte_meter_srtcm egress_meter; | |
3593 | }; | |
3594 | ||
78bd47cf DDP |
3595 | static void |
3596 | egress_policer_details_to_param(const struct smap *details, | |
3597 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 3598 | { |
78bd47cf DDP |
3599 | memset(params, 0, sizeof *params); |
3600 | params->cir = smap_get_ullong(details, "cir", 0); | |
3601 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
3602 | params->ebs = 0; | |
0bf765f7 IS |
3603 | } |
3604 | ||
3605 | static int | |
78bd47cf DDP |
3606 | egress_policer_qos_construct(const struct smap *details, |
3607 | struct qos_conf **conf) | |
0bf765f7 | 3608 | { |
0bf765f7 | 3609 | struct egress_policer *policer; |
0bf765f7 IS |
3610 | int err = 0; |
3611 | ||
0bf765f7 IS |
3612 | policer = xmalloc(sizeof *policer); |
3613 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 3614 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
0bf765f7 | 3615 | err = rte_meter_srtcm_config(&policer->egress_meter, |
78bd47cf DDP |
3616 | &policer->app_srtcm_params); |
3617 | if (!err) { | |
3618 | *conf = &policer->qos_conf; | |
3619 | } else { | |
7ea266e9 | 3620 | free(policer); |
78bd47cf | 3621 | *conf = NULL; |
7ea266e9 IS |
3622 | err = -err; |
3623 | } | |
0bf765f7 IS |
3624 | |
3625 | return err; | |
3626 | } | |
3627 | ||
3628 | static void | |
78bd47cf | 3629 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
3630 | { |
3631 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 3632 | qos_conf); |
0bf765f7 IS |
3633 | free(policer); |
3634 | } | |
3635 | ||
3636 | static int | |
78bd47cf | 3637 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 3638 | { |
78bd47cf DDP |
3639 | struct egress_policer *policer = |
3640 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3641 | ||
3642 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
3643 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 3644 | |
0bf765f7 IS |
3645 | return 0; |
3646 | } | |
3647 | ||
78bd47cf | 3648 | static bool |
47a45d86 KT |
3649 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
3650 | const struct smap *details) | |
0bf765f7 | 3651 | { |
78bd47cf DDP |
3652 | struct egress_policer *policer = |
3653 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3654 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 3655 | |
78bd47cf | 3656 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 3657 | |
78bd47cf | 3658 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
3659 | } |
3660 | ||
0bf765f7 | 3661 | static int |
3e90f7d7 | 3662 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt, |
7d7ded7a | 3663 | bool should_steal) |
0bf765f7 | 3664 | { |
0bf765f7 | 3665 | int cnt = 0; |
78bd47cf DDP |
3666 | struct egress_policer *policer = |
3667 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 3668 | |
3e90f7d7 | 3669 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, |
7d7ded7a | 3670 | pkt_cnt, should_steal); |
0bf765f7 IS |
3671 | |
3672 | return cnt; | |
3673 | } | |
3674 | ||
3675 | static const struct dpdk_qos_ops egress_policer_ops = { | |
3676 | "egress-policer", /* qos_name */ | |
3677 | egress_policer_qos_construct, | |
3678 | egress_policer_qos_destruct, | |
3679 | egress_policer_qos_get, | |
78bd47cf | 3680 | egress_policer_qos_is_equal, |
0bf765f7 IS |
3681 | egress_policer_run |
3682 | }; | |
3683 | ||
050c60bf DDP |
3684 | static int |
3685 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
3686 | { | |
3687 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3688 | int err = 0; | |
3689 | ||
050c60bf DDP |
3690 | ovs_mutex_lock(&dev->mutex); |
3691 | ||
3692 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 3693 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b | 3694 | && dev->mtu == dev->requested_mtu |
f8b64a61 | 3695 | && dev->lsc_interrupt_mode == dev->requested_lsc_interrupt_mode |
b685696b | 3696 | && dev->rxq_size == dev->requested_rxq_size |
bd4e172b | 3697 | && dev->txq_size == dev->requested_txq_size |
606f6650 EC |
3698 | && dev->socket_id == dev->requested_socket_id |
3699 | && dev->started) { | |
050c60bf DDP |
3700 | /* Reconfiguration is unnecessary */ |
3701 | ||
3702 | goto out; | |
3703 | } | |
3704 | ||
3705 | rte_eth_dev_stop(dev->port_id); | |
606f6650 | 3706 | dev->started = false; |
050c60bf | 3707 | |
d555d9bd | 3708 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 | 3709 | if (err && err != EEXIST) { |
d555d9bd | 3710 | goto out; |
0072e931 MK |
3711 | } |
3712 | ||
f8b64a61 RM |
3713 | dev->lsc_interrupt_mode = dev->requested_lsc_interrupt_mode; |
3714 | ||
050c60bf DDP |
3715 | netdev->n_txq = dev->requested_n_txq; |
3716 | netdev->n_rxq = dev->requested_n_rxq; | |
3717 | ||
b685696b CL |
3718 | dev->rxq_size = dev->requested_rxq_size; |
3719 | dev->txq_size = dev->requested_txq_size; | |
3720 | ||
050c60bf DDP |
3721 | rte_free(dev->tx_q); |
3722 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
3723 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
3724 | if (!dev->tx_q) { | |
3725 | err = ENOMEM; | |
3726 | } | |
050c60bf | 3727 | |
0072e931 MK |
3728 | netdev_change_seq_changed(netdev); |
3729 | ||
050c60bf | 3730 | out: |
050c60bf | 3731 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
3732 | return err; |
3733 | } | |
3734 | ||
7f381c2e | 3735 | static int |
2d24d165 | 3736 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 3737 | OVS_REQUIRES(dev->mutex) |
050c60bf | 3738 | { |
2d24d165 CL |
3739 | dev->up.n_txq = dev->requested_n_txq; |
3740 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 3741 | int err; |
050c60bf | 3742 | |
81acebda IM |
3743 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
3744 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
3745 | dev->tx_q[0].map = 0; | |
3746 | } | |
3747 | ||
3748 | netdev_dpdk_remap_txqs(dev); | |
3749 | ||
d555d9bd | 3750 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 FA |
3751 | if (!err) { |
3752 | /* A new mempool was created. */ | |
d555d9bd | 3753 | netdev_change_seq_changed(&dev->up); |
b6b26021 FA |
3754 | } else if (err != EEXIST){ |
3755 | return err; | |
db8f13b0 | 3756 | } |
0a0f39df | 3757 | if (netdev_dpdk_get_vid(dev) >= 0) { |
894af647 | 3758 | if (dev->vhost_reconfigured == false) { |
3759 | dev->vhost_reconfigured = true; | |
3760 | /* Carrier status may need updating. */ | |
3761 | netdev_change_seq_changed(&dev->up); | |
3762 | } | |
81acebda | 3763 | } |
7f381c2e DDP |
3764 | |
3765 | return 0; | |
2d24d165 CL |
3766 | } |
3767 | ||
3768 | static int | |
3769 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
3770 | { | |
3771 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3772 | int err; |
2d24d165 | 3773 | |
2d24d165 | 3774 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 3775 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 3776 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
3777 | |
3778 | return err; | |
2d24d165 CL |
3779 | } |
3780 | ||
3781 | static int | |
3782 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3783 | { | |
3784 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3785 | int err; |
a14d1cc8 | 3786 | uint64_t vhost_flags = 0; |
10087cba | 3787 | bool zc_enabled; |
2d24d165 | 3788 | |
2d24d165 CL |
3789 | ovs_mutex_lock(&dev->mutex); |
3790 | ||
c1ff66ac CL |
3791 | /* Configure vHost client mode if requested and if the following criteria |
3792 | * are met: | |
2d24d165 CL |
3793 | * 1. Device hasn't been registered yet. |
3794 | * 2. A path has been specified. | |
c1ff66ac CL |
3795 | */ |
3796 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 | 3797 | && strlen(dev->vhost_id)) { |
a14d1cc8 MK |
3798 | /* Register client-mode device. */ |
3799 | vhost_flags |= RTE_VHOST_USER_CLIENT; | |
3800 | ||
3801 | /* Enable IOMMU support, if explicitly requested. */ | |
3802 | if (dpdk_vhost_iommu_enabled()) { | |
3803 | vhost_flags |= RTE_VHOST_USER_IOMMU_SUPPORT; | |
3804 | } | |
10087cba CL |
3805 | |
3806 | zc_enabled = dev->vhost_driver_flags | |
3807 | & RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
3808 | /* Enable zero copy flag, if requested */ | |
3809 | if (zc_enabled) { | |
3810 | vhost_flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
3811 | } | |
3812 | ||
a14d1cc8 | 3813 | err = rte_vhost_driver_register(dev->vhost_id, vhost_flags); |
c1ff66ac | 3814 | if (err) { |
2d24d165 CL |
3815 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3816 | dev->vhost_id); | |
7f381c2e | 3817 | goto unlock; |
c1ff66ac | 3818 | } else { |
2d24d165 | 3819 | /* Configuration successful */ |
a14d1cc8 | 3820 | dev->vhost_driver_flags |= vhost_flags; |
2d24d165 CL |
3821 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " |
3822 | "using client socket '%s'", | |
3823 | dev->up.name, dev->vhost_id); | |
10087cba CL |
3824 | if (zc_enabled) { |
3825 | VLOG_INFO("Zero copy enabled for vHost port %s", dev->up.name); | |
3826 | } | |
c1ff66ac | 3827 | } |
f3e7ec25 MW |
3828 | |
3829 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
3830 | &virtio_net_device_ops); | |
3831 | if (err) { | |
3832 | VLOG_ERR("rte_vhost_driver_callback_register failed for " | |
3833 | "vhost user client port: %s\n", dev->up.name); | |
3834 | goto unlock; | |
3835 | } | |
3836 | ||
3837 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
3838 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
3839 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
3840 | | 1ULL << VIRTIO_NET_F_CSUM); | |
3841 | if (err) { | |
3842 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
3843 | "client port: %s\n", dev->up.name); | |
3844 | goto unlock; | |
3845 | } | |
3846 | ||
3847 | err = rte_vhost_driver_start(dev->vhost_id); | |
3848 | if (err) { | |
3849 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
3850 | "client port: %s\n", dev->up.name); | |
3851 | goto unlock; | |
3852 | } | |
c1ff66ac CL |
3853 | } |
3854 | ||
7f381c2e DDP |
3855 | err = dpdk_vhost_reconfigure_helper(dev); |
3856 | ||
3857 | unlock: | |
050c60bf | 3858 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 3859 | |
7f381c2e | 3860 | return err; |
050c60bf DDP |
3861 | } |
3862 | ||
ecc1a34e | 3863 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
81acebda | 3864 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
971f4b39 MW |
3865 | GET_CARRIER, GET_STATS, \ |
3866 | GET_CUSTOM_STATS, \ | |
81acebda IM |
3867 | GET_FEATURES, GET_STATUS, \ |
3868 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3869 | { \ |
3870 | NAME, \ | |
118c77b1 | 3871 | true, /* is_pmd */ \ |
ecc1a34e | 3872 | INIT, /* init */ \ |
95fb793a | 3873 | NULL, /* netdev_dpdk_run */ \ |
3874 | NULL, /* netdev_dpdk_wait */ \ | |
3875 | \ | |
3876 | netdev_dpdk_alloc, \ | |
3877 | CONSTRUCT, \ | |
58397e6c | 3878 | DESTRUCT, \ |
95fb793a | 3879 | netdev_dpdk_dealloc, \ |
3880 | netdev_dpdk_get_config, \ | |
81acebda | 3881 | SET_CONFIG, \ |
95fb793a | 3882 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3883 | NULL, /* build header */ \ |
3884 | NULL, /* push header */ \ | |
3885 | NULL, /* pop header */ \ | |
7dec44fe | 3886 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3887 | SET_TX_MULTIQ, \ |
95fb793a | 3888 | \ |
7251515e | 3889 | SEND, /* send */ \ |
95fb793a | 3890 | NULL, /* send_wait */ \ |
3891 | \ | |
3892 | netdev_dpdk_set_etheraddr, \ | |
3893 | netdev_dpdk_get_etheraddr, \ | |
3894 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3895 | netdev_dpdk_set_mtu, \ |
95fb793a | 3896 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3897 | GET_CARRIER, \ |
95fb793a | 3898 | netdev_dpdk_get_carrier_resets, \ |
3899 | netdev_dpdk_set_miimon, \ | |
58397e6c | 3900 | GET_STATS, \ |
971f4b39 | 3901 | GET_CUSTOM_STATS, \ |
58397e6c | 3902 | GET_FEATURES, \ |
95fb793a | 3903 | NULL, /* set_advertisements */ \ |
875ab130 | 3904 | NULL, /* get_pt_mode */ \ |
95fb793a | 3905 | \ |
9509913a | 3906 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3907 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3908 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3909 | netdev_dpdk_get_qos, \ |
3910 | netdev_dpdk_set_qos, \ | |
95fb793a | 3911 | NULL, /* get_queue */ \ |
3912 | NULL, /* set_queue */ \ | |
3913 | NULL, /* delete_queue */ \ | |
3914 | NULL, /* get_queue_stats */ \ | |
3915 | NULL, /* queue_dump_start */ \ | |
3916 | NULL, /* queue_dump_next */ \ | |
3917 | NULL, /* queue_dump_done */ \ | |
3918 | NULL, /* dump_queue_stats */ \ | |
3919 | \ | |
95fb793a | 3920 | NULL, /* set_in4 */ \ |
a8704b50 | 3921 | NULL, /* get_addr_list */ \ |
95fb793a | 3922 | NULL, /* add_router */ \ |
3923 | NULL, /* get_next_hop */ \ | |
58397e6c | 3924 | GET_STATUS, \ |
95fb793a | 3925 | NULL, /* arp_lookup */ \ |
3926 | \ | |
3927 | netdev_dpdk_update_flags, \ | |
050c60bf | 3928 | RECONFIGURE, \ |
95fb793a | 3929 | \ |
3930 | netdev_dpdk_rxq_alloc, \ | |
3931 | netdev_dpdk_rxq_construct, \ | |
3932 | netdev_dpdk_rxq_destruct, \ | |
3933 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3934 | RXQ_RECV, \ |
95fb793a | 3935 | NULL, /* rx_wait */ \ |
3936 | NULL, /* rxq_drain */ \ | |
18ebd48c | 3937 | NO_OFFLOAD_API \ |
95fb793a | 3938 | } |
8a9562d2 | 3939 | |
bce01e3a | 3940 | static const struct netdev_class dpdk_class = |
95fb793a | 3941 | NETDEV_DPDK_CLASS( |
3942 | "dpdk", | |
ecc1a34e | 3943 | netdev_dpdk_class_init, |
5496878c | 3944 | netdev_dpdk_construct, |
58397e6c | 3945 | netdev_dpdk_destruct, |
81acebda IM |
3946 | netdev_dpdk_set_config, |
3947 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3948 | netdev_dpdk_eth_send, |
3949 | netdev_dpdk_get_carrier, | |
3950 | netdev_dpdk_get_stats, | |
971f4b39 | 3951 | netdev_dpdk_get_custom_stats, |
58397e6c KT |
3952 | netdev_dpdk_get_features, |
3953 | netdev_dpdk_get_status, | |
050c60bf | 3954 | netdev_dpdk_reconfigure, |
58397e6c | 3955 | netdev_dpdk_rxq_recv); |
95fb793a | 3956 | |
bce01e3a | 3957 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3958 | NETDEV_DPDK_CLASS( |
3959 | "dpdkr", | |
ecc1a34e | 3960 | netdev_dpdk_class_init, |
5496878c | 3961 | netdev_dpdk_ring_construct, |
58397e6c | 3962 | netdev_dpdk_destruct, |
c3d062a7 | 3963 | netdev_dpdk_ring_set_config, |
81acebda | 3964 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3965 | netdev_dpdk_ring_send, |
3966 | netdev_dpdk_get_carrier, | |
3967 | netdev_dpdk_get_stats, | |
971f4b39 | 3968 | netdev_dpdk_get_custom_stats, |
58397e6c KT |
3969 | netdev_dpdk_get_features, |
3970 | netdev_dpdk_get_status, | |
050c60bf | 3971 | netdev_dpdk_reconfigure, |
58397e6c KT |
3972 | netdev_dpdk_rxq_recv); |
3973 | ||
53f50d24 | 3974 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3975 | NETDEV_DPDK_CLASS( |
3976 | "dpdkvhostuser", | |
f3e7ec25 | 3977 | NULL, |
53f50d24 | 3978 | netdev_dpdk_vhost_construct, |
58397e6c | 3979 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3980 | NULL, |
81acebda | 3981 | NULL, |
58397e6c KT |
3982 | netdev_dpdk_vhost_send, |
3983 | netdev_dpdk_vhost_get_carrier, | |
3984 | netdev_dpdk_vhost_get_stats, | |
3985 | NULL, | |
7251515e | 3986 | NULL, |
b2e8b12f | 3987 | netdev_dpdk_vhost_user_get_status, |
53f50d24 | 3988 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3989 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3990 | static const struct netdev_class dpdk_vhost_client_class = |
3991 | NETDEV_DPDK_CLASS( | |
3992 | "dpdkvhostuserclient", | |
f3e7ec25 | 3993 | NULL, |
2d24d165 CL |
3994 | netdev_dpdk_vhost_client_construct, |
3995 | netdev_dpdk_vhost_destruct, | |
3996 | netdev_dpdk_vhost_client_set_config, | |
3997 | NULL, | |
3998 | netdev_dpdk_vhost_send, | |
3999 | netdev_dpdk_vhost_get_carrier, | |
4000 | netdev_dpdk_vhost_get_stats, | |
4001 | NULL, | |
4002 | NULL, | |
b2e8b12f | 4003 | netdev_dpdk_vhost_user_get_status, |
2d24d165 CL |
4004 | netdev_dpdk_vhost_client_reconfigure, |
4005 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 4006 | |
8a9562d2 PS |
4007 | void |
4008 | netdev_dpdk_register(void) | |
4009 | { | |
bab69409 AC |
4010 | netdev_register_provider(&dpdk_class); |
4011 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 4012 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 4013 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 4014 | } |