]>
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 DDP |
28 | |
29 | #include <rte_config.h> | |
30 | #include <rte_cycles.h> | |
31 | #include <rte_errno.h> | |
32 | #include <rte_eth_ring.h> | |
33 | #include <rte_ethdev.h> | |
34 | #include <rte_malloc.h> | |
35 | #include <rte_mbuf.h> | |
36 | #include <rte_meter.h> | |
fc56f5e0 | 37 | #include <rte_pci.h> |
f3e7ec25 | 38 | #include <rte_vhost.h> |
8a9562d2 | 39 | |
7d1ced01 | 40 | #include "dirs.h" |
e14deea0 | 41 | #include "dp-packet.h" |
01961bbd | 42 | #include "dpdk.h" |
8a9562d2 | 43 | #include "dpif-netdev.h" |
e5c0f5a4 | 44 | #include "fatal-signal.h" |
8a9562d2 PS |
45 | #include "netdev-provider.h" |
46 | #include "netdev-vport.h" | |
47 | #include "odp-util.h" | |
eac84432 | 48 | #include "openvswitch/dynamic-string.h" |
25d436fb BW |
49 | #include "openvswitch/list.h" |
50 | #include "openvswitch/ofp-print.h" | |
51 | #include "openvswitch/vlog.h" | |
94143fc4 | 52 | #include "ovs-numa.h" |
8a9562d2 PS |
53 | #include "ovs-thread.h" |
54 | #include "ovs-rcu.h" | |
55 | #include "packets.h" | |
ee89ea7b | 56 | #include "openvswitch/shash.h" |
0bf765f7 | 57 | #include "smap.h" |
8a9562d2 PS |
58 | #include "sset.h" |
59 | #include "unaligned.h" | |
60 | #include "timeval.h" | |
61 | #include "unixctl.h" | |
8a9562d2 | 62 | |
f3e7ec25 MW |
63 | enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM}; |
64 | ||
05b49df6 | 65 | VLOG_DEFINE_THIS_MODULE(netdev_dpdk); |
8a9562d2 PS |
66 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
67 | ||
68 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
69 | ||
70 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
71 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
72 | ||
73 | /* | |
74 | * need to reserve tons of extra space in the mbufs so we can align the | |
75 | * DMA addresses to 4KB. | |
18f777b2 TP |
76 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
77 | * performance for standard Ethernet MTU. | |
8a9562d2 | 78 | */ |
58be5c0e MK |
79 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN \ |
80 | + (2 * VLAN_HEADER_LEN)) | |
4be4d22c MK |
81 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) |
82 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
58be5c0e MK |
83 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len) \ |
84 | - ETHER_HDR_LEN - ETHER_CRC_LEN) | |
31b88c97 SS |
85 | #define MBUF_SIZE(mtu) ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) \ |
86 | + sizeof(struct dp_packet) \ | |
87 | + RTE_PKTMBUF_HEADROOM), \ | |
88 | RTE_CACHE_LINE_SIZE) | |
4be4d22c | 89 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 90 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 91 | |
da79ce2b DDP |
92 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
93 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
94 | * enough hugepages) we keep halving the number until the allocation succeeds | |
95 | * or we reach MIN_NB_MBUF */ | |
96 | ||
97 | #define MAX_NB_MBUF (4096 * 64) | |
98 | #define MIN_NB_MBUF (4096 * 4) | |
99 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
100 | ||
101 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
102 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
103 | ||
104 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
105 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
106 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
107 | % MP_CACHE_SZ == 0); | |
108 | ||
d6e3feb5 | 109 | /* |
110 | * DPDK XSTATS Counter names definition | |
111 | */ | |
112 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
113 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
114 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
115 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
116 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
117 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
118 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
119 | ||
120 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
121 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
122 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
123 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
124 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
125 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
126 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
127 | ||
d57f777f | 128 | #define XSTAT_RX_MULTICAST_PACKETS "rx_multicast_packets" |
d6e3feb5 | 129 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" |
130 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
131 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
132 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
133 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
134 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
135 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
136 | ||
8a9562d2 PS |
137 | #define SOCKET0 0 |
138 | ||
b685696b CL |
139 | /* Default size of Physical NIC RXQ */ |
140 | #define NIC_PORT_DEFAULT_RXQ_SIZE 2048 | |
141 | /* Default size of Physical NIC TXQ */ | |
142 | #define NIC_PORT_DEFAULT_TXQ_SIZE 2048 | |
143 | /* Maximum size of Physical NIC Queues */ | |
144 | #define NIC_PORT_MAX_Q_SIZE 4096 | |
79f5354c | 145 | |
585a5bea | 146 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
147 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
148 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
149 | * yet mapped to another queue. */ | |
585a5bea | 150 | |
bb37956a IM |
151 | #define DPDK_ETH_PORT_ID_INVALID RTE_MAX_ETHPORTS |
152 | ||
153 | /* DPDK library uses uint8_t for port_id. */ | |
154 | typedef uint8_t dpdk_port_t; | |
155 | ||
31871ee3 | 156 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 157 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 158 | |
8a9562d2 | 159 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
160 | .rxmode = { |
161 | .mq_mode = ETH_MQ_RX_RSS, | |
162 | .split_hdr_size = 0, | |
163 | .header_split = 0, /* Header Split disabled */ | |
164 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
165 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
166 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
167 | .hw_strip_crc = 0, | |
168 | }, | |
169 | .rx_adv_conf = { | |
170 | .rss_conf = { | |
171 | .rss_key = NULL, | |
543342a4 | 172 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 173 | }, |
a28ddd11 DDP |
174 | }, |
175 | .txmode = { | |
176 | .mq_mode = ETH_MQ_TX_NONE, | |
177 | }, | |
8a9562d2 PS |
178 | }; |
179 | ||
f3e7ec25 MW |
180 | /* |
181 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
182 | * configuration has been fully completed. | |
183 | */ | |
184 | static int new_device(int vid); | |
185 | static void destroy_device(int vid); | |
186 | static int vring_state_changed(int vid, uint16_t queue_id, int enable); | |
187 | static const struct vhost_device_ops virtio_net_device_ops = | |
188 | { | |
189 | .new_device = new_device, | |
190 | .destroy_device = destroy_device, | |
191 | .vring_state_changed = vring_state_changed, | |
192 | .features_changed = NULL | |
193 | }; | |
194 | ||
58f7c37b DDP |
195 | enum { DPDK_RING_SIZE = 256 }; |
196 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
197 | enum { DRAIN_TSC = 200000ULL }; |
198 | ||
58397e6c KT |
199 | enum dpdk_dev_type { |
200 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 201 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
202 | }; |
203 | ||
0bf765f7 IS |
204 | /* Quality of Service */ |
205 | ||
206 | /* An instance of a QoS configuration. Always associated with a particular | |
207 | * network device. | |
208 | * | |
209 | * Each QoS implementation subclasses this with whatever additional data it | |
210 | * needs. | |
211 | */ | |
212 | struct qos_conf { | |
213 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 214 | rte_spinlock_t lock; |
0bf765f7 IS |
215 | }; |
216 | ||
217 | /* A particular implementation of dpdk QoS operations. | |
218 | * | |
219 | * The functions below return 0 if successful or a positive errno value on | |
220 | * failure, except where otherwise noted. All of them must be provided, except | |
221 | * where otherwise noted. | |
222 | */ | |
223 | struct dpdk_qos_ops { | |
224 | ||
225 | /* Name of the QoS type */ | |
226 | const char *qos_name; | |
227 | ||
78bd47cf DDP |
228 | /* Called to construct a qos_conf object. The implementation should make |
229 | * the appropriate calls to configure QoS according to 'details'. | |
0bf765f7 IS |
230 | * |
231 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
232 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
233 | * (which is built as ovs-vswitchd.conf.db(8)). | |
234 | * | |
78bd47cf DDP |
235 | * This function must return 0 if and only if it sets '*conf' to an |
236 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
237 | * |
238 | * For all QoS implementations it should always be non-null. | |
239 | */ | |
78bd47cf | 240 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
241 | |
242 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 243 | * 'qos_conf'. |
0bf765f7 IS |
244 | * |
245 | * For all QoS implementations it should always be non-null. | |
246 | */ | |
78bd47cf | 247 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 248 | |
78bd47cf | 249 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
250 | * |
251 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
252 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
253 | * (which is built as ovs-vswitchd.conf.db(8)). | |
254 | */ | |
78bd47cf | 255 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 256 | |
78bd47cf | 257 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
258 | * |
259 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
260 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
261 | * (which is built as ovs-vswitchd.conf.db(8)). | |
262 | * | |
78bd47cf | 263 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 264 | */ |
78bd47cf DDP |
265 | bool (*qos_is_equal)(const struct qos_conf *conf, |
266 | const struct smap *details); | |
0bf765f7 IS |
267 | |
268 | /* Modify an array of rte_mbufs. The modification is specific to | |
269 | * each qos implementation. | |
270 | * | |
271 | * The function should take and array of mbufs and an int representing | |
272 | * the current number of mbufs present in the array. | |
273 | * | |
274 | * After the function has performed a qos modification to the array of | |
275 | * mbufs it returns an int representing the number of mbufs now present in | |
276 | * the array. This value is can then be passed to the port send function | |
277 | * along with the modified array for transmission. | |
278 | * | |
279 | * For all QoS implementations it should always be non-null. | |
280 | */ | |
78bd47cf | 281 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
3e90f7d7 | 282 | int pkt_cnt, bool may_steal); |
0bf765f7 IS |
283 | }; |
284 | ||
285 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
286 | static const struct dpdk_qos_ops egress_policer_ops; | |
287 | ||
288 | /* | |
289 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
290 | * operations. | |
291 | */ | |
292 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
293 | &egress_policer_ops, | |
294 | NULL | |
295 | }; | |
296 | ||
c2adb102 IM |
297 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
298 | ||
8a9562d2 | 299 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 300 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 301 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 302 | |
c2adb102 IM |
303 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
304 | = OVS_MUTEX_INITIALIZER; | |
305 | ||
8a9562d2 PS |
306 | struct dpdk_mp { |
307 | struct rte_mempool *mp; | |
308 | int mtu; | |
309 | int socket_id; | |
d555d9bd RW |
310 | char if_name[IFNAMSIZ]; |
311 | unsigned mp_size; | |
c2adb102 | 312 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mp_mutex); |
8a9562d2 PS |
313 | }; |
314 | ||
5a034064 AW |
315 | /* There should be one 'struct dpdk_tx_queue' created for |
316 | * each cpu core. */ | |
8a9562d2 | 317 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
318 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
319 | * from concurrent access. It is used only | |
320 | * if the queue is shared among different | |
324c8374 | 321 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
322 | int map; /* Mapping of configured vhost-user queues |
323 | * to enabled by guest. */ | |
8a9562d2 PS |
324 | }; |
325 | ||
95fb793a | 326 | /* dpdk has no way to remove dpdk ring ethernet devices |
327 | so we have to keep them around once they've been created | |
328 | */ | |
329 | ||
ca6ba700 | 330 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 331 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 332 | |
333 | struct dpdk_ring { | |
334 | /* For the client rings */ | |
335 | struct rte_ring *cring_tx; | |
336 | struct rte_ring *cring_rx; | |
b83a2df1 | 337 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
bb37956a | 338 | dpdk_port_t eth_port_id; /* ethernet device port id */ |
ca6ba700 | 339 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 340 | }; |
341 | ||
9509913a IS |
342 | struct ingress_policer { |
343 | struct rte_meter_srtcm_params app_srtcm_params; | |
344 | struct rte_meter_srtcm in_policer; | |
345 | rte_spinlock_t policer_lock; | |
346 | }; | |
347 | ||
1a2bb118 SC |
348 | enum dpdk_hw_ol_features { |
349 | NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0, | |
350 | }; | |
351 | ||
8a9562d2 PS |
352 | struct netdev_dpdk { |
353 | struct netdev up; | |
bb37956a | 354 | dpdk_port_t port_id; |
8a9562d2 | 355 | int max_packet_len; |
58397e6c | 356 | enum dpdk_dev_type type; |
8a9562d2 | 357 | |
5a034064 | 358 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
359 | |
360 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
361 | ||
362 | struct dpdk_mp *dpdk_mp; | |
363 | int mtu; | |
364 | int socket_id; | |
365 | int buf_size; | |
8a9562d2 | 366 | struct netdev_stats stats; |
45d947c4 DDP |
367 | /* Protects stats */ |
368 | rte_spinlock_t stats_lock; | |
8a9562d2 | 369 | |
74ff3298 | 370 | struct eth_addr hwaddr; |
8a9562d2 PS |
371 | enum netdev_flags flags; |
372 | ||
373 | struct rte_eth_link link; | |
374 | int link_reset_cnt; | |
375 | ||
0a0f39df CL |
376 | /* virtio identifier for vhost devices */ |
377 | ovsrcu_index vid; | |
378 | ||
379 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
380 | bool vhost_reconfigured; | |
58397e6c | 381 | |
2d24d165 CL |
382 | /* Identifier used to distinguish vhost devices from each other. */ |
383 | char vhost_id[PATH_MAX]; | |
7d1ced01 | 384 | |
55e075e6 CL |
385 | /* Device arguments for dpdk ports */ |
386 | char *devargs; | |
387 | ||
5dcde09c IM |
388 | /* If true, device was attached by rte_eth_dev_attach(). */ |
389 | bool attached; | |
390 | ||
8a9562d2 | 391 | /* In dpdk_list. */ |
ca6ba700 | 392 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
0bf765f7 IS |
393 | |
394 | /* QoS configuration and lock for the device */ | |
78bd47cf | 395 | OVSRCU_TYPE(struct qos_conf *) qos_conf; |
0bf765f7 | 396 | |
050c60bf DDP |
397 | /* The following properties cannot be changed when a device is running, |
398 | * so we remember the request and update them next time | |
399 | * netdev_dpdk*_reconfigure() is called */ | |
0072e931 | 400 | int requested_mtu; |
050c60bf DDP |
401 | int requested_n_txq; |
402 | int requested_n_rxq; | |
b685696b CL |
403 | int requested_rxq_size; |
404 | int requested_txq_size; | |
405 | ||
406 | /* Number of rx/tx descriptors for physical devices */ | |
407 | int rxq_size; | |
408 | int txq_size; | |
9509913a | 409 | |
db8f13b0 CL |
410 | /* Socket ID detected when vHost device is brought up */ |
411 | int requested_socket_id; | |
412 | ||
c1ff66ac CL |
413 | /* Denotes whether vHost port is client/server mode */ |
414 | uint64_t vhost_driver_flags; | |
415 | ||
9509913a IS |
416 | /* Ingress Policer */ |
417 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
418 | uint32_t policer_rate; | |
419 | uint32_t policer_burst; | |
9fd39370 SC |
420 | |
421 | /* DPDK-ETH Flow control */ | |
422 | struct rte_eth_fc_conf fc_conf; | |
1a2bb118 SC |
423 | |
424 | /* DPDK-ETH hardware offload features, | |
425 | * from the enum set 'dpdk_hw_ol_features' */ | |
426 | uint32_t hw_ol_features; | |
8a9562d2 PS |
427 | }; |
428 | ||
429 | struct netdev_rxq_dpdk { | |
430 | struct netdev_rxq up; | |
bb37956a | 431 | dpdk_port_t port_id; |
8a9562d2 PS |
432 | }; |
433 | ||
f3e7ec25 MW |
434 | static void netdev_dpdk_destruct(struct netdev *netdev); |
435 | static void netdev_dpdk_vhost_destruct(struct netdev *netdev); | |
8a9562d2 | 436 | |
0a0f39df | 437 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 438 | |
9509913a IS |
439 | struct ingress_policer * |
440 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
441 | ||
8a9562d2 PS |
442 | static bool |
443 | is_dpdk_class(const struct netdev_class *class) | |
444 | { | |
f3e7ec25 MW |
445 | return class->destruct == netdev_dpdk_destruct |
446 | || class->destruct == netdev_dpdk_vhost_destruct; | |
8a9562d2 PS |
447 | } |
448 | ||
4be4d22c MK |
449 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
450 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
451 | * value, insufficient buffers are allocated to accomodate the packet in its | |
452 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
453 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
454 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
455 | * behaviour, which reduces performance. To prevent this, use a buffer size |
456 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
457 | */ |
458 | static uint32_t | |
459 | dpdk_buf_size(int mtu) | |
460 | { | |
461 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
462 | NETDEV_DPDK_MBUF_ALIGN); | |
463 | } | |
464 | ||
eff23640 DDP |
465 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
466 | * | |
467 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
468 | static void * |
469 | dpdk_rte_mzalloc(size_t sz) | |
470 | { | |
eff23640 | 471 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
472 | } |
473 | ||
474 | void | |
e14deea0 | 475 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 476 | { |
db73f716 | 477 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 478 | |
b00b4a81 | 479 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
480 | } |
481 | ||
b3cd9f9d | 482 | static void |
401b70d6 | 483 | ovs_rte_pktmbuf_init(struct rte_mempool *mp OVS_UNUSED, |
b3cd9f9d | 484 | void *opaque_arg OVS_UNUSED, |
2391135c | 485 | void *_p, |
b3cd9f9d PS |
486 | unsigned i OVS_UNUSED) |
487 | { | |
2391135c | 488 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 489 | |
2391135c | 490 | dp_packet_init_dpdk((struct dp_packet *) pkt, pkt->buf_len); |
b3cd9f9d PS |
491 | } |
492 | ||
d555d9bd RW |
493 | /* |
494 | * Full DPDK memory pool name must be unique | |
495 | * and cannot be longer than RTE_MEMPOOL_NAMESIZE | |
496 | */ | |
497 | static char * | |
498 | dpdk_mp_name(struct dpdk_mp *dmp) | |
499 | { | |
500 | uint32_t h = hash_string(dmp->if_name, 0); | |
501 | char *mp_name = xcalloc(RTE_MEMPOOL_NAMESIZE, sizeof *mp_name); | |
502 | int ret = snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_%x_%d_%u", | |
503 | h, dmp->mtu, dmp->mp_size); | |
504 | if (ret < 0 || ret >= RTE_MEMPOOL_NAMESIZE) { | |
505 | return NULL; | |
506 | } | |
507 | return mp_name; | |
508 | } | |
509 | ||
8a9562d2 | 510 | static struct dpdk_mp * |
d555d9bd | 511 | dpdk_mp_create(struct netdev_dpdk *dev, int mtu) |
8a9562d2 | 512 | { |
d555d9bd | 513 | struct dpdk_mp *dmp = dpdk_rte_mzalloc(sizeof *dmp); |
eff23640 DDP |
514 | if (!dmp) { |
515 | return NULL; | |
516 | } | |
d555d9bd | 517 | dmp->socket_id = dev->requested_socket_id; |
8a9562d2 | 518 | dmp->mtu = mtu; |
d555d9bd RW |
519 | ovs_strzcpy(dmp->if_name, dev->up.name, IFNAMSIZ); |
520 | ||
521 | /* | |
522 | * XXX: rough estimation of memory required for port: | |
523 | * <packets required to fill the device rxqs> | |
524 | * + <packets that could be stuck on other ports txqs> | |
525 | * + <packets in the pmd threads> | |
526 | * + <additional memory for corner cases> | |
0072e931 | 527 | */ |
d555d9bd RW |
528 | dmp->mp_size = dev->requested_n_rxq * dev->requested_rxq_size |
529 | + dev->requested_n_txq * dev->requested_txq_size | |
530 | + MIN(RTE_MAX_LCORE, dev->requested_n_rxq) * NETDEV_MAX_BURST | |
531 | + MIN_NB_MBUF; | |
532 | ||
533 | bool mp_exists = false; | |
8a9562d2 | 534 | |
da79ce2b | 535 | do { |
d555d9bd | 536 | char *mp_name = dpdk_mp_name(dmp); |
95fb793a | 537 | |
d555d9bd RW |
538 | VLOG_DBG("Requesting a mempool of %u mbufs for netdev %s " |
539 | "with %d Rx and %d Tx queues.", | |
540 | dmp->mp_size, dev->up.name, | |
541 | dev->requested_n_rxq, dev->requested_n_txq); | |
542 | ||
543 | dmp->mp = rte_pktmbuf_pool_create(mp_name, dmp->mp_size, | |
401b70d6 HA |
544 | MP_CACHE_SZ, |
545 | sizeof (struct dp_packet) | |
546 | - sizeof (struct rte_mbuf), | |
547 | MBUF_SIZE(mtu) | |
548 | - sizeof(struct dp_packet), | |
d555d9bd | 549 | dmp->socket_id); |
0c6f39e5 | 550 | if (dmp->mp) { |
d555d9bd RW |
551 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", mp_name, |
552 | dmp->mp_size); | |
553 | } else if (rte_errno == EEXIST) { | |
554 | /* A mempool with the same name already exists. We just | |
555 | * retrieve its pointer to be returned to the caller. */ | |
556 | dmp->mp = rte_mempool_lookup(mp_name); | |
557 | VLOG_DBG("A mempool with name %s already exists at %p.", | |
558 | mp_name, dmp->mp); | |
559 | /* As the mempool create returned EEXIST we can expect the | |
560 | * lookup has returned a valid pointer. If for some reason | |
561 | * that's not the case we keep track of it. */ | |
562 | mp_exists = true; | |
563 | } else { | |
564 | VLOG_ERR("Failed mempool \"%s\" create request of %u mbufs", | |
565 | mp_name, dmp->mp_size); | |
0c6f39e5 DDP |
566 | } |
567 | free(mp_name); | |
568 | if (dmp->mp) { | |
401b70d6 HA |
569 | /* rte_pktmbuf_pool_create has done some initialization of the |
570 | * rte_mbuf part of each dp_packet, while ovs_rte_pktmbuf_init | |
571 | * initializes some OVS specific fields of dp_packet. | |
572 | */ | |
573 | rte_mempool_obj_iter(dmp->mp, ovs_rte_pktmbuf_init, NULL); | |
0c6f39e5 DDP |
574 | return dmp; |
575 | } | |
d555d9bd RW |
576 | } while (!mp_exists && |
577 | (rte_errno == ENOMEM && (dmp->mp_size /= 2) >= MIN_NB_MBUF)); | |
2ae3d542 | 578 | |
2ae3d542 DDP |
579 | rte_free(dmp); |
580 | return NULL; | |
581 | } | |
582 | ||
583 | static struct dpdk_mp * | |
d555d9bd | 584 | dpdk_mp_get(struct netdev_dpdk *dev, int mtu) |
2ae3d542 DDP |
585 | { |
586 | struct dpdk_mp *dmp; | |
587 | ||
588 | ovs_mutex_lock(&dpdk_mp_mutex); | |
d555d9bd | 589 | dmp = dpdk_mp_create(dev, mtu); |
c2adb102 IM |
590 | ovs_mutex_unlock(&dpdk_mp_mutex); |
591 | ||
c2adb102 | 592 | return dmp; |
8a9562d2 PS |
593 | } |
594 | ||
595 | static void | |
c2adb102 | 596 | dpdk_mp_put(struct dpdk_mp *dmp) |
8a9562d2 | 597 | { |
d555d9bd RW |
598 | char *mp_name; |
599 | ||
8a9562d2 PS |
600 | if (!dmp) { |
601 | return; | |
602 | } | |
603 | ||
c2adb102 | 604 | ovs_mutex_lock(&dpdk_mp_mutex); |
d555d9bd RW |
605 | mp_name = dpdk_mp_name(dmp); |
606 | VLOG_DBG("Releasing \"%s\" mempool", mp_name); | |
607 | free(mp_name); | |
608 | rte_mempool_free(dmp->mp); | |
609 | rte_free(dmp); | |
c2adb102 | 610 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
611 | } |
612 | ||
0072e931 MK |
613 | /* Tries to allocate new mempool on requested_socket_id with |
614 | * mbuf size corresponding to requested_mtu. | |
615 | * On success new configuration will be applied. | |
616 | * On error, device will be left unchanged. */ | |
617 | static int | |
618 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
619 | OVS_REQUIRES(dev->mutex) |
620 | { | |
621 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
622 | struct dpdk_mp *mp; | |
623 | ||
d555d9bd | 624 | mp = dpdk_mp_get(dev, FRAME_LEN_TO_MTU(buf_size)); |
0072e931 | 625 | if (!mp) { |
c67e46c0 MK |
626 | VLOG_ERR("Failed to create memory pool for netdev " |
627 | "%s, with MTU %d on socket %d: %s\n", | |
628 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
629 | rte_strerror(rte_errno)); | |
630 | return rte_errno; | |
0072e931 MK |
631 | } else { |
632 | dpdk_mp_put(dev->dpdk_mp); | |
633 | dev->dpdk_mp = mp; | |
634 | dev->mtu = dev->requested_mtu; | |
635 | dev->socket_id = dev->requested_socket_id; | |
636 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
637 | } | |
638 | ||
639 | return 0; | |
640 | } | |
641 | ||
8a9562d2 PS |
642 | static void |
643 | check_link_status(struct netdev_dpdk *dev) | |
644 | { | |
645 | struct rte_eth_link link; | |
646 | ||
647 | rte_eth_link_get_nowait(dev->port_id, &link); | |
648 | ||
649 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 650 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
651 | |
652 | dev->link_reset_cnt++; | |
653 | dev->link = link; | |
654 | if (dev->link.link_status) { | |
bb37956a | 655 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Up - speed %u Mbps - %s", |
58be5c0e | 656 | dev->port_id, (unsigned) dev->link.link_speed, |
8a9562d2 PS |
657 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? |
658 | ("full-duplex") : ("half-duplex")); | |
659 | } else { | |
bb37956a | 660 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Down", dev->port_id); |
8a9562d2 PS |
661 | } |
662 | } | |
663 | } | |
664 | ||
665 | static void * | |
666 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
667 | { | |
668 | struct netdev_dpdk *dev; | |
669 | ||
670 | pthread_detach(pthread_self()); | |
671 | ||
672 | for (;;) { | |
673 | ovs_mutex_lock(&dpdk_mutex); | |
674 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
675 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
676 | if (dev->type == DPDK_DEV_ETH) { |
677 | check_link_status(dev); | |
678 | } | |
8a9562d2 PS |
679 | ovs_mutex_unlock(&dev->mutex); |
680 | } | |
681 | ovs_mutex_unlock(&dpdk_mutex); | |
682 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
683 | } | |
684 | ||
685 | return NULL; | |
686 | } | |
687 | ||
b98d7669 DDP |
688 | static int |
689 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
690 | { | |
691 | int diag = 0; | |
692 | int i; | |
0072e931 | 693 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 694 | |
67fe6d63 MK |
695 | /* For some NICs (e.g. Niantic), scatter_rx mode needs to be explicitly |
696 | * enabled. */ | |
0072e931 | 697 | if (dev->mtu > ETHER_MTU) { |
67fe6d63 | 698 | conf.rxmode.enable_scatter = 1; |
0072e931 | 699 | } |
67fe6d63 | 700 | |
1a2bb118 SC |
701 | conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & |
702 | NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
b98d7669 DDP |
703 | /* A device may report more queues than it makes available (this has |
704 | * been observed for Intel xl710, which reserves some of them for | |
705 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
706 | * available. When this happens we can retry the configuration | |
707 | * and request less queues */ | |
708 | while (n_rxq && n_txq) { | |
709 | if (diag) { | |
710 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
711 | } | |
712 | ||
0072e931 | 713 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 714 | if (diag) { |
0072e931 MK |
715 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
716 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
717 | break; |
718 | } | |
719 | ||
67fe6d63 MK |
720 | diag = rte_eth_dev_set_mtu(dev->port_id, dev->mtu); |
721 | if (diag) { | |
722 | VLOG_ERR("Interface %s MTU (%d) setup error: %s", | |
723 | dev->up.name, dev->mtu, rte_strerror(-diag)); | |
724 | break; | |
725 | } | |
726 | ||
b98d7669 | 727 | for (i = 0; i < n_txq; i++) { |
b685696b | 728 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
729 | dev->socket_id, NULL); |
730 | if (diag) { | |
731 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
732 | dev->up.name, i, rte_strerror(-diag)); | |
733 | break; | |
734 | } | |
735 | } | |
736 | ||
737 | if (i != n_txq) { | |
738 | /* Retry with less tx queues */ | |
739 | n_txq = i; | |
740 | continue; | |
741 | } | |
742 | ||
743 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 744 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
b98d7669 DDP |
745 | dev->socket_id, NULL, |
746 | dev->dpdk_mp->mp); | |
747 | if (diag) { | |
748 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
749 | dev->up.name, i, rte_strerror(-diag)); | |
750 | break; | |
751 | } | |
752 | } | |
753 | ||
754 | if (i != n_rxq) { | |
755 | /* Retry with less rx queues */ | |
756 | n_rxq = i; | |
757 | continue; | |
758 | } | |
759 | ||
760 | dev->up.n_rxq = n_rxq; | |
81acebda | 761 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
762 | |
763 | return 0; | |
764 | } | |
765 | ||
766 | return diag; | |
767 | } | |
768 | ||
9fd39370 SC |
769 | static void |
770 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
771 | { | |
772 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
bb37956a IM |
773 | VLOG_WARN("Failed to enable flow control on device %"PRIu8, |
774 | dev->port_id); | |
9fd39370 SC |
775 | } |
776 | } | |
b98d7669 | 777 | |
8a9562d2 | 778 | static int |
c2adb102 IM |
779 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
780 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
781 | { |
782 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 783 | struct rte_eth_dev_info info; |
8a9562d2 PS |
784 | struct ether_addr eth_addr; |
785 | int diag; | |
b98d7669 | 786 | int n_rxq, n_txq; |
d4f5282c KT |
787 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | |
788 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
789 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
8a9562d2 | 790 | |
a0cb2d66 | 791 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 792 | |
d4f5282c KT |
793 | if ((info.rx_offload_capa & rx_chksm_offload_capa) != |
794 | rx_chksm_offload_capa) { | |
2cfe866f | 795 | VLOG_WARN("Rx checksum offload is not supported on port %"PRIu8, |
d4f5282c KT |
796 | dev->port_id); |
797 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; | |
798 | } else { | |
799 | dev->hw_ol_features |= NETDEV_RX_CHECKSUM_OFFLOAD; | |
800 | } | |
801 | ||
b98d7669 DDP |
802 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
803 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
804 | ||
805 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 806 | if (diag) { |
b98d7669 DDP |
807 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
808 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 809 | return -diag; |
8a9562d2 PS |
810 | } |
811 | ||
8a9562d2 PS |
812 | diag = rte_eth_dev_start(dev->port_id); |
813 | if (diag) { | |
b98d7669 DDP |
814 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
815 | rte_strerror(-diag)); | |
95fb793a | 816 | return -diag; |
8a9562d2 PS |
817 | } |
818 | ||
819 | rte_eth_promiscuous_enable(dev->port_id); | |
820 | rte_eth_allmulticast_enable(dev->port_id); | |
821 | ||
822 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
823 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
bb37956a | 824 | VLOG_INFO_RL(&rl, "Port %"PRIu8": "ETH_ADDR_FMT, |
ca92d173 | 825 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 826 | |
ca92d173 | 827 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
828 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
829 | ||
830 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
831 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
832 | ||
9fd39370 SC |
833 | /* Get the Flow control configuration for DPDK-ETH */ |
834 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
835 | if (diag) { | |
bb37956a | 836 | VLOG_DBG("cannot get flow control parameters on port=%"PRIu8", err=%d", |
9fd39370 SC |
837 | dev->port_id, diag); |
838 | } | |
839 | ||
8a9562d2 PS |
840 | return 0; |
841 | } | |
842 | ||
843 | static struct netdev_dpdk * | |
844 | netdev_dpdk_cast(const struct netdev *netdev) | |
845 | { | |
846 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
847 | } | |
848 | ||
849 | static struct netdev * | |
850 | netdev_dpdk_alloc(void) | |
851 | { | |
bab69409 AC |
852 | struct netdev_dpdk *dev; |
853 | ||
65e19e70 DDP |
854 | dev = dpdk_rte_mzalloc(sizeof *dev); |
855 | if (dev) { | |
856 | return &dev->up; | |
bab69409 | 857 | } |
65e19e70 | 858 | |
bab69409 | 859 | return NULL; |
8a9562d2 PS |
860 | } |
861 | ||
eff23640 DDP |
862 | static struct dpdk_tx_queue * |
863 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 864 | { |
eff23640 | 865 | struct dpdk_tx_queue *txqs; |
bd5131ba | 866 | unsigned i; |
5a034064 | 867 | |
eff23640 DDP |
868 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
869 | if (txqs) { | |
870 | for (i = 0; i < n_txqs; i++) { | |
871 | /* Initialize map for vhost devices. */ | |
872 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
873 | rte_spinlock_init(&txqs[i].tx_lock); | |
874 | } | |
5a034064 | 875 | } |
eff23640 DDP |
876 | |
877 | return txqs; | |
5a034064 AW |
878 | } |
879 | ||
8a9562d2 | 880 | static int |
bb37956a | 881 | common_construct(struct netdev *netdev, dpdk_port_t port_no, |
1ce30dfd | 882 | enum dpdk_dev_type type, int socket_id) |
5a034064 | 883 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 884 | { |
d46285a2 | 885 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 886 | |
d46285a2 | 887 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 888 | |
d46285a2 | 889 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 890 | |
1b7a04e0 AW |
891 | /* If the 'sid' is negative, it means that the kernel fails |
892 | * to obtain the pci numa info. In that situation, always | |
893 | * use 'SOCKET0'. */ | |
1ce30dfd | 894 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 895 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
896 | dev->port_id = port_no; |
897 | dev->type = type; | |
898 | dev->flags = 0; | |
7f381c2e | 899 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 900 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
901 | ovsrcu_index_init(&dev->vid, -1); |
902 | dev->vhost_reconfigured = false; | |
5dcde09c | 903 | dev->attached = false; |
8a9562d2 | 904 | |
78bd47cf | 905 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 906 | |
9509913a IS |
907 | ovsrcu_init(&dev->ingress_policer, NULL); |
908 | dev->policer_rate = 0; | |
909 | dev->policer_burst = 0; | |
910 | ||
7f381c2e DDP |
911 | netdev->n_rxq = 0; |
912 | netdev->n_txq = 0; | |
913 | dev->requested_n_rxq = NR_QUEUE; | |
914 | dev->requested_n_txq = NR_QUEUE; | |
915 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
916 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 917 | |
9fd39370 SC |
918 | /* Initialize the flow control to NULL */ |
919 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
920 | |
921 | /* Initilize the hardware offload flags to 0 */ | |
922 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
923 | |
924 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
925 | ||
d46285a2 | 926 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 927 | |
7f381c2e DDP |
928 | netdev_request_reconfigure(netdev); |
929 | ||
1ce30dfd | 930 | return 0; |
95fb793a | 931 | } |
932 | ||
b83a2df1 MV |
933 | /* dev_name must be the prefix followed by a positive decimal number. |
934 | * (no leading + or - signs are allowed) */ | |
95fb793a | 935 | static int |
936 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
937 | unsigned int *port_no) | |
938 | { | |
939 | const char *cport; | |
940 | ||
941 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
942 | return ENODEV; | |
943 | } | |
944 | ||
945 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
946 | |
947 | if (str_to_uint(cport, 10, port_no)) { | |
948 | return 0; | |
949 | } else { | |
950 | return ENODEV; | |
951 | } | |
95fb793a | 952 | } |
953 | ||
1ce30dfd DDP |
954 | static int |
955 | vhost_common_construct(struct netdev *netdev) | |
956 | OVS_REQUIRES(dpdk_mutex) | |
957 | { | |
958 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
959 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
960 | ||
961 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); | |
962 | if (!dev->tx_q) { | |
963 | return ENOMEM; | |
964 | } | |
965 | ||
bb37956a IM |
966 | return common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
967 | DPDK_DEV_VHOST, socket_id); | |
1ce30dfd DDP |
968 | } |
969 | ||
7d1ced01 | 970 | static int |
53f50d24 | 971 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 972 | { |
d46285a2 DDP |
973 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
974 | const char *name = netdev->name; | |
7d1ced01 | 975 | int err; |
a0cb2d66 | 976 | |
1af27e8a DDP |
977 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
978 | * the file system. '/' or '\' would traverse directories, so they're not | |
979 | * acceptable in 'name'. */ | |
980 | if (strchr(name, '/') || strchr(name, '\\')) { | |
981 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
982 | "A valid name must not include '/' or '\\'", | |
983 | name); | |
984 | return EINVAL; | |
985 | } | |
986 | ||
7d1ced01 CL |
987 | ovs_mutex_lock(&dpdk_mutex); |
988 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 989 | * the socket is to be created, then register the socket. |
7d1ced01 | 990 | */ |
2d24d165 | 991 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 992 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 993 | |
2d24d165 CL |
994 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
995 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
996 | if (err) { |
997 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 998 | dev->vhost_id); |
f3e7ec25 | 999 | goto out; |
e5c0f5a4 | 1000 | } else { |
2d24d165 CL |
1001 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
1002 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
1003 | dev->vhost_id, name); | |
1004 | } | |
f3e7ec25 MW |
1005 | |
1006 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
1007 | &virtio_net_device_ops); | |
1008 | if (err) { | |
1009 | VLOG_ERR("rte_vhost_driver_callback_register failed for vhost user " | |
1010 | "port: %s\n", name); | |
1011 | goto out; | |
1012 | } | |
1013 | ||
1014 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
1015 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
1016 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
1017 | | 1ULL << VIRTIO_NET_F_CSUM); | |
1018 | if (err) { | |
1019 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
1020 | "port: %s\n", name); | |
1021 | goto out; | |
1022 | } | |
1023 | ||
1024 | err = rte_vhost_driver_start(dev->vhost_id); | |
1025 | if (err) { | |
1026 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
1027 | "port: %s\n", name); | |
1028 | goto out; | |
1029 | } | |
1030 | ||
1ce30dfd | 1031 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1032 | if (err) { |
1033 | VLOG_ERR("vhost_common_construct failed for vhost user " | |
1034 | "port: %s\n", name); | |
1035 | } | |
2d24d165 | 1036 | |
f3e7ec25 | 1037 | out: |
2d24d165 | 1038 | ovs_mutex_unlock(&dpdk_mutex); |
28ca969e AC |
1039 | VLOG_WARN_ONCE("dpdkvhostuser ports are considered deprecated; " |
1040 | "please migrate to dpdkvhostuserclient ports."); | |
2d24d165 CL |
1041 | return err; |
1042 | } | |
1043 | ||
1044 | static int | |
1045 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
1046 | { | |
1047 | int err; | |
1048 | ||
2d24d165 | 1049 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 1050 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1051 | if (err) { |
1052 | VLOG_ERR("vhost_common_construct failed for vhost user client" | |
1053 | "port: %s\n", netdev->name); | |
1054 | } | |
7d1ced01 | 1055 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
1056 | return err; |
1057 | } | |
1058 | ||
95fb793a | 1059 | static int |
1060 | netdev_dpdk_construct(struct netdev *netdev) | |
1061 | { | |
95fb793a | 1062 | int err; |
1063 | ||
95fb793a | 1064 | ovs_mutex_lock(&dpdk_mutex); |
bb37956a IM |
1065 | err = common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1066 | DPDK_DEV_ETH, SOCKET0); | |
8a9562d2 PS |
1067 | ovs_mutex_unlock(&dpdk_mutex); |
1068 | return err; | |
1069 | } | |
1070 | ||
1ce30dfd DDP |
1071 | static void |
1072 | common_destruct(struct netdev_dpdk *dev) | |
1073 | OVS_REQUIRES(dpdk_mutex) | |
1074 | OVS_EXCLUDED(dev->mutex) | |
1075 | { | |
1076 | rte_free(dev->tx_q); | |
1077 | dpdk_mp_put(dev->dpdk_mp); | |
1078 | ||
1079 | ovs_list_remove(&dev->list_node); | |
1080 | free(ovsrcu_get_protected(struct ingress_policer *, | |
1081 | &dev->ingress_policer)); | |
1082 | ovs_mutex_destroy(&dev->mutex); | |
1083 | } | |
1084 | ||
8a9562d2 | 1085 | static void |
d46285a2 | 1086 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 1087 | { |
d46285a2 | 1088 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5dcde09c | 1089 | char devname[RTE_ETH_NAME_MAX_LEN]; |
8a9562d2 | 1090 | |
8d38823b | 1091 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1092 | |
8a9562d2 | 1093 | rte_eth_dev_stop(dev->port_id); |
5dcde09c IM |
1094 | |
1095 | if (dev->attached) { | |
1096 | rte_eth_dev_close(dev->port_id); | |
1097 | if (rte_eth_dev_detach(dev->port_id, devname) < 0) { | |
1098 | VLOG_ERR("Device '%s' can not be detached", dev->devargs); | |
1099 | } else { | |
0ee821c2 | 1100 | VLOG_INFO("Device '%s' has been detached", devname); |
5dcde09c IM |
1101 | } |
1102 | } | |
1103 | ||
55e075e6 | 1104 | free(dev->devargs); |
1ce30dfd | 1105 | common_destruct(dev); |
8d38823b | 1106 | |
8a9562d2 | 1107 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1108 | } |
8a9562d2 | 1109 | |
3f891bbe DDP |
1110 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1111 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1112 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1113 | static int | |
c1ff66ac CL |
1114 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1115 | char *vhost_id) | |
3f891bbe DDP |
1116 | OVS_EXCLUDED(dpdk_mutex) |
1117 | OVS_EXCLUDED(dev->mutex) | |
1118 | { | |
c1ff66ac | 1119 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1120 | } |
1121 | ||
58397e6c | 1122 | static void |
d46285a2 | 1123 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1124 | { |
d46285a2 | 1125 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1126 | char *vhost_id; |
58397e6c | 1127 | |
8d38823b | 1128 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1129 | |
c62da695 | 1130 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1131 | if (netdev_dpdk_get_vid(dev) >= 0 |
1132 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1133 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1134 | netdev->name); |
58be5c0e MK |
1135 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1136 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1137 | } |
1138 | ||
2d24d165 | 1139 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1140 | |
1ce30dfd DDP |
1141 | common_destruct(dev); |
1142 | ||
58397e6c | 1143 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1144 | |
569c26da | 1145 | if (!vhost_id[0]) { |
821b8664 IM |
1146 | goto out; |
1147 | } | |
1148 | ||
c1ff66ac | 1149 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1150 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1151 | netdev->name, vhost_id); | |
c1ff66ac CL |
1152 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1153 | /* OVS server mode - remove this socket from list for deletion */ | |
1154 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1155 | } |
821b8664 | 1156 | out: |
c1ff66ac | 1157 | free(vhost_id); |
8a9562d2 PS |
1158 | } |
1159 | ||
1160 | static void | |
d46285a2 | 1161 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1162 | { |
d46285a2 | 1163 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1164 | |
d46285a2 | 1165 | rte_free(dev); |
8a9562d2 PS |
1166 | } |
1167 | ||
1168 | static int | |
a14b8947 | 1169 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1170 | { |
a14b8947 | 1171 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1172 | |
1173 | ovs_mutex_lock(&dev->mutex); | |
1174 | ||
050c60bf | 1175 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1176 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1177 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1178 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1179 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1180 | |
1181 | if (dev->type == DPDK_DEV_ETH) { | |
1182 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1183 | dev->requested_rxq_size); | |
1184 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1185 | dev->rxq_size); | |
1186 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1187 | dev->requested_txq_size); | |
1188 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1189 | dev->txq_size); | |
1a2bb118 SC |
1190 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1191 | smap_add(args, "rx_csum_offload", "true"); | |
8155ab7e KT |
1192 | } else { |
1193 | smap_add(args, "rx_csum_offload", "false"); | |
1a2bb118 | 1194 | } |
451f26fd | 1195 | } |
8a9562d2 PS |
1196 | ovs_mutex_unlock(&dev->mutex); |
1197 | ||
1198 | return 0; | |
1199 | } | |
1200 | ||
55e075e6 | 1201 | static struct netdev_dpdk * |
bb37956a | 1202 | netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id) |
55e075e6 CL |
1203 | OVS_REQUIRES(dpdk_mutex) |
1204 | { | |
1205 | struct netdev_dpdk *dev; | |
1206 | ||
1207 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1208 | if (dev->port_id == port_id) { | |
1209 | return dev; | |
1210 | } | |
1211 | } | |
1212 | ||
1213 | return NULL; | |
1214 | } | |
1215 | ||
bb37956a | 1216 | static dpdk_port_t |
5dcde09c IM |
1217 | netdev_dpdk_process_devargs(struct netdev_dpdk *dev, |
1218 | const char *devargs, char **errp) | |
55e075e6 | 1219 | { |
03d6399e IM |
1220 | /* Get the name up to the first comma. */ |
1221 | char *name = xmemdup0(devargs, strcspn(devargs, ",")); | |
bb37956a | 1222 | dpdk_port_t new_port_id = DPDK_ETH_PORT_ID_INVALID; |
55e075e6 | 1223 | |
69876ed7 | 1224 | if (!rte_eth_dev_count() |
03d6399e | 1225 | || rte_eth_dev_get_port_by_name(name, &new_port_id) |
69876ed7 CL |
1226 | || !rte_eth_dev_is_valid_port(new_port_id)) { |
1227 | /* Device not found in DPDK, attempt to attach it */ | |
1228 | if (!rte_eth_dev_attach(devargs, &new_port_id)) { | |
1229 | /* Attach successful */ | |
5dcde09c | 1230 | dev->attached = true; |
69876ed7 CL |
1231 | VLOG_INFO("Device '%s' attached to DPDK", devargs); |
1232 | } else { | |
1233 | /* Attach unsuccessful */ | |
bb37956a | 1234 | new_port_id = DPDK_ETH_PORT_ID_INVALID; |
47a45d86 KT |
1235 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", |
1236 | devargs); | |
55e075e6 CL |
1237 | } |
1238 | } | |
1239 | ||
03d6399e | 1240 | free(name); |
55e075e6 CL |
1241 | return new_port_id; |
1242 | } | |
1243 | ||
c3d062a7 CL |
1244 | static void |
1245 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1246 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1247 | { |
050c60bf | 1248 | int new_n_rxq; |
a14b8947 | 1249 | |
2a21e757 | 1250 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1251 | if (new_n_rxq != dev->requested_n_rxq) { |
1252 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1253 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1254 | } |
c3d062a7 CL |
1255 | } |
1256 | ||
b685696b CL |
1257 | static void |
1258 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1259 | const char *flag, int default_size, int *new_size) | |
1260 | { | |
1261 | int queue_size = smap_get_int(args, flag, default_size); | |
1262 | ||
1263 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1264 | || !is_pow2(queue_size)) { | |
1265 | queue_size = default_size; | |
1266 | } | |
1267 | ||
1268 | if (queue_size != *new_size) { | |
1269 | *new_size = queue_size; | |
1270 | netdev_request_reconfigure(netdev); | |
1271 | } | |
1272 | } | |
1273 | ||
c3d062a7 | 1274 | static int |
9fff138e DDP |
1275 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1276 | char **errp) | |
c3d062a7 CL |
1277 | { |
1278 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
b614c894 IM |
1279 | bool rx_fc_en, tx_fc_en, autoneg; |
1280 | enum rte_eth_fc_mode fc_mode; | |
1281 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1282 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1283 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1284 | }; | |
55e075e6 CL |
1285 | const char *new_devargs; |
1286 | int err = 0; | |
c3d062a7 | 1287 | |
55e075e6 | 1288 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1289 | ovs_mutex_lock(&dev->mutex); |
1290 | ||
1291 | dpdk_set_rxq_config(dev, args); | |
1292 | ||
b685696b CL |
1293 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1294 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1295 | &dev->requested_rxq_size); | |
1296 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1297 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1298 | &dev->requested_txq_size); | |
1299 | ||
55e075e6 CL |
1300 | new_devargs = smap_get(args, "dpdk-devargs"); |
1301 | ||
1302 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1303 | /* The user requested a new device. If we return error, the caller | |
1304 | * will delete this netdev and try to recreate it. */ | |
1305 | err = EAGAIN; | |
1306 | goto out; | |
1307 | } | |
1308 | ||
1309 | /* dpdk-devargs is required for device configuration */ | |
1310 | if (new_devargs && new_devargs[0]) { | |
1311 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1312 | * is valid */ | |
1313 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1314 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
bb37956a IM |
1315 | dpdk_port_t new_port_id = netdev_dpdk_process_devargs(dev, |
1316 | new_devargs, | |
1317 | errp); | |
55e075e6 CL |
1318 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1319 | err = EINVAL; | |
1320 | } else if (new_port_id == dev->port_id) { | |
1321 | /* Already configured, do not reconfigure again */ | |
1322 | err = 0; | |
1323 | } else { | |
1324 | struct netdev_dpdk *dup_dev; | |
bb37956a | 1325 | |
55e075e6 CL |
1326 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); |
1327 | if (dup_dev) { | |
9fff138e DDP |
1328 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
1329 | "which is already in use by '%s'", | |
1330 | netdev_get_name(netdev), new_devargs, | |
1331 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1332 | err = EADDRINUSE; |
1333 | } else { | |
bd4e172b | 1334 | int sid = rte_eth_dev_socket_id(new_port_id); |
bb37956a | 1335 | |
bd4e172b | 1336 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; |
55e075e6 CL |
1337 | dev->devargs = xstrdup(new_devargs); |
1338 | dev->port_id = new_port_id; | |
1339 | netdev_request_reconfigure(&dev->up); | |
1340 | err = 0; | |
1341 | } | |
1342 | } | |
1343 | } | |
1344 | } else { | |
9fff138e DDP |
1345 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1346 | "The old 'dpdk<port_id>' names are not supported", | |
1347 | netdev_get_name(netdev)); | |
55e075e6 CL |
1348 | err = EINVAL; |
1349 | } | |
1350 | ||
1351 | if (err) { | |
1352 | goto out; | |
1353 | } | |
1354 | ||
c3d062a7 CL |
1355 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1356 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1357 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1358 | |
b614c894 IM |
1359 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1360 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1361 | dev->fc_conf.mode = fc_mode; | |
1362 | dev->fc_conf.autoneg = autoneg; | |
1363 | dpdk_eth_flow_ctrl_setup(dev); | |
1364 | } | |
9fd39370 | 1365 | |
55e075e6 | 1366 | out: |
c3d062a7 | 1367 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1368 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1369 | |
55e075e6 | 1370 | return err; |
c3d062a7 CL |
1371 | } |
1372 | ||
1373 | static int | |
9fff138e DDP |
1374 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1375 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1376 | { |
1377 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1378 | ||
1379 | ovs_mutex_lock(&dev->mutex); | |
1380 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1381 | ovs_mutex_unlock(&dev->mutex); |
1382 | ||
1383 | return 0; | |
1384 | } | |
1385 | ||
c1ff66ac | 1386 | static int |
2d24d165 | 1387 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1388 | const struct smap *args, |
1389 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1390 | { |
1391 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1392 | const char *path; | |
1393 | ||
6881885a | 1394 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1395 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1396 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1397 | if (path && strcmp(path, dev->vhost_id)) { |
1398 | strcpy(dev->vhost_id, path); | |
c1ff66ac CL |
1399 | netdev_request_reconfigure(netdev); |
1400 | } | |
1401 | } | |
6881885a | 1402 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1403 | |
1404 | return 0; | |
1405 | } | |
1406 | ||
7dec44fe | 1407 | static int |
d46285a2 | 1408 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1409 | { |
d46285a2 | 1410 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1411 | |
d46285a2 | 1412 | return dev->socket_id; |
7dec44fe AW |
1413 | } |
1414 | ||
050c60bf | 1415 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1416 | static int |
050c60bf | 1417 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1418 | { |
d46285a2 | 1419 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1420 | |
d46285a2 | 1421 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1422 | |
050c60bf DDP |
1423 | if (dev->requested_n_txq == n_txq) { |
1424 | goto out; | |
4573fbd3 FL |
1425 | } |
1426 | ||
050c60bf DDP |
1427 | dev->requested_n_txq = n_txq; |
1428 | netdev_request_reconfigure(netdev); | |
58397e6c | 1429 | |
050c60bf | 1430 | out: |
d46285a2 | 1431 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1432 | return 0; |
58397e6c KT |
1433 | } |
1434 | ||
8a9562d2 PS |
1435 | static struct netdev_rxq * |
1436 | netdev_dpdk_rxq_alloc(void) | |
1437 | { | |
1438 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1439 | ||
eff23640 DDP |
1440 | if (rx) { |
1441 | return &rx->up; | |
1442 | } | |
1443 | ||
1444 | return NULL; | |
8a9562d2 PS |
1445 | } |
1446 | ||
1447 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1448 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1449 | { |
d46285a2 | 1450 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1451 | } |
1452 | ||
1453 | static int | |
d46285a2 | 1454 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1455 | { |
d46285a2 DDP |
1456 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1457 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1458 | |
d46285a2 DDP |
1459 | ovs_mutex_lock(&dev->mutex); |
1460 | rx->port_id = dev->port_id; | |
1461 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1462 | |
1463 | return 0; | |
1464 | } | |
1465 | ||
1466 | static void | |
d46285a2 | 1467 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1468 | { |
1469 | } | |
1470 | ||
1471 | static void | |
d46285a2 | 1472 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1473 | { |
d46285a2 | 1474 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1475 | |
1476 | rte_free(rx); | |
1477 | } | |
1478 | ||
819f13bd DDP |
1479 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1480 | * 'pkts', even in case of failure. | |
1481 | * | |
1482 | * Returns the number of packets that weren't transmitted. */ | |
1483 | static inline int | |
b59cc14e | 1484 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 1485 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 1486 | { |
1304f1f8 DDP |
1487 | uint32_t nb_tx = 0; |
1488 | ||
b59cc14e | 1489 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1490 | uint32_t ret; |
1491 | ||
b59cc14e | 1492 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1493 | if (!ret) { |
1494 | break; | |
1495 | } | |
1496 | ||
1497 | nb_tx += ret; | |
1498 | } | |
8a9562d2 | 1499 | |
b59cc14e | 1500 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 1501 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
1502 | * packet could come from a different mempool) */ |
1503 | int i; | |
1504 | ||
b59cc14e IM |
1505 | for (i = nb_tx; i < cnt; i++) { |
1506 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1507 | } |
8a9562d2 | 1508 | } |
819f13bd DDP |
1509 | |
1510 | return cnt - nb_tx; | |
8a9562d2 PS |
1511 | } |
1512 | ||
f3926f29 IS |
1513 | static inline bool |
1514 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1515 | struct rte_mbuf *pkt, uint64_t time) | |
1516 | { | |
1517 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1518 | ||
1519 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1520 | e_RTE_METER_GREEN; | |
1521 | } | |
1522 | ||
1523 | static int | |
1524 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
3e90f7d7 GZ |
1525 | struct rte_mbuf **pkts, int pkt_cnt, |
1526 | bool may_steal) | |
f3926f29 IS |
1527 | { |
1528 | int i = 0; | |
1529 | int cnt = 0; | |
1530 | struct rte_mbuf *pkt = NULL; | |
1531 | uint64_t current_time = rte_rdtsc(); | |
1532 | ||
1533 | for (i = 0; i < pkt_cnt; i++) { | |
1534 | pkt = pkts[i]; | |
1535 | /* Handle current packet */ | |
1536 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1537 | if (cnt != i) { | |
1538 | pkts[cnt] = pkt; | |
1539 | } | |
1540 | cnt++; | |
1541 | } else { | |
3e90f7d7 GZ |
1542 | if (may_steal) { |
1543 | rte_pktmbuf_free(pkt); | |
1544 | } | |
f3926f29 IS |
1545 | } |
1546 | } | |
1547 | ||
1548 | return cnt; | |
1549 | } | |
1550 | ||
9509913a IS |
1551 | static int |
1552 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
3e90f7d7 | 1553 | int pkt_cnt, bool may_steal) |
9509913a IS |
1554 | { |
1555 | int cnt = 0; | |
1556 | ||
1557 | rte_spinlock_lock(&policer->policer_lock); | |
3e90f7d7 GZ |
1558 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, |
1559 | pkt_cnt, may_steal); | |
9509913a IS |
1560 | rte_spinlock_unlock(&policer->policer_lock); |
1561 | ||
1562 | return cnt; | |
1563 | } | |
1564 | ||
58397e6c | 1565 | static bool |
0a0f39df | 1566 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1567 | { |
0a0f39df | 1568 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1569 | } |
1570 | ||
d6e3feb5 | 1571 | static inline void |
1572 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1573 | unsigned int packet_size) | |
1574 | { | |
1575 | /* Hard-coded search for the size bucket. */ | |
1576 | if (packet_size < 256) { | |
1577 | if (packet_size >= 128) { | |
1578 | stats->rx_128_to_255_packets++; | |
1579 | } else if (packet_size <= 64) { | |
1580 | stats->rx_1_to_64_packets++; | |
1581 | } else { | |
1582 | stats->rx_65_to_127_packets++; | |
1583 | } | |
1584 | } else { | |
1585 | if (packet_size >= 1523) { | |
1586 | stats->rx_1523_to_max_packets++; | |
1587 | } else if (packet_size >= 1024) { | |
1588 | stats->rx_1024_to_1522_packets++; | |
1589 | } else if (packet_size < 512) { | |
1590 | stats->rx_256_to_511_packets++; | |
1591 | } else { | |
1592 | stats->rx_512_to_1023_packets++; | |
1593 | } | |
1594 | } | |
1595 | } | |
1596 | ||
9e3ddd45 TP |
1597 | static inline void |
1598 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1599 | struct dp_packet **packets, int count, |
1600 | int dropped) | |
9e3ddd45 TP |
1601 | { |
1602 | int i; | |
d6e3feb5 | 1603 | unsigned int packet_size; |
9e3ddd45 TP |
1604 | struct dp_packet *packet; |
1605 | ||
1606 | stats->rx_packets += count; | |
9509913a | 1607 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1608 | for (i = 0; i < count; i++) { |
1609 | packet = packets[i]; | |
d6e3feb5 | 1610 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1611 | |
d6e3feb5 | 1612 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1613 | /* This only protects the following multicast counting from |
1614 | * too short packets, but it does not stop the packet from | |
1615 | * further processing. */ | |
1616 | stats->rx_errors++; | |
1617 | stats->rx_length_errors++; | |
1618 | continue; | |
1619 | } | |
1620 | ||
d6e3feb5 | 1621 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1622 | ||
9e3ddd45 TP |
1623 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1624 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1625 | stats->multicast++; | |
1626 | } | |
1627 | ||
d6e3feb5 | 1628 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1629 | } |
1630 | } | |
1631 | ||
58397e6c KT |
1632 | /* |
1633 | * The receive path for the vhost port is the TX path out from guest. | |
1634 | */ | |
1635 | static int | |
d46285a2 | 1636 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1637 | struct dp_packet_batch *batch) |
58397e6c | 1638 | { |
d46285a2 | 1639 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
d46285a2 | 1640 | int qid = rxq->queue_id; |
9509913a | 1641 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1642 | uint16_t nb_rx = 0; |
9509913a | 1643 | uint16_t dropped = 0; |
58397e6c | 1644 | |
0a0f39df | 1645 | if (OVS_UNLIKELY(!is_vhost_running(dev) |
e543851d | 1646 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1647 | return EAGAIN; |
1648 | } | |
1649 | ||
0a0f39df CL |
1650 | nb_rx = rte_vhost_dequeue_burst(netdev_dpdk_get_vid(dev), |
1651 | qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
d46285a2 | 1652 | dev->dpdk_mp->mp, |
64839cf4 | 1653 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1654 | NETDEV_MAX_BURST); |
58397e6c KT |
1655 | if (!nb_rx) { |
1656 | return EAGAIN; | |
1657 | } | |
1658 | ||
9509913a IS |
1659 | if (policer) { |
1660 | dropped = nb_rx; | |
64839cf4 WT |
1661 | nb_rx = ingress_policer_run(policer, |
1662 | (struct rte_mbuf **) batch->packets, | |
3e90f7d7 | 1663 | nb_rx, true); |
9509913a IS |
1664 | dropped -= nb_rx; |
1665 | } | |
1666 | ||
d46285a2 | 1667 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1668 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1669 | nb_rx, dropped); | |
d46285a2 | 1670 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1671 | |
75fb9148 ZB |
1672 | batch->count = nb_rx; |
1673 | dp_packet_batch_init_packet_fields(batch); | |
1674 | ||
58397e6c KT |
1675 | return 0; |
1676 | } | |
1677 | ||
8a9562d2 | 1678 | static int |
64839cf4 | 1679 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1680 | { |
d46285a2 DDP |
1681 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1682 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1683 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1684 | int nb_rx; |
9509913a | 1685 | int dropped = 0; |
8a9562d2 | 1686 | |
3b1fb077 DDP |
1687 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1688 | return EAGAIN; | |
1689 | } | |
1690 | ||
d46285a2 | 1691 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1692 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1693 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1694 | if (!nb_rx) { |
1695 | return EAGAIN; | |
1696 | } | |
1697 | ||
9509913a IS |
1698 | if (policer) { |
1699 | dropped = nb_rx; | |
64839cf4 | 1700 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 1701 | (struct rte_mbuf **) batch->packets, |
3e90f7d7 | 1702 | nb_rx, true); |
9509913a IS |
1703 | dropped -= nb_rx; |
1704 | } | |
1705 | ||
1706 | /* Update stats to reflect dropped packets */ | |
1707 | if (OVS_UNLIKELY(dropped)) { | |
1708 | rte_spinlock_lock(&dev->stats_lock); | |
1709 | dev->stats.rx_dropped += dropped; | |
1710 | rte_spinlock_unlock(&dev->stats_lock); | |
1711 | } | |
1712 | ||
64839cf4 | 1713 | batch->count = nb_rx; |
75fb9148 | 1714 | dp_packet_batch_init_packet_fields(batch); |
8a9562d2 PS |
1715 | |
1716 | return 0; | |
1717 | } | |
1718 | ||
0bf765f7 | 1719 | static inline int |
78bd47cf | 1720 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
3e90f7d7 | 1721 | int cnt, bool may_steal) |
0bf765f7 | 1722 | { |
78bd47cf | 1723 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 1724 | |
78bd47cf DDP |
1725 | if (qos_conf) { |
1726 | rte_spinlock_lock(&qos_conf->lock); | |
3e90f7d7 | 1727 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt, may_steal); |
78bd47cf | 1728 | rte_spinlock_unlock(&qos_conf->lock); |
0bf765f7 IS |
1729 | } |
1730 | ||
1731 | return cnt; | |
1732 | } | |
1733 | ||
c6ec9d17 IM |
1734 | static int |
1735 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1736 | int pkt_cnt) | |
1737 | { | |
1738 | int i = 0; | |
1739 | int cnt = 0; | |
1740 | struct rte_mbuf *pkt; | |
1741 | ||
1742 | for (i = 0; i < pkt_cnt; i++) { | |
1743 | pkt = pkts[i]; | |
1744 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
1745 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
1746 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
1747 | rte_pktmbuf_free(pkt); | |
1748 | continue; | |
1749 | } | |
1750 | ||
1751 | if (OVS_UNLIKELY(i != cnt)) { | |
1752 | pkts[cnt] = pkt; | |
1753 | } | |
1754 | cnt++; | |
1755 | } | |
1756 | ||
1757 | return cnt; | |
1758 | } | |
1759 | ||
9e3ddd45 TP |
1760 | static inline void |
1761 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1762 | struct dp_packet **packets, | |
1763 | int attempted, | |
1764 | int dropped) | |
1765 | { | |
1766 | int i; | |
1767 | int sent = attempted - dropped; | |
1768 | ||
1769 | stats->tx_packets += sent; | |
1770 | stats->tx_dropped += dropped; | |
1771 | ||
1772 | for (i = 0; i < sent; i++) { | |
1773 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1774 | } | |
1775 | } | |
1776 | ||
58397e6c | 1777 | static void |
4573fbd3 | 1778 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 1779 | struct dp_packet **pkts, int cnt) |
58397e6c | 1780 | { |
d46285a2 | 1781 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1782 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1783 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 1784 | unsigned int dropped = 0; |
dd52de45 | 1785 | int i, retries = 0; |
58397e6c | 1786 | |
81acebda | 1787 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1788 | |
0a0f39df | 1789 | if (OVS_UNLIKELY(!is_vhost_running(dev) || qid < 0 |
e543851d | 1790 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1791 | rte_spinlock_lock(&dev->stats_lock); |
1792 | dev->stats.tx_dropped+= cnt; | |
1793 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1794 | goto out; |
58397e6c KT |
1795 | } |
1796 | ||
d46285a2 | 1797 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1798 | |
c6ec9d17 | 1799 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 1800 | /* Check has QoS has been configured for the netdev */ |
3e90f7d7 | 1801 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt, true); |
c6ec9d17 | 1802 | dropped = total_pkts - cnt; |
0bf765f7 | 1803 | |
95e9881f | 1804 | do { |
4573fbd3 | 1805 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1806 | unsigned int tx_pkts; |
1807 | ||
0a0f39df | 1808 | tx_pkts = rte_vhost_enqueue_burst(netdev_dpdk_get_vid(dev), |
c6ec9d17 | 1809 | vhost_qid, cur_pkts, cnt); |
95e9881f KT |
1810 | if (OVS_LIKELY(tx_pkts)) { |
1811 | /* Packets have been sent.*/ | |
1812 | cnt -= tx_pkts; | |
31871ee3 | 1813 | /* Prepare for possible retry.*/ |
95e9881f KT |
1814 | cur_pkts = &cur_pkts[tx_pkts]; |
1815 | } else { | |
31871ee3 KT |
1816 | /* No packets sent - do not retry.*/ |
1817 | break; | |
95e9881f | 1818 | } |
c6ec9d17 | 1819 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1820 | |
d46285a2 | 1821 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1822 | |
d46285a2 | 1823 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 1824 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 1825 | cnt + dropped); |
d46285a2 | 1826 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1827 | |
1828 | out: | |
c6ec9d17 | 1829 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 1830 | dp_packet_delete(pkts[i]); |
58397e6c KT |
1831 | } |
1832 | } | |
1833 | ||
8a9562d2 PS |
1834 | /* Tx function. Transmit packets indefinitely */ |
1835 | static void | |
64839cf4 | 1836 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1837 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1838 | { |
8a14bd7b | 1839 | const size_t batch_cnt = dp_packet_batch_size(batch); |
bce01e3a | 1840 | #if !defined(__CHECKER__) && !defined(_WIN32) |
8a14bd7b | 1841 | const size_t PKT_ARRAY_SIZE = batch_cnt; |
bce01e3a EJ |
1842 | #else |
1843 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1844 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1845 | #endif |
8a9562d2 | 1846 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 1847 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
8a14bd7b | 1848 | uint32_t cnt = batch_cnt; |
3e90f7d7 GZ |
1849 | uint32_t dropped = 0; |
1850 | ||
1851 | if (dev->type != DPDK_DEV_VHOST) { | |
1852 | /* Check if QoS has been configured for this netdev. */ | |
1853 | cnt = netdev_dpdk_qos_run(dev, (struct rte_mbuf **) batch->packets, | |
8a14bd7b BB |
1854 | batch_cnt, false); |
1855 | dropped += batch_cnt - cnt; | |
3e90f7d7 | 1856 | } |
8a9562d2 | 1857 | |
7d6d1a40 WT |
1858 | dp_packet_batch_apply_cutlen(batch); |
1859 | ||
3e90f7d7 GZ |
1860 | uint32_t txcnt = 0; |
1861 | ||
1862 | for (uint32_t i = 0; i < cnt; i++) { | |
8a14bd7b BB |
1863 | struct dp_packet *packet = batch->packets[i]; |
1864 | uint32_t size = dp_packet_size(packet); | |
95fb793a | 1865 | |
f98d7864 | 1866 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
3e90f7d7 GZ |
1867 | VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d", |
1868 | size, dev->max_packet_len); | |
f4fd623c | 1869 | |
175cf4de | 1870 | dropped++; |
f4fd623c DDP |
1871 | continue; |
1872 | } | |
8a9562d2 | 1873 | |
3e90f7d7 | 1874 | pkts[txcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a14bd7b | 1875 | if (OVS_UNLIKELY(!pkts[txcnt])) { |
3e90f7d7 | 1876 | dropped += cnt - i; |
175cf4de | 1877 | break; |
f4fd623c DDP |
1878 | } |
1879 | ||
1880 | /* We have to do a copy for now */ | |
3e90f7d7 | 1881 | memcpy(rte_pktmbuf_mtod(pkts[txcnt], void *), |
8a14bd7b BB |
1882 | dp_packet_data(packet), size); |
1883 | dp_packet_set_size((struct dp_packet *)pkts[txcnt], size); | |
f4fd623c | 1884 | |
3e90f7d7 | 1885 | txcnt++; |
f4fd623c | 1886 | } |
8a9562d2 | 1887 | |
3e90f7d7 GZ |
1888 | if (OVS_LIKELY(txcnt)) { |
1889 | if (dev->type == DPDK_DEV_VHOST) { | |
1890 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, | |
1891 | txcnt); | |
1892 | } else { | |
1893 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, txcnt); | |
1894 | } | |
58397e6c | 1895 | } |
db73f716 | 1896 | |
0bf765f7 IS |
1897 | if (OVS_UNLIKELY(dropped)) { |
1898 | rte_spinlock_lock(&dev->stats_lock); | |
1899 | dev->stats.tx_dropped += dropped; | |
1900 | rte_spinlock_unlock(&dev->stats_lock); | |
1901 | } | |
8a9562d2 PS |
1902 | } |
1903 | ||
58397e6c | 1904 | static int |
64839cf4 WT |
1905 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1906 | struct dp_packet_batch *batch, | |
324c8374 | 1907 | bool may_steal, bool concurrent_txq OVS_UNUSED) |
58397e6c | 1908 | { |
58397e6c | 1909 | |
dd52de45 | 1910 | if (OVS_UNLIKELY(!may_steal || batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 WT |
1911 | dpdk_do_tx_copy(netdev, qid, batch); |
1912 | dp_packet_delete_batch(batch, may_steal); | |
58397e6c | 1913 | } else { |
64839cf4 | 1914 | dp_packet_batch_apply_cutlen(batch); |
dd52de45 | 1915 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
1916 | } |
1917 | return 0; | |
1918 | } | |
1919 | ||
7251515e DV |
1920 | static inline void |
1921 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
324c8374 IM |
1922 | struct dp_packet_batch *batch, bool may_steal, |
1923 | bool concurrent_txq) | |
8a9562d2 | 1924 | { |
3b1fb077 DDP |
1925 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1926 | dp_packet_delete_batch(batch, may_steal); | |
1927 | return; | |
1928 | } | |
1929 | ||
324c8374 | 1930 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 1931 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
1932 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
1933 | } | |
1934 | ||
7251515e | 1935 | if (OVS_UNLIKELY(!may_steal || |
64839cf4 | 1936 | batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1937 | struct netdev *netdev = &dev->up; |
1938 | ||
64839cf4 WT |
1939 | dpdk_do_tx_copy(netdev, qid, batch); |
1940 | dp_packet_delete_batch(batch, may_steal); | |
8a9562d2 | 1941 | } else { |
fd57eeba BB |
1942 | int tx_cnt, dropped; |
1943 | int batch_cnt = dp_packet_batch_size(batch); | |
2391135c | 1944 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 1945 | |
7d6d1a40 WT |
1946 | dp_packet_batch_apply_cutlen(batch); |
1947 | ||
fd57eeba BB |
1948 | tx_cnt = netdev_dpdk_filter_packet_len(dev, pkts, batch_cnt); |
1949 | tx_cnt = netdev_dpdk_qos_run(dev, pkts, tx_cnt, true); | |
1950 | dropped = batch_cnt - tx_cnt; | |
1b99bb05 | 1951 | |
fd57eeba | 1952 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, tx_cnt); |
8a9562d2 | 1953 | |
f4fd623c | 1954 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1955 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1956 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1957 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1958 | } |
8a9562d2 | 1959 | } |
a0cb2d66 | 1960 | |
324c8374 | 1961 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
1962 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
1963 | } | |
7251515e DV |
1964 | } |
1965 | ||
1966 | static int | |
1967 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
324c8374 IM |
1968 | struct dp_packet_batch *batch, bool may_steal, |
1969 | bool concurrent_txq) | |
7251515e DV |
1970 | { |
1971 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1972 | |
324c8374 | 1973 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e | 1974 | return 0; |
8a9562d2 PS |
1975 | } |
1976 | ||
1977 | static int | |
74ff3298 | 1978 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1979 | { |
1980 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1981 | ||
1982 | ovs_mutex_lock(&dev->mutex); | |
1983 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1984 | dev->hwaddr = mac; |
045c0d1a | 1985 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1986 | } |
1987 | ovs_mutex_unlock(&dev->mutex); | |
1988 | ||
1989 | return 0; | |
1990 | } | |
1991 | ||
1992 | static int | |
74ff3298 | 1993 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1994 | { |
1995 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1996 | ||
1997 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1998 | *mac = dev->hwaddr; |
8a9562d2 PS |
1999 | ovs_mutex_unlock(&dev->mutex); |
2000 | ||
2001 | return 0; | |
2002 | } | |
2003 | ||
2004 | static int | |
2005 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
2006 | { | |
2007 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2008 | ||
2009 | ovs_mutex_lock(&dev->mutex); | |
2010 | *mtup = dev->mtu; | |
2011 | ovs_mutex_unlock(&dev->mutex); | |
2012 | ||
2013 | return 0; | |
2014 | } | |
2015 | ||
0072e931 MK |
2016 | static int |
2017 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
2018 | { | |
2019 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2020 | ||
2021 | if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
2022 | || mtu < ETHER_MIN_MTU) { | |
2023 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
2024 | return EINVAL; | |
2025 | } | |
2026 | ||
2027 | ovs_mutex_lock(&dev->mutex); | |
2028 | if (dev->requested_mtu != mtu) { | |
2029 | dev->requested_mtu = mtu; | |
2030 | netdev_request_reconfigure(netdev); | |
2031 | } | |
2032 | ovs_mutex_unlock(&dev->mutex); | |
2033 | ||
2034 | return 0; | |
2035 | } | |
2036 | ||
8a9562d2 | 2037 | static int |
d46285a2 | 2038 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 2039 | |
58397e6c KT |
2040 | static int |
2041 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
2042 | struct netdev_stats *stats) | |
2043 | { | |
2044 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2045 | ||
2046 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 2047 | |
45d947c4 | 2048 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c | 2049 | /* Supported Stats */ |
50986e78 | 2050 | stats->rx_packets = dev->stats.rx_packets; |
2051 | stats->tx_packets = dev->stats.tx_packets; | |
9509913a | 2052 | stats->rx_dropped = dev->stats.rx_dropped; |
50986e78 | 2053 | stats->tx_dropped = dev->stats.tx_dropped; |
9e3ddd45 TP |
2054 | stats->multicast = dev->stats.multicast; |
2055 | stats->rx_bytes = dev->stats.rx_bytes; | |
2056 | stats->tx_bytes = dev->stats.tx_bytes; | |
2057 | stats->rx_errors = dev->stats.rx_errors; | |
2058 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 2059 | |
2060 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
2061 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
2062 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
2063 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
2064 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
2065 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
2066 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
2067 | ||
45d947c4 | 2068 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 2069 | |
58397e6c KT |
2070 | ovs_mutex_unlock(&dev->mutex); |
2071 | ||
2072 | return 0; | |
2073 | } | |
2074 | ||
d6e3feb5 | 2075 | static void |
2076 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
2077 | const struct rte_eth_xstat *xstats, |
2078 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 2079 | const unsigned int size) |
2080 | { | |
d6e3feb5 | 2081 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 2082 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2083 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2084 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2085 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2086 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2087 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2088 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2089 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2090 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2091 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2092 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2093 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2094 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2095 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 2096 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2097 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2098 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2099 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2100 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2101 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2102 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2103 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2104 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2105 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2106 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2107 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2108 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2109 | stats->tx_1523_to_max_packets = xstats[i].value; |
d57f777f PS |
2110 | } else if (strcmp(XSTAT_RX_MULTICAST_PACKETS, names[i].name) == 0) { |
2111 | stats->multicast = xstats[i].value; | |
0a0f39df | 2112 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2113 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 2114 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2115 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2116 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2117 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2118 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2119 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2120 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2121 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2122 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2123 | stats->rx_jabber_errors = xstats[i].value; |
2124 | } | |
2125 | } | |
2126 | } | |
2127 | ||
8a9562d2 PS |
2128 | static int |
2129 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2130 | { | |
2131 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2132 | struct rte_eth_stats rte_stats; | |
2133 | bool gg; | |
2134 | ||
2135 | netdev_dpdk_get_carrier(netdev, &gg); | |
2136 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2137 | |
0a0f39df CL |
2138 | struct rte_eth_xstat *rte_xstats = NULL; |
2139 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2140 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2141 | |
2142 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
bb37956a | 2143 | VLOG_ERR("Can't get ETH statistics for port: %"PRIu8, dev->port_id); |
f9256822 | 2144 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2145 | return EPROTO; |
2146 | } | |
2147 | ||
0a0f39df CL |
2148 | /* Get length of statistics */ |
2149 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2150 | if (rte_xstats_len < 0) { | |
bb37956a | 2151 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2152 | goto out; |
2153 | } | |
2154 | /* Reserve memory for xstats names and values */ | |
2155 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2156 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2157 | ||
2158 | /* Retreive xstats names */ | |
2159 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2160 | rte_xstats_names, | |
2161 | rte_xstats_len); | |
2162 | if (rte_xstats_new_len != rte_xstats_len) { | |
bb37956a | 2163 | VLOG_WARN("Cannot get XSTATS names for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2164 | goto out; |
2165 | } | |
2166 | /* Retreive xstats values */ | |
2167 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2168 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2169 | rte_xstats_len); | |
2170 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2171 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2172 | rte_xstats_len); | |
d6e3feb5 | 2173 | } else { |
bb37956a | 2174 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
d6e3feb5 | 2175 | } |
8a9562d2 | 2176 | |
0a0f39df CL |
2177 | out: |
2178 | free(rte_xstats); | |
2179 | free(rte_xstats_names); | |
2180 | ||
2f9dd77f PS |
2181 | stats->rx_packets = rte_stats.ipackets; |
2182 | stats->tx_packets = rte_stats.opackets; | |
2183 | stats->rx_bytes = rte_stats.ibytes; | |
2184 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2185 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2186 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2187 | |
45d947c4 | 2188 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2189 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2190 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2191 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2192 | |
2193 | /* These are the available DPDK counters for packets not received due to | |
2194 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2195 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2196 | stats->rx_missed_errors = rte_stats.imissed; |
2197 | ||
8a9562d2 PS |
2198 | ovs_mutex_unlock(&dev->mutex); |
2199 | ||
2200 | return 0; | |
2201 | } | |
2202 | ||
2203 | static int | |
d46285a2 | 2204 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2205 | enum netdev_features *current, |
ca3d4f55 BX |
2206 | enum netdev_features *advertised, |
2207 | enum netdev_features *supported, | |
2208 | enum netdev_features *peer) | |
8a9562d2 | 2209 | { |
d46285a2 | 2210 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2211 | struct rte_eth_link link; |
2212 | ||
2213 | ovs_mutex_lock(&dev->mutex); | |
2214 | link = dev->link; | |
2215 | ovs_mutex_unlock(&dev->mutex); | |
2216 | ||
362ca396 | 2217 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2218 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
2219 | *current = NETDEV_F_10MB_HD; |
2220 | } | |
362ca396 | 2221 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2222 | *current = NETDEV_F_100MB_HD; |
2223 | } | |
362ca396 | 2224 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2225 | *current = NETDEV_F_1GB_HD; |
2226 | } | |
2227 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 2228 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
2229 | *current = NETDEV_F_10MB_FD; |
2230 | } | |
362ca396 | 2231 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2232 | *current = NETDEV_F_100MB_FD; |
2233 | } | |
362ca396 | 2234 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2235 | *current = NETDEV_F_1GB_FD; |
2236 | } | |
362ca396 | 2237 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
2238 | *current = NETDEV_F_10GB_FD; |
2239 | } | |
2240 | } | |
2241 | ||
362ca396 | 2242 | if (link.link_autoneg) { |
2243 | *current |= NETDEV_F_AUTONEG; | |
2244 | } | |
2245 | ||
ca3d4f55 BX |
2246 | *advertised = *supported = *peer = 0; |
2247 | ||
8a9562d2 PS |
2248 | return 0; |
2249 | } | |
2250 | ||
9509913a IS |
2251 | static struct ingress_policer * |
2252 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2253 | { | |
2254 | struct ingress_policer *policer = NULL; | |
2255 | uint64_t rate_bytes; | |
2256 | uint64_t burst_bytes; | |
2257 | int err = 0; | |
2258 | ||
2259 | policer = xmalloc(sizeof *policer); | |
2260 | rte_spinlock_init(&policer->policer_lock); | |
2261 | ||
2262 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
602c8668 LR |
2263 | rate_bytes = rate * 1000ULL / 8; |
2264 | burst_bytes = burst * 1000ULL / 8; | |
9509913a IS |
2265 | |
2266 | policer->app_srtcm_params.cir = rate_bytes; | |
2267 | policer->app_srtcm_params.cbs = burst_bytes; | |
2268 | policer->app_srtcm_params.ebs = 0; | |
2269 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2270 | &policer->app_srtcm_params); | |
58be5c0e | 2271 | if (err) { |
9509913a IS |
2272 | VLOG_ERR("Could not create rte meter for ingress policer"); |
2273 | return NULL; | |
2274 | } | |
2275 | ||
2276 | return policer; | |
2277 | } | |
2278 | ||
2279 | static int | |
2280 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2281 | uint32_t policer_burst) | |
2282 | { | |
2283 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2284 | struct ingress_policer *policer; | |
2285 | ||
2286 | /* Force to 0 if no rate specified, | |
2287 | * default to 8000 kbits if burst is 0, | |
2288 | * else stick with user-specified value. | |
2289 | */ | |
2290 | policer_burst = (!policer_rate ? 0 | |
2291 | : !policer_burst ? 8000 | |
2292 | : policer_burst); | |
2293 | ||
2294 | ovs_mutex_lock(&dev->mutex); | |
2295 | ||
2296 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2297 | &dev->ingress_policer); | |
2298 | ||
2299 | if (dev->policer_rate == policer_rate && | |
2300 | dev->policer_burst == policer_burst) { | |
2301 | /* Assume that settings haven't changed since we last set them. */ | |
2302 | ovs_mutex_unlock(&dev->mutex); | |
2303 | return 0; | |
2304 | } | |
2305 | ||
2306 | /* Destroy any existing ingress policer for the device if one exists */ | |
2307 | if (policer) { | |
2308 | ovsrcu_postpone(free, policer); | |
2309 | } | |
2310 | ||
2311 | if (policer_rate != 0) { | |
2312 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2313 | } else { | |
2314 | policer = NULL; | |
2315 | } | |
2316 | ovsrcu_set(&dev->ingress_policer, policer); | |
2317 | dev->policer_rate = policer_rate; | |
2318 | dev->policer_burst = policer_burst; | |
2319 | ovs_mutex_unlock(&dev->mutex); | |
2320 | ||
2321 | return 0; | |
2322 | } | |
2323 | ||
8a9562d2 PS |
2324 | static int |
2325 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2326 | { | |
2327 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
2328 | |
2329 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
2330 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
2331 | * postive integer to meet RFC 2863 recommendations. | |
2332 | */ | |
2333 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
2334 | ovs_mutex_unlock(&dev->mutex); |
2335 | ||
2336 | return ifindex; | |
2337 | } | |
2338 | ||
2339 | static int | |
d46285a2 | 2340 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2341 | { |
d46285a2 | 2342 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2343 | |
2344 | ovs_mutex_lock(&dev->mutex); | |
2345 | check_link_status(dev); | |
2346 | *carrier = dev->link.link_status; | |
58397e6c KT |
2347 | |
2348 | ovs_mutex_unlock(&dev->mutex); | |
2349 | ||
2350 | return 0; | |
2351 | } | |
2352 | ||
2353 | static int | |
d46285a2 | 2354 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2355 | { |
d46285a2 | 2356 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2357 | |
2358 | ovs_mutex_lock(&dev->mutex); | |
2359 | ||
0a0f39df | 2360 | if (is_vhost_running(dev)) { |
58397e6c KT |
2361 | *carrier = 1; |
2362 | } else { | |
2363 | *carrier = 0; | |
2364 | } | |
2365 | ||
8a9562d2 PS |
2366 | ovs_mutex_unlock(&dev->mutex); |
2367 | ||
2368 | return 0; | |
2369 | } | |
2370 | ||
2371 | static long long int | |
d46285a2 | 2372 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2373 | { |
d46285a2 | 2374 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2375 | long long int carrier_resets; |
2376 | ||
2377 | ovs_mutex_lock(&dev->mutex); | |
2378 | carrier_resets = dev->link_reset_cnt; | |
2379 | ovs_mutex_unlock(&dev->mutex); | |
2380 | ||
2381 | return carrier_resets; | |
2382 | } | |
2383 | ||
2384 | static int | |
d46285a2 | 2385 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2386 | long long int interval OVS_UNUSED) |
2387 | { | |
ee32150e | 2388 | return EOPNOTSUPP; |
8a9562d2 PS |
2389 | } |
2390 | ||
2391 | static int | |
2392 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2393 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2394 | enum netdev_flags *old_flagsp) |
2395 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 2396 | { |
8a9562d2 PS |
2397 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
2398 | return EINVAL; | |
2399 | } | |
2400 | ||
2401 | *old_flagsp = dev->flags; | |
2402 | dev->flags |= on; | |
2403 | dev->flags &= ~off; | |
2404 | ||
2405 | if (dev->flags == *old_flagsp) { | |
2406 | return 0; | |
2407 | } | |
2408 | ||
58397e6c | 2409 | if (dev->type == DPDK_DEV_ETH) { |
58397e6c KT |
2410 | if (dev->flags & NETDEV_PROMISC) { |
2411 | rte_eth_promiscuous_enable(dev->port_id); | |
2412 | } | |
8a9562d2 | 2413 | |
314fb5ad | 2414 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
2415 | } else { |
2416 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2417 | * running then change netdev's change_seq to trigger link state | |
2418 | * update. */ | |
e543851d ZB |
2419 | |
2420 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2421 | && is_vhost_running(dev)) { |
e543851d ZB |
2422 | netdev_change_seq_changed(&dev->up); |
2423 | ||
2424 | /* Clear statistics if device is getting up. */ | |
2425 | if (NETDEV_UP & on) { | |
2426 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 2427 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
2428 | rte_spinlock_unlock(&dev->stats_lock); |
2429 | } | |
2430 | } | |
8a9562d2 PS |
2431 | } |
2432 | ||
2433 | return 0; | |
2434 | } | |
2435 | ||
2436 | static int | |
d46285a2 | 2437 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2438 | enum netdev_flags off, enum netdev_flags on, |
2439 | enum netdev_flags *old_flagsp) | |
2440 | { | |
d46285a2 | 2441 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2442 | int error; |
2443 | ||
d46285a2 DDP |
2444 | ovs_mutex_lock(&dev->mutex); |
2445 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2446 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2447 | |
2448 | return error; | |
2449 | } | |
2450 | ||
2451 | static int | |
d46285a2 | 2452 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2453 | { |
d46285a2 | 2454 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2455 | struct rte_eth_dev_info dev_info; |
2456 | ||
7cd1261d | 2457 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2458 | return ENODEV; |
7cd1261d | 2459 | } |
8a9562d2 PS |
2460 | |
2461 | ovs_mutex_lock(&dev->mutex); | |
2462 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2463 | ovs_mutex_unlock(&dev->mutex); | |
2464 | ||
95fb793a | 2465 | smap_add_format(args, "port_no", "%d", dev->port_id); |
58be5c0e MK |
2466 | smap_add_format(args, "numa_id", "%d", |
2467 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
2468 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
2469 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2470 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2471 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2472 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2473 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
2474 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
2475 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
2476 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
2477 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2478 | ||
39c2baa9 | 2479 | if (dev_info.pci_dev) { |
2480 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2481 | dev_info.pci_dev->id.vendor_id); | |
2482 | smap_add_format(args, "pci-device_id", "0x%x", | |
2483 | dev_info.pci_dev->id.device_id); | |
2484 | } | |
8a9562d2 PS |
2485 | |
2486 | return 0; | |
2487 | } | |
2488 | ||
2489 | static void | |
2490 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2491 | OVS_REQUIRES(dev->mutex) | |
2492 | { | |
2493 | enum netdev_flags old_flags; | |
2494 | ||
2495 | if (admin_state) { | |
2496 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2497 | } else { | |
2498 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2499 | } | |
2500 | } | |
2501 | ||
2502 | static void | |
2503 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2504 | const char *argv[], void *aux OVS_UNUSED) | |
2505 | { | |
2506 | bool up; | |
2507 | ||
2508 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2509 | up = true; | |
2510 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2511 | up = false; | |
2512 | } else { | |
2513 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2514 | return; | |
2515 | } | |
2516 | ||
2517 | if (argc > 2) { | |
2518 | struct netdev *netdev = netdev_from_name(argv[1]); | |
2519 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
2520 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
2521 | ||
2522 | ovs_mutex_lock(&dpdk_dev->mutex); | |
2523 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
2524 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
2525 | ||
2526 | netdev_close(netdev); | |
2527 | } else { | |
2528 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2529 | netdev_close(netdev); | |
2530 | return; | |
2531 | } | |
2532 | } else { | |
2533 | struct netdev_dpdk *netdev; | |
2534 | ||
2535 | ovs_mutex_lock(&dpdk_mutex); | |
2536 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
2537 | ovs_mutex_lock(&netdev->mutex); | |
2538 | netdev_dpdk_set_admin_state__(netdev, up); | |
2539 | ovs_mutex_unlock(&netdev->mutex); | |
2540 | } | |
2541 | ovs_mutex_unlock(&dpdk_mutex); | |
2542 | } | |
2543 | unixctl_command_reply(conn, "OK"); | |
2544 | } | |
2545 | ||
0ee821c2 DB |
2546 | static void |
2547 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2548 | const char *argv[], void *aux OVS_UNUSED) | |
2549 | { | |
2550 | int ret; | |
2551 | char *response; | |
2552 | uint8_t port_id; | |
2553 | char devname[RTE_ETH_NAME_MAX_LEN]; | |
2554 | struct netdev_dpdk *dev; | |
2555 | ||
2556 | ovs_mutex_lock(&dpdk_mutex); | |
2557 | ||
2558 | if (!rte_eth_dev_count() || rte_eth_dev_get_port_by_name(argv[1], | |
2559 | &port_id)) { | |
2560 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); | |
2561 | goto error; | |
2562 | } | |
2563 | ||
2564 | dev = netdev_dpdk_lookup_by_port_id(port_id); | |
2565 | if (dev) { | |
2566 | response = xasprintf("Device '%s' is being used by interface '%s'. " | |
2567 | "Remove it before detaching", | |
2568 | argv[1], netdev_get_name(&dev->up)); | |
2569 | goto error; | |
2570 | } | |
2571 | ||
2572 | rte_eth_dev_close(port_id); | |
2573 | ||
2574 | ret = rte_eth_dev_detach(port_id, devname); | |
2575 | if (ret < 0) { | |
2576 | response = xasprintf("Device '%s' can not be detached", argv[1]); | |
2577 | goto error; | |
2578 | } | |
2579 | ||
2580 | response = xasprintf("Device '%s' has been detached", argv[1]); | |
2581 | ||
2582 | ovs_mutex_unlock(&dpdk_mutex); | |
2583 | unixctl_command_reply(conn, response); | |
2584 | free(response); | |
2585 | return; | |
2586 | ||
2587 | error: | |
2588 | ovs_mutex_unlock(&dpdk_mutex); | |
2589 | unixctl_command_reply_error(conn, response); | |
2590 | free(response); | |
2591 | } | |
2592 | ||
58397e6c KT |
2593 | /* |
2594 | * Set virtqueue flags so that we do not receive interrupts. | |
2595 | */ | |
2596 | static void | |
0a0f39df | 2597 | set_irq_status(int vid) |
58397e6c | 2598 | { |
4573fbd3 | 2599 | uint32_t i; |
4573fbd3 | 2600 | |
f3e7ec25 MW |
2601 | for (i = 0; i < rte_vhost_get_vring_num(vid); i++) { |
2602 | rte_vhost_enable_guest_notification(vid, i, 0); | |
4573fbd3 FL |
2603 | } |
2604 | } | |
2605 | ||
585a5bea IM |
2606 | /* |
2607 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2608 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2609 | */ |
2610 | static void | |
d46285a2 DDP |
2611 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2612 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2613 | { |
2614 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2615 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 2616 | |
eff23640 | 2617 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
2618 | |
2619 | for (i = 0; i < total_txqs; i++) { | |
2620 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2621 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2622 | enabled_queues[n_enabled++] = i; |
2623 | } | |
2624 | } | |
2625 | ||
2626 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2627 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2628 | n_enabled = 1; |
2629 | } | |
2630 | ||
2631 | k = 0; | |
2632 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2633 | if (dev->tx_q[i].map != i) { |
2634 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2635 | k = (k + 1) % n_enabled; |
2636 | } | |
2637 | } | |
2638 | ||
2d24d165 | 2639 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2640 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2641 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2642 | } |
2643 | ||
eff23640 | 2644 | free(enabled_queues); |
585a5bea | 2645 | } |
4573fbd3 | 2646 | |
58397e6c KT |
2647 | /* |
2648 | * A new virtio-net device is added to a vhost port. | |
2649 | */ | |
2650 | static int | |
0a0f39df | 2651 | new_device(int vid) |
58397e6c | 2652 | { |
d46285a2 | 2653 | struct netdev_dpdk *dev; |
58397e6c | 2654 | bool exists = false; |
db8f13b0 | 2655 | int newnode = 0; |
0a0f39df CL |
2656 | char ifname[IF_NAME_SZ]; |
2657 | ||
58be5c0e | 2658 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2659 | |
2660 | ovs_mutex_lock(&dpdk_mutex); | |
2661 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2662 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 2663 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2664 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
f3e7ec25 | 2665 | uint32_t qp_num = rte_vhost_get_vring_num(vid)/VIRTIO_QNUM; |
db8f13b0 CL |
2666 | |
2667 | /* Get NUMA information */ | |
0a0f39df CL |
2668 | newnode = rte_vhost_get_numa_node(vid); |
2669 | if (newnode == -1) { | |
5b9bf9e0 | 2670 | #ifdef VHOST_NUMA |
db8f13b0 | 2671 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2672 | ifname); |
5b9bf9e0 | 2673 | #endif |
db8f13b0 | 2674 | newnode = dev->socket_id; |
db8f13b0 CL |
2675 | } |
2676 | ||
7f5f2bd0 IM |
2677 | if (dev->requested_n_txq != qp_num |
2678 | || dev->requested_n_rxq != qp_num | |
2679 | || dev->requested_socket_id != newnode) { | |
2680 | dev->requested_socket_id = newnode; | |
2681 | dev->requested_n_rxq = qp_num; | |
2682 | dev->requested_n_txq = qp_num; | |
2683 | netdev_request_reconfigure(&dev->up); | |
2684 | } else { | |
2685 | /* Reconfiguration not required. */ | |
2686 | dev->vhost_reconfigured = true; | |
2687 | } | |
81acebda | 2688 | |
0a0f39df | 2689 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2690 | exists = true; |
2691 | ||
58397e6c | 2692 | /* Disable notifications. */ |
0a0f39df | 2693 | set_irq_status(vid); |
e543851d | 2694 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2695 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2696 | break; |
2697 | } | |
c1ff66ac | 2698 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2699 | } |
2700 | ovs_mutex_unlock(&dpdk_mutex); | |
2701 | ||
2702 | if (!exists) { | |
0a0f39df | 2703 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2704 | |
2705 | return -1; | |
2706 | } | |
2707 | ||
0a0f39df CL |
2708 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2709 | ifname, newnode); | |
2710 | ||
58397e6c KT |
2711 | return 0; |
2712 | } | |
2713 | ||
f3ea2ad2 IM |
2714 | /* Clears mapping for all available queues of vhost interface. */ |
2715 | static void | |
2716 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2717 | OVS_REQUIRES(dev->mutex) | |
2718 | { | |
2719 | int i; | |
2720 | ||
81acebda | 2721 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
2722 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
2723 | } | |
2724 | } | |
2725 | ||
58397e6c KT |
2726 | /* |
2727 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2728 | * flag to stop any more packets from being sent or received to/from a VM and | |
2729 | * ensure all currently queued packets have been sent/received before removing | |
2730 | * the device. | |
2731 | */ | |
2732 | static void | |
0a0f39df | 2733 | destroy_device(int vid) |
58397e6c | 2734 | { |
d46285a2 | 2735 | struct netdev_dpdk *dev; |
afee281f | 2736 | bool exists = false; |
0a0f39df CL |
2737 | char ifname[IF_NAME_SZ]; |
2738 | ||
58be5c0e | 2739 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2740 | |
2741 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2742 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2743 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 2744 | |
d46285a2 | 2745 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
2746 | dev->vhost_reconfigured = false; |
2747 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 2748 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 2749 | |
e543851d | 2750 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2751 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 2752 | exists = true; |
afee281f | 2753 | break; |
58397e6c KT |
2754 | } |
2755 | } | |
afee281f | 2756 | |
58397e6c KT |
2757 | ovs_mutex_unlock(&dpdk_mutex); |
2758 | ||
0a0f39df | 2759 | if (exists) { |
afee281f KT |
2760 | /* |
2761 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2762 | * to NULL, before returning. | |
2763 | */ | |
2764 | ovsrcu_synchronize(); | |
2765 | /* | |
2766 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2767 | * put thread back into quiescent state before returning. | |
2768 | */ | |
2769 | ovsrcu_quiesce_start(); | |
0a0f39df | 2770 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 2771 | } else { |
0a0f39df | 2772 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 2773 | } |
58397e6c KT |
2774 | } |
2775 | ||
585a5bea | 2776 | static int |
0a0f39df | 2777 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 2778 | { |
d46285a2 | 2779 | struct netdev_dpdk *dev; |
585a5bea IM |
2780 | bool exists = false; |
2781 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
2782 | char ifname[IF_NAME_SZ]; |
2783 | ||
58be5c0e | 2784 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
2785 | |
2786 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2787 | return 0; | |
2788 | } | |
2789 | ||
2790 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2791 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 2792 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2793 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 2794 | if (enable) { |
d46285a2 | 2795 | dev->tx_q[qid].map = qid; |
585a5bea | 2796 | } else { |
d46285a2 | 2797 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2798 | } |
d46285a2 | 2799 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2800 | exists = true; |
d46285a2 | 2801 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2802 | break; |
2803 | } | |
c1ff66ac | 2804 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2805 | } |
2806 | ovs_mutex_unlock(&dpdk_mutex); | |
2807 | ||
2808 | if (exists) { | |
0a0f39df CL |
2809 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
2810 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 2811 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 2812 | } else { |
0a0f39df | 2813 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
2814 | return -1; |
2815 | } | |
2816 | ||
2817 | return 0; | |
2818 | } | |
2819 | ||
0a0f39df CL |
2820 | int |
2821 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 2822 | { |
0a0f39df | 2823 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
2824 | } |
2825 | ||
9509913a IS |
2826 | struct ingress_policer * |
2827 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
2828 | { | |
2829 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
2830 | } | |
2831 | ||
58397e6c | 2832 | static int |
ecc1a34e | 2833 | netdev_dpdk_class_init(void) |
7d1ced01 | 2834 | { |
ecc1a34e DDP |
2835 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2836 | ||
2837 | /* This function can be called for different classes. The initialization | |
2838 | * needs to be done only once */ | |
2839 | if (ovsthread_once_start(&once)) { | |
2840 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
2841 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2842 | "[netdev] up|down", 1, 2, | |
2843 | netdev_dpdk_set_admin_state, NULL); | |
2844 | ||
0ee821c2 DB |
2845 | unixctl_command_register("netdev-dpdk/detach", |
2846 | "pci address of device", 1, 1, | |
2847 | netdev_dpdk_detach, NULL); | |
2848 | ||
ecc1a34e DDP |
2849 | ovsthread_once_done(&once); |
2850 | } | |
362ca396 | 2851 | |
7d1ced01 CL |
2852 | return 0; |
2853 | } | |
2854 | ||
033e9df2 | 2855 | |
95fb793a | 2856 | /* Client Rings */ |
2857 | ||
95fb793a | 2858 | static int |
2859 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
bb37956a | 2860 | dpdk_port_t *eth_port_id) |
95fb793a | 2861 | { |
48fffdee | 2862 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 2863 | char *ring_name; |
b8374d0d | 2864 | int port_id; |
95fb793a | 2865 | |
48fffdee KT |
2866 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
2867 | if (!ring_pair) { | |
95fb793a | 2868 | return ENOMEM; |
2869 | } | |
2870 | ||
7251515e | 2871 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 2872 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 2873 | |
8f0a76c9 | 2874 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 2875 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2876 | RING_F_SP_ENQ); |
0c6f39e5 | 2877 | free(ring_name); |
48fffdee KT |
2878 | if (ring_pair->cring_tx == NULL) { |
2879 | rte_free(ring_pair); | |
95fb793a | 2880 | return ENOMEM; |
2881 | } | |
2882 | ||
0c6f39e5 | 2883 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 2884 | |
8f0a76c9 | 2885 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 2886 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2887 | RING_F_SC_DEQ); |
0c6f39e5 | 2888 | free(ring_name); |
48fffdee KT |
2889 | if (ring_pair->cring_rx == NULL) { |
2890 | rte_free(ring_pair); | |
95fb793a | 2891 | return ENOMEM; |
2892 | } | |
2893 | ||
b8374d0d MV |
2894 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
2895 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 2896 | |
b8374d0d | 2897 | if (port_id < 0) { |
48fffdee | 2898 | rte_free(ring_pair); |
95fb793a | 2899 | return ENODEV; |
2900 | } | |
2901 | ||
48fffdee | 2902 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
2903 | ring_pair->eth_port_id = port_id; |
2904 | *eth_port_id = port_id; | |
2905 | ||
48fffdee | 2906 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 2907 | |
95fb793a | 2908 | return 0; |
2909 | } | |
2910 | ||
2911 | static int | |
bb37956a | 2912 | dpdk_ring_open(const char dev_name[], dpdk_port_t *eth_port_id) |
64839cf4 | 2913 | OVS_REQUIRES(dpdk_mutex) |
95fb793a | 2914 | { |
48fffdee | 2915 | struct dpdk_ring *ring_pair; |
95fb793a | 2916 | unsigned int port_no; |
2917 | int err = 0; | |
2918 | ||
2919 | /* Names always start with "dpdkr" */ | |
2920 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2921 | if (err) { | |
2922 | return err; | |
2923 | } | |
2924 | ||
58be5c0e | 2925 | /* Look through our list to find the device */ |
48fffdee KT |
2926 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
2927 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 2928 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 2929 | /* Really all that is needed */ |
48fffdee | 2930 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 2931 | return 0; |
2932 | } | |
2933 | } | |
2934 | /* Need to create the device rings */ | |
2935 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2936 | } | |
2937 | ||
7251515e | 2938 | static int |
d46285a2 | 2939 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
324c8374 IM |
2940 | struct dp_packet_batch *batch, bool may_steal, |
2941 | bool concurrent_txq) | |
7251515e | 2942 | { |
d46285a2 | 2943 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a543eb0 | 2944 | struct dp_packet *packet; |
1b99bb05 | 2945 | |
58be5c0e MK |
2946 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
2947 | * the rss hash field is clear. This is because the same mbuf may be | |
2948 | * modified by the consumer of the ring and return into the datapath | |
2949 | * without recalculating the RSS hash. */ | |
8a543eb0 BB |
2950 | DP_PACKET_BATCH_FOR_EACH (packet, batch) { |
2951 | dp_packet_mbuf_rss_flag_reset(packet); | |
1b99bb05 | 2952 | } |
7251515e | 2953 | |
324c8374 | 2954 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e DV |
2955 | return 0; |
2956 | } | |
2957 | ||
95fb793a | 2958 | static int |
2959 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2960 | { | |
bb37956a | 2961 | dpdk_port_t port_no = 0; |
95fb793a | 2962 | int err = 0; |
2963 | ||
95fb793a | 2964 | ovs_mutex_lock(&dpdk_mutex); |
2965 | ||
2966 | err = dpdk_ring_open(netdev->name, &port_no); | |
2967 | if (err) { | |
2968 | goto unlock_dpdk; | |
2969 | } | |
2970 | ||
1ce30dfd DDP |
2971 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
2972 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 2973 | unlock_dpdk: |
2974 | ovs_mutex_unlock(&dpdk_mutex); | |
2975 | return err; | |
2976 | } | |
2977 | ||
0bf765f7 IS |
2978 | /* QoS Functions */ |
2979 | ||
2980 | /* | |
2981 | * Initialize QoS configuration operations. | |
2982 | */ | |
2983 | static void | |
2984 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
2985 | { | |
2986 | conf->ops = ops; | |
78bd47cf | 2987 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
2988 | } |
2989 | ||
2990 | /* | |
2991 | * Search existing QoS operations in qos_ops and compare each set of | |
2992 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
2993 | * else return NULL | |
2994 | */ | |
2995 | static const struct dpdk_qos_ops * | |
2996 | qos_lookup_name(const char *name) | |
2997 | { | |
2998 | const struct dpdk_qos_ops *const *opsp; | |
2999 | ||
3000 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3001 | const struct dpdk_qos_ops *ops = *opsp; | |
3002 | if (!strcmp(name, ops->qos_name)) { | |
3003 | return ops; | |
3004 | } | |
3005 | } | |
3006 | return NULL; | |
3007 | } | |
3008 | ||
0bf765f7 IS |
3009 | static int |
3010 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
3011 | struct sset *types) | |
3012 | { | |
3013 | const struct dpdk_qos_ops *const *opsp; | |
3014 | ||
3015 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3016 | const struct dpdk_qos_ops *ops = *opsp; | |
3017 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
3018 | sset_add(types, ops->qos_name); | |
3019 | } | |
3020 | } | |
3021 | return 0; | |
3022 | } | |
3023 | ||
3024 | static int | |
d46285a2 | 3025 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
3026 | const char **typep, struct smap *details) |
3027 | { | |
d46285a2 | 3028 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 3029 | struct qos_conf *qos_conf; |
0bf765f7 IS |
3030 | int error = 0; |
3031 | ||
d46285a2 | 3032 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
3033 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
3034 | if (qos_conf) { | |
3035 | *typep = qos_conf->ops->qos_name; | |
3036 | error = (qos_conf->ops->qos_get | |
3037 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
3038 | } else { |
3039 | /* No QoS configuration set, return an empty string */ | |
3040 | *typep = ""; | |
0bf765f7 | 3041 | } |
d46285a2 | 3042 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
3043 | |
3044 | return error; | |
3045 | } | |
3046 | ||
3047 | static int | |
78bd47cf DDP |
3048 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
3049 | const struct smap *details) | |
0bf765f7 | 3050 | { |
d46285a2 | 3051 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 3052 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 3053 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
3054 | int error = 0; |
3055 | ||
d46285a2 | 3056 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 3057 | |
78bd47cf | 3058 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 3059 | |
78bd47cf DDP |
3060 | new_ops = qos_lookup_name(type); |
3061 | ||
3062 | if (!new_ops || !new_ops->qos_construct) { | |
3063 | new_qos_conf = NULL; | |
3064 | if (type && type[0]) { | |
3065 | error = EOPNOTSUPP; | |
0bf765f7 | 3066 | } |
44975bb0 | 3067 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
3068 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
3069 | new_qos_conf = qos_conf; | |
0bf765f7 | 3070 | } else { |
78bd47cf | 3071 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
3072 | } |
3073 | ||
7ea266e9 | 3074 | if (error) { |
78bd47cf DDP |
3075 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
3076 | type, netdev->name, rte_strerror(error)); | |
3077 | } | |
3078 | ||
3079 | if (new_qos_conf != qos_conf) { | |
3080 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
3081 | if (qos_conf) { | |
3082 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
3083 | } | |
0bf765f7 IS |
3084 | } |
3085 | ||
d46285a2 | 3086 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 3087 | |
0bf765f7 IS |
3088 | return error; |
3089 | } | |
3090 | ||
3091 | /* egress-policer details */ | |
3092 | ||
3093 | struct egress_policer { | |
3094 | struct qos_conf qos_conf; | |
3095 | struct rte_meter_srtcm_params app_srtcm_params; | |
3096 | struct rte_meter_srtcm egress_meter; | |
3097 | }; | |
3098 | ||
78bd47cf DDP |
3099 | static void |
3100 | egress_policer_details_to_param(const struct smap *details, | |
3101 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 3102 | { |
78bd47cf DDP |
3103 | memset(params, 0, sizeof *params); |
3104 | params->cir = smap_get_ullong(details, "cir", 0); | |
3105 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
3106 | params->ebs = 0; | |
0bf765f7 IS |
3107 | } |
3108 | ||
3109 | static int | |
78bd47cf DDP |
3110 | egress_policer_qos_construct(const struct smap *details, |
3111 | struct qos_conf **conf) | |
0bf765f7 | 3112 | { |
0bf765f7 | 3113 | struct egress_policer *policer; |
0bf765f7 IS |
3114 | int err = 0; |
3115 | ||
0bf765f7 IS |
3116 | policer = xmalloc(sizeof *policer); |
3117 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 3118 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
0bf765f7 | 3119 | err = rte_meter_srtcm_config(&policer->egress_meter, |
78bd47cf DDP |
3120 | &policer->app_srtcm_params); |
3121 | if (!err) { | |
3122 | *conf = &policer->qos_conf; | |
3123 | } else { | |
7ea266e9 | 3124 | free(policer); |
78bd47cf | 3125 | *conf = NULL; |
7ea266e9 IS |
3126 | err = -err; |
3127 | } | |
0bf765f7 IS |
3128 | |
3129 | return err; | |
3130 | } | |
3131 | ||
3132 | static void | |
78bd47cf | 3133 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
3134 | { |
3135 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 3136 | qos_conf); |
0bf765f7 IS |
3137 | free(policer); |
3138 | } | |
3139 | ||
3140 | static int | |
78bd47cf | 3141 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 3142 | { |
78bd47cf DDP |
3143 | struct egress_policer *policer = |
3144 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3145 | ||
3146 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
3147 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 3148 | |
0bf765f7 IS |
3149 | return 0; |
3150 | } | |
3151 | ||
78bd47cf | 3152 | static bool |
47a45d86 KT |
3153 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
3154 | const struct smap *details) | |
0bf765f7 | 3155 | { |
78bd47cf DDP |
3156 | struct egress_policer *policer = |
3157 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3158 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 3159 | |
78bd47cf | 3160 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 3161 | |
78bd47cf | 3162 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
3163 | } |
3164 | ||
0bf765f7 | 3165 | static int |
3e90f7d7 GZ |
3166 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt, |
3167 | bool may_steal) | |
0bf765f7 | 3168 | { |
0bf765f7 | 3169 | int cnt = 0; |
78bd47cf DDP |
3170 | struct egress_policer *policer = |
3171 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 3172 | |
3e90f7d7 GZ |
3173 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, |
3174 | pkt_cnt, may_steal); | |
0bf765f7 IS |
3175 | |
3176 | return cnt; | |
3177 | } | |
3178 | ||
3179 | static const struct dpdk_qos_ops egress_policer_ops = { | |
3180 | "egress-policer", /* qos_name */ | |
3181 | egress_policer_qos_construct, | |
3182 | egress_policer_qos_destruct, | |
3183 | egress_policer_qos_get, | |
78bd47cf | 3184 | egress_policer_qos_is_equal, |
0bf765f7 IS |
3185 | egress_policer_run |
3186 | }; | |
3187 | ||
050c60bf DDP |
3188 | static int |
3189 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
3190 | { | |
3191 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3192 | int err = 0; | |
3193 | ||
050c60bf DDP |
3194 | ovs_mutex_lock(&dev->mutex); |
3195 | ||
3196 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 3197 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b CL |
3198 | && dev->mtu == dev->requested_mtu |
3199 | && dev->rxq_size == dev->requested_rxq_size | |
bd4e172b | 3200 | && dev->txq_size == dev->requested_txq_size |
3201 | && dev->socket_id == dev->requested_socket_id) { | |
050c60bf DDP |
3202 | /* Reconfiguration is unnecessary */ |
3203 | ||
3204 | goto out; | |
3205 | } | |
3206 | ||
3207 | rte_eth_dev_stop(dev->port_id); | |
3208 | ||
d555d9bd RW |
3209 | err = netdev_dpdk_mempool_configure(dev); |
3210 | if (err) { | |
3211 | goto out; | |
0072e931 MK |
3212 | } |
3213 | ||
050c60bf DDP |
3214 | netdev->n_txq = dev->requested_n_txq; |
3215 | netdev->n_rxq = dev->requested_n_rxq; | |
3216 | ||
b685696b CL |
3217 | dev->rxq_size = dev->requested_rxq_size; |
3218 | dev->txq_size = dev->requested_txq_size; | |
3219 | ||
050c60bf DDP |
3220 | rte_free(dev->tx_q); |
3221 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
3222 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
3223 | if (!dev->tx_q) { | |
3224 | err = ENOMEM; | |
3225 | } | |
050c60bf | 3226 | |
0072e931 MK |
3227 | netdev_change_seq_changed(netdev); |
3228 | ||
050c60bf | 3229 | out: |
050c60bf | 3230 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
3231 | return err; |
3232 | } | |
3233 | ||
7f381c2e | 3234 | static int |
2d24d165 | 3235 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 3236 | OVS_REQUIRES(dev->mutex) |
050c60bf | 3237 | { |
2d24d165 CL |
3238 | dev->up.n_txq = dev->requested_n_txq; |
3239 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 3240 | int err; |
050c60bf | 3241 | |
81acebda IM |
3242 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
3243 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
3244 | dev->tx_q[0].map = 0; | |
3245 | } | |
3246 | ||
3247 | netdev_dpdk_remap_txqs(dev); | |
3248 | ||
d555d9bd RW |
3249 | err = netdev_dpdk_mempool_configure(dev); |
3250 | if (err) { | |
3251 | return err; | |
3252 | } else { | |
3253 | netdev_change_seq_changed(&dev->up); | |
db8f13b0 CL |
3254 | } |
3255 | ||
0a0f39df | 3256 | if (netdev_dpdk_get_vid(dev) >= 0) { |
894af647 | 3257 | if (dev->vhost_reconfigured == false) { |
3258 | dev->vhost_reconfigured = true; | |
3259 | /* Carrier status may need updating. */ | |
3260 | netdev_change_seq_changed(&dev->up); | |
3261 | } | |
81acebda | 3262 | } |
7f381c2e DDP |
3263 | |
3264 | return 0; | |
2d24d165 CL |
3265 | } |
3266 | ||
3267 | static int | |
3268 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
3269 | { | |
3270 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3271 | int err; |
2d24d165 | 3272 | |
2d24d165 | 3273 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 3274 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 3275 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
3276 | |
3277 | return err; | |
2d24d165 CL |
3278 | } |
3279 | ||
3280 | static int | |
3281 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3282 | { | |
3283 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3284 | int err; |
2d24d165 | 3285 | |
2d24d165 CL |
3286 | ovs_mutex_lock(&dev->mutex); |
3287 | ||
c1ff66ac CL |
3288 | /* Configure vHost client mode if requested and if the following criteria |
3289 | * are met: | |
2d24d165 CL |
3290 | * 1. Device hasn't been registered yet. |
3291 | * 2. A path has been specified. | |
c1ff66ac CL |
3292 | */ |
3293 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 CL |
3294 | && strlen(dev->vhost_id)) { |
3295 | /* Register client-mode device */ | |
3296 | err = rte_vhost_driver_register(dev->vhost_id, | |
3297 | RTE_VHOST_USER_CLIENT); | |
c1ff66ac | 3298 | if (err) { |
2d24d165 CL |
3299 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3300 | dev->vhost_id); | |
7f381c2e | 3301 | goto unlock; |
c1ff66ac | 3302 | } else { |
2d24d165 CL |
3303 | /* Configuration successful */ |
3304 | dev->vhost_driver_flags |= RTE_VHOST_USER_CLIENT; | |
3305 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " | |
3306 | "using client socket '%s'", | |
3307 | dev->up.name, dev->vhost_id); | |
c1ff66ac | 3308 | } |
f3e7ec25 MW |
3309 | |
3310 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
3311 | &virtio_net_device_ops); | |
3312 | if (err) { | |
3313 | VLOG_ERR("rte_vhost_driver_callback_register failed for " | |
3314 | "vhost user client port: %s\n", dev->up.name); | |
3315 | goto unlock; | |
3316 | } | |
3317 | ||
3318 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
3319 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
3320 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
3321 | | 1ULL << VIRTIO_NET_F_CSUM); | |
3322 | if (err) { | |
3323 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
3324 | "client port: %s\n", dev->up.name); | |
3325 | goto unlock; | |
3326 | } | |
3327 | ||
3328 | err = rte_vhost_driver_start(dev->vhost_id); | |
3329 | if (err) { | |
3330 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
3331 | "client port: %s\n", dev->up.name); | |
3332 | goto unlock; | |
3333 | } | |
c1ff66ac CL |
3334 | } |
3335 | ||
7f381c2e DDP |
3336 | err = dpdk_vhost_reconfigure_helper(dev); |
3337 | ||
3338 | unlock: | |
050c60bf | 3339 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 3340 | |
7f381c2e | 3341 | return err; |
050c60bf DDP |
3342 | } |
3343 | ||
ecc1a34e | 3344 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
81acebda IM |
3345 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
3346 | GET_CARRIER, GET_STATS, \ | |
3347 | GET_FEATURES, GET_STATUS, \ | |
3348 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3349 | { \ |
3350 | NAME, \ | |
118c77b1 | 3351 | true, /* is_pmd */ \ |
ecc1a34e | 3352 | INIT, /* init */ \ |
95fb793a | 3353 | NULL, /* netdev_dpdk_run */ \ |
3354 | NULL, /* netdev_dpdk_wait */ \ | |
3355 | \ | |
3356 | netdev_dpdk_alloc, \ | |
3357 | CONSTRUCT, \ | |
58397e6c | 3358 | DESTRUCT, \ |
95fb793a | 3359 | netdev_dpdk_dealloc, \ |
3360 | netdev_dpdk_get_config, \ | |
81acebda | 3361 | SET_CONFIG, \ |
95fb793a | 3362 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3363 | NULL, /* build header */ \ |
3364 | NULL, /* push header */ \ | |
3365 | NULL, /* pop header */ \ | |
7dec44fe | 3366 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3367 | SET_TX_MULTIQ, \ |
95fb793a | 3368 | \ |
7251515e | 3369 | SEND, /* send */ \ |
95fb793a | 3370 | NULL, /* send_wait */ \ |
3371 | \ | |
3372 | netdev_dpdk_set_etheraddr, \ | |
3373 | netdev_dpdk_get_etheraddr, \ | |
3374 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3375 | netdev_dpdk_set_mtu, \ |
95fb793a | 3376 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3377 | GET_CARRIER, \ |
95fb793a | 3378 | netdev_dpdk_get_carrier_resets, \ |
3379 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
3380 | GET_STATS, \ |
3381 | GET_FEATURES, \ | |
95fb793a | 3382 | NULL, /* set_advertisements */ \ |
875ab130 | 3383 | NULL, /* get_pt_mode */ \ |
95fb793a | 3384 | \ |
9509913a | 3385 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3386 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3387 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3388 | netdev_dpdk_get_qos, \ |
3389 | netdev_dpdk_set_qos, \ | |
95fb793a | 3390 | NULL, /* get_queue */ \ |
3391 | NULL, /* set_queue */ \ | |
3392 | NULL, /* delete_queue */ \ | |
3393 | NULL, /* get_queue_stats */ \ | |
3394 | NULL, /* queue_dump_start */ \ | |
3395 | NULL, /* queue_dump_next */ \ | |
3396 | NULL, /* queue_dump_done */ \ | |
3397 | NULL, /* dump_queue_stats */ \ | |
3398 | \ | |
95fb793a | 3399 | NULL, /* set_in4 */ \ |
a8704b50 | 3400 | NULL, /* get_addr_list */ \ |
95fb793a | 3401 | NULL, /* add_router */ \ |
3402 | NULL, /* get_next_hop */ \ | |
58397e6c | 3403 | GET_STATUS, \ |
95fb793a | 3404 | NULL, /* arp_lookup */ \ |
3405 | \ | |
3406 | netdev_dpdk_update_flags, \ | |
050c60bf | 3407 | RECONFIGURE, \ |
95fb793a | 3408 | \ |
3409 | netdev_dpdk_rxq_alloc, \ | |
3410 | netdev_dpdk_rxq_construct, \ | |
3411 | netdev_dpdk_rxq_destruct, \ | |
3412 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3413 | RXQ_RECV, \ |
95fb793a | 3414 | NULL, /* rx_wait */ \ |
3415 | NULL, /* rxq_drain */ \ | |
18ebd48c | 3416 | NO_OFFLOAD_API \ |
95fb793a | 3417 | } |
8a9562d2 | 3418 | |
bce01e3a | 3419 | static const struct netdev_class dpdk_class = |
95fb793a | 3420 | NETDEV_DPDK_CLASS( |
3421 | "dpdk", | |
ecc1a34e | 3422 | netdev_dpdk_class_init, |
5496878c | 3423 | netdev_dpdk_construct, |
58397e6c | 3424 | netdev_dpdk_destruct, |
81acebda IM |
3425 | netdev_dpdk_set_config, |
3426 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3427 | netdev_dpdk_eth_send, |
3428 | netdev_dpdk_get_carrier, | |
3429 | netdev_dpdk_get_stats, | |
3430 | netdev_dpdk_get_features, | |
3431 | netdev_dpdk_get_status, | |
050c60bf | 3432 | netdev_dpdk_reconfigure, |
58397e6c | 3433 | netdev_dpdk_rxq_recv); |
95fb793a | 3434 | |
bce01e3a | 3435 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3436 | NETDEV_DPDK_CLASS( |
3437 | "dpdkr", | |
ecc1a34e | 3438 | netdev_dpdk_class_init, |
5496878c | 3439 | netdev_dpdk_ring_construct, |
58397e6c | 3440 | netdev_dpdk_destruct, |
c3d062a7 | 3441 | netdev_dpdk_ring_set_config, |
81acebda | 3442 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3443 | netdev_dpdk_ring_send, |
3444 | netdev_dpdk_get_carrier, | |
3445 | netdev_dpdk_get_stats, | |
3446 | netdev_dpdk_get_features, | |
3447 | netdev_dpdk_get_status, | |
050c60bf | 3448 | netdev_dpdk_reconfigure, |
58397e6c KT |
3449 | netdev_dpdk_rxq_recv); |
3450 | ||
53f50d24 | 3451 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3452 | NETDEV_DPDK_CLASS( |
3453 | "dpdkvhostuser", | |
f3e7ec25 | 3454 | NULL, |
53f50d24 | 3455 | netdev_dpdk_vhost_construct, |
58397e6c | 3456 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3457 | NULL, |
81acebda | 3458 | NULL, |
58397e6c KT |
3459 | netdev_dpdk_vhost_send, |
3460 | netdev_dpdk_vhost_get_carrier, | |
3461 | netdev_dpdk_vhost_get_stats, | |
3462 | NULL, | |
7251515e | 3463 | NULL, |
53f50d24 | 3464 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3465 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3466 | static const struct netdev_class dpdk_vhost_client_class = |
3467 | NETDEV_DPDK_CLASS( | |
3468 | "dpdkvhostuserclient", | |
f3e7ec25 | 3469 | NULL, |
2d24d165 CL |
3470 | netdev_dpdk_vhost_client_construct, |
3471 | netdev_dpdk_vhost_destruct, | |
3472 | netdev_dpdk_vhost_client_set_config, | |
3473 | NULL, | |
3474 | netdev_dpdk_vhost_send, | |
3475 | netdev_dpdk_vhost_get_carrier, | |
3476 | netdev_dpdk_vhost_get_stats, | |
3477 | NULL, | |
3478 | NULL, | |
3479 | netdev_dpdk_vhost_client_reconfigure, | |
3480 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 3481 | |
8a9562d2 PS |
3482 | void |
3483 | netdev_dpdk_register(void) | |
3484 | { | |
bab69409 AC |
3485 | netdev_register_provider(&dpdk_class); |
3486 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3487 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 3488 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 3489 | } |