]>
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 | 310 | char if_name[IFNAMSIZ]; |
ad9b5b9b | 311 | unsigned n_mbufs; /* Number of mbufs inside the mempool. */ |
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); | |
f06546a5 | 502 | int ret = snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_%x_%d_%d_%u", |
ad9b5b9b | 503 | h, dmp->socket_id, dmp->mtu, dmp->n_mbufs); |
d555d9bd | 504 | if (ret < 0 || ret >= RTE_MEMPOOL_NAMESIZE) { |
65056fd7 FA |
505 | VLOG_DBG("snprintf returned %d. Failed to generate a mempool " |
506 | "name for \"%s\". Hash:0x%x, mtu:%d, mbufs:%u.", | |
ad9b5b9b | 507 | ret, dmp->if_name, h, dmp->mtu, dmp->n_mbufs); |
d555d9bd RW |
508 | return NULL; |
509 | } | |
510 | return mp_name; | |
511 | } | |
512 | ||
8a9562d2 | 513 | static struct dpdk_mp * |
b6b26021 | 514 | dpdk_mp_create(struct netdev_dpdk *dev, int mtu, bool *mp_exists) |
8a9562d2 | 515 | { |
d555d9bd | 516 | struct dpdk_mp *dmp = dpdk_rte_mzalloc(sizeof *dmp); |
eff23640 DDP |
517 | if (!dmp) { |
518 | return NULL; | |
519 | } | |
b6b26021 | 520 | *mp_exists = false; |
d555d9bd | 521 | dmp->socket_id = dev->requested_socket_id; |
8a9562d2 | 522 | dmp->mtu = mtu; |
d555d9bd RW |
523 | ovs_strzcpy(dmp->if_name, dev->up.name, IFNAMSIZ); |
524 | ||
525 | /* | |
ad9b5b9b | 526 | * XXX: rough estimation of number of mbufs required for this port: |
d555d9bd RW |
527 | * <packets required to fill the device rxqs> |
528 | * + <packets that could be stuck on other ports txqs> | |
529 | * + <packets in the pmd threads> | |
530 | * + <additional memory for corner cases> | |
0072e931 | 531 | */ |
ad9b5b9b | 532 | dmp->n_mbufs = dev->requested_n_rxq * dev->requested_rxq_size |
d555d9bd RW |
533 | + dev->requested_n_txq * dev->requested_txq_size |
534 | + MIN(RTE_MAX_LCORE, dev->requested_n_rxq) * NETDEV_MAX_BURST | |
535 | + MIN_NB_MBUF; | |
536 | ||
da79ce2b | 537 | do { |
d555d9bd | 538 | char *mp_name = dpdk_mp_name(dmp); |
65056fd7 FA |
539 | if (!mp_name) { |
540 | rte_free(dmp); | |
541 | return NULL; | |
542 | } | |
95fb793a | 543 | |
f06546a5 FA |
544 | VLOG_DBG("Port %s: Requesting a mempool of %u mbufs " |
545 | "on socket %d for %d Rx and %d Tx queues.", | |
ad9b5b9b | 546 | dev->up.name, dmp->n_mbufs, |
f06546a5 FA |
547 | dev->requested_socket_id, |
548 | dev->requested_n_rxq, dev->requested_n_txq); | |
d555d9bd | 549 | |
ad9b5b9b | 550 | dmp->mp = rte_pktmbuf_pool_create(mp_name, dmp->n_mbufs, |
401b70d6 HA |
551 | MP_CACHE_SZ, |
552 | sizeof (struct dp_packet) | |
553 | - sizeof (struct rte_mbuf), | |
554 | MBUF_SIZE(mtu) | |
555 | - sizeof(struct dp_packet), | |
d555d9bd | 556 | dmp->socket_id); |
0c6f39e5 | 557 | if (dmp->mp) { |
d555d9bd | 558 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", mp_name, |
ad9b5b9b | 559 | dmp->n_mbufs); |
837c1761 FA |
560 | /* rte_pktmbuf_pool_create has done some initialization of the |
561 | * rte_mbuf part of each dp_packet. Some OvS specific fields | |
562 | * of the packet still need to be initialized by | |
563 | * ovs_rte_pktmbuf_init. */ | |
564 | rte_mempool_obj_iter(dmp->mp, ovs_rte_pktmbuf_init, NULL); | |
d555d9bd RW |
565 | } else if (rte_errno == EEXIST) { |
566 | /* A mempool with the same name already exists. We just | |
567 | * retrieve its pointer to be returned to the caller. */ | |
568 | dmp->mp = rte_mempool_lookup(mp_name); | |
569 | VLOG_DBG("A mempool with name %s already exists at %p.", | |
570 | mp_name, dmp->mp); | |
571 | /* As the mempool create returned EEXIST we can expect the | |
572 | * lookup has returned a valid pointer. If for some reason | |
573 | * that's not the case we keep track of it. */ | |
b6b26021 | 574 | *mp_exists = true; |
d555d9bd RW |
575 | } else { |
576 | VLOG_ERR("Failed mempool \"%s\" create request of %u mbufs", | |
ad9b5b9b | 577 | mp_name, dmp->n_mbufs); |
0c6f39e5 DDP |
578 | } |
579 | free(mp_name); | |
580 | if (dmp->mp) { | |
581 | return dmp; | |
582 | } | |
b6b26021 | 583 | } while (!(*mp_exists) && |
ad9b5b9b | 584 | (rte_errno == ENOMEM && (dmp->n_mbufs /= 2) >= MIN_NB_MBUF)); |
2ae3d542 | 585 | |
2ae3d542 DDP |
586 | rte_free(dmp); |
587 | return NULL; | |
588 | } | |
589 | ||
b6b26021 FA |
590 | /* Returns a valid pointer when either of the following is true: |
591 | * - a new mempool was just created; | |
592 | * - a matching mempool already exists. */ | |
2ae3d542 | 593 | static struct dpdk_mp * |
b6b26021 | 594 | dpdk_mp_get(struct netdev_dpdk *dev, int mtu, bool *mp_exists) |
2ae3d542 DDP |
595 | { |
596 | struct dpdk_mp *dmp; | |
597 | ||
598 | ovs_mutex_lock(&dpdk_mp_mutex); | |
b6b26021 | 599 | dmp = dpdk_mp_create(dev, mtu, mp_exists); |
c2adb102 IM |
600 | ovs_mutex_unlock(&dpdk_mp_mutex); |
601 | ||
c2adb102 | 602 | return dmp; |
8a9562d2 PS |
603 | } |
604 | ||
a08a115d | 605 | /* Release an existing mempool. */ |
8a9562d2 | 606 | static void |
a08a115d | 607 | dpdk_mp_free(struct dpdk_mp *dmp) |
8a9562d2 | 608 | { |
d555d9bd RW |
609 | char *mp_name; |
610 | ||
8a9562d2 PS |
611 | if (!dmp) { |
612 | return; | |
613 | } | |
614 | ||
c2adb102 | 615 | ovs_mutex_lock(&dpdk_mp_mutex); |
d555d9bd RW |
616 | mp_name = dpdk_mp_name(dmp); |
617 | VLOG_DBG("Releasing \"%s\" mempool", mp_name); | |
618 | free(mp_name); | |
619 | rte_mempool_free(dmp->mp); | |
620 | rte_free(dmp); | |
c2adb102 | 621 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
622 | } |
623 | ||
b6b26021 FA |
624 | /* Tries to allocate a new mempool - or re-use an existing one where |
625 | * appropriate - on requested_socket_id with a size determined by | |
626 | * requested_mtu and requested Rx/Tx queues. | |
627 | * On success - or when re-using an existing mempool - the new configuration | |
628 | * will be applied. | |
0072e931 MK |
629 | * On error, device will be left unchanged. */ |
630 | static int | |
631 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
632 | OVS_REQUIRES(dev->mutex) |
633 | { | |
634 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
635 | struct dpdk_mp *mp; | |
b6b26021 | 636 | bool mp_exists; |
0072e931 | 637 | |
b6b26021 | 638 | mp = dpdk_mp_get(dev, FRAME_LEN_TO_MTU(buf_size), &mp_exists); |
0072e931 | 639 | if (!mp) { |
c67e46c0 MK |
640 | VLOG_ERR("Failed to create memory pool for netdev " |
641 | "%s, with MTU %d on socket %d: %s\n", | |
642 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
643 | rte_strerror(rte_errno)); | |
644 | return rte_errno; | |
b6b26021 FA |
645 | } else if (mp_exists) { |
646 | /* If a new MTU was requested and its rounded value equals the one | |
647 | * that is currently used, then the existing mempool is returned. | |
648 | * Update dev with the new values. */ | |
649 | dev->mtu = dev->requested_mtu; | |
650 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
651 | return EEXIST; | |
0072e931 | 652 | } else { |
a08a115d FA |
653 | /* A new mempool was created, release the previous one. */ |
654 | dpdk_mp_free(dev->dpdk_mp); | |
0072e931 MK |
655 | dev->dpdk_mp = mp; |
656 | dev->mtu = dev->requested_mtu; | |
657 | dev->socket_id = dev->requested_socket_id; | |
658 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
659 | } | |
660 | ||
661 | return 0; | |
662 | } | |
663 | ||
8a9562d2 PS |
664 | static void |
665 | check_link_status(struct netdev_dpdk *dev) | |
666 | { | |
667 | struct rte_eth_link link; | |
668 | ||
669 | rte_eth_link_get_nowait(dev->port_id, &link); | |
670 | ||
671 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 672 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
673 | |
674 | dev->link_reset_cnt++; | |
675 | dev->link = link; | |
676 | if (dev->link.link_status) { | |
bb37956a | 677 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Up - speed %u Mbps - %s", |
58be5c0e | 678 | dev->port_id, (unsigned) dev->link.link_speed, |
8a9562d2 PS |
679 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? |
680 | ("full-duplex") : ("half-duplex")); | |
681 | } else { | |
bb37956a | 682 | VLOG_DBG_RL(&rl, "Port %"PRIu8" Link Down", dev->port_id); |
8a9562d2 PS |
683 | } |
684 | } | |
685 | } | |
686 | ||
687 | static void * | |
688 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
689 | { | |
690 | struct netdev_dpdk *dev; | |
691 | ||
692 | pthread_detach(pthread_self()); | |
693 | ||
694 | for (;;) { | |
695 | ovs_mutex_lock(&dpdk_mutex); | |
696 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
697 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
698 | if (dev->type == DPDK_DEV_ETH) { |
699 | check_link_status(dev); | |
700 | } | |
8a9562d2 PS |
701 | ovs_mutex_unlock(&dev->mutex); |
702 | } | |
703 | ovs_mutex_unlock(&dpdk_mutex); | |
704 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
705 | } | |
706 | ||
707 | return NULL; | |
708 | } | |
709 | ||
b98d7669 DDP |
710 | static int |
711 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
712 | { | |
713 | int diag = 0; | |
714 | int i; | |
0072e931 | 715 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 716 | |
67fe6d63 MK |
717 | /* For some NICs (e.g. Niantic), scatter_rx mode needs to be explicitly |
718 | * enabled. */ | |
0072e931 | 719 | if (dev->mtu > ETHER_MTU) { |
67fe6d63 | 720 | conf.rxmode.enable_scatter = 1; |
0072e931 | 721 | } |
67fe6d63 | 722 | |
1a2bb118 SC |
723 | conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & |
724 | NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
b98d7669 DDP |
725 | /* A device may report more queues than it makes available (this has |
726 | * been observed for Intel xl710, which reserves some of them for | |
727 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
728 | * available. When this happens we can retry the configuration | |
729 | * and request less queues */ | |
730 | while (n_rxq && n_txq) { | |
731 | if (diag) { | |
732 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
733 | } | |
734 | ||
0072e931 | 735 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 736 | if (diag) { |
0072e931 MK |
737 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
738 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
739 | break; |
740 | } | |
741 | ||
67fe6d63 MK |
742 | diag = rte_eth_dev_set_mtu(dev->port_id, dev->mtu); |
743 | if (diag) { | |
744 | VLOG_ERR("Interface %s MTU (%d) setup error: %s", | |
745 | dev->up.name, dev->mtu, rte_strerror(-diag)); | |
746 | break; | |
747 | } | |
748 | ||
b98d7669 | 749 | for (i = 0; i < n_txq; i++) { |
b685696b | 750 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
751 | dev->socket_id, NULL); |
752 | if (diag) { | |
753 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
754 | dev->up.name, i, rte_strerror(-diag)); | |
755 | break; | |
756 | } | |
757 | } | |
758 | ||
759 | if (i != n_txq) { | |
760 | /* Retry with less tx queues */ | |
761 | n_txq = i; | |
762 | continue; | |
763 | } | |
764 | ||
765 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 766 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
b98d7669 DDP |
767 | dev->socket_id, NULL, |
768 | dev->dpdk_mp->mp); | |
769 | if (diag) { | |
770 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
771 | dev->up.name, i, rte_strerror(-diag)); | |
772 | break; | |
773 | } | |
774 | } | |
775 | ||
776 | if (i != n_rxq) { | |
777 | /* Retry with less rx queues */ | |
778 | n_rxq = i; | |
779 | continue; | |
780 | } | |
781 | ||
782 | dev->up.n_rxq = n_rxq; | |
81acebda | 783 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
784 | |
785 | return 0; | |
786 | } | |
787 | ||
788 | return diag; | |
789 | } | |
790 | ||
9fd39370 SC |
791 | static void |
792 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
793 | { | |
794 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
bb37956a IM |
795 | VLOG_WARN("Failed to enable flow control on device %"PRIu8, |
796 | dev->port_id); | |
9fd39370 SC |
797 | } |
798 | } | |
b98d7669 | 799 | |
8a9562d2 | 800 | static int |
c2adb102 IM |
801 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
802 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
803 | { |
804 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 805 | struct rte_eth_dev_info info; |
8a9562d2 PS |
806 | struct ether_addr eth_addr; |
807 | int diag; | |
b98d7669 | 808 | int n_rxq, n_txq; |
d4f5282c KT |
809 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | |
810 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
811 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
8a9562d2 | 812 | |
a0cb2d66 | 813 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 814 | |
d4f5282c KT |
815 | if ((info.rx_offload_capa & rx_chksm_offload_capa) != |
816 | rx_chksm_offload_capa) { | |
2cfe866f | 817 | VLOG_WARN("Rx checksum offload is not supported on port %"PRIu8, |
d4f5282c KT |
818 | dev->port_id); |
819 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; | |
820 | } else { | |
821 | dev->hw_ol_features |= NETDEV_RX_CHECKSUM_OFFLOAD; | |
822 | } | |
823 | ||
b98d7669 DDP |
824 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
825 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
826 | ||
827 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 828 | if (diag) { |
b98d7669 DDP |
829 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
830 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 831 | return -diag; |
8a9562d2 PS |
832 | } |
833 | ||
8a9562d2 PS |
834 | diag = rte_eth_dev_start(dev->port_id); |
835 | if (diag) { | |
b98d7669 DDP |
836 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
837 | rte_strerror(-diag)); | |
95fb793a | 838 | return -diag; |
8a9562d2 PS |
839 | } |
840 | ||
841 | rte_eth_promiscuous_enable(dev->port_id); | |
842 | rte_eth_allmulticast_enable(dev->port_id); | |
843 | ||
844 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
845 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
bb37956a | 846 | VLOG_INFO_RL(&rl, "Port %"PRIu8": "ETH_ADDR_FMT, |
ca92d173 | 847 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 848 | |
ca92d173 | 849 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
850 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
851 | ||
852 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
853 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
854 | ||
9fd39370 SC |
855 | /* Get the Flow control configuration for DPDK-ETH */ |
856 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
857 | if (diag) { | |
bb37956a | 858 | VLOG_DBG("cannot get flow control parameters on port=%"PRIu8", err=%d", |
9fd39370 SC |
859 | dev->port_id, diag); |
860 | } | |
861 | ||
8a9562d2 PS |
862 | return 0; |
863 | } | |
864 | ||
865 | static struct netdev_dpdk * | |
866 | netdev_dpdk_cast(const struct netdev *netdev) | |
867 | { | |
868 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
869 | } | |
870 | ||
871 | static struct netdev * | |
872 | netdev_dpdk_alloc(void) | |
873 | { | |
bab69409 AC |
874 | struct netdev_dpdk *dev; |
875 | ||
65e19e70 DDP |
876 | dev = dpdk_rte_mzalloc(sizeof *dev); |
877 | if (dev) { | |
878 | return &dev->up; | |
bab69409 | 879 | } |
65e19e70 | 880 | |
bab69409 | 881 | return NULL; |
8a9562d2 PS |
882 | } |
883 | ||
eff23640 DDP |
884 | static struct dpdk_tx_queue * |
885 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 886 | { |
eff23640 | 887 | struct dpdk_tx_queue *txqs; |
bd5131ba | 888 | unsigned i; |
5a034064 | 889 | |
eff23640 DDP |
890 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
891 | if (txqs) { | |
892 | for (i = 0; i < n_txqs; i++) { | |
893 | /* Initialize map for vhost devices. */ | |
894 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
895 | rte_spinlock_init(&txqs[i].tx_lock); | |
896 | } | |
5a034064 | 897 | } |
eff23640 DDP |
898 | |
899 | return txqs; | |
5a034064 AW |
900 | } |
901 | ||
8a9562d2 | 902 | static int |
bb37956a | 903 | common_construct(struct netdev *netdev, dpdk_port_t port_no, |
1ce30dfd | 904 | enum dpdk_dev_type type, int socket_id) |
5a034064 | 905 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 906 | { |
d46285a2 | 907 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 908 | |
d46285a2 | 909 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 910 | |
d46285a2 | 911 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 912 | |
1b7a04e0 AW |
913 | /* If the 'sid' is negative, it means that the kernel fails |
914 | * to obtain the pci numa info. In that situation, always | |
915 | * use 'SOCKET0'. */ | |
1ce30dfd | 916 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 917 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
918 | dev->port_id = port_no; |
919 | dev->type = type; | |
920 | dev->flags = 0; | |
7f381c2e | 921 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 922 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
923 | ovsrcu_index_init(&dev->vid, -1); |
924 | dev->vhost_reconfigured = false; | |
5dcde09c | 925 | dev->attached = false; |
8a9562d2 | 926 | |
78bd47cf | 927 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 928 | |
9509913a IS |
929 | ovsrcu_init(&dev->ingress_policer, NULL); |
930 | dev->policer_rate = 0; | |
931 | dev->policer_burst = 0; | |
932 | ||
7f381c2e DDP |
933 | netdev->n_rxq = 0; |
934 | netdev->n_txq = 0; | |
935 | dev->requested_n_rxq = NR_QUEUE; | |
936 | dev->requested_n_txq = NR_QUEUE; | |
937 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
938 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 939 | |
9fd39370 SC |
940 | /* Initialize the flow control to NULL */ |
941 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
942 | |
943 | /* Initilize the hardware offload flags to 0 */ | |
944 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
945 | |
946 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
947 | ||
d46285a2 | 948 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 949 | |
7f381c2e DDP |
950 | netdev_request_reconfigure(netdev); |
951 | ||
1ce30dfd | 952 | return 0; |
95fb793a | 953 | } |
954 | ||
b83a2df1 MV |
955 | /* dev_name must be the prefix followed by a positive decimal number. |
956 | * (no leading + or - signs are allowed) */ | |
95fb793a | 957 | static int |
958 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
959 | unsigned int *port_no) | |
960 | { | |
961 | const char *cport; | |
962 | ||
963 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
964 | return ENODEV; | |
965 | } | |
966 | ||
967 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
968 | |
969 | if (str_to_uint(cport, 10, port_no)) { | |
970 | return 0; | |
971 | } else { | |
972 | return ENODEV; | |
973 | } | |
95fb793a | 974 | } |
975 | ||
1ce30dfd DDP |
976 | static int |
977 | vhost_common_construct(struct netdev *netdev) | |
978 | OVS_REQUIRES(dpdk_mutex) | |
979 | { | |
980 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
981 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
982 | ||
983 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); | |
984 | if (!dev->tx_q) { | |
985 | return ENOMEM; | |
986 | } | |
987 | ||
bb37956a IM |
988 | return common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
989 | DPDK_DEV_VHOST, socket_id); | |
1ce30dfd DDP |
990 | } |
991 | ||
7d1ced01 | 992 | static int |
53f50d24 | 993 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 994 | { |
d46285a2 DDP |
995 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
996 | const char *name = netdev->name; | |
7d1ced01 | 997 | int err; |
a0cb2d66 | 998 | |
1af27e8a DDP |
999 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
1000 | * the file system. '/' or '\' would traverse directories, so they're not | |
1001 | * acceptable in 'name'. */ | |
1002 | if (strchr(name, '/') || strchr(name, '\\')) { | |
1003 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
1004 | "A valid name must not include '/' or '\\'", | |
1005 | name); | |
1006 | return EINVAL; | |
1007 | } | |
1008 | ||
7d1ced01 CL |
1009 | ovs_mutex_lock(&dpdk_mutex); |
1010 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 1011 | * the socket is to be created, then register the socket. |
7d1ced01 | 1012 | */ |
2d24d165 | 1013 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 1014 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 1015 | |
2d24d165 CL |
1016 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
1017 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
1018 | if (err) { |
1019 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 1020 | dev->vhost_id); |
f3e7ec25 | 1021 | goto out; |
e5c0f5a4 | 1022 | } else { |
2d24d165 CL |
1023 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
1024 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
1025 | dev->vhost_id, name); | |
1026 | } | |
f3e7ec25 MW |
1027 | |
1028 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
1029 | &virtio_net_device_ops); | |
1030 | if (err) { | |
1031 | VLOG_ERR("rte_vhost_driver_callback_register failed for vhost user " | |
1032 | "port: %s\n", name); | |
1033 | goto out; | |
1034 | } | |
1035 | ||
1036 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
1037 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
1038 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
1039 | | 1ULL << VIRTIO_NET_F_CSUM); | |
1040 | if (err) { | |
1041 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
1042 | "port: %s\n", name); | |
1043 | goto out; | |
1044 | } | |
1045 | ||
1046 | err = rte_vhost_driver_start(dev->vhost_id); | |
1047 | if (err) { | |
1048 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
1049 | "port: %s\n", name); | |
1050 | goto out; | |
1051 | } | |
1052 | ||
1ce30dfd | 1053 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1054 | if (err) { |
1055 | VLOG_ERR("vhost_common_construct failed for vhost user " | |
1056 | "port: %s\n", name); | |
1057 | } | |
2d24d165 | 1058 | |
f3e7ec25 | 1059 | out: |
2d24d165 | 1060 | ovs_mutex_unlock(&dpdk_mutex); |
28ca969e AC |
1061 | VLOG_WARN_ONCE("dpdkvhostuser ports are considered deprecated; " |
1062 | "please migrate to dpdkvhostuserclient ports."); | |
2d24d165 CL |
1063 | return err; |
1064 | } | |
1065 | ||
1066 | static int | |
1067 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
1068 | { | |
1069 | int err; | |
1070 | ||
2d24d165 | 1071 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 1072 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1073 | if (err) { |
1074 | VLOG_ERR("vhost_common_construct failed for vhost user client" | |
1075 | "port: %s\n", netdev->name); | |
1076 | } | |
7d1ced01 | 1077 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
1078 | return err; |
1079 | } | |
1080 | ||
95fb793a | 1081 | static int |
1082 | netdev_dpdk_construct(struct netdev *netdev) | |
1083 | { | |
95fb793a | 1084 | int err; |
1085 | ||
95fb793a | 1086 | ovs_mutex_lock(&dpdk_mutex); |
bb37956a IM |
1087 | err = common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1088 | DPDK_DEV_ETH, SOCKET0); | |
8a9562d2 PS |
1089 | ovs_mutex_unlock(&dpdk_mutex); |
1090 | return err; | |
1091 | } | |
1092 | ||
1ce30dfd DDP |
1093 | static void |
1094 | common_destruct(struct netdev_dpdk *dev) | |
1095 | OVS_REQUIRES(dpdk_mutex) | |
1096 | OVS_EXCLUDED(dev->mutex) | |
1097 | { | |
1098 | rte_free(dev->tx_q); | |
a08a115d | 1099 | dpdk_mp_free(dev->dpdk_mp); |
1ce30dfd DDP |
1100 | |
1101 | ovs_list_remove(&dev->list_node); | |
1102 | free(ovsrcu_get_protected(struct ingress_policer *, | |
1103 | &dev->ingress_policer)); | |
1104 | ovs_mutex_destroy(&dev->mutex); | |
1105 | } | |
1106 | ||
8a9562d2 | 1107 | static void |
d46285a2 | 1108 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 1109 | { |
d46285a2 | 1110 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5dcde09c | 1111 | char devname[RTE_ETH_NAME_MAX_LEN]; |
8a9562d2 | 1112 | |
8d38823b | 1113 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1114 | |
8a9562d2 | 1115 | rte_eth_dev_stop(dev->port_id); |
5dcde09c IM |
1116 | |
1117 | if (dev->attached) { | |
1118 | rte_eth_dev_close(dev->port_id); | |
1119 | if (rte_eth_dev_detach(dev->port_id, devname) < 0) { | |
1120 | VLOG_ERR("Device '%s' can not be detached", dev->devargs); | |
1121 | } else { | |
0ee821c2 | 1122 | VLOG_INFO("Device '%s' has been detached", devname); |
5dcde09c IM |
1123 | } |
1124 | } | |
1125 | ||
55e075e6 | 1126 | free(dev->devargs); |
1ce30dfd | 1127 | common_destruct(dev); |
8d38823b | 1128 | |
8a9562d2 | 1129 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1130 | } |
8a9562d2 | 1131 | |
3f891bbe DDP |
1132 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1133 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1134 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1135 | static int | |
c1ff66ac CL |
1136 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1137 | char *vhost_id) | |
3f891bbe DDP |
1138 | OVS_EXCLUDED(dpdk_mutex) |
1139 | OVS_EXCLUDED(dev->mutex) | |
1140 | { | |
c1ff66ac | 1141 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1142 | } |
1143 | ||
58397e6c | 1144 | static void |
d46285a2 | 1145 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1146 | { |
d46285a2 | 1147 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1148 | char *vhost_id; |
58397e6c | 1149 | |
8d38823b | 1150 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1151 | |
c62da695 | 1152 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1153 | if (netdev_dpdk_get_vid(dev) >= 0 |
1154 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1155 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1156 | netdev->name); |
58be5c0e MK |
1157 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1158 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1159 | } |
1160 | ||
2d24d165 | 1161 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1162 | |
1ce30dfd DDP |
1163 | common_destruct(dev); |
1164 | ||
58397e6c | 1165 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1166 | |
569c26da | 1167 | if (!vhost_id[0]) { |
821b8664 IM |
1168 | goto out; |
1169 | } | |
1170 | ||
c1ff66ac | 1171 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1172 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1173 | netdev->name, vhost_id); | |
c1ff66ac CL |
1174 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1175 | /* OVS server mode - remove this socket from list for deletion */ | |
1176 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1177 | } |
821b8664 | 1178 | out: |
c1ff66ac | 1179 | free(vhost_id); |
8a9562d2 PS |
1180 | } |
1181 | ||
1182 | static void | |
d46285a2 | 1183 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1184 | { |
d46285a2 | 1185 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1186 | |
d46285a2 | 1187 | rte_free(dev); |
8a9562d2 PS |
1188 | } |
1189 | ||
1190 | static int | |
a14b8947 | 1191 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1192 | { |
a14b8947 | 1193 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1194 | |
1195 | ovs_mutex_lock(&dev->mutex); | |
1196 | ||
050c60bf | 1197 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1198 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1199 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1200 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1201 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1202 | |
1203 | if (dev->type == DPDK_DEV_ETH) { | |
1204 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1205 | dev->requested_rxq_size); | |
1206 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1207 | dev->rxq_size); | |
1208 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1209 | dev->requested_txq_size); | |
1210 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1211 | dev->txq_size); | |
1a2bb118 SC |
1212 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1213 | smap_add(args, "rx_csum_offload", "true"); | |
8155ab7e KT |
1214 | } else { |
1215 | smap_add(args, "rx_csum_offload", "false"); | |
1a2bb118 | 1216 | } |
451f26fd | 1217 | } |
8a9562d2 PS |
1218 | ovs_mutex_unlock(&dev->mutex); |
1219 | ||
1220 | return 0; | |
1221 | } | |
1222 | ||
55e075e6 | 1223 | static struct netdev_dpdk * |
bb37956a | 1224 | netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id) |
55e075e6 CL |
1225 | OVS_REQUIRES(dpdk_mutex) |
1226 | { | |
1227 | struct netdev_dpdk *dev; | |
1228 | ||
1229 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1230 | if (dev->port_id == port_id) { | |
1231 | return dev; | |
1232 | } | |
1233 | } | |
1234 | ||
1235 | return NULL; | |
1236 | } | |
1237 | ||
bb37956a | 1238 | static dpdk_port_t |
5dcde09c IM |
1239 | netdev_dpdk_process_devargs(struct netdev_dpdk *dev, |
1240 | const char *devargs, char **errp) | |
55e075e6 | 1241 | { |
03d6399e IM |
1242 | /* Get the name up to the first comma. */ |
1243 | char *name = xmemdup0(devargs, strcspn(devargs, ",")); | |
bb37956a | 1244 | dpdk_port_t new_port_id = DPDK_ETH_PORT_ID_INVALID; |
55e075e6 | 1245 | |
69876ed7 | 1246 | if (!rte_eth_dev_count() |
03d6399e | 1247 | || rte_eth_dev_get_port_by_name(name, &new_port_id) |
69876ed7 CL |
1248 | || !rte_eth_dev_is_valid_port(new_port_id)) { |
1249 | /* Device not found in DPDK, attempt to attach it */ | |
1250 | if (!rte_eth_dev_attach(devargs, &new_port_id)) { | |
1251 | /* Attach successful */ | |
5dcde09c | 1252 | dev->attached = true; |
69876ed7 CL |
1253 | VLOG_INFO("Device '%s' attached to DPDK", devargs); |
1254 | } else { | |
1255 | /* Attach unsuccessful */ | |
bb37956a | 1256 | new_port_id = DPDK_ETH_PORT_ID_INVALID; |
47a45d86 KT |
1257 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", |
1258 | devargs); | |
55e075e6 CL |
1259 | } |
1260 | } | |
1261 | ||
03d6399e | 1262 | free(name); |
55e075e6 CL |
1263 | return new_port_id; |
1264 | } | |
1265 | ||
c3d062a7 CL |
1266 | static void |
1267 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1268 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1269 | { |
050c60bf | 1270 | int new_n_rxq; |
a14b8947 | 1271 | |
2a21e757 | 1272 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1273 | if (new_n_rxq != dev->requested_n_rxq) { |
1274 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1275 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1276 | } |
c3d062a7 CL |
1277 | } |
1278 | ||
b685696b CL |
1279 | static void |
1280 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1281 | const char *flag, int default_size, int *new_size) | |
1282 | { | |
1283 | int queue_size = smap_get_int(args, flag, default_size); | |
1284 | ||
1285 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1286 | || !is_pow2(queue_size)) { | |
1287 | queue_size = default_size; | |
1288 | } | |
1289 | ||
1290 | if (queue_size != *new_size) { | |
1291 | *new_size = queue_size; | |
1292 | netdev_request_reconfigure(netdev); | |
1293 | } | |
1294 | } | |
1295 | ||
c3d062a7 | 1296 | static int |
9fff138e DDP |
1297 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1298 | char **errp) | |
c3d062a7 CL |
1299 | { |
1300 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
b614c894 IM |
1301 | bool rx_fc_en, tx_fc_en, autoneg; |
1302 | enum rte_eth_fc_mode fc_mode; | |
1303 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1304 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1305 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1306 | }; | |
55e075e6 CL |
1307 | const char *new_devargs; |
1308 | int err = 0; | |
c3d062a7 | 1309 | |
55e075e6 | 1310 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1311 | ovs_mutex_lock(&dev->mutex); |
1312 | ||
1313 | dpdk_set_rxq_config(dev, args); | |
1314 | ||
b685696b CL |
1315 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1316 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1317 | &dev->requested_rxq_size); | |
1318 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1319 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1320 | &dev->requested_txq_size); | |
1321 | ||
55e075e6 CL |
1322 | new_devargs = smap_get(args, "dpdk-devargs"); |
1323 | ||
1324 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1325 | /* The user requested a new device. If we return error, the caller | |
1326 | * will delete this netdev and try to recreate it. */ | |
1327 | err = EAGAIN; | |
1328 | goto out; | |
1329 | } | |
1330 | ||
1331 | /* dpdk-devargs is required for device configuration */ | |
1332 | if (new_devargs && new_devargs[0]) { | |
1333 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1334 | * is valid */ | |
1335 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1336 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
bb37956a IM |
1337 | dpdk_port_t new_port_id = netdev_dpdk_process_devargs(dev, |
1338 | new_devargs, | |
1339 | errp); | |
55e075e6 CL |
1340 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1341 | err = EINVAL; | |
1342 | } else if (new_port_id == dev->port_id) { | |
1343 | /* Already configured, do not reconfigure again */ | |
1344 | err = 0; | |
1345 | } else { | |
1346 | struct netdev_dpdk *dup_dev; | |
bb37956a | 1347 | |
55e075e6 CL |
1348 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); |
1349 | if (dup_dev) { | |
9fff138e DDP |
1350 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
1351 | "which is already in use by '%s'", | |
1352 | netdev_get_name(netdev), new_devargs, | |
1353 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1354 | err = EADDRINUSE; |
1355 | } else { | |
bd4e172b | 1356 | int sid = rte_eth_dev_socket_id(new_port_id); |
bb37956a | 1357 | |
bd4e172b | 1358 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; |
55e075e6 CL |
1359 | dev->devargs = xstrdup(new_devargs); |
1360 | dev->port_id = new_port_id; | |
1361 | netdev_request_reconfigure(&dev->up); | |
1362 | err = 0; | |
1363 | } | |
1364 | } | |
1365 | } | |
1366 | } else { | |
9fff138e DDP |
1367 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1368 | "The old 'dpdk<port_id>' names are not supported", | |
1369 | netdev_get_name(netdev)); | |
55e075e6 CL |
1370 | err = EINVAL; |
1371 | } | |
1372 | ||
1373 | if (err) { | |
1374 | goto out; | |
1375 | } | |
1376 | ||
c3d062a7 CL |
1377 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1378 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1379 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1380 | |
b614c894 IM |
1381 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1382 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1383 | dev->fc_conf.mode = fc_mode; | |
1384 | dev->fc_conf.autoneg = autoneg; | |
1385 | dpdk_eth_flow_ctrl_setup(dev); | |
1386 | } | |
9fd39370 | 1387 | |
55e075e6 | 1388 | out: |
c3d062a7 | 1389 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1390 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1391 | |
55e075e6 | 1392 | return err; |
c3d062a7 CL |
1393 | } |
1394 | ||
1395 | static int | |
9fff138e DDP |
1396 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1397 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1398 | { |
1399 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1400 | ||
1401 | ovs_mutex_lock(&dev->mutex); | |
1402 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1403 | ovs_mutex_unlock(&dev->mutex); |
1404 | ||
1405 | return 0; | |
1406 | } | |
1407 | ||
c1ff66ac | 1408 | static int |
2d24d165 | 1409 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1410 | const struct smap *args, |
1411 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1412 | { |
1413 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1414 | const char *path; | |
1415 | ||
6881885a | 1416 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1417 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1418 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1419 | if (path && strcmp(path, dev->vhost_id)) { |
1420 | strcpy(dev->vhost_id, path); | |
c1ff66ac CL |
1421 | netdev_request_reconfigure(netdev); |
1422 | } | |
1423 | } | |
6881885a | 1424 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1425 | |
1426 | return 0; | |
1427 | } | |
1428 | ||
7dec44fe | 1429 | static int |
d46285a2 | 1430 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1431 | { |
d46285a2 | 1432 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1433 | |
d46285a2 | 1434 | return dev->socket_id; |
7dec44fe AW |
1435 | } |
1436 | ||
050c60bf | 1437 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1438 | static int |
050c60bf | 1439 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1440 | { |
d46285a2 | 1441 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1442 | |
d46285a2 | 1443 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1444 | |
050c60bf DDP |
1445 | if (dev->requested_n_txq == n_txq) { |
1446 | goto out; | |
4573fbd3 FL |
1447 | } |
1448 | ||
050c60bf DDP |
1449 | dev->requested_n_txq = n_txq; |
1450 | netdev_request_reconfigure(netdev); | |
58397e6c | 1451 | |
050c60bf | 1452 | out: |
d46285a2 | 1453 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1454 | return 0; |
58397e6c KT |
1455 | } |
1456 | ||
8a9562d2 PS |
1457 | static struct netdev_rxq * |
1458 | netdev_dpdk_rxq_alloc(void) | |
1459 | { | |
1460 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1461 | ||
eff23640 DDP |
1462 | if (rx) { |
1463 | return &rx->up; | |
1464 | } | |
1465 | ||
1466 | return NULL; | |
8a9562d2 PS |
1467 | } |
1468 | ||
1469 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1470 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1471 | { |
d46285a2 | 1472 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1473 | } |
1474 | ||
1475 | static int | |
d46285a2 | 1476 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1477 | { |
d46285a2 DDP |
1478 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1479 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1480 | |
d46285a2 DDP |
1481 | ovs_mutex_lock(&dev->mutex); |
1482 | rx->port_id = dev->port_id; | |
1483 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1484 | |
1485 | return 0; | |
1486 | } | |
1487 | ||
1488 | static void | |
d46285a2 | 1489 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1490 | { |
1491 | } | |
1492 | ||
1493 | static void | |
d46285a2 | 1494 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1495 | { |
d46285a2 | 1496 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1497 | |
1498 | rte_free(rx); | |
1499 | } | |
1500 | ||
819f13bd DDP |
1501 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1502 | * 'pkts', even in case of failure. | |
1503 | * | |
1504 | * Returns the number of packets that weren't transmitted. */ | |
1505 | static inline int | |
b59cc14e | 1506 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 1507 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 1508 | { |
1304f1f8 DDP |
1509 | uint32_t nb_tx = 0; |
1510 | ||
b59cc14e | 1511 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1512 | uint32_t ret; |
1513 | ||
b59cc14e | 1514 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1515 | if (!ret) { |
1516 | break; | |
1517 | } | |
1518 | ||
1519 | nb_tx += ret; | |
1520 | } | |
8a9562d2 | 1521 | |
b59cc14e | 1522 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 1523 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
1524 | * packet could come from a different mempool) */ |
1525 | int i; | |
1526 | ||
b59cc14e IM |
1527 | for (i = nb_tx; i < cnt; i++) { |
1528 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1529 | } |
8a9562d2 | 1530 | } |
819f13bd DDP |
1531 | |
1532 | return cnt - nb_tx; | |
8a9562d2 PS |
1533 | } |
1534 | ||
f3926f29 IS |
1535 | static inline bool |
1536 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1537 | struct rte_mbuf *pkt, uint64_t time) | |
1538 | { | |
1539 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1540 | ||
1541 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1542 | e_RTE_METER_GREEN; | |
1543 | } | |
1544 | ||
1545 | static int | |
1546 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
3e90f7d7 GZ |
1547 | struct rte_mbuf **pkts, int pkt_cnt, |
1548 | bool may_steal) | |
f3926f29 IS |
1549 | { |
1550 | int i = 0; | |
1551 | int cnt = 0; | |
1552 | struct rte_mbuf *pkt = NULL; | |
1553 | uint64_t current_time = rte_rdtsc(); | |
1554 | ||
1555 | for (i = 0; i < pkt_cnt; i++) { | |
1556 | pkt = pkts[i]; | |
1557 | /* Handle current packet */ | |
1558 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1559 | if (cnt != i) { | |
1560 | pkts[cnt] = pkt; | |
1561 | } | |
1562 | cnt++; | |
1563 | } else { | |
3e90f7d7 GZ |
1564 | if (may_steal) { |
1565 | rte_pktmbuf_free(pkt); | |
1566 | } | |
f3926f29 IS |
1567 | } |
1568 | } | |
1569 | ||
1570 | return cnt; | |
1571 | } | |
1572 | ||
9509913a IS |
1573 | static int |
1574 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
3e90f7d7 | 1575 | int pkt_cnt, bool may_steal) |
9509913a IS |
1576 | { |
1577 | int cnt = 0; | |
1578 | ||
1579 | rte_spinlock_lock(&policer->policer_lock); | |
3e90f7d7 GZ |
1580 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, |
1581 | pkt_cnt, may_steal); | |
9509913a IS |
1582 | rte_spinlock_unlock(&policer->policer_lock); |
1583 | ||
1584 | return cnt; | |
1585 | } | |
1586 | ||
58397e6c | 1587 | static bool |
0a0f39df | 1588 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1589 | { |
0a0f39df | 1590 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1591 | } |
1592 | ||
d6e3feb5 | 1593 | static inline void |
1594 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1595 | unsigned int packet_size) | |
1596 | { | |
1597 | /* Hard-coded search for the size bucket. */ | |
1598 | if (packet_size < 256) { | |
1599 | if (packet_size >= 128) { | |
1600 | stats->rx_128_to_255_packets++; | |
1601 | } else if (packet_size <= 64) { | |
1602 | stats->rx_1_to_64_packets++; | |
1603 | } else { | |
1604 | stats->rx_65_to_127_packets++; | |
1605 | } | |
1606 | } else { | |
1607 | if (packet_size >= 1523) { | |
1608 | stats->rx_1523_to_max_packets++; | |
1609 | } else if (packet_size >= 1024) { | |
1610 | stats->rx_1024_to_1522_packets++; | |
1611 | } else if (packet_size < 512) { | |
1612 | stats->rx_256_to_511_packets++; | |
1613 | } else { | |
1614 | stats->rx_512_to_1023_packets++; | |
1615 | } | |
1616 | } | |
1617 | } | |
1618 | ||
9e3ddd45 TP |
1619 | static inline void |
1620 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1621 | struct dp_packet **packets, int count, |
1622 | int dropped) | |
9e3ddd45 TP |
1623 | { |
1624 | int i; | |
d6e3feb5 | 1625 | unsigned int packet_size; |
9e3ddd45 TP |
1626 | struct dp_packet *packet; |
1627 | ||
1628 | stats->rx_packets += count; | |
9509913a | 1629 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1630 | for (i = 0; i < count; i++) { |
1631 | packet = packets[i]; | |
d6e3feb5 | 1632 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1633 | |
d6e3feb5 | 1634 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1635 | /* This only protects the following multicast counting from |
1636 | * too short packets, but it does not stop the packet from | |
1637 | * further processing. */ | |
1638 | stats->rx_errors++; | |
1639 | stats->rx_length_errors++; | |
1640 | continue; | |
1641 | } | |
1642 | ||
d6e3feb5 | 1643 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1644 | ||
9e3ddd45 TP |
1645 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1646 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1647 | stats->multicast++; | |
1648 | } | |
1649 | ||
d6e3feb5 | 1650 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1651 | } |
1652 | } | |
1653 | ||
58397e6c KT |
1654 | /* |
1655 | * The receive path for the vhost port is the TX path out from guest. | |
1656 | */ | |
1657 | static int | |
d46285a2 | 1658 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1659 | struct dp_packet_batch *batch) |
58397e6c | 1660 | { |
d46285a2 | 1661 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
d46285a2 | 1662 | int qid = rxq->queue_id; |
9509913a | 1663 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1664 | uint16_t nb_rx = 0; |
9509913a | 1665 | uint16_t dropped = 0; |
58397e6c | 1666 | |
0a0f39df | 1667 | if (OVS_UNLIKELY(!is_vhost_running(dev) |
e543851d | 1668 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1669 | return EAGAIN; |
1670 | } | |
1671 | ||
0a0f39df CL |
1672 | nb_rx = rte_vhost_dequeue_burst(netdev_dpdk_get_vid(dev), |
1673 | qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
d46285a2 | 1674 | dev->dpdk_mp->mp, |
64839cf4 | 1675 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1676 | NETDEV_MAX_BURST); |
58397e6c KT |
1677 | if (!nb_rx) { |
1678 | return EAGAIN; | |
1679 | } | |
1680 | ||
9509913a IS |
1681 | if (policer) { |
1682 | dropped = nb_rx; | |
64839cf4 WT |
1683 | nb_rx = ingress_policer_run(policer, |
1684 | (struct rte_mbuf **) batch->packets, | |
3e90f7d7 | 1685 | nb_rx, true); |
9509913a IS |
1686 | dropped -= nb_rx; |
1687 | } | |
1688 | ||
d46285a2 | 1689 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1690 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1691 | nb_rx, dropped); | |
d46285a2 | 1692 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1693 | |
75fb9148 ZB |
1694 | batch->count = nb_rx; |
1695 | dp_packet_batch_init_packet_fields(batch); | |
1696 | ||
58397e6c KT |
1697 | return 0; |
1698 | } | |
1699 | ||
8a9562d2 | 1700 | static int |
64839cf4 | 1701 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1702 | { |
d46285a2 DDP |
1703 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1704 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1705 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1706 | int nb_rx; |
9509913a | 1707 | int dropped = 0; |
8a9562d2 | 1708 | |
3b1fb077 DDP |
1709 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1710 | return EAGAIN; | |
1711 | } | |
1712 | ||
d46285a2 | 1713 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1714 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1715 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1716 | if (!nb_rx) { |
1717 | return EAGAIN; | |
1718 | } | |
1719 | ||
9509913a IS |
1720 | if (policer) { |
1721 | dropped = nb_rx; | |
64839cf4 | 1722 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 1723 | (struct rte_mbuf **) batch->packets, |
3e90f7d7 | 1724 | nb_rx, true); |
9509913a IS |
1725 | dropped -= nb_rx; |
1726 | } | |
1727 | ||
1728 | /* Update stats to reflect dropped packets */ | |
1729 | if (OVS_UNLIKELY(dropped)) { | |
1730 | rte_spinlock_lock(&dev->stats_lock); | |
1731 | dev->stats.rx_dropped += dropped; | |
1732 | rte_spinlock_unlock(&dev->stats_lock); | |
1733 | } | |
1734 | ||
64839cf4 | 1735 | batch->count = nb_rx; |
75fb9148 | 1736 | dp_packet_batch_init_packet_fields(batch); |
8a9562d2 PS |
1737 | |
1738 | return 0; | |
1739 | } | |
1740 | ||
0bf765f7 | 1741 | static inline int |
78bd47cf | 1742 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
3e90f7d7 | 1743 | int cnt, bool may_steal) |
0bf765f7 | 1744 | { |
78bd47cf | 1745 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 1746 | |
78bd47cf DDP |
1747 | if (qos_conf) { |
1748 | rte_spinlock_lock(&qos_conf->lock); | |
3e90f7d7 | 1749 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt, may_steal); |
78bd47cf | 1750 | rte_spinlock_unlock(&qos_conf->lock); |
0bf765f7 IS |
1751 | } |
1752 | ||
1753 | return cnt; | |
1754 | } | |
1755 | ||
c6ec9d17 IM |
1756 | static int |
1757 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1758 | int pkt_cnt) | |
1759 | { | |
1760 | int i = 0; | |
1761 | int cnt = 0; | |
1762 | struct rte_mbuf *pkt; | |
1763 | ||
1764 | for (i = 0; i < pkt_cnt; i++) { | |
1765 | pkt = pkts[i]; | |
1766 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
1767 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
1768 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
1769 | rte_pktmbuf_free(pkt); | |
1770 | continue; | |
1771 | } | |
1772 | ||
1773 | if (OVS_UNLIKELY(i != cnt)) { | |
1774 | pkts[cnt] = pkt; | |
1775 | } | |
1776 | cnt++; | |
1777 | } | |
1778 | ||
1779 | return cnt; | |
1780 | } | |
1781 | ||
9e3ddd45 TP |
1782 | static inline void |
1783 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1784 | struct dp_packet **packets, | |
1785 | int attempted, | |
1786 | int dropped) | |
1787 | { | |
1788 | int i; | |
1789 | int sent = attempted - dropped; | |
1790 | ||
1791 | stats->tx_packets += sent; | |
1792 | stats->tx_dropped += dropped; | |
1793 | ||
1794 | for (i = 0; i < sent; i++) { | |
1795 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1796 | } | |
1797 | } | |
1798 | ||
58397e6c | 1799 | static void |
4573fbd3 | 1800 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 1801 | struct dp_packet **pkts, int cnt) |
58397e6c | 1802 | { |
d46285a2 | 1803 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1804 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1805 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 1806 | unsigned int dropped = 0; |
dd52de45 | 1807 | int i, retries = 0; |
58397e6c | 1808 | |
81acebda | 1809 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1810 | |
0a0f39df | 1811 | if (OVS_UNLIKELY(!is_vhost_running(dev) || qid < 0 |
e543851d | 1812 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1813 | rte_spinlock_lock(&dev->stats_lock); |
1814 | dev->stats.tx_dropped+= cnt; | |
1815 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1816 | goto out; |
58397e6c KT |
1817 | } |
1818 | ||
d46285a2 | 1819 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1820 | |
c6ec9d17 | 1821 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 1822 | /* Check has QoS has been configured for the netdev */ |
3e90f7d7 | 1823 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt, true); |
c6ec9d17 | 1824 | dropped = total_pkts - cnt; |
0bf765f7 | 1825 | |
95e9881f | 1826 | do { |
4573fbd3 | 1827 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1828 | unsigned int tx_pkts; |
1829 | ||
0a0f39df | 1830 | tx_pkts = rte_vhost_enqueue_burst(netdev_dpdk_get_vid(dev), |
c6ec9d17 | 1831 | vhost_qid, cur_pkts, cnt); |
95e9881f KT |
1832 | if (OVS_LIKELY(tx_pkts)) { |
1833 | /* Packets have been sent.*/ | |
1834 | cnt -= tx_pkts; | |
31871ee3 | 1835 | /* Prepare for possible retry.*/ |
95e9881f KT |
1836 | cur_pkts = &cur_pkts[tx_pkts]; |
1837 | } else { | |
31871ee3 KT |
1838 | /* No packets sent - do not retry.*/ |
1839 | break; | |
95e9881f | 1840 | } |
c6ec9d17 | 1841 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1842 | |
d46285a2 | 1843 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1844 | |
d46285a2 | 1845 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 1846 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 1847 | cnt + dropped); |
d46285a2 | 1848 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1849 | |
1850 | out: | |
c6ec9d17 | 1851 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 1852 | dp_packet_delete(pkts[i]); |
58397e6c KT |
1853 | } |
1854 | } | |
1855 | ||
8a9562d2 PS |
1856 | /* Tx function. Transmit packets indefinitely */ |
1857 | static void | |
64839cf4 | 1858 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1859 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1860 | { |
8a14bd7b | 1861 | const size_t batch_cnt = dp_packet_batch_size(batch); |
bce01e3a | 1862 | #if !defined(__CHECKER__) && !defined(_WIN32) |
8a14bd7b | 1863 | const size_t PKT_ARRAY_SIZE = batch_cnt; |
bce01e3a EJ |
1864 | #else |
1865 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1866 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1867 | #endif |
8a9562d2 | 1868 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 1869 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
8a14bd7b | 1870 | uint32_t cnt = batch_cnt; |
3e90f7d7 GZ |
1871 | uint32_t dropped = 0; |
1872 | ||
1873 | if (dev->type != DPDK_DEV_VHOST) { | |
1874 | /* Check if QoS has been configured for this netdev. */ | |
1875 | cnt = netdev_dpdk_qos_run(dev, (struct rte_mbuf **) batch->packets, | |
8a14bd7b BB |
1876 | batch_cnt, false); |
1877 | dropped += batch_cnt - cnt; | |
3e90f7d7 | 1878 | } |
8a9562d2 | 1879 | |
7d6d1a40 WT |
1880 | dp_packet_batch_apply_cutlen(batch); |
1881 | ||
3e90f7d7 GZ |
1882 | uint32_t txcnt = 0; |
1883 | ||
1884 | for (uint32_t i = 0; i < cnt; i++) { | |
8a14bd7b BB |
1885 | struct dp_packet *packet = batch->packets[i]; |
1886 | uint32_t size = dp_packet_size(packet); | |
95fb793a | 1887 | |
f98d7864 | 1888 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
3e90f7d7 GZ |
1889 | VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d", |
1890 | size, dev->max_packet_len); | |
f4fd623c | 1891 | |
175cf4de | 1892 | dropped++; |
f4fd623c DDP |
1893 | continue; |
1894 | } | |
8a9562d2 | 1895 | |
3e90f7d7 | 1896 | pkts[txcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a14bd7b | 1897 | if (OVS_UNLIKELY(!pkts[txcnt])) { |
3e90f7d7 | 1898 | dropped += cnt - i; |
175cf4de | 1899 | break; |
f4fd623c DDP |
1900 | } |
1901 | ||
1902 | /* We have to do a copy for now */ | |
3e90f7d7 | 1903 | memcpy(rte_pktmbuf_mtod(pkts[txcnt], void *), |
8a14bd7b BB |
1904 | dp_packet_data(packet), size); |
1905 | dp_packet_set_size((struct dp_packet *)pkts[txcnt], size); | |
f4fd623c | 1906 | |
3e90f7d7 | 1907 | txcnt++; |
f4fd623c | 1908 | } |
8a9562d2 | 1909 | |
3e90f7d7 GZ |
1910 | if (OVS_LIKELY(txcnt)) { |
1911 | if (dev->type == DPDK_DEV_VHOST) { | |
1912 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, | |
1913 | txcnt); | |
1914 | } else { | |
1915 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, txcnt); | |
1916 | } | |
58397e6c | 1917 | } |
db73f716 | 1918 | |
0bf765f7 IS |
1919 | if (OVS_UNLIKELY(dropped)) { |
1920 | rte_spinlock_lock(&dev->stats_lock); | |
1921 | dev->stats.tx_dropped += dropped; | |
1922 | rte_spinlock_unlock(&dev->stats_lock); | |
1923 | } | |
8a9562d2 PS |
1924 | } |
1925 | ||
58397e6c | 1926 | static int |
64839cf4 WT |
1927 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1928 | struct dp_packet_batch *batch, | |
324c8374 | 1929 | bool may_steal, bool concurrent_txq OVS_UNUSED) |
58397e6c | 1930 | { |
58397e6c | 1931 | |
dd52de45 | 1932 | if (OVS_UNLIKELY(!may_steal || batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 WT |
1933 | dpdk_do_tx_copy(netdev, qid, batch); |
1934 | dp_packet_delete_batch(batch, may_steal); | |
58397e6c | 1935 | } else { |
64839cf4 | 1936 | dp_packet_batch_apply_cutlen(batch); |
dd52de45 | 1937 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
1938 | } |
1939 | return 0; | |
1940 | } | |
1941 | ||
7251515e DV |
1942 | static inline void |
1943 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
324c8374 IM |
1944 | struct dp_packet_batch *batch, bool may_steal, |
1945 | bool concurrent_txq) | |
8a9562d2 | 1946 | { |
3b1fb077 DDP |
1947 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1948 | dp_packet_delete_batch(batch, may_steal); | |
1949 | return; | |
1950 | } | |
1951 | ||
324c8374 | 1952 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 1953 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
1954 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
1955 | } | |
1956 | ||
7251515e | 1957 | if (OVS_UNLIKELY(!may_steal || |
64839cf4 | 1958 | batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1959 | struct netdev *netdev = &dev->up; |
1960 | ||
64839cf4 WT |
1961 | dpdk_do_tx_copy(netdev, qid, batch); |
1962 | dp_packet_delete_batch(batch, may_steal); | |
8a9562d2 | 1963 | } else { |
fd57eeba BB |
1964 | int tx_cnt, dropped; |
1965 | int batch_cnt = dp_packet_batch_size(batch); | |
2391135c | 1966 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 1967 | |
7d6d1a40 WT |
1968 | dp_packet_batch_apply_cutlen(batch); |
1969 | ||
fd57eeba BB |
1970 | tx_cnt = netdev_dpdk_filter_packet_len(dev, pkts, batch_cnt); |
1971 | tx_cnt = netdev_dpdk_qos_run(dev, pkts, tx_cnt, true); | |
1972 | dropped = batch_cnt - tx_cnt; | |
1b99bb05 | 1973 | |
fd57eeba | 1974 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, tx_cnt); |
8a9562d2 | 1975 | |
f4fd623c | 1976 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1977 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1978 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1979 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1980 | } |
8a9562d2 | 1981 | } |
a0cb2d66 | 1982 | |
324c8374 | 1983 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
1984 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
1985 | } | |
7251515e DV |
1986 | } |
1987 | ||
1988 | static int | |
1989 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
324c8374 IM |
1990 | struct dp_packet_batch *batch, bool may_steal, |
1991 | bool concurrent_txq) | |
7251515e DV |
1992 | { |
1993 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1994 | |
324c8374 | 1995 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e | 1996 | return 0; |
8a9562d2 PS |
1997 | } |
1998 | ||
1999 | static int | |
74ff3298 | 2000 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
2001 | { |
2002 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2003 | ||
2004 | ovs_mutex_lock(&dev->mutex); | |
2005 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 2006 | dev->hwaddr = mac; |
045c0d1a | 2007 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
2008 | } |
2009 | ovs_mutex_unlock(&dev->mutex); | |
2010 | ||
2011 | return 0; | |
2012 | } | |
2013 | ||
2014 | static int | |
74ff3298 | 2015 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
2016 | { |
2017 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2018 | ||
2019 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 2020 | *mac = dev->hwaddr; |
8a9562d2 PS |
2021 | ovs_mutex_unlock(&dev->mutex); |
2022 | ||
2023 | return 0; | |
2024 | } | |
2025 | ||
2026 | static int | |
2027 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
2028 | { | |
2029 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2030 | ||
2031 | ovs_mutex_lock(&dev->mutex); | |
2032 | *mtup = dev->mtu; | |
2033 | ovs_mutex_unlock(&dev->mutex); | |
2034 | ||
2035 | return 0; | |
2036 | } | |
2037 | ||
0072e931 MK |
2038 | static int |
2039 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
2040 | { | |
2041 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2042 | ||
2043 | if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
2044 | || mtu < ETHER_MIN_MTU) { | |
2045 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
2046 | return EINVAL; | |
2047 | } | |
2048 | ||
2049 | ovs_mutex_lock(&dev->mutex); | |
2050 | if (dev->requested_mtu != mtu) { | |
2051 | dev->requested_mtu = mtu; | |
2052 | netdev_request_reconfigure(netdev); | |
2053 | } | |
2054 | ovs_mutex_unlock(&dev->mutex); | |
2055 | ||
2056 | return 0; | |
2057 | } | |
2058 | ||
8a9562d2 | 2059 | static int |
d46285a2 | 2060 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 2061 | |
58397e6c KT |
2062 | static int |
2063 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
2064 | struct netdev_stats *stats) | |
2065 | { | |
2066 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2067 | ||
2068 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 2069 | |
45d947c4 | 2070 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c | 2071 | /* Supported Stats */ |
50986e78 | 2072 | stats->rx_packets = dev->stats.rx_packets; |
2073 | stats->tx_packets = dev->stats.tx_packets; | |
9509913a | 2074 | stats->rx_dropped = dev->stats.rx_dropped; |
50986e78 | 2075 | stats->tx_dropped = dev->stats.tx_dropped; |
9e3ddd45 TP |
2076 | stats->multicast = dev->stats.multicast; |
2077 | stats->rx_bytes = dev->stats.rx_bytes; | |
2078 | stats->tx_bytes = dev->stats.tx_bytes; | |
2079 | stats->rx_errors = dev->stats.rx_errors; | |
2080 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 2081 | |
2082 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
2083 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
2084 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
2085 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
2086 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
2087 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
2088 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
2089 | ||
45d947c4 | 2090 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 2091 | |
58397e6c KT |
2092 | ovs_mutex_unlock(&dev->mutex); |
2093 | ||
2094 | return 0; | |
2095 | } | |
2096 | ||
d6e3feb5 | 2097 | static void |
2098 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
2099 | const struct rte_eth_xstat *xstats, |
2100 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 2101 | const unsigned int size) |
2102 | { | |
d6e3feb5 | 2103 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 2104 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2105 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2106 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2107 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2108 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2109 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2110 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2111 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2112 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2113 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2114 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2115 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2116 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2117 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 2118 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2119 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2120 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2121 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2122 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2123 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2124 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2125 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2126 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2127 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2128 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2129 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2130 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2131 | stats->tx_1523_to_max_packets = xstats[i].value; |
d57f777f PS |
2132 | } else if (strcmp(XSTAT_RX_MULTICAST_PACKETS, names[i].name) == 0) { |
2133 | stats->multicast = xstats[i].value; | |
0a0f39df | 2134 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2135 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 2136 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2137 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2138 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2139 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2140 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2141 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2142 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2143 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2144 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2145 | stats->rx_jabber_errors = xstats[i].value; |
2146 | } | |
2147 | } | |
2148 | } | |
2149 | ||
8a9562d2 PS |
2150 | static int |
2151 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2152 | { | |
2153 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2154 | struct rte_eth_stats rte_stats; | |
2155 | bool gg; | |
2156 | ||
2157 | netdev_dpdk_get_carrier(netdev, &gg); | |
2158 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2159 | |
0a0f39df CL |
2160 | struct rte_eth_xstat *rte_xstats = NULL; |
2161 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2162 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2163 | |
2164 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
bb37956a | 2165 | VLOG_ERR("Can't get ETH statistics for port: %"PRIu8, dev->port_id); |
f9256822 | 2166 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2167 | return EPROTO; |
2168 | } | |
2169 | ||
0a0f39df CL |
2170 | /* Get length of statistics */ |
2171 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2172 | if (rte_xstats_len < 0) { | |
bb37956a | 2173 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2174 | goto out; |
2175 | } | |
2176 | /* Reserve memory for xstats names and values */ | |
2177 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2178 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2179 | ||
2180 | /* Retreive xstats names */ | |
2181 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2182 | rte_xstats_names, | |
2183 | rte_xstats_len); | |
2184 | if (rte_xstats_new_len != rte_xstats_len) { | |
bb37956a | 2185 | VLOG_WARN("Cannot get XSTATS names for port: %"PRIu8, dev->port_id); |
0a0f39df CL |
2186 | goto out; |
2187 | } | |
2188 | /* Retreive xstats values */ | |
2189 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2190 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2191 | rte_xstats_len); | |
2192 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2193 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2194 | rte_xstats_len); | |
d6e3feb5 | 2195 | } else { |
bb37956a | 2196 | VLOG_WARN("Cannot get XSTATS values for port: %"PRIu8, dev->port_id); |
d6e3feb5 | 2197 | } |
8a9562d2 | 2198 | |
0a0f39df CL |
2199 | out: |
2200 | free(rte_xstats); | |
2201 | free(rte_xstats_names); | |
2202 | ||
2f9dd77f PS |
2203 | stats->rx_packets = rte_stats.ipackets; |
2204 | stats->tx_packets = rte_stats.opackets; | |
2205 | stats->rx_bytes = rte_stats.ibytes; | |
2206 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2207 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2208 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2209 | |
45d947c4 | 2210 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2211 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2212 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2213 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2214 | |
2215 | /* These are the available DPDK counters for packets not received due to | |
2216 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2217 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2218 | stats->rx_missed_errors = rte_stats.imissed; |
2219 | ||
8a9562d2 PS |
2220 | ovs_mutex_unlock(&dev->mutex); |
2221 | ||
2222 | return 0; | |
2223 | } | |
2224 | ||
2225 | static int | |
d46285a2 | 2226 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2227 | enum netdev_features *current, |
ca3d4f55 BX |
2228 | enum netdev_features *advertised, |
2229 | enum netdev_features *supported, | |
2230 | enum netdev_features *peer) | |
8a9562d2 | 2231 | { |
d46285a2 | 2232 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2233 | struct rte_eth_link link; |
2234 | ||
2235 | ovs_mutex_lock(&dev->mutex); | |
2236 | link = dev->link; | |
2237 | ovs_mutex_unlock(&dev->mutex); | |
2238 | ||
362ca396 | 2239 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2240 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
2241 | *current = NETDEV_F_10MB_HD; |
2242 | } | |
362ca396 | 2243 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2244 | *current = NETDEV_F_100MB_HD; |
2245 | } | |
362ca396 | 2246 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2247 | *current = NETDEV_F_1GB_HD; |
2248 | } | |
2249 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 2250 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
2251 | *current = NETDEV_F_10MB_FD; |
2252 | } | |
362ca396 | 2253 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2254 | *current = NETDEV_F_100MB_FD; |
2255 | } | |
362ca396 | 2256 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2257 | *current = NETDEV_F_1GB_FD; |
2258 | } | |
362ca396 | 2259 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
2260 | *current = NETDEV_F_10GB_FD; |
2261 | } | |
2262 | } | |
2263 | ||
362ca396 | 2264 | if (link.link_autoneg) { |
2265 | *current |= NETDEV_F_AUTONEG; | |
2266 | } | |
2267 | ||
ca3d4f55 BX |
2268 | *advertised = *supported = *peer = 0; |
2269 | ||
8a9562d2 PS |
2270 | return 0; |
2271 | } | |
2272 | ||
9509913a IS |
2273 | static struct ingress_policer * |
2274 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2275 | { | |
2276 | struct ingress_policer *policer = NULL; | |
2277 | uint64_t rate_bytes; | |
2278 | uint64_t burst_bytes; | |
2279 | int err = 0; | |
2280 | ||
2281 | policer = xmalloc(sizeof *policer); | |
2282 | rte_spinlock_init(&policer->policer_lock); | |
2283 | ||
2284 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
602c8668 LR |
2285 | rate_bytes = rate * 1000ULL / 8; |
2286 | burst_bytes = burst * 1000ULL / 8; | |
9509913a IS |
2287 | |
2288 | policer->app_srtcm_params.cir = rate_bytes; | |
2289 | policer->app_srtcm_params.cbs = burst_bytes; | |
2290 | policer->app_srtcm_params.ebs = 0; | |
2291 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2292 | &policer->app_srtcm_params); | |
58be5c0e | 2293 | if (err) { |
9509913a IS |
2294 | VLOG_ERR("Could not create rte meter for ingress policer"); |
2295 | return NULL; | |
2296 | } | |
2297 | ||
2298 | return policer; | |
2299 | } | |
2300 | ||
2301 | static int | |
2302 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2303 | uint32_t policer_burst) | |
2304 | { | |
2305 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2306 | struct ingress_policer *policer; | |
2307 | ||
2308 | /* Force to 0 if no rate specified, | |
2309 | * default to 8000 kbits if burst is 0, | |
2310 | * else stick with user-specified value. | |
2311 | */ | |
2312 | policer_burst = (!policer_rate ? 0 | |
2313 | : !policer_burst ? 8000 | |
2314 | : policer_burst); | |
2315 | ||
2316 | ovs_mutex_lock(&dev->mutex); | |
2317 | ||
2318 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2319 | &dev->ingress_policer); | |
2320 | ||
2321 | if (dev->policer_rate == policer_rate && | |
2322 | dev->policer_burst == policer_burst) { | |
2323 | /* Assume that settings haven't changed since we last set them. */ | |
2324 | ovs_mutex_unlock(&dev->mutex); | |
2325 | return 0; | |
2326 | } | |
2327 | ||
2328 | /* Destroy any existing ingress policer for the device if one exists */ | |
2329 | if (policer) { | |
2330 | ovsrcu_postpone(free, policer); | |
2331 | } | |
2332 | ||
2333 | if (policer_rate != 0) { | |
2334 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2335 | } else { | |
2336 | policer = NULL; | |
2337 | } | |
2338 | ovsrcu_set(&dev->ingress_policer, policer); | |
2339 | dev->policer_rate = policer_rate; | |
2340 | dev->policer_burst = policer_burst; | |
2341 | ovs_mutex_unlock(&dev->mutex); | |
2342 | ||
2343 | return 0; | |
2344 | } | |
2345 | ||
8a9562d2 PS |
2346 | static int |
2347 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2348 | { | |
2349 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
2350 | |
2351 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
2352 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
2353 | * postive integer to meet RFC 2863 recommendations. | |
2354 | */ | |
2355 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
2356 | ovs_mutex_unlock(&dev->mutex); |
2357 | ||
2358 | return ifindex; | |
2359 | } | |
2360 | ||
2361 | static int | |
d46285a2 | 2362 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2363 | { |
d46285a2 | 2364 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2365 | |
2366 | ovs_mutex_lock(&dev->mutex); | |
2367 | check_link_status(dev); | |
2368 | *carrier = dev->link.link_status; | |
58397e6c KT |
2369 | |
2370 | ovs_mutex_unlock(&dev->mutex); | |
2371 | ||
2372 | return 0; | |
2373 | } | |
2374 | ||
2375 | static int | |
d46285a2 | 2376 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2377 | { |
d46285a2 | 2378 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2379 | |
2380 | ovs_mutex_lock(&dev->mutex); | |
2381 | ||
0a0f39df | 2382 | if (is_vhost_running(dev)) { |
58397e6c KT |
2383 | *carrier = 1; |
2384 | } else { | |
2385 | *carrier = 0; | |
2386 | } | |
2387 | ||
8a9562d2 PS |
2388 | ovs_mutex_unlock(&dev->mutex); |
2389 | ||
2390 | return 0; | |
2391 | } | |
2392 | ||
2393 | static long long int | |
d46285a2 | 2394 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2395 | { |
d46285a2 | 2396 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2397 | long long int carrier_resets; |
2398 | ||
2399 | ovs_mutex_lock(&dev->mutex); | |
2400 | carrier_resets = dev->link_reset_cnt; | |
2401 | ovs_mutex_unlock(&dev->mutex); | |
2402 | ||
2403 | return carrier_resets; | |
2404 | } | |
2405 | ||
2406 | static int | |
d46285a2 | 2407 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2408 | long long int interval OVS_UNUSED) |
2409 | { | |
ee32150e | 2410 | return EOPNOTSUPP; |
8a9562d2 PS |
2411 | } |
2412 | ||
2413 | static int | |
2414 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2415 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2416 | enum netdev_flags *old_flagsp) |
2417 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 2418 | { |
8a9562d2 PS |
2419 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
2420 | return EINVAL; | |
2421 | } | |
2422 | ||
2423 | *old_flagsp = dev->flags; | |
2424 | dev->flags |= on; | |
2425 | dev->flags &= ~off; | |
2426 | ||
2427 | if (dev->flags == *old_flagsp) { | |
2428 | return 0; | |
2429 | } | |
2430 | ||
58397e6c | 2431 | if (dev->type == DPDK_DEV_ETH) { |
58397e6c KT |
2432 | if (dev->flags & NETDEV_PROMISC) { |
2433 | rte_eth_promiscuous_enable(dev->port_id); | |
2434 | } | |
8a9562d2 | 2435 | |
314fb5ad | 2436 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
2437 | } else { |
2438 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2439 | * running then change netdev's change_seq to trigger link state | |
2440 | * update. */ | |
e543851d ZB |
2441 | |
2442 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2443 | && is_vhost_running(dev)) { |
e543851d ZB |
2444 | netdev_change_seq_changed(&dev->up); |
2445 | ||
2446 | /* Clear statistics if device is getting up. */ | |
2447 | if (NETDEV_UP & on) { | |
2448 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 2449 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
2450 | rte_spinlock_unlock(&dev->stats_lock); |
2451 | } | |
2452 | } | |
8a9562d2 PS |
2453 | } |
2454 | ||
2455 | return 0; | |
2456 | } | |
2457 | ||
2458 | static int | |
d46285a2 | 2459 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2460 | enum netdev_flags off, enum netdev_flags on, |
2461 | enum netdev_flags *old_flagsp) | |
2462 | { | |
d46285a2 | 2463 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2464 | int error; |
2465 | ||
d46285a2 DDP |
2466 | ovs_mutex_lock(&dev->mutex); |
2467 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2468 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2469 | |
2470 | return error; | |
2471 | } | |
2472 | ||
2473 | static int | |
d46285a2 | 2474 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2475 | { |
d46285a2 | 2476 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2477 | struct rte_eth_dev_info dev_info; |
2478 | ||
7cd1261d | 2479 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2480 | return ENODEV; |
7cd1261d | 2481 | } |
8a9562d2 PS |
2482 | |
2483 | ovs_mutex_lock(&dev->mutex); | |
2484 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2485 | ovs_mutex_unlock(&dev->mutex); | |
2486 | ||
95fb793a | 2487 | smap_add_format(args, "port_no", "%d", dev->port_id); |
58be5c0e MK |
2488 | smap_add_format(args, "numa_id", "%d", |
2489 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
2490 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
2491 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2492 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2493 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2494 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2495 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
2496 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
2497 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
2498 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
2499 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2500 | ||
39c2baa9 | 2501 | if (dev_info.pci_dev) { |
2502 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2503 | dev_info.pci_dev->id.vendor_id); | |
2504 | smap_add_format(args, "pci-device_id", "0x%x", | |
2505 | dev_info.pci_dev->id.device_id); | |
2506 | } | |
8a9562d2 PS |
2507 | |
2508 | return 0; | |
2509 | } | |
2510 | ||
2511 | static void | |
2512 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2513 | OVS_REQUIRES(dev->mutex) | |
2514 | { | |
2515 | enum netdev_flags old_flags; | |
2516 | ||
2517 | if (admin_state) { | |
2518 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2519 | } else { | |
2520 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2521 | } | |
2522 | } | |
2523 | ||
2524 | static void | |
2525 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2526 | const char *argv[], void *aux OVS_UNUSED) | |
2527 | { | |
2528 | bool up; | |
2529 | ||
2530 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2531 | up = true; | |
2532 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2533 | up = false; | |
2534 | } else { | |
2535 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2536 | return; | |
2537 | } | |
2538 | ||
2539 | if (argc > 2) { | |
2540 | struct netdev *netdev = netdev_from_name(argv[1]); | |
2541 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
2542 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
2543 | ||
2544 | ovs_mutex_lock(&dpdk_dev->mutex); | |
2545 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
2546 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
2547 | ||
2548 | netdev_close(netdev); | |
2549 | } else { | |
2550 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2551 | netdev_close(netdev); | |
2552 | return; | |
2553 | } | |
2554 | } else { | |
2555 | struct netdev_dpdk *netdev; | |
2556 | ||
2557 | ovs_mutex_lock(&dpdk_mutex); | |
2558 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
2559 | ovs_mutex_lock(&netdev->mutex); | |
2560 | netdev_dpdk_set_admin_state__(netdev, up); | |
2561 | ovs_mutex_unlock(&netdev->mutex); | |
2562 | } | |
2563 | ovs_mutex_unlock(&dpdk_mutex); | |
2564 | } | |
2565 | unixctl_command_reply(conn, "OK"); | |
2566 | } | |
2567 | ||
0ee821c2 DB |
2568 | static void |
2569 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2570 | const char *argv[], void *aux OVS_UNUSED) | |
2571 | { | |
2572 | int ret; | |
2573 | char *response; | |
2574 | uint8_t port_id; | |
2575 | char devname[RTE_ETH_NAME_MAX_LEN]; | |
2576 | struct netdev_dpdk *dev; | |
2577 | ||
2578 | ovs_mutex_lock(&dpdk_mutex); | |
2579 | ||
2580 | if (!rte_eth_dev_count() || rte_eth_dev_get_port_by_name(argv[1], | |
2581 | &port_id)) { | |
2582 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); | |
2583 | goto error; | |
2584 | } | |
2585 | ||
2586 | dev = netdev_dpdk_lookup_by_port_id(port_id); | |
2587 | if (dev) { | |
2588 | response = xasprintf("Device '%s' is being used by interface '%s'. " | |
2589 | "Remove it before detaching", | |
2590 | argv[1], netdev_get_name(&dev->up)); | |
2591 | goto error; | |
2592 | } | |
2593 | ||
2594 | rte_eth_dev_close(port_id); | |
2595 | ||
2596 | ret = rte_eth_dev_detach(port_id, devname); | |
2597 | if (ret < 0) { | |
2598 | response = xasprintf("Device '%s' can not be detached", argv[1]); | |
2599 | goto error; | |
2600 | } | |
2601 | ||
2602 | response = xasprintf("Device '%s' has been detached", argv[1]); | |
2603 | ||
2604 | ovs_mutex_unlock(&dpdk_mutex); | |
2605 | unixctl_command_reply(conn, response); | |
2606 | free(response); | |
2607 | return; | |
2608 | ||
2609 | error: | |
2610 | ovs_mutex_unlock(&dpdk_mutex); | |
2611 | unixctl_command_reply_error(conn, response); | |
2612 | free(response); | |
2613 | } | |
2614 | ||
58397e6c KT |
2615 | /* |
2616 | * Set virtqueue flags so that we do not receive interrupts. | |
2617 | */ | |
2618 | static void | |
0a0f39df | 2619 | set_irq_status(int vid) |
58397e6c | 2620 | { |
4573fbd3 | 2621 | uint32_t i; |
4573fbd3 | 2622 | |
f3e7ec25 MW |
2623 | for (i = 0; i < rte_vhost_get_vring_num(vid); i++) { |
2624 | rte_vhost_enable_guest_notification(vid, i, 0); | |
4573fbd3 FL |
2625 | } |
2626 | } | |
2627 | ||
585a5bea IM |
2628 | /* |
2629 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2630 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2631 | */ |
2632 | static void | |
d46285a2 DDP |
2633 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2634 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2635 | { |
2636 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2637 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 2638 | |
eff23640 | 2639 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
2640 | |
2641 | for (i = 0; i < total_txqs; i++) { | |
2642 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2643 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2644 | enabled_queues[n_enabled++] = i; |
2645 | } | |
2646 | } | |
2647 | ||
2648 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2649 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2650 | n_enabled = 1; |
2651 | } | |
2652 | ||
2653 | k = 0; | |
2654 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2655 | if (dev->tx_q[i].map != i) { |
2656 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2657 | k = (k + 1) % n_enabled; |
2658 | } | |
2659 | } | |
2660 | ||
2d24d165 | 2661 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2662 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2663 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2664 | } |
2665 | ||
eff23640 | 2666 | free(enabled_queues); |
585a5bea | 2667 | } |
4573fbd3 | 2668 | |
58397e6c KT |
2669 | /* |
2670 | * A new virtio-net device is added to a vhost port. | |
2671 | */ | |
2672 | static int | |
0a0f39df | 2673 | new_device(int vid) |
58397e6c | 2674 | { |
d46285a2 | 2675 | struct netdev_dpdk *dev; |
58397e6c | 2676 | bool exists = false; |
db8f13b0 | 2677 | int newnode = 0; |
0a0f39df CL |
2678 | char ifname[IF_NAME_SZ]; |
2679 | ||
58be5c0e | 2680 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2681 | |
2682 | ovs_mutex_lock(&dpdk_mutex); | |
2683 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2684 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 2685 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2686 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
f3e7ec25 | 2687 | uint32_t qp_num = rte_vhost_get_vring_num(vid)/VIRTIO_QNUM; |
db8f13b0 CL |
2688 | |
2689 | /* Get NUMA information */ | |
0a0f39df CL |
2690 | newnode = rte_vhost_get_numa_node(vid); |
2691 | if (newnode == -1) { | |
5b9bf9e0 | 2692 | #ifdef VHOST_NUMA |
db8f13b0 | 2693 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2694 | ifname); |
5b9bf9e0 | 2695 | #endif |
db8f13b0 | 2696 | newnode = dev->socket_id; |
db8f13b0 CL |
2697 | } |
2698 | ||
7f5f2bd0 IM |
2699 | if (dev->requested_n_txq != qp_num |
2700 | || dev->requested_n_rxq != qp_num | |
2701 | || dev->requested_socket_id != newnode) { | |
2702 | dev->requested_socket_id = newnode; | |
2703 | dev->requested_n_rxq = qp_num; | |
2704 | dev->requested_n_txq = qp_num; | |
2705 | netdev_request_reconfigure(&dev->up); | |
2706 | } else { | |
2707 | /* Reconfiguration not required. */ | |
2708 | dev->vhost_reconfigured = true; | |
2709 | } | |
81acebda | 2710 | |
0a0f39df | 2711 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2712 | exists = true; |
2713 | ||
58397e6c | 2714 | /* Disable notifications. */ |
0a0f39df | 2715 | set_irq_status(vid); |
e543851d | 2716 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2717 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2718 | break; |
2719 | } | |
c1ff66ac | 2720 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2721 | } |
2722 | ovs_mutex_unlock(&dpdk_mutex); | |
2723 | ||
2724 | if (!exists) { | |
0a0f39df | 2725 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2726 | |
2727 | return -1; | |
2728 | } | |
2729 | ||
0a0f39df CL |
2730 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2731 | ifname, newnode); | |
2732 | ||
58397e6c KT |
2733 | return 0; |
2734 | } | |
2735 | ||
f3ea2ad2 IM |
2736 | /* Clears mapping for all available queues of vhost interface. */ |
2737 | static void | |
2738 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2739 | OVS_REQUIRES(dev->mutex) | |
2740 | { | |
2741 | int i; | |
2742 | ||
81acebda | 2743 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
2744 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
2745 | } | |
2746 | } | |
2747 | ||
58397e6c KT |
2748 | /* |
2749 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2750 | * flag to stop any more packets from being sent or received to/from a VM and | |
2751 | * ensure all currently queued packets have been sent/received before removing | |
2752 | * the device. | |
2753 | */ | |
2754 | static void | |
0a0f39df | 2755 | destroy_device(int vid) |
58397e6c | 2756 | { |
d46285a2 | 2757 | struct netdev_dpdk *dev; |
afee281f | 2758 | bool exists = false; |
0a0f39df CL |
2759 | char ifname[IF_NAME_SZ]; |
2760 | ||
58be5c0e | 2761 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2762 | |
2763 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2764 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2765 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 2766 | |
d46285a2 | 2767 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
2768 | dev->vhost_reconfigured = false; |
2769 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 2770 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 2771 | |
e543851d | 2772 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2773 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 2774 | exists = true; |
afee281f | 2775 | break; |
58397e6c KT |
2776 | } |
2777 | } | |
afee281f | 2778 | |
58397e6c KT |
2779 | ovs_mutex_unlock(&dpdk_mutex); |
2780 | ||
0a0f39df | 2781 | if (exists) { |
afee281f KT |
2782 | /* |
2783 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2784 | * to NULL, before returning. | |
2785 | */ | |
2786 | ovsrcu_synchronize(); | |
2787 | /* | |
2788 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2789 | * put thread back into quiescent state before returning. | |
2790 | */ | |
2791 | ovsrcu_quiesce_start(); | |
0a0f39df | 2792 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 2793 | } else { |
0a0f39df | 2794 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 2795 | } |
58397e6c KT |
2796 | } |
2797 | ||
585a5bea | 2798 | static int |
0a0f39df | 2799 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 2800 | { |
d46285a2 | 2801 | struct netdev_dpdk *dev; |
585a5bea IM |
2802 | bool exists = false; |
2803 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
2804 | char ifname[IF_NAME_SZ]; |
2805 | ||
58be5c0e | 2806 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
2807 | |
2808 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2809 | return 0; | |
2810 | } | |
2811 | ||
2812 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2813 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 2814 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2815 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 2816 | if (enable) { |
d46285a2 | 2817 | dev->tx_q[qid].map = qid; |
585a5bea | 2818 | } else { |
d46285a2 | 2819 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2820 | } |
d46285a2 | 2821 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2822 | exists = true; |
d46285a2 | 2823 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2824 | break; |
2825 | } | |
c1ff66ac | 2826 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2827 | } |
2828 | ovs_mutex_unlock(&dpdk_mutex); | |
2829 | ||
2830 | if (exists) { | |
0a0f39df CL |
2831 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
2832 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 2833 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 2834 | } else { |
0a0f39df | 2835 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
2836 | return -1; |
2837 | } | |
2838 | ||
2839 | return 0; | |
2840 | } | |
2841 | ||
0a0f39df CL |
2842 | int |
2843 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 2844 | { |
0a0f39df | 2845 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
2846 | } |
2847 | ||
9509913a IS |
2848 | struct ingress_policer * |
2849 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
2850 | { | |
2851 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
2852 | } | |
2853 | ||
58397e6c | 2854 | static int |
ecc1a34e | 2855 | netdev_dpdk_class_init(void) |
7d1ced01 | 2856 | { |
ecc1a34e DDP |
2857 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2858 | ||
2859 | /* This function can be called for different classes. The initialization | |
2860 | * needs to be done only once */ | |
2861 | if (ovsthread_once_start(&once)) { | |
2862 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
2863 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2864 | "[netdev] up|down", 1, 2, | |
2865 | netdev_dpdk_set_admin_state, NULL); | |
2866 | ||
0ee821c2 DB |
2867 | unixctl_command_register("netdev-dpdk/detach", |
2868 | "pci address of device", 1, 1, | |
2869 | netdev_dpdk_detach, NULL); | |
2870 | ||
ecc1a34e DDP |
2871 | ovsthread_once_done(&once); |
2872 | } | |
362ca396 | 2873 | |
7d1ced01 CL |
2874 | return 0; |
2875 | } | |
2876 | ||
033e9df2 | 2877 | |
95fb793a | 2878 | /* Client Rings */ |
2879 | ||
95fb793a | 2880 | static int |
2881 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
bb37956a | 2882 | dpdk_port_t *eth_port_id) |
95fb793a | 2883 | { |
48fffdee | 2884 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 2885 | char *ring_name; |
b8374d0d | 2886 | int port_id; |
95fb793a | 2887 | |
48fffdee KT |
2888 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
2889 | if (!ring_pair) { | |
95fb793a | 2890 | return ENOMEM; |
2891 | } | |
2892 | ||
7251515e | 2893 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 2894 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 2895 | |
8f0a76c9 | 2896 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 2897 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2898 | RING_F_SP_ENQ); |
0c6f39e5 | 2899 | free(ring_name); |
48fffdee KT |
2900 | if (ring_pair->cring_tx == NULL) { |
2901 | rte_free(ring_pair); | |
95fb793a | 2902 | return ENOMEM; |
2903 | } | |
2904 | ||
0c6f39e5 | 2905 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 2906 | |
8f0a76c9 | 2907 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 2908 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2909 | RING_F_SC_DEQ); |
0c6f39e5 | 2910 | free(ring_name); |
48fffdee KT |
2911 | if (ring_pair->cring_rx == NULL) { |
2912 | rte_free(ring_pair); | |
95fb793a | 2913 | return ENOMEM; |
2914 | } | |
2915 | ||
b8374d0d MV |
2916 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
2917 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 2918 | |
b8374d0d | 2919 | if (port_id < 0) { |
48fffdee | 2920 | rte_free(ring_pair); |
95fb793a | 2921 | return ENODEV; |
2922 | } | |
2923 | ||
48fffdee | 2924 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
2925 | ring_pair->eth_port_id = port_id; |
2926 | *eth_port_id = port_id; | |
2927 | ||
48fffdee | 2928 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 2929 | |
95fb793a | 2930 | return 0; |
2931 | } | |
2932 | ||
2933 | static int | |
bb37956a | 2934 | dpdk_ring_open(const char dev_name[], dpdk_port_t *eth_port_id) |
64839cf4 | 2935 | OVS_REQUIRES(dpdk_mutex) |
95fb793a | 2936 | { |
48fffdee | 2937 | struct dpdk_ring *ring_pair; |
95fb793a | 2938 | unsigned int port_no; |
2939 | int err = 0; | |
2940 | ||
2941 | /* Names always start with "dpdkr" */ | |
2942 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2943 | if (err) { | |
2944 | return err; | |
2945 | } | |
2946 | ||
58be5c0e | 2947 | /* Look through our list to find the device */ |
48fffdee KT |
2948 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
2949 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 2950 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 2951 | /* Really all that is needed */ |
48fffdee | 2952 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 2953 | return 0; |
2954 | } | |
2955 | } | |
2956 | /* Need to create the device rings */ | |
2957 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2958 | } | |
2959 | ||
7251515e | 2960 | static int |
d46285a2 | 2961 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
324c8374 IM |
2962 | struct dp_packet_batch *batch, bool may_steal, |
2963 | bool concurrent_txq) | |
7251515e | 2964 | { |
d46285a2 | 2965 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a543eb0 | 2966 | struct dp_packet *packet; |
1b99bb05 | 2967 | |
58be5c0e MK |
2968 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
2969 | * the rss hash field is clear. This is because the same mbuf may be | |
2970 | * modified by the consumer of the ring and return into the datapath | |
2971 | * without recalculating the RSS hash. */ | |
8a543eb0 BB |
2972 | DP_PACKET_BATCH_FOR_EACH (packet, batch) { |
2973 | dp_packet_mbuf_rss_flag_reset(packet); | |
1b99bb05 | 2974 | } |
7251515e | 2975 | |
324c8374 | 2976 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e DV |
2977 | return 0; |
2978 | } | |
2979 | ||
95fb793a | 2980 | static int |
2981 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2982 | { | |
bb37956a | 2983 | dpdk_port_t port_no = 0; |
95fb793a | 2984 | int err = 0; |
2985 | ||
95fb793a | 2986 | ovs_mutex_lock(&dpdk_mutex); |
2987 | ||
2988 | err = dpdk_ring_open(netdev->name, &port_no); | |
2989 | if (err) { | |
2990 | goto unlock_dpdk; | |
2991 | } | |
2992 | ||
1ce30dfd DDP |
2993 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
2994 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 2995 | unlock_dpdk: |
2996 | ovs_mutex_unlock(&dpdk_mutex); | |
2997 | return err; | |
2998 | } | |
2999 | ||
0bf765f7 IS |
3000 | /* QoS Functions */ |
3001 | ||
3002 | /* | |
3003 | * Initialize QoS configuration operations. | |
3004 | */ | |
3005 | static void | |
3006 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
3007 | { | |
3008 | conf->ops = ops; | |
78bd47cf | 3009 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
3010 | } |
3011 | ||
3012 | /* | |
3013 | * Search existing QoS operations in qos_ops and compare each set of | |
3014 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
3015 | * else return NULL | |
3016 | */ | |
3017 | static const struct dpdk_qos_ops * | |
3018 | qos_lookup_name(const char *name) | |
3019 | { | |
3020 | const struct dpdk_qos_ops *const *opsp; | |
3021 | ||
3022 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3023 | const struct dpdk_qos_ops *ops = *opsp; | |
3024 | if (!strcmp(name, ops->qos_name)) { | |
3025 | return ops; | |
3026 | } | |
3027 | } | |
3028 | return NULL; | |
3029 | } | |
3030 | ||
0bf765f7 IS |
3031 | static int |
3032 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
3033 | struct sset *types) | |
3034 | { | |
3035 | const struct dpdk_qos_ops *const *opsp; | |
3036 | ||
3037 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3038 | const struct dpdk_qos_ops *ops = *opsp; | |
3039 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
3040 | sset_add(types, ops->qos_name); | |
3041 | } | |
3042 | } | |
3043 | return 0; | |
3044 | } | |
3045 | ||
3046 | static int | |
d46285a2 | 3047 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
3048 | const char **typep, struct smap *details) |
3049 | { | |
d46285a2 | 3050 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 3051 | struct qos_conf *qos_conf; |
0bf765f7 IS |
3052 | int error = 0; |
3053 | ||
d46285a2 | 3054 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
3055 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
3056 | if (qos_conf) { | |
3057 | *typep = qos_conf->ops->qos_name; | |
3058 | error = (qos_conf->ops->qos_get | |
3059 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
3060 | } else { |
3061 | /* No QoS configuration set, return an empty string */ | |
3062 | *typep = ""; | |
0bf765f7 | 3063 | } |
d46285a2 | 3064 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
3065 | |
3066 | return error; | |
3067 | } | |
3068 | ||
3069 | static int | |
78bd47cf DDP |
3070 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
3071 | const struct smap *details) | |
0bf765f7 | 3072 | { |
d46285a2 | 3073 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 3074 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 3075 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
3076 | int error = 0; |
3077 | ||
d46285a2 | 3078 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 3079 | |
78bd47cf | 3080 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 3081 | |
78bd47cf DDP |
3082 | new_ops = qos_lookup_name(type); |
3083 | ||
3084 | if (!new_ops || !new_ops->qos_construct) { | |
3085 | new_qos_conf = NULL; | |
3086 | if (type && type[0]) { | |
3087 | error = EOPNOTSUPP; | |
0bf765f7 | 3088 | } |
44975bb0 | 3089 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
3090 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
3091 | new_qos_conf = qos_conf; | |
0bf765f7 | 3092 | } else { |
78bd47cf | 3093 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
3094 | } |
3095 | ||
7ea266e9 | 3096 | if (error) { |
78bd47cf DDP |
3097 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
3098 | type, netdev->name, rte_strerror(error)); | |
3099 | } | |
3100 | ||
3101 | if (new_qos_conf != qos_conf) { | |
3102 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
3103 | if (qos_conf) { | |
3104 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
3105 | } | |
0bf765f7 IS |
3106 | } |
3107 | ||
d46285a2 | 3108 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 3109 | |
0bf765f7 IS |
3110 | return error; |
3111 | } | |
3112 | ||
3113 | /* egress-policer details */ | |
3114 | ||
3115 | struct egress_policer { | |
3116 | struct qos_conf qos_conf; | |
3117 | struct rte_meter_srtcm_params app_srtcm_params; | |
3118 | struct rte_meter_srtcm egress_meter; | |
3119 | }; | |
3120 | ||
78bd47cf DDP |
3121 | static void |
3122 | egress_policer_details_to_param(const struct smap *details, | |
3123 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 3124 | { |
78bd47cf DDP |
3125 | memset(params, 0, sizeof *params); |
3126 | params->cir = smap_get_ullong(details, "cir", 0); | |
3127 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
3128 | params->ebs = 0; | |
0bf765f7 IS |
3129 | } |
3130 | ||
3131 | static int | |
78bd47cf DDP |
3132 | egress_policer_qos_construct(const struct smap *details, |
3133 | struct qos_conf **conf) | |
0bf765f7 | 3134 | { |
0bf765f7 | 3135 | struct egress_policer *policer; |
0bf765f7 IS |
3136 | int err = 0; |
3137 | ||
0bf765f7 IS |
3138 | policer = xmalloc(sizeof *policer); |
3139 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 3140 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
0bf765f7 | 3141 | err = rte_meter_srtcm_config(&policer->egress_meter, |
78bd47cf DDP |
3142 | &policer->app_srtcm_params); |
3143 | if (!err) { | |
3144 | *conf = &policer->qos_conf; | |
3145 | } else { | |
7ea266e9 | 3146 | free(policer); |
78bd47cf | 3147 | *conf = NULL; |
7ea266e9 IS |
3148 | err = -err; |
3149 | } | |
0bf765f7 IS |
3150 | |
3151 | return err; | |
3152 | } | |
3153 | ||
3154 | static void | |
78bd47cf | 3155 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
3156 | { |
3157 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 3158 | qos_conf); |
0bf765f7 IS |
3159 | free(policer); |
3160 | } | |
3161 | ||
3162 | static int | |
78bd47cf | 3163 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 3164 | { |
78bd47cf DDP |
3165 | struct egress_policer *policer = |
3166 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3167 | ||
3168 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
3169 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 3170 | |
0bf765f7 IS |
3171 | return 0; |
3172 | } | |
3173 | ||
78bd47cf | 3174 | static bool |
47a45d86 KT |
3175 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
3176 | const struct smap *details) | |
0bf765f7 | 3177 | { |
78bd47cf DDP |
3178 | struct egress_policer *policer = |
3179 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3180 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 3181 | |
78bd47cf | 3182 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 3183 | |
78bd47cf | 3184 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
3185 | } |
3186 | ||
0bf765f7 | 3187 | static int |
3e90f7d7 GZ |
3188 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt, |
3189 | bool may_steal) | |
0bf765f7 | 3190 | { |
0bf765f7 | 3191 | int cnt = 0; |
78bd47cf DDP |
3192 | struct egress_policer *policer = |
3193 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 3194 | |
3e90f7d7 GZ |
3195 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, |
3196 | pkt_cnt, may_steal); | |
0bf765f7 IS |
3197 | |
3198 | return cnt; | |
3199 | } | |
3200 | ||
3201 | static const struct dpdk_qos_ops egress_policer_ops = { | |
3202 | "egress-policer", /* qos_name */ | |
3203 | egress_policer_qos_construct, | |
3204 | egress_policer_qos_destruct, | |
3205 | egress_policer_qos_get, | |
78bd47cf | 3206 | egress_policer_qos_is_equal, |
0bf765f7 IS |
3207 | egress_policer_run |
3208 | }; | |
3209 | ||
050c60bf DDP |
3210 | static int |
3211 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
3212 | { | |
3213 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3214 | int err = 0; | |
3215 | ||
050c60bf DDP |
3216 | ovs_mutex_lock(&dev->mutex); |
3217 | ||
3218 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 3219 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b CL |
3220 | && dev->mtu == dev->requested_mtu |
3221 | && dev->rxq_size == dev->requested_rxq_size | |
bd4e172b | 3222 | && dev->txq_size == dev->requested_txq_size |
3223 | && dev->socket_id == dev->requested_socket_id) { | |
050c60bf DDP |
3224 | /* Reconfiguration is unnecessary */ |
3225 | ||
3226 | goto out; | |
3227 | } | |
3228 | ||
3229 | rte_eth_dev_stop(dev->port_id); | |
3230 | ||
d555d9bd | 3231 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 | 3232 | if (err && err != EEXIST) { |
d555d9bd | 3233 | goto out; |
0072e931 MK |
3234 | } |
3235 | ||
050c60bf DDP |
3236 | netdev->n_txq = dev->requested_n_txq; |
3237 | netdev->n_rxq = dev->requested_n_rxq; | |
3238 | ||
b685696b CL |
3239 | dev->rxq_size = dev->requested_rxq_size; |
3240 | dev->txq_size = dev->requested_txq_size; | |
3241 | ||
050c60bf DDP |
3242 | rte_free(dev->tx_q); |
3243 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
3244 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
3245 | if (!dev->tx_q) { | |
3246 | err = ENOMEM; | |
3247 | } | |
050c60bf | 3248 | |
0072e931 MK |
3249 | netdev_change_seq_changed(netdev); |
3250 | ||
050c60bf | 3251 | out: |
050c60bf | 3252 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
3253 | return err; |
3254 | } | |
3255 | ||
7f381c2e | 3256 | static int |
2d24d165 | 3257 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 3258 | OVS_REQUIRES(dev->mutex) |
050c60bf | 3259 | { |
2d24d165 CL |
3260 | dev->up.n_txq = dev->requested_n_txq; |
3261 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 3262 | int err; |
050c60bf | 3263 | |
81acebda IM |
3264 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
3265 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
3266 | dev->tx_q[0].map = 0; | |
3267 | } | |
3268 | ||
3269 | netdev_dpdk_remap_txqs(dev); | |
3270 | ||
d555d9bd | 3271 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 FA |
3272 | if (!err) { |
3273 | /* A new mempool was created. */ | |
d555d9bd | 3274 | netdev_change_seq_changed(&dev->up); |
b6b26021 FA |
3275 | } else if (err != EEXIST){ |
3276 | return err; | |
db8f13b0 | 3277 | } |
0a0f39df | 3278 | if (netdev_dpdk_get_vid(dev) >= 0) { |
894af647 | 3279 | if (dev->vhost_reconfigured == false) { |
3280 | dev->vhost_reconfigured = true; | |
3281 | /* Carrier status may need updating. */ | |
3282 | netdev_change_seq_changed(&dev->up); | |
3283 | } | |
81acebda | 3284 | } |
7f381c2e DDP |
3285 | |
3286 | return 0; | |
2d24d165 CL |
3287 | } |
3288 | ||
3289 | static int | |
3290 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
3291 | { | |
3292 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3293 | int err; |
2d24d165 | 3294 | |
2d24d165 | 3295 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 3296 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 3297 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
3298 | |
3299 | return err; | |
2d24d165 CL |
3300 | } |
3301 | ||
3302 | static int | |
3303 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3304 | { | |
3305 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3306 | int err; |
2d24d165 | 3307 | |
2d24d165 CL |
3308 | ovs_mutex_lock(&dev->mutex); |
3309 | ||
c1ff66ac CL |
3310 | /* Configure vHost client mode if requested and if the following criteria |
3311 | * are met: | |
2d24d165 CL |
3312 | * 1. Device hasn't been registered yet. |
3313 | * 2. A path has been specified. | |
c1ff66ac CL |
3314 | */ |
3315 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 CL |
3316 | && strlen(dev->vhost_id)) { |
3317 | /* Register client-mode device */ | |
3318 | err = rte_vhost_driver_register(dev->vhost_id, | |
3319 | RTE_VHOST_USER_CLIENT); | |
c1ff66ac | 3320 | if (err) { |
2d24d165 CL |
3321 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3322 | dev->vhost_id); | |
7f381c2e | 3323 | goto unlock; |
c1ff66ac | 3324 | } else { |
2d24d165 CL |
3325 | /* Configuration successful */ |
3326 | dev->vhost_driver_flags |= RTE_VHOST_USER_CLIENT; | |
3327 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " | |
3328 | "using client socket '%s'", | |
3329 | dev->up.name, dev->vhost_id); | |
c1ff66ac | 3330 | } |
f3e7ec25 MW |
3331 | |
3332 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
3333 | &virtio_net_device_ops); | |
3334 | if (err) { | |
3335 | VLOG_ERR("rte_vhost_driver_callback_register failed for " | |
3336 | "vhost user client port: %s\n", dev->up.name); | |
3337 | goto unlock; | |
3338 | } | |
3339 | ||
3340 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
3341 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
3342 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
3343 | | 1ULL << VIRTIO_NET_F_CSUM); | |
3344 | if (err) { | |
3345 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
3346 | "client port: %s\n", dev->up.name); | |
3347 | goto unlock; | |
3348 | } | |
3349 | ||
3350 | err = rte_vhost_driver_start(dev->vhost_id); | |
3351 | if (err) { | |
3352 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
3353 | "client port: %s\n", dev->up.name); | |
3354 | goto unlock; | |
3355 | } | |
c1ff66ac CL |
3356 | } |
3357 | ||
7f381c2e DDP |
3358 | err = dpdk_vhost_reconfigure_helper(dev); |
3359 | ||
3360 | unlock: | |
050c60bf | 3361 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 3362 | |
7f381c2e | 3363 | return err; |
050c60bf DDP |
3364 | } |
3365 | ||
ecc1a34e | 3366 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
81acebda IM |
3367 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
3368 | GET_CARRIER, GET_STATS, \ | |
3369 | GET_FEATURES, GET_STATUS, \ | |
3370 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3371 | { \ |
3372 | NAME, \ | |
118c77b1 | 3373 | true, /* is_pmd */ \ |
ecc1a34e | 3374 | INIT, /* init */ \ |
95fb793a | 3375 | NULL, /* netdev_dpdk_run */ \ |
3376 | NULL, /* netdev_dpdk_wait */ \ | |
3377 | \ | |
3378 | netdev_dpdk_alloc, \ | |
3379 | CONSTRUCT, \ | |
58397e6c | 3380 | DESTRUCT, \ |
95fb793a | 3381 | netdev_dpdk_dealloc, \ |
3382 | netdev_dpdk_get_config, \ | |
81acebda | 3383 | SET_CONFIG, \ |
95fb793a | 3384 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3385 | NULL, /* build header */ \ |
3386 | NULL, /* push header */ \ | |
3387 | NULL, /* pop header */ \ | |
7dec44fe | 3388 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3389 | SET_TX_MULTIQ, \ |
95fb793a | 3390 | \ |
7251515e | 3391 | SEND, /* send */ \ |
95fb793a | 3392 | NULL, /* send_wait */ \ |
3393 | \ | |
3394 | netdev_dpdk_set_etheraddr, \ | |
3395 | netdev_dpdk_get_etheraddr, \ | |
3396 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3397 | netdev_dpdk_set_mtu, \ |
95fb793a | 3398 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3399 | GET_CARRIER, \ |
95fb793a | 3400 | netdev_dpdk_get_carrier_resets, \ |
3401 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
3402 | GET_STATS, \ |
3403 | GET_FEATURES, \ | |
95fb793a | 3404 | NULL, /* set_advertisements */ \ |
875ab130 | 3405 | NULL, /* get_pt_mode */ \ |
95fb793a | 3406 | \ |
9509913a | 3407 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3408 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3409 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3410 | netdev_dpdk_get_qos, \ |
3411 | netdev_dpdk_set_qos, \ | |
95fb793a | 3412 | NULL, /* get_queue */ \ |
3413 | NULL, /* set_queue */ \ | |
3414 | NULL, /* delete_queue */ \ | |
3415 | NULL, /* get_queue_stats */ \ | |
3416 | NULL, /* queue_dump_start */ \ | |
3417 | NULL, /* queue_dump_next */ \ | |
3418 | NULL, /* queue_dump_done */ \ | |
3419 | NULL, /* dump_queue_stats */ \ | |
3420 | \ | |
95fb793a | 3421 | NULL, /* set_in4 */ \ |
a8704b50 | 3422 | NULL, /* get_addr_list */ \ |
95fb793a | 3423 | NULL, /* add_router */ \ |
3424 | NULL, /* get_next_hop */ \ | |
58397e6c | 3425 | GET_STATUS, \ |
95fb793a | 3426 | NULL, /* arp_lookup */ \ |
3427 | \ | |
3428 | netdev_dpdk_update_flags, \ | |
050c60bf | 3429 | RECONFIGURE, \ |
95fb793a | 3430 | \ |
3431 | netdev_dpdk_rxq_alloc, \ | |
3432 | netdev_dpdk_rxq_construct, \ | |
3433 | netdev_dpdk_rxq_destruct, \ | |
3434 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3435 | RXQ_RECV, \ |
95fb793a | 3436 | NULL, /* rx_wait */ \ |
3437 | NULL, /* rxq_drain */ \ | |
18ebd48c | 3438 | NO_OFFLOAD_API \ |
95fb793a | 3439 | } |
8a9562d2 | 3440 | |
bce01e3a | 3441 | static const struct netdev_class dpdk_class = |
95fb793a | 3442 | NETDEV_DPDK_CLASS( |
3443 | "dpdk", | |
ecc1a34e | 3444 | netdev_dpdk_class_init, |
5496878c | 3445 | netdev_dpdk_construct, |
58397e6c | 3446 | netdev_dpdk_destruct, |
81acebda IM |
3447 | netdev_dpdk_set_config, |
3448 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3449 | netdev_dpdk_eth_send, |
3450 | netdev_dpdk_get_carrier, | |
3451 | netdev_dpdk_get_stats, | |
3452 | netdev_dpdk_get_features, | |
3453 | netdev_dpdk_get_status, | |
050c60bf | 3454 | netdev_dpdk_reconfigure, |
58397e6c | 3455 | netdev_dpdk_rxq_recv); |
95fb793a | 3456 | |
bce01e3a | 3457 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3458 | NETDEV_DPDK_CLASS( |
3459 | "dpdkr", | |
ecc1a34e | 3460 | netdev_dpdk_class_init, |
5496878c | 3461 | netdev_dpdk_ring_construct, |
58397e6c | 3462 | netdev_dpdk_destruct, |
c3d062a7 | 3463 | netdev_dpdk_ring_set_config, |
81acebda | 3464 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3465 | netdev_dpdk_ring_send, |
3466 | netdev_dpdk_get_carrier, | |
3467 | netdev_dpdk_get_stats, | |
3468 | netdev_dpdk_get_features, | |
3469 | netdev_dpdk_get_status, | |
050c60bf | 3470 | netdev_dpdk_reconfigure, |
58397e6c KT |
3471 | netdev_dpdk_rxq_recv); |
3472 | ||
53f50d24 | 3473 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3474 | NETDEV_DPDK_CLASS( |
3475 | "dpdkvhostuser", | |
f3e7ec25 | 3476 | NULL, |
53f50d24 | 3477 | netdev_dpdk_vhost_construct, |
58397e6c | 3478 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3479 | NULL, |
81acebda | 3480 | NULL, |
58397e6c KT |
3481 | netdev_dpdk_vhost_send, |
3482 | netdev_dpdk_vhost_get_carrier, | |
3483 | netdev_dpdk_vhost_get_stats, | |
3484 | NULL, | |
7251515e | 3485 | NULL, |
53f50d24 | 3486 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3487 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3488 | static const struct netdev_class dpdk_vhost_client_class = |
3489 | NETDEV_DPDK_CLASS( | |
3490 | "dpdkvhostuserclient", | |
f3e7ec25 | 3491 | NULL, |
2d24d165 CL |
3492 | netdev_dpdk_vhost_client_construct, |
3493 | netdev_dpdk_vhost_destruct, | |
3494 | netdev_dpdk_vhost_client_set_config, | |
3495 | NULL, | |
3496 | netdev_dpdk_vhost_send, | |
3497 | netdev_dpdk_vhost_get_carrier, | |
3498 | netdev_dpdk_vhost_get_stats, | |
3499 | NULL, | |
3500 | NULL, | |
3501 | netdev_dpdk_vhost_client_reconfigure, | |
3502 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 3503 | |
8a9562d2 PS |
3504 | void |
3505 | netdev_dpdk_register(void) | |
3506 | { | |
bab69409 AC |
3507 | netdev_register_provider(&dpdk_class); |
3508 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3509 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 3510 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 3511 | } |