]>
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 | |
6ebc4b09 | 20 | #include <errno.h> |
8a9562d2 PS |
21 | #include <signal.h> |
22 | #include <stdlib.h> | |
6ebc4b09 | 23 | #include <string.h> |
8a9562d2 | 24 | #include <unistd.h> |
f3e7ec25 MW |
25 | #include <linux/virtio_net.h> |
26 | #include <sys/socket.h> | |
27 | #include <linux/if.h> | |
01961bbd | 28 | |
5e925ccc | 29 | #include <rte_bus_pci.h> |
01961bbd DDP |
30 | #include <rte_config.h> |
31 | #include <rte_cycles.h> | |
32 | #include <rte_errno.h> | |
33 | #include <rte_eth_ring.h> | |
34 | #include <rte_ethdev.h> | |
6ebc4b09 | 35 | #include <rte_flow.h> |
01961bbd DDP |
36 | #include <rte_malloc.h> |
37 | #include <rte_mbuf.h> | |
38 | #include <rte_meter.h> | |
fc56f5e0 | 39 | #include <rte_pci.h> |
3eb8d4fa | 40 | #include <rte_version.h> |
6ebc4b09 | 41 | #include <rte_vhost.h> |
8a9562d2 | 42 | |
e8a2b5bf | 43 | #include "cmap.h" |
7d1ced01 | 44 | #include "dirs.h" |
e14deea0 | 45 | #include "dp-packet.h" |
01961bbd | 46 | #include "dpdk.h" |
8a9562d2 | 47 | #include "dpif-netdev.h" |
e5c0f5a4 | 48 | #include "fatal-signal.h" |
8a9562d2 PS |
49 | #include "netdev-provider.h" |
50 | #include "netdev-vport.h" | |
51 | #include "odp-util.h" | |
eac84432 | 52 | #include "openvswitch/dynamic-string.h" |
25d436fb | 53 | #include "openvswitch/list.h" |
6ebc4b09 | 54 | #include "openvswitch/match.h" |
25d436fb | 55 | #include "openvswitch/ofp-print.h" |
6ebc4b09 | 56 | #include "openvswitch/shash.h" |
25d436fb | 57 | #include "openvswitch/vlog.h" |
94143fc4 | 58 | #include "ovs-numa.h" |
8a9562d2 | 59 | #include "ovs-rcu.h" |
6ebc4b09 | 60 | #include "ovs-thread.h" |
8a9562d2 | 61 | #include "packets.h" |
0bf765f7 | 62 | #include "smap.h" |
8a9562d2 | 63 | #include "sset.h" |
8a9562d2 | 64 | #include "timeval.h" |
6ebc4b09 | 65 | #include "unaligned.h" |
8a9562d2 | 66 | #include "unixctl.h" |
6ebc4b09 IM |
67 | #include "util.h" |
68 | #include "uuid.h" | |
8a9562d2 | 69 | |
f3e7ec25 MW |
70 | enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM}; |
71 | ||
05b49df6 | 72 | VLOG_DEFINE_THIS_MODULE(netdev_dpdk); |
8a9562d2 PS |
73 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
74 | ||
75 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
76 | ||
77 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
78 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
79 | ||
80 | /* | |
81 | * need to reserve tons of extra space in the mbufs so we can align the | |
82 | * DMA addresses to 4KB. | |
18f777b2 TP |
83 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
84 | * performance for standard Ethernet MTU. | |
8a9562d2 | 85 | */ |
58be5c0e MK |
86 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN \ |
87 | + (2 * VLAN_HEADER_LEN)) | |
4be4d22c MK |
88 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) |
89 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
58be5c0e MK |
90 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len) \ |
91 | - ETHER_HDR_LEN - ETHER_CRC_LEN) | |
4be4d22c | 92 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 93 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 94 | |
43307ad0 IS |
95 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
96 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
97 | * enough hugepages) we keep halving the number until the allocation succeeds | |
98 | * or we reach MIN_NB_MBUF */ | |
99 | ||
100 | #define MAX_NB_MBUF (4096 * 64) | |
da79ce2b DDP |
101 | #define MIN_NB_MBUF (4096 * 4) |
102 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
103 | ||
43307ad0 IS |
104 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ |
105 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF / MIN_NB_MBUF) | |
106 | == 0); | |
107 | ||
108 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
109 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
110 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF / MIN_NB_MBUF)) | |
111 | % MP_CACHE_SZ == 0); | |
112 | ||
d6e3feb5 | 113 | /* |
114 | * DPDK XSTATS Counter names definition | |
115 | */ | |
116 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
117 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
118 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
119 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
120 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
121 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
122 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
123 | ||
124 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
125 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
126 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
127 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
128 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
129 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
130 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
131 | ||
d57f777f | 132 | #define XSTAT_RX_MULTICAST_PACKETS "rx_multicast_packets" |
d6e3feb5 | 133 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" |
134 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
135 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
136 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
137 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
138 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
139 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
140 | ||
c161357d KT |
141 | /* Size of vHost custom stats. */ |
142 | #define VHOST_CUSTOM_STATS_SIZE 1 | |
143 | ||
144 | /* Names of vHost custom stats. */ | |
145 | #define VHOST_STAT_TX_RETRIES "tx_retries" | |
146 | ||
8a9562d2 PS |
147 | #define SOCKET0 0 |
148 | ||
b685696b CL |
149 | /* Default size of Physical NIC RXQ */ |
150 | #define NIC_PORT_DEFAULT_RXQ_SIZE 2048 | |
151 | /* Default size of Physical NIC TXQ */ | |
152 | #define NIC_PORT_DEFAULT_TXQ_SIZE 2048 | |
153 | /* Maximum size of Physical NIC Queues */ | |
154 | #define NIC_PORT_MAX_Q_SIZE 4096 | |
79f5354c | 155 | |
585a5bea | 156 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
157 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
158 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
159 | * yet mapped to another queue. */ | |
585a5bea | 160 | |
bb37956a IM |
161 | #define DPDK_ETH_PORT_ID_INVALID RTE_MAX_ETHPORTS |
162 | ||
5e925ccc MK |
163 | /* DPDK library uses uint16_t for port_id. */ |
164 | typedef uint16_t dpdk_port_t; | |
fa9f4eeb | 165 | #define DPDK_PORT_ID_FMT "%"PRIu16 |
bb37956a | 166 | |
31871ee3 | 167 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 168 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 169 | |
8a9562d2 | 170 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
171 | .rxmode = { |
172 | .mq_mode = ETH_MQ_RX_RSS, | |
173 | .split_hdr_size = 0, | |
03f3f9c0 | 174 | .offloads = 0, |
a28ddd11 DDP |
175 | }, |
176 | .rx_adv_conf = { | |
177 | .rss_conf = { | |
178 | .rss_key = NULL, | |
543342a4 | 179 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 180 | }, |
a28ddd11 DDP |
181 | }, |
182 | .txmode = { | |
183 | .mq_mode = ETH_MQ_TX_NONE, | |
184 | }, | |
8a9562d2 PS |
185 | }; |
186 | ||
f3e7ec25 MW |
187 | /* |
188 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
189 | * configuration has been fully completed. | |
190 | */ | |
191 | static int new_device(int vid); | |
192 | static void destroy_device(int vid); | |
193 | static int vring_state_changed(int vid, uint16_t queue_id, int enable); | |
61473a0e | 194 | static void destroy_connection(int vid); |
f3e7ec25 MW |
195 | static const struct vhost_device_ops virtio_net_device_ops = |
196 | { | |
197 | .new_device = new_device, | |
198 | .destroy_device = destroy_device, | |
199 | .vring_state_changed = vring_state_changed, | |
61473a0e DM |
200 | .features_changed = NULL, |
201 | .new_connection = NULL, | |
202 | .destroy_connection = destroy_connection, | |
f3e7ec25 MW |
203 | }; |
204 | ||
58f7c37b DDP |
205 | enum { DPDK_RING_SIZE = 256 }; |
206 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
207 | enum { DRAIN_TSC = 200000ULL }; |
208 | ||
58397e6c KT |
209 | enum dpdk_dev_type { |
210 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 211 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
212 | }; |
213 | ||
0bf765f7 IS |
214 | /* Quality of Service */ |
215 | ||
216 | /* An instance of a QoS configuration. Always associated with a particular | |
217 | * network device. | |
218 | * | |
219 | * Each QoS implementation subclasses this with whatever additional data it | |
220 | * needs. | |
221 | */ | |
222 | struct qos_conf { | |
223 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 224 | rte_spinlock_t lock; |
0bf765f7 IS |
225 | }; |
226 | ||
227 | /* A particular implementation of dpdk QoS operations. | |
228 | * | |
229 | * The functions below return 0 if successful or a positive errno value on | |
230 | * failure, except where otherwise noted. All of them must be provided, except | |
231 | * where otherwise noted. | |
232 | */ | |
233 | struct dpdk_qos_ops { | |
234 | ||
235 | /* Name of the QoS type */ | |
236 | const char *qos_name; | |
237 | ||
78bd47cf DDP |
238 | /* Called to construct a qos_conf object. The implementation should make |
239 | * the appropriate calls to configure QoS according to 'details'. | |
0bf765f7 IS |
240 | * |
241 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
242 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
243 | * (which is built as ovs-vswitchd.conf.db(8)). | |
244 | * | |
78bd47cf DDP |
245 | * This function must return 0 if and only if it sets '*conf' to an |
246 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
247 | * |
248 | * For all QoS implementations it should always be non-null. | |
249 | */ | |
78bd47cf | 250 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
251 | |
252 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 253 | * 'qos_conf'. |
0bf765f7 IS |
254 | * |
255 | * For all QoS implementations it should always be non-null. | |
256 | */ | |
78bd47cf | 257 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 258 | |
78bd47cf | 259 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
260 | * |
261 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
262 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
263 | * (which is built as ovs-vswitchd.conf.db(8)). | |
264 | */ | |
78bd47cf | 265 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 266 | |
78bd47cf | 267 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
268 | * |
269 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
270 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
271 | * (which is built as ovs-vswitchd.conf.db(8)). | |
272 | * | |
78bd47cf | 273 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 274 | */ |
78bd47cf DDP |
275 | bool (*qos_is_equal)(const struct qos_conf *conf, |
276 | const struct smap *details); | |
0bf765f7 IS |
277 | |
278 | /* Modify an array of rte_mbufs. The modification is specific to | |
279 | * each qos implementation. | |
280 | * | |
281 | * The function should take and array of mbufs and an int representing | |
282 | * the current number of mbufs present in the array. | |
283 | * | |
284 | * After the function has performed a qos modification to the array of | |
285 | * mbufs it returns an int representing the number of mbufs now present in | |
286 | * the array. This value is can then be passed to the port send function | |
287 | * along with the modified array for transmission. | |
288 | * | |
289 | * For all QoS implementations it should always be non-null. | |
290 | */ | |
78bd47cf | 291 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
7d7ded7a | 292 | int pkt_cnt, bool should_steal); |
0bf765f7 IS |
293 | }; |
294 | ||
295 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
296 | static const struct dpdk_qos_ops egress_policer_ops; | |
297 | ||
298 | /* | |
299 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
300 | * operations. | |
301 | */ | |
302 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
303 | &egress_policer_ops, | |
304 | NULL | |
305 | }; | |
306 | ||
c2adb102 IM |
307 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
308 | ||
8a9562d2 | 309 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 310 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 311 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 312 | |
c2adb102 IM |
313 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
314 | = OVS_MUTEX_INITIALIZER; | |
315 | ||
91fccdad | 316 | /* Contains all 'struct dpdk_mp's. */ |
43307ad0 IS |
317 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mp_mutex) |
318 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); | |
91fccdad | 319 | |
91fccdad KT |
320 | struct dpdk_mp { |
321 | struct rte_mempool *mp; | |
43307ad0 IS |
322 | int mtu; |
323 | int socket_id; | |
324 | int refcount; | |
91fccdad KT |
325 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mp_mutex); |
326 | }; | |
327 | ||
5a034064 AW |
328 | /* There should be one 'struct dpdk_tx_queue' created for |
329 | * each cpu core. */ | |
8a9562d2 | 330 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
331 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
332 | * from concurrent access. It is used only | |
333 | * if the queue is shared among different | |
324c8374 | 334 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
335 | int map; /* Mapping of configured vhost-user queues |
336 | * to enabled by guest. */ | |
8a9562d2 PS |
337 | }; |
338 | ||
95fb793a | 339 | /* dpdk has no way to remove dpdk ring ethernet devices |
340 | so we have to keep them around once they've been created | |
341 | */ | |
342 | ||
ca6ba700 | 343 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 344 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 345 | |
346 | struct dpdk_ring { | |
347 | /* For the client rings */ | |
348 | struct rte_ring *cring_tx; | |
349 | struct rte_ring *cring_rx; | |
b83a2df1 | 350 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
bb37956a | 351 | dpdk_port_t eth_port_id; /* ethernet device port id */ |
ca6ba700 | 352 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 353 | }; |
354 | ||
9509913a IS |
355 | struct ingress_policer { |
356 | struct rte_meter_srtcm_params app_srtcm_params; | |
357 | struct rte_meter_srtcm in_policer; | |
03f3f9c0 | 358 | struct rte_meter_srtcm_profile in_prof; |
9509913a IS |
359 | rte_spinlock_t policer_lock; |
360 | }; | |
361 | ||
1a2bb118 SC |
362 | enum dpdk_hw_ol_features { |
363 | NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0, | |
e10ca8b9 | 364 | NETDEV_RX_HW_CRC_STRIP = 1 << 1, |
03f3f9c0 | 365 | NETDEV_RX_HW_SCATTER = 1 << 2 |
1a2bb118 SC |
366 | }; |
367 | ||
b2e72a9c IM |
368 | /* |
369 | * In order to avoid confusion in variables names, following naming convention | |
370 | * should be used, if possible: | |
371 | * | |
372 | * 'struct netdev' : 'netdev' | |
373 | * 'struct netdev_dpdk' : 'dev' | |
374 | * 'struct netdev_rxq' : 'rxq' | |
375 | * 'struct netdev_rxq_dpdk' : 'rx' | |
376 | * | |
377 | * Example: | |
378 | * struct netdev *netdev = netdev_from_name(name); | |
379 | * struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
380 | * | |
381 | * Also, 'netdev' should be used instead of 'dev->up', where 'netdev' was | |
382 | * already defined. | |
383 | */ | |
384 | ||
8a9562d2 | 385 | struct netdev_dpdk { |
23d4d53f BB |
386 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline0, |
387 | dpdk_port_t port_id; | |
388 | ||
389 | /* If true, device was attached by rte_eth_dev_attach(). */ | |
390 | bool attached; | |
606f6650 EC |
391 | /* If true, rte_eth_dev_start() was successfully called */ |
392 | bool started; | |
23d4d53f BB |
393 | struct eth_addr hwaddr; |
394 | int mtu; | |
395 | int socket_id; | |
396 | int buf_size; | |
397 | int max_packet_len; | |
398 | enum dpdk_dev_type type; | |
399 | enum netdev_flags flags; | |
eaa43581 | 400 | int link_reset_cnt; |
bb9d2623 IM |
401 | union { |
402 | /* Device arguments for dpdk ports. */ | |
403 | char *devargs; | |
404 | /* Identifier used to distinguish vhost devices from each other. */ | |
405 | char *vhost_id; | |
406 | }; | |
23d4d53f BB |
407 | struct dpdk_tx_queue *tx_q; |
408 | struct rte_eth_link link; | |
23d4d53f BB |
409 | ); |
410 | ||
411 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline1, | |
412 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
43307ad0 | 413 | struct dpdk_mp *dpdk_mp; |
23d4d53f BB |
414 | |
415 | /* virtio identifier for vhost devices */ | |
416 | ovsrcu_index vid; | |
417 | ||
418 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
419 | bool vhost_reconfigured; | |
420 | /* 3 pad bytes here. */ | |
421 | ); | |
422 | ||
23d4d53f BB |
423 | PADDED_MEMBERS(CACHE_LINE_SIZE, |
424 | struct netdev up; | |
425 | /* In dpdk_list. */ | |
426 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); | |
427 | ||
428 | /* QoS configuration and lock for the device */ | |
429 | OVSRCU_TYPE(struct qos_conf *) qos_conf; | |
430 | ||
431 | /* Ingress Policer */ | |
432 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
433 | uint32_t policer_rate; | |
434 | uint32_t policer_burst; | |
35c91567 DM |
435 | |
436 | /* Array of vhost rxq states, see vring_state_changed. */ | |
437 | bool *vhost_rxq_enabled; | |
23d4d53f BB |
438 | ); |
439 | ||
440 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
441 | struct netdev_stats stats; | |
c161357d KT |
442 | /* Custom stat for retries when unable to transmit. */ |
443 | uint64_t tx_retries; | |
23d4d53f BB |
444 | /* Protects stats */ |
445 | rte_spinlock_t stats_lock; | |
c161357d | 446 | /* 4 pad bytes here. */ |
23d4d53f BB |
447 | ); |
448 | ||
449 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
450 | /* The following properties cannot be changed when a device is running, | |
451 | * so we remember the request and update them next time | |
452 | * netdev_dpdk*_reconfigure() is called */ | |
453 | int requested_mtu; | |
454 | int requested_n_txq; | |
455 | int requested_n_rxq; | |
456 | int requested_rxq_size; | |
457 | int requested_txq_size; | |
458 | ||
459 | /* Number of rx/tx descriptors for physical devices */ | |
460 | int rxq_size; | |
461 | int txq_size; | |
462 | ||
463 | /* Socket ID detected when vHost device is brought up */ | |
464 | int requested_socket_id; | |
465 | ||
466 | /* Denotes whether vHost port is client/server mode */ | |
467 | uint64_t vhost_driver_flags; | |
468 | ||
469 | /* DPDK-ETH Flow control */ | |
470 | struct rte_eth_fc_conf fc_conf; | |
471 | ||
472 | /* DPDK-ETH hardware offload features, | |
473 | * from the enum set 'dpdk_hw_ol_features' */ | |
474 | uint32_t hw_ol_features; | |
f8b64a61 RM |
475 | |
476 | /* Properties for link state change detection mode. | |
477 | * If lsc_interrupt_mode is set to false, poll mode is used, | |
478 | * otherwise interrupt mode is used. */ | |
479 | bool requested_lsc_interrupt_mode; | |
480 | bool lsc_interrupt_mode; | |
23d4d53f | 481 | ); |
971f4b39 MW |
482 | |
483 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
484 | /* Names of all XSTATS counters */ | |
485 | struct rte_eth_xstat_name *rte_xstats_names; | |
486 | int rte_xstats_names_size; | |
487 | int rte_xstats_ids_size; | |
488 | uint64_t *rte_xstats_ids; | |
489 | ); | |
8a9562d2 PS |
490 | }; |
491 | ||
492 | struct netdev_rxq_dpdk { | |
493 | struct netdev_rxq up; | |
bb37956a | 494 | dpdk_port_t port_id; |
8a9562d2 PS |
495 | }; |
496 | ||
f3e7ec25 MW |
497 | static void netdev_dpdk_destruct(struct netdev *netdev); |
498 | static void netdev_dpdk_vhost_destruct(struct netdev *netdev); | |
8a9562d2 | 499 | |
ac1a9bb9 IM |
500 | static void netdev_dpdk_clear_xstats(struct netdev_dpdk *dev); |
501 | ||
0a0f39df | 502 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 503 | |
9509913a IS |
504 | struct ingress_policer * |
505 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
506 | ||
8a9562d2 PS |
507 | static bool |
508 | is_dpdk_class(const struct netdev_class *class) | |
509 | { | |
f3e7ec25 MW |
510 | return class->destruct == netdev_dpdk_destruct |
511 | || class->destruct == netdev_dpdk_vhost_destruct; | |
8a9562d2 PS |
512 | } |
513 | ||
4be4d22c MK |
514 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
515 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
516 | * value, insufficient buffers are allocated to accomodate the packet in its | |
517 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
518 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
519 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
520 | * behaviour, which reduces performance. To prevent this, use a buffer size |
521 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
522 | */ |
523 | static uint32_t | |
524 | dpdk_buf_size(int mtu) | |
525 | { | |
a32bab26 TL |
526 | return ROUND_UP(MTU_TO_MAX_FRAME_LEN(mtu), NETDEV_DPDK_MBUF_ALIGN) |
527 | + RTE_PKTMBUF_HEADROOM; | |
4be4d22c MK |
528 | } |
529 | ||
eff23640 DDP |
530 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
531 | * | |
532 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
533 | static void * |
534 | dpdk_rte_mzalloc(size_t sz) | |
535 | { | |
eff23640 | 536 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
537 | } |
538 | ||
539 | void | |
e14deea0 | 540 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 541 | { |
db73f716 | 542 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 543 | |
b00b4a81 | 544 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
545 | } |
546 | ||
b3cd9f9d | 547 | static void |
401b70d6 | 548 | ovs_rte_pktmbuf_init(struct rte_mempool *mp OVS_UNUSED, |
b3cd9f9d | 549 | void *opaque_arg OVS_UNUSED, |
2391135c | 550 | void *_p, |
b3cd9f9d PS |
551 | unsigned i OVS_UNUSED) |
552 | { | |
2391135c | 553 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 554 | |
3aaa6201 | 555 | dp_packet_init_dpdk((struct dp_packet *) pkt); |
b3cd9f9d PS |
556 | } |
557 | ||
91fccdad KT |
558 | static int |
559 | dpdk_mp_full(const struct rte_mempool *mp) OVS_REQUIRES(dpdk_mp_mutex) | |
560 | { | |
1f84a2d5 KT |
561 | /* At this point we want to know if all the mbufs are back |
562 | * in the mempool. rte_mempool_full() is not atomic but it's | |
563 | * the best available and as we are no longer requesting mbufs | |
564 | * from the mempool, it means mbufs will not move from | |
565 | * 'mempool ring' --> 'mempool cache'. In rte_mempool_full() | |
566 | * the ring is counted before caches, so we won't get false | |
567 | * positives in this use case and we handle false negatives. | |
568 | * | |
569 | * If future implementations of rte_mempool_full() were to change | |
570 | * it could be possible for a false positive. Even that would | |
571 | * likely be ok, as there are additional checks during mempool | |
572 | * freeing but it would make things racey. | |
91fccdad | 573 | */ |
1f84a2d5 | 574 | return rte_mempool_full(mp); |
91fccdad KT |
575 | } |
576 | ||
577 | /* Free unused mempools. */ | |
578 | static void | |
43307ad0 | 579 | dpdk_mp_sweep(void) OVS_REQUIRES(dpdk_mp_mutex) |
91fccdad KT |
580 | { |
581 | struct dpdk_mp *dmp, *next; | |
582 | ||
43307ad0 IS |
583 | LIST_FOR_EACH_SAFE (dmp, next, list_node, &dpdk_mp_list) { |
584 | if (!dmp->refcount && dpdk_mp_full(dmp->mp)) { | |
91fccdad KT |
585 | VLOG_DBG("Freeing mempool \"%s\"", dmp->mp->name); |
586 | ovs_list_remove(&dmp->list_node); | |
587 | rte_mempool_free(dmp->mp); | |
588 | rte_free(dmp); | |
589 | } | |
590 | } | |
91fccdad KT |
591 | } |
592 | ||
43307ad0 IS |
593 | /* Calculating the required number of mbufs differs depending on the |
594 | * mempool model being used. Check if per port memory is in use before | |
595 | * calculating. | |
596 | */ | |
597 | static uint32_t | |
598 | dpdk_calculate_mbufs(struct netdev_dpdk *dev, int mtu, bool per_port_mp) | |
91fccdad | 599 | { |
43307ad0 | 600 | uint32_t n_mbufs; |
91fccdad | 601 | |
43307ad0 IS |
602 | if (!per_port_mp) { |
603 | /* Shared memory are being used. | |
604 | * XXX: this is a really rough method of provisioning memory. | |
605 | * It's impossible to determine what the exact memory requirements are | |
606 | * when the number of ports and rxqs that utilize a particular mempool | |
607 | * can change dynamically at runtime. For now, use this rough | |
608 | * heurisitic. | |
609 | */ | |
610 | if (mtu >= ETHER_MTU) { | |
611 | n_mbufs = MAX_NB_MBUF; | |
612 | } else { | |
613 | n_mbufs = MIN_NB_MBUF; | |
91fccdad | 614 | } |
43307ad0 IS |
615 | } else { |
616 | /* Per port memory is being used. | |
617 | * XXX: rough estimation of number of mbufs required for this port: | |
618 | * <packets required to fill the device rxqs> | |
619 | * + <packets that could be stuck on other ports txqs> | |
620 | * + <packets in the pmd threads> | |
621 | * + <additional memory for corner cases> | |
622 | */ | |
623 | n_mbufs = dev->requested_n_rxq * dev->requested_rxq_size | |
624 | + dev->requested_n_txq * dev->requested_txq_size | |
625 | + MIN(RTE_MAX_LCORE, dev->requested_n_rxq) * NETDEV_MAX_BURST | |
626 | + MIN_NB_MBUF; | |
91fccdad | 627 | } |
43307ad0 IS |
628 | |
629 | return n_mbufs; | |
91fccdad KT |
630 | } |
631 | ||
43307ad0 IS |
632 | static struct dpdk_mp * |
633 | dpdk_mp_create(struct netdev_dpdk *dev, int mtu, bool per_port_mp) | |
8a9562d2 | 634 | { |
24e78f93 IM |
635 | char mp_name[RTE_MEMPOOL_NAMESIZE]; |
636 | const char *netdev_name = netdev_get_name(&dev->up); | |
637 | int socket_id = dev->requested_socket_id; | |
dfaf00e8 MK |
638 | uint32_t n_mbufs = 0; |
639 | uint32_t mbuf_size = 0; | |
640 | uint32_t aligned_mbuf_size = 0; | |
641 | uint32_t mbuf_priv_data_len = 0; | |
642 | uint32_t pkt_size = 0; | |
24e78f93 | 643 | uint32_t hash = hash_string(netdev_name, 0); |
43307ad0 IS |
644 | struct dpdk_mp *dmp = NULL; |
645 | int ret; | |
646 | ||
647 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
648 | if (!dmp) { | |
649 | return NULL; | |
650 | } | |
651 | dmp->socket_id = socket_id; | |
652 | dmp->mtu = mtu; | |
653 | dmp->refcount = 1; | |
654 | ||
dfaf00e8 MK |
655 | /* Get the size of each mbuf, based on the MTU */ |
656 | mbuf_size = MTU_TO_FRAME_LEN(mtu); | |
657 | ||
43307ad0 | 658 | n_mbufs = dpdk_calculate_mbufs(dev, mtu, per_port_mp); |
d555d9bd | 659 | |
da79ce2b | 660 | do { |
24e78f93 | 661 | /* Full DPDK memory pool name must be unique and cannot be |
43307ad0 IS |
662 | * longer than RTE_MEMPOOL_NAMESIZE. Note that for the shared |
663 | * mempool case this can result in one device using a mempool | |
664 | * which references a different device in it's name. However as | |
665 | * mempool names are hashed, the device name will not be readable | |
666 | * so this is not an issue for tasks such as debugging. | |
667 | */ | |
668 | ret = snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, | |
dfaf00e8 MK |
669 | "ovs%08x%02d%05d%07u", |
670 | hash, socket_id, mtu, n_mbufs); | |
24e78f93 IM |
671 | if (ret < 0 || ret >= RTE_MEMPOOL_NAMESIZE) { |
672 | VLOG_DBG("snprintf returned %d. " | |
673 | "Failed to generate a mempool name for \"%s\". " | |
674 | "Hash:0x%x, socket_id: %d, mtu:%d, mbufs:%u.", | |
675 | ret, netdev_name, hash, socket_id, mtu, n_mbufs); | |
676 | break; | |
65056fd7 | 677 | } |
95fb793a | 678 | |
dfaf00e8 MK |
679 | VLOG_DBG("Port %s: Requesting a mempool of %u mbufs of size %u " |
680 | "on socket %d for %d Rx and %d Tx queues, " | |
681 | "cache line size of %u", | |
682 | netdev_name, n_mbufs, mbuf_size, socket_id, | |
683 | dev->requested_n_rxq, dev->requested_n_txq, | |
684 | RTE_CACHE_LINE_SIZE); | |
685 | ||
a32bab26 TL |
686 | /* The size of the mbuf's private area (i.e. area that holds OvS' |
687 | * dp_packet data)*/ | |
dfaf00e8 MK |
688 | mbuf_priv_data_len = sizeof(struct dp_packet) - |
689 | sizeof(struct rte_mbuf); | |
690 | /* The size of the entire dp_packet. */ | |
691 | pkt_size = sizeof(struct dp_packet) + mbuf_size; | |
692 | /* mbuf size, rounded up to cacheline size. */ | |
693 | aligned_mbuf_size = ROUND_UP(pkt_size, RTE_CACHE_LINE_SIZE); | |
694 | /* If there is a size discrepancy, add padding to mbuf_priv_data_len. | |
695 | * This maintains mbuf size cache alignment, while also honoring RX | |
696 | * buffer alignment in the data portion of the mbuf. If this adjustment | |
697 | * is not made, there is a possiblity later on that for an element of | |
698 | * the mempool, buf, buf->data_len < (buf->buf_len - buf->data_off). | |
699 | * This is problematic in the case of multi-segment mbufs, particularly | |
700 | * when an mbuf segment needs to be resized (when [push|popp]ing a VLAN | |
701 | * header, for example. | |
702 | */ | |
703 | mbuf_priv_data_len += (aligned_mbuf_size - pkt_size); | |
704 | ||
705 | dmp->mp = rte_pktmbuf_pool_create(mp_name, n_mbufs, MP_CACHE_SZ, | |
706 | mbuf_priv_data_len, | |
707 | mbuf_size, | |
43307ad0 | 708 | socket_id); |
24e78f93 | 709 | |
43307ad0 | 710 | if (dmp->mp) { |
24e78f93 IM |
711 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", |
712 | mp_name, n_mbufs); | |
837c1761 | 713 | /* rte_pktmbuf_pool_create has done some initialization of the |
43307ad0 IS |
714 | * rte_mbuf part of each dp_packet, while ovs_rte_pktmbuf_init |
715 | * initializes some OVS specific fields of dp_packet. | |
716 | */ | |
717 | rte_mempool_obj_iter(dmp->mp, ovs_rte_pktmbuf_init, NULL); | |
718 | return dmp; | |
d555d9bd RW |
719 | } else if (rte_errno == EEXIST) { |
720 | /* A mempool with the same name already exists. We just | |
721 | * retrieve its pointer to be returned to the caller. */ | |
43307ad0 | 722 | dmp->mp = rte_mempool_lookup(mp_name); |
d555d9bd RW |
723 | /* As the mempool create returned EEXIST we can expect the |
724 | * lookup has returned a valid pointer. If for some reason | |
725 | * that's not the case we keep track of it. */ | |
24e78f93 | 726 | VLOG_DBG("A mempool with name \"%s\" already exists at %p.", |
43307ad0 IS |
727 | mp_name, dmp->mp); |
728 | return dmp; | |
d555d9bd | 729 | } else { |
43307ad0 IS |
730 | VLOG_DBG("Failed to create mempool \"%s\" with a request of " |
731 | "%u mbufs, retrying with %u mbufs", | |
732 | mp_name, n_mbufs, n_mbufs / 2); | |
0c6f39e5 | 733 | } |
43307ad0 | 734 | } while (!dmp->mp && rte_errno == ENOMEM && (n_mbufs /= 2) >= MIN_NB_MBUF); |
2ae3d542 | 735 | |
43307ad0 IS |
736 | VLOG_ERR("Failed to create mempool \"%s\" with a request of %u mbufs", |
737 | mp_name, n_mbufs); | |
738 | ||
739 | rte_free(dmp); | |
740 | return NULL; | |
8a9562d2 PS |
741 | } |
742 | ||
43307ad0 IS |
743 | static struct dpdk_mp * |
744 | dpdk_mp_get(struct netdev_dpdk *dev, int mtu, bool per_port_mp) | |
8a9562d2 | 745 | { |
43307ad0 IS |
746 | struct dpdk_mp *dmp, *next; |
747 | bool reuse = false; | |
8a9562d2 | 748 | |
c2adb102 | 749 | ovs_mutex_lock(&dpdk_mp_mutex); |
43307ad0 IS |
750 | /* Check if shared memory is being used, if so check existing mempools |
751 | * to see if reuse is possible. */ | |
752 | if (!per_port_mp) { | |
753 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
754 | if (dmp->socket_id == dev->requested_socket_id | |
755 | && dmp->mtu == mtu) { | |
756 | VLOG_DBG("Reusing mempool \"%s\"", dmp->mp->name); | |
757 | dmp->refcount++; | |
758 | reuse = true; | |
759 | break; | |
760 | } | |
761 | } | |
762 | } | |
763 | /* Sweep mempools after reuse or before create. */ | |
764 | dpdk_mp_sweep(); | |
91fccdad | 765 | |
43307ad0 IS |
766 | if (!reuse) { |
767 | dmp = dpdk_mp_create(dev, mtu, per_port_mp); | |
91fccdad | 768 | if (dmp) { |
43307ad0 IS |
769 | /* Shared memory will hit the reuse case above so will not |
770 | * request a mempool that already exists but we need to check | |
771 | * for the EEXIST case for per port memory case. Compare the | |
772 | * mempool returned by dmp to each entry in dpdk_mp_list. If a | |
773 | * match is found, free dmp as a new entry is not required, set | |
774 | * dmp to point to the existing entry and increment the refcount | |
775 | * to avoid being freed at a later stage. | |
776 | */ | |
777 | if (per_port_mp && rte_errno == EEXIST) { | |
778 | LIST_FOR_EACH (next, list_node, &dpdk_mp_list) { | |
779 | if (dmp->mp == next->mp) { | |
780 | rte_free(dmp); | |
781 | dmp = next; | |
782 | dmp->refcount++; | |
783 | } | |
784 | } | |
785 | } else { | |
786 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); | |
787 | } | |
91fccdad KT |
788 | } |
789 | } | |
43307ad0 | 790 | |
43307ad0 IS |
791 | ovs_mutex_unlock(&dpdk_mp_mutex); |
792 | ||
793 | return dmp; | |
794 | } | |
795 | ||
796 | /* Decrement reference to a mempool. */ | |
797 | static void | |
798 | dpdk_mp_put(struct dpdk_mp *dmp) | |
799 | { | |
800 | if (!dmp) { | |
801 | return; | |
802 | } | |
803 | ||
804 | ovs_mutex_lock(&dpdk_mp_mutex); | |
805 | ovs_assert(dmp->refcount); | |
806 | dmp->refcount--; | |
c2adb102 | 807 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
808 | } |
809 | ||
43307ad0 IS |
810 | /* Depending on the memory model being used this function tries to |
811 | * identify and reuse an existing mempool or tries to allocate a new | |
812 | * mempool on requested_socket_id with mbuf size corresponding to the | |
813 | * requested_mtu. On success, a new configuration will be applied. | |
0072e931 MK |
814 | * On error, device will be left unchanged. */ |
815 | static int | |
816 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
817 | OVS_REQUIRES(dev->mutex) |
818 | { | |
819 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
43307ad0 | 820 | struct dpdk_mp *dmp; |
24e78f93 | 821 | int ret = 0; |
43307ad0 | 822 | bool per_port_mp = dpdk_per_port_memory(); |
0072e931 | 823 | |
43307ad0 IS |
824 | /* With shared memory we do not need to configure a mempool if the MTU |
825 | * and socket ID have not changed, the previous configuration is still | |
826 | * valid so return 0 */ | |
827 | if (!per_port_mp && dev->mtu == dev->requested_mtu | |
828 | && dev->socket_id == dev->requested_socket_id) { | |
829 | return ret; | |
830 | } | |
91fccdad | 831 | |
43307ad0 IS |
832 | dmp = dpdk_mp_get(dev, FRAME_LEN_TO_MTU(buf_size), per_port_mp); |
833 | if (!dmp) { | |
c67e46c0 MK |
834 | VLOG_ERR("Failed to create memory pool for netdev " |
835 | "%s, with MTU %d on socket %d: %s\n", | |
836 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
837 | rte_strerror(rte_errno)); | |
24e78f93 | 838 | ret = rte_errno; |
0072e931 | 839 | } else { |
43307ad0 IS |
840 | /* Check for any pre-existing dpdk_mp for the device before accessing |
841 | * the associated mempool. | |
842 | */ | |
843 | if (dev->dpdk_mp != NULL) { | |
844 | /* A new MTU was requested, decrement the reference count for the | |
845 | * devices current dpdk_mp. This is required even if a pointer to | |
846 | * same dpdk_mp is returned by dpdk_mp_get. The refcount for dmp | |
847 | * has already been incremented by dpdk_mp_get at this stage so it | |
848 | * must be decremented to keep an accurate refcount for the | |
849 | * dpdk_mp. | |
850 | */ | |
851 | dpdk_mp_put(dev->dpdk_mp); | |
852 | } | |
853 | dev->dpdk_mp = dmp; | |
0072e931 MK |
854 | dev->mtu = dev->requested_mtu; |
855 | dev->socket_id = dev->requested_socket_id; | |
856 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
857 | } | |
858 | ||
24e78f93 | 859 | return ret; |
0072e931 MK |
860 | } |
861 | ||
8a9562d2 PS |
862 | static void |
863 | check_link_status(struct netdev_dpdk *dev) | |
864 | { | |
865 | struct rte_eth_link link; | |
866 | ||
867 | rte_eth_link_get_nowait(dev->port_id, &link); | |
868 | ||
869 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 870 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
871 | |
872 | dev->link_reset_cnt++; | |
873 | dev->link = link; | |
874 | if (dev->link.link_status) { | |
fa9f4eeb IM |
875 | VLOG_DBG_RL(&rl, |
876 | "Port "DPDK_PORT_ID_FMT" Link Up - speed %u Mbps - %s", | |
58be5c0e | 877 | dev->port_id, (unsigned) dev->link.link_speed, |
fa9f4eeb IM |
878 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) |
879 | ? "full-duplex" : "half-duplex"); | |
8a9562d2 | 880 | } else { |
fa9f4eeb IM |
881 | VLOG_DBG_RL(&rl, "Port "DPDK_PORT_ID_FMT" Link Down", |
882 | dev->port_id); | |
8a9562d2 PS |
883 | } |
884 | } | |
885 | } | |
886 | ||
887 | static void * | |
888 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
889 | { | |
890 | struct netdev_dpdk *dev; | |
891 | ||
892 | pthread_detach(pthread_self()); | |
893 | ||
894 | for (;;) { | |
895 | ovs_mutex_lock(&dpdk_mutex); | |
896 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
897 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
898 | if (dev->type == DPDK_DEV_ETH) { |
899 | check_link_status(dev); | |
900 | } | |
8a9562d2 PS |
901 | ovs_mutex_unlock(&dev->mutex); |
902 | } | |
903 | ovs_mutex_unlock(&dpdk_mutex); | |
904 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
905 | } | |
906 | ||
907 | return NULL; | |
908 | } | |
909 | ||
b98d7669 | 910 | static int |
f8b64a61 | 911 | dpdk_eth_dev_port_config(struct netdev_dpdk *dev, int n_rxq, int n_txq) |
b98d7669 DDP |
912 | { |
913 | int diag = 0; | |
914 | int i; | |
0072e931 | 915 | struct rte_eth_conf conf = port_conf; |
65a87968 | 916 | struct rte_eth_dev_info info; |
4dd16ca0 | 917 | uint16_t conf_mtu; |
65a87968 | 918 | |
03f3f9c0 OM |
919 | rte_eth_dev_info_get(dev->port_id, &info); |
920 | ||
65a87968 | 921 | /* As of DPDK 17.11.1 a few PMDs require to explicitly enable |
03f3f9c0 OM |
922 | * scatter to support jumbo RX. |
923 | * Setting scatter for the device is done after checking for | |
924 | * scatter support in the device capabilites. */ | |
0072e931 | 925 | if (dev->mtu > ETHER_MTU) { |
03f3f9c0 OM |
926 | if (dev->hw_ol_features & NETDEV_RX_HW_SCATTER) { |
927 | conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER; | |
65a87968 | 928 | } |
0072e931 | 929 | } |
67fe6d63 | 930 | |
f8b64a61 | 931 | conf.intr_conf.lsc = dev->lsc_interrupt_mode; |
e10ca8b9 | 932 | |
03f3f9c0 OM |
933 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
934 | conf.rxmode.offloads |= DEV_RX_OFFLOAD_CHECKSUM; | |
935 | } | |
936 | ||
937 | if (!(dev->hw_ol_features & NETDEV_RX_HW_CRC_STRIP) | |
938 | && info.rx_offload_capa & DEV_RX_OFFLOAD_KEEP_CRC) { | |
939 | conf.rxmode.offloads |= DEV_RX_OFFLOAD_KEEP_CRC; | |
e10ca8b9 MW |
940 | } |
941 | ||
03f3f9c0 OM |
942 | /* Limit configured rss hash functions to only those supported |
943 | * by the eth device. */ | |
944 | conf.rx_adv_conf.rss_conf.rss_hf &= info.flow_type_rss_offloads; | |
945 | ||
b98d7669 DDP |
946 | /* A device may report more queues than it makes available (this has |
947 | * been observed for Intel xl710, which reserves some of them for | |
948 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
949 | * available. When this happens we can retry the configuration | |
950 | * and request less queues */ | |
951 | while (n_rxq && n_txq) { | |
952 | if (diag) { | |
953 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
954 | } | |
955 | ||
0072e931 | 956 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 957 | if (diag) { |
0072e931 MK |
958 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
959 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
960 | break; |
961 | } | |
962 | ||
67fe6d63 MK |
963 | diag = rte_eth_dev_set_mtu(dev->port_id, dev->mtu); |
964 | if (diag) { | |
4dd16ca0 IS |
965 | /* A device may not support rte_eth_dev_set_mtu, in this case |
966 | * flag a warning to the user and include the devices configured | |
967 | * MTU value that will be used instead. */ | |
968 | if (-ENOTSUP == diag) { | |
969 | rte_eth_dev_get_mtu(dev->port_id, &conf_mtu); | |
970 | VLOG_WARN("Interface %s does not support MTU configuration, " | |
971 | "max packet size supported is %"PRIu16".", | |
972 | dev->up.name, conf_mtu); | |
973 | } else { | |
974 | VLOG_ERR("Interface %s MTU (%d) setup error: %s", | |
975 | dev->up.name, dev->mtu, rte_strerror(-diag)); | |
976 | break; | |
977 | } | |
67fe6d63 MK |
978 | } |
979 | ||
b98d7669 | 980 | for (i = 0; i < n_txq; i++) { |
b685696b | 981 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
982 | dev->socket_id, NULL); |
983 | if (diag) { | |
1dfebee9 | 984 | VLOG_INFO("Interface %s unable to setup txq(%d): %s", |
b98d7669 DDP |
985 | dev->up.name, i, rte_strerror(-diag)); |
986 | break; | |
987 | } | |
988 | } | |
989 | ||
990 | if (i != n_txq) { | |
991 | /* Retry with less tx queues */ | |
992 | n_txq = i; | |
993 | continue; | |
994 | } | |
995 | ||
996 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 997 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
43307ad0 IS |
998 | dev->socket_id, NULL, |
999 | dev->dpdk_mp->mp); | |
b98d7669 | 1000 | if (diag) { |
1dfebee9 | 1001 | VLOG_INFO("Interface %s unable to setup rxq(%d): %s", |
b98d7669 DDP |
1002 | dev->up.name, i, rte_strerror(-diag)); |
1003 | break; | |
1004 | } | |
1005 | } | |
1006 | ||
1007 | if (i != n_rxq) { | |
1008 | /* Retry with less rx queues */ | |
1009 | n_rxq = i; | |
1010 | continue; | |
1011 | } | |
1012 | ||
1013 | dev->up.n_rxq = n_rxq; | |
81acebda | 1014 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
1015 | |
1016 | return 0; | |
1017 | } | |
1018 | ||
1019 | return diag; | |
1020 | } | |
1021 | ||
9fd39370 SC |
1022 | static void |
1023 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
1024 | { | |
1025 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
fa9f4eeb | 1026 | VLOG_WARN("Failed to enable flow control on device "DPDK_PORT_ID_FMT, |
bb37956a | 1027 | dev->port_id); |
9fd39370 SC |
1028 | } |
1029 | } | |
b98d7669 | 1030 | |
8a9562d2 | 1031 | static int |
c2adb102 IM |
1032 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
1033 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
1034 | { |
1035 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 1036 | struct rte_eth_dev_info info; |
8a9562d2 PS |
1037 | struct ether_addr eth_addr; |
1038 | int diag; | |
b98d7669 | 1039 | int n_rxq, n_txq; |
d4f5282c KT |
1040 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | |
1041 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
1042 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
8a9562d2 | 1043 | |
a0cb2d66 | 1044 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 1045 | |
e10ca8b9 MW |
1046 | if (strstr(info.driver_name, "vf") != NULL) { |
1047 | VLOG_INFO("Virtual function detected, HW_CRC_STRIP will be enabled"); | |
1048 | dev->hw_ol_features |= NETDEV_RX_HW_CRC_STRIP; | |
1049 | } else { | |
1050 | dev->hw_ol_features &= ~NETDEV_RX_HW_CRC_STRIP; | |
1051 | } | |
1052 | ||
d4f5282c KT |
1053 | if ((info.rx_offload_capa & rx_chksm_offload_capa) != |
1054 | rx_chksm_offload_capa) { | |
fa9f4eeb IM |
1055 | VLOG_WARN("Rx checksum offload is not supported on port " |
1056 | DPDK_PORT_ID_FMT, dev->port_id); | |
d4f5282c KT |
1057 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; |
1058 | } else { | |
1059 | dev->hw_ol_features |= NETDEV_RX_CHECKSUM_OFFLOAD; | |
1060 | } | |
1061 | ||
03f3f9c0 OM |
1062 | if (info.rx_offload_capa & DEV_RX_OFFLOAD_SCATTER) { |
1063 | dev->hw_ol_features |= NETDEV_RX_HW_SCATTER; | |
1064 | } else { | |
1065 | /* Do not warn on lack of scatter support */ | |
1066 | dev->hw_ol_features &= ~NETDEV_RX_HW_SCATTER; | |
1067 | } | |
1068 | ||
b98d7669 DDP |
1069 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
1070 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
1071 | ||
f8b64a61 | 1072 | diag = dpdk_eth_dev_port_config(dev, n_rxq, n_txq); |
8a9562d2 | 1073 | if (diag) { |
f8b64a61 RM |
1074 | VLOG_ERR("Interface %s(rxq:%d txq:%d lsc interrupt mode:%s) " |
1075 | "configure error: %s", | |
1076 | dev->up.name, n_rxq, n_txq, | |
1077 | dev->lsc_interrupt_mode ? "true" : "false", | |
1078 | rte_strerror(-diag)); | |
95fb793a | 1079 | return -diag; |
8a9562d2 PS |
1080 | } |
1081 | ||
8a9562d2 PS |
1082 | diag = rte_eth_dev_start(dev->port_id); |
1083 | if (diag) { | |
b98d7669 DDP |
1084 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
1085 | rte_strerror(-diag)); | |
95fb793a | 1086 | return -diag; |
8a9562d2 | 1087 | } |
606f6650 | 1088 | dev->started = true; |
8a9562d2 PS |
1089 | |
1090 | rte_eth_promiscuous_enable(dev->port_id); | |
1091 | rte_eth_allmulticast_enable(dev->port_id); | |
1092 | ||
1093 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
1094 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
fa9f4eeb IM |
1095 | VLOG_INFO_RL(&rl, "Port "DPDK_PORT_ID_FMT": "ETH_ADDR_FMT, |
1096 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); | |
8a9562d2 | 1097 | |
ca92d173 | 1098 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
1099 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
1100 | ||
43307ad0 | 1101 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); |
8a9562d2 | 1102 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; |
8a9562d2 PS |
1103 | return 0; |
1104 | } | |
1105 | ||
1106 | static struct netdev_dpdk * | |
1107 | netdev_dpdk_cast(const struct netdev *netdev) | |
1108 | { | |
1109 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
1110 | } | |
1111 | ||
1112 | static struct netdev * | |
1113 | netdev_dpdk_alloc(void) | |
1114 | { | |
bab69409 AC |
1115 | struct netdev_dpdk *dev; |
1116 | ||
65e19e70 DDP |
1117 | dev = dpdk_rte_mzalloc(sizeof *dev); |
1118 | if (dev) { | |
1119 | return &dev->up; | |
bab69409 | 1120 | } |
65e19e70 | 1121 | |
bab69409 | 1122 | return NULL; |
8a9562d2 PS |
1123 | } |
1124 | ||
eff23640 DDP |
1125 | static struct dpdk_tx_queue * |
1126 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 1127 | { |
eff23640 | 1128 | struct dpdk_tx_queue *txqs; |
bd5131ba | 1129 | unsigned i; |
5a034064 | 1130 | |
eff23640 DDP |
1131 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
1132 | if (txqs) { | |
1133 | for (i = 0; i < n_txqs; i++) { | |
1134 | /* Initialize map for vhost devices. */ | |
1135 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
1136 | rte_spinlock_init(&txqs[i].tx_lock); | |
1137 | } | |
5a034064 | 1138 | } |
eff23640 DDP |
1139 | |
1140 | return txqs; | |
5a034064 AW |
1141 | } |
1142 | ||
8a9562d2 | 1143 | static int |
bb37956a | 1144 | common_construct(struct netdev *netdev, dpdk_port_t port_no, |
1ce30dfd | 1145 | enum dpdk_dev_type type, int socket_id) |
5a034064 | 1146 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 1147 | { |
d46285a2 | 1148 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1149 | |
d46285a2 | 1150 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 1151 | |
d46285a2 | 1152 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 1153 | |
1b7a04e0 AW |
1154 | /* If the 'sid' is negative, it means that the kernel fails |
1155 | * to obtain the pci numa info. In that situation, always | |
1156 | * use 'SOCKET0'. */ | |
1ce30dfd | 1157 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 1158 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
1159 | dev->port_id = port_no; |
1160 | dev->type = type; | |
1161 | dev->flags = 0; | |
7f381c2e | 1162 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 1163 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
f8b64a61 | 1164 | dev->requested_lsc_interrupt_mode = 0; |
0a0f39df CL |
1165 | ovsrcu_index_init(&dev->vid, -1); |
1166 | dev->vhost_reconfigured = false; | |
5dcde09c | 1167 | dev->attached = false; |
8a9562d2 | 1168 | |
78bd47cf | 1169 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 1170 | |
9509913a IS |
1171 | ovsrcu_init(&dev->ingress_policer, NULL); |
1172 | dev->policer_rate = 0; | |
1173 | dev->policer_burst = 0; | |
1174 | ||
7f381c2e DDP |
1175 | netdev->n_rxq = 0; |
1176 | netdev->n_txq = 0; | |
1177 | dev->requested_n_rxq = NR_QUEUE; | |
1178 | dev->requested_n_txq = NR_QUEUE; | |
1179 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
1180 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 1181 | |
9fd39370 SC |
1182 | /* Initialize the flow control to NULL */ |
1183 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
1184 | |
1185 | /* Initilize the hardware offload flags to 0 */ | |
1186 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
1187 | |
1188 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
1189 | ||
d46285a2 | 1190 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 1191 | |
7f381c2e DDP |
1192 | netdev_request_reconfigure(netdev); |
1193 | ||
971f4b39 MW |
1194 | dev->rte_xstats_names = NULL; |
1195 | dev->rte_xstats_names_size = 0; | |
1196 | ||
1197 | dev->rte_xstats_ids = NULL; | |
1198 | dev->rte_xstats_ids_size = 0; | |
1199 | ||
c161357d KT |
1200 | dev->tx_retries = 0; |
1201 | ||
1ce30dfd | 1202 | return 0; |
95fb793a | 1203 | } |
1204 | ||
b83a2df1 MV |
1205 | /* dev_name must be the prefix followed by a positive decimal number. |
1206 | * (no leading + or - signs are allowed) */ | |
95fb793a | 1207 | static int |
1208 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
1209 | unsigned int *port_no) | |
1210 | { | |
1211 | const char *cport; | |
1212 | ||
1213 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
1214 | return ENODEV; | |
1215 | } | |
1216 | ||
1217 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
1218 | |
1219 | if (str_to_uint(cport, 10, port_no)) { | |
1220 | return 0; | |
1221 | } else { | |
1222 | return ENODEV; | |
1223 | } | |
95fb793a | 1224 | } |
1225 | ||
40e940e4 OM |
1226 | /* Get the number of OVS interfaces which have the same DPDK |
1227 | * rte device (e.g. same pci bus address). | |
1228 | * FIXME: avoid direct access to DPDK internal array rte_eth_devices. | |
1229 | */ | |
1230 | static int | |
1231 | netdev_dpdk_get_num_ports(struct rte_device *device) | |
1232 | OVS_REQUIRES(dpdk_mutex) | |
1233 | { | |
1234 | struct netdev_dpdk *dev; | |
1235 | int count = 0; | |
1236 | ||
1237 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1238 | if (rte_eth_devices[dev->port_id].device == device | |
1239 | && rte_eth_devices[dev->port_id].state != RTE_ETH_DEV_UNUSED) { | |
1240 | count++; | |
1241 | } | |
1242 | } | |
1243 | return count; | |
1244 | } | |
1245 | ||
1ce30dfd DDP |
1246 | static int |
1247 | vhost_common_construct(struct netdev *netdev) | |
1248 | OVS_REQUIRES(dpdk_mutex) | |
1249 | { | |
1250 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
1251 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1252 | ||
35c91567 DM |
1253 | dev->vhost_rxq_enabled = dpdk_rte_mzalloc(OVS_VHOST_MAX_QUEUE_NUM * |
1254 | sizeof *dev->vhost_rxq_enabled); | |
1255 | if (!dev->vhost_rxq_enabled) { | |
1256 | return ENOMEM; | |
1257 | } | |
1ce30dfd DDP |
1258 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); |
1259 | if (!dev->tx_q) { | |
35c91567 | 1260 | rte_free(dev->vhost_rxq_enabled); |
1ce30dfd DDP |
1261 | return ENOMEM; |
1262 | } | |
1263 | ||
bb37956a IM |
1264 | return common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1265 | DPDK_DEV_VHOST, socket_id); | |
1ce30dfd DDP |
1266 | } |
1267 | ||
7d1ced01 | 1268 | static int |
53f50d24 | 1269 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 1270 | { |
d46285a2 DDP |
1271 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1272 | const char *name = netdev->name; | |
7d1ced01 | 1273 | int err; |
a0cb2d66 | 1274 | |
1af27e8a DDP |
1275 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
1276 | * the file system. '/' or '\' would traverse directories, so they're not | |
1277 | * acceptable in 'name'. */ | |
1278 | if (strchr(name, '/') || strchr(name, '\\')) { | |
1279 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
1280 | "A valid name must not include '/' or '\\'", | |
1281 | name); | |
1282 | return EINVAL; | |
1283 | } | |
1284 | ||
7d1ced01 CL |
1285 | ovs_mutex_lock(&dpdk_mutex); |
1286 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 1287 | * the socket is to be created, then register the socket. |
7d1ced01 | 1288 | */ |
bb9d2623 | 1289 | dev->vhost_id = xasprintf("%s/%s", dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 1290 | |
2d24d165 CL |
1291 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
1292 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
1293 | if (err) { |
1294 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 1295 | dev->vhost_id); |
f3e7ec25 | 1296 | goto out; |
e5c0f5a4 | 1297 | } else { |
2d24d165 CL |
1298 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
1299 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
1300 | dev->vhost_id, name); | |
1301 | } | |
f3e7ec25 MW |
1302 | |
1303 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
1304 | &virtio_net_device_ops); | |
1305 | if (err) { | |
1306 | VLOG_ERR("rte_vhost_driver_callback_register failed for vhost user " | |
1307 | "port: %s\n", name); | |
1308 | goto out; | |
1309 | } | |
1310 | ||
1311 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
1312 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
1313 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
1314 | | 1ULL << VIRTIO_NET_F_CSUM); | |
1315 | if (err) { | |
1316 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
1317 | "port: %s\n", name); | |
1318 | goto out; | |
1319 | } | |
1320 | ||
1321 | err = rte_vhost_driver_start(dev->vhost_id); | |
1322 | if (err) { | |
1323 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
1324 | "port: %s\n", name); | |
1325 | goto out; | |
1326 | } | |
1327 | ||
1ce30dfd | 1328 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1329 | if (err) { |
1330 | VLOG_ERR("vhost_common_construct failed for vhost user " | |
1331 | "port: %s\n", name); | |
1332 | } | |
2d24d165 | 1333 | |
f3e7ec25 | 1334 | out: |
bb9d2623 IM |
1335 | if (err) { |
1336 | free(dev->vhost_id); | |
1337 | dev->vhost_id = NULL; | |
1338 | } | |
1339 | ||
2d24d165 | 1340 | ovs_mutex_unlock(&dpdk_mutex); |
28ca969e AC |
1341 | VLOG_WARN_ONCE("dpdkvhostuser ports are considered deprecated; " |
1342 | "please migrate to dpdkvhostuserclient ports."); | |
2d24d165 CL |
1343 | return err; |
1344 | } | |
1345 | ||
1346 | static int | |
1347 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
1348 | { | |
1349 | int err; | |
1350 | ||
2d24d165 | 1351 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 1352 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1353 | if (err) { |
1354 | VLOG_ERR("vhost_common_construct failed for vhost user client" | |
1355 | "port: %s\n", netdev->name); | |
1356 | } | |
7d1ced01 | 1357 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
1358 | return err; |
1359 | } | |
1360 | ||
95fb793a | 1361 | static int |
1362 | netdev_dpdk_construct(struct netdev *netdev) | |
1363 | { | |
95fb793a | 1364 | int err; |
1365 | ||
95fb793a | 1366 | ovs_mutex_lock(&dpdk_mutex); |
bb37956a IM |
1367 | err = common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1368 | DPDK_DEV_ETH, SOCKET0); | |
8a9562d2 PS |
1369 | ovs_mutex_unlock(&dpdk_mutex); |
1370 | return err; | |
1371 | } | |
1372 | ||
1ce30dfd DDP |
1373 | static void |
1374 | common_destruct(struct netdev_dpdk *dev) | |
1375 | OVS_REQUIRES(dpdk_mutex) | |
1376 | OVS_EXCLUDED(dev->mutex) | |
1377 | { | |
1378 | rte_free(dev->tx_q); | |
43307ad0 | 1379 | dpdk_mp_put(dev->dpdk_mp); |
1ce30dfd DDP |
1380 | |
1381 | ovs_list_remove(&dev->list_node); | |
1382 | free(ovsrcu_get_protected(struct ingress_policer *, | |
1383 | &dev->ingress_policer)); | |
1384 | ovs_mutex_destroy(&dev->mutex); | |
1385 | } | |
1386 | ||
8a9562d2 | 1387 | static void |
d46285a2 | 1388 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 1389 | { |
d46285a2 | 1390 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
40e940e4 OM |
1391 | struct rte_device *rte_dev; |
1392 | struct rte_eth_dev *eth_dev; | |
1393 | bool remove_on_close; | |
8a9562d2 | 1394 | |
8d38823b | 1395 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1396 | |
8a9562d2 | 1397 | rte_eth_dev_stop(dev->port_id); |
606f6650 | 1398 | dev->started = false; |
5dcde09c IM |
1399 | |
1400 | if (dev->attached) { | |
40e940e4 OM |
1401 | /* Retrieve eth device data before closing it. |
1402 | * FIXME: avoid direct access to DPDK internal array rte_eth_devices. | |
1403 | */ | |
1404 | eth_dev = &rte_eth_devices[dev->port_id]; | |
1405 | remove_on_close = | |
1406 | eth_dev->data && | |
1407 | (eth_dev->data->dev_flags & RTE_ETH_DEV_CLOSE_REMOVE); | |
1408 | rte_dev = eth_dev->device; | |
1409 | ||
1410 | /* Remove the eth device. */ | |
5dcde09c | 1411 | rte_eth_dev_close(dev->port_id); |
40e940e4 OM |
1412 | |
1413 | /* Remove this rte device and all its eth devices if flag | |
1414 | * RTE_ETH_DEV_CLOSE_REMOVE is not supported (which means representors | |
1415 | * are not supported), or if all the eth devices belonging to the rte | |
1416 | * device are closed. | |
1417 | */ | |
1418 | if (!remove_on_close || !netdev_dpdk_get_num_ports(rte_dev)) { | |
595ce47c IM |
1419 | int ret = rte_dev_remove(rte_dev); |
1420 | ||
1421 | if (ret < 0) { | |
1422 | VLOG_ERR("Device '%s' can not be detached: %s.", | |
1423 | dev->devargs, rte_strerror(-ret)); | |
40e940e4 OM |
1424 | } else { |
1425 | /* Device was closed and detached. */ | |
1426 | VLOG_INFO("Device '%s' has been removed and detached", | |
1427 | dev->devargs); | |
1428 | } | |
5dcde09c | 1429 | } else { |
40e940e4 OM |
1430 | /* Device was only closed. rte_dev_remove() was not called. */ |
1431 | VLOG_INFO("Device '%s' has been removed", dev->devargs); | |
5dcde09c IM |
1432 | } |
1433 | } | |
1434 | ||
ac1a9bb9 | 1435 | netdev_dpdk_clear_xstats(dev); |
55e075e6 | 1436 | free(dev->devargs); |
1ce30dfd | 1437 | common_destruct(dev); |
8d38823b | 1438 | |
8a9562d2 | 1439 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1440 | } |
8a9562d2 | 1441 | |
3f891bbe DDP |
1442 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1443 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1444 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1445 | static int | |
c1ff66ac CL |
1446 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1447 | char *vhost_id) | |
3f891bbe DDP |
1448 | OVS_EXCLUDED(dpdk_mutex) |
1449 | OVS_EXCLUDED(dev->mutex) | |
1450 | { | |
c1ff66ac | 1451 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1452 | } |
1453 | ||
58397e6c | 1454 | static void |
d46285a2 | 1455 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1456 | { |
d46285a2 | 1457 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1458 | char *vhost_id; |
58397e6c | 1459 | |
8d38823b | 1460 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1461 | |
c62da695 | 1462 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1463 | if (netdev_dpdk_get_vid(dev) >= 0 |
1464 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1465 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1466 | netdev->name); |
58be5c0e MK |
1467 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1468 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1469 | } |
1470 | ||
bb9d2623 IM |
1471 | vhost_id = dev->vhost_id; |
1472 | dev->vhost_id = NULL; | |
35c91567 | 1473 | rte_free(dev->vhost_rxq_enabled); |
c1ff66ac | 1474 | |
1ce30dfd DDP |
1475 | common_destruct(dev); |
1476 | ||
58397e6c | 1477 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1478 | |
bb9d2623 | 1479 | if (!vhost_id) { |
821b8664 IM |
1480 | goto out; |
1481 | } | |
1482 | ||
c1ff66ac | 1483 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1484 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1485 | netdev->name, vhost_id); | |
c1ff66ac CL |
1486 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1487 | /* OVS server mode - remove this socket from list for deletion */ | |
1488 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1489 | } |
821b8664 | 1490 | out: |
c1ff66ac | 1491 | free(vhost_id); |
8a9562d2 PS |
1492 | } |
1493 | ||
1494 | static void | |
d46285a2 | 1495 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1496 | { |
d46285a2 | 1497 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1498 | |
d46285a2 | 1499 | rte_free(dev); |
8a9562d2 PS |
1500 | } |
1501 | ||
971f4b39 | 1502 | static void |
ac1a9bb9 | 1503 | netdev_dpdk_clear_xstats(struct netdev_dpdk *dev) |
971f4b39 MW |
1504 | { |
1505 | /* If statistics are already allocated, we have to | |
1506 | * reconfigure, as port_id could have been changed. */ | |
1507 | if (dev->rte_xstats_names) { | |
1508 | free(dev->rte_xstats_names); | |
1509 | dev->rte_xstats_names = NULL; | |
1510 | dev->rte_xstats_names_size = 0; | |
1511 | } | |
1512 | if (dev->rte_xstats_ids) { | |
1513 | free(dev->rte_xstats_ids); | |
1514 | dev->rte_xstats_ids = NULL; | |
1515 | dev->rte_xstats_ids_size = 0; | |
1516 | } | |
1517 | } | |
1518 | ||
1519 | static const char* | |
1520 | netdev_dpdk_get_xstat_name(struct netdev_dpdk *dev, uint64_t id) | |
1521 | { | |
1522 | if (id >= dev->rte_xstats_names_size) { | |
1523 | return "UNKNOWN"; | |
1524 | } | |
1525 | return dev->rte_xstats_names[id].name; | |
1526 | } | |
1527 | ||
1528 | static bool | |
1529 | netdev_dpdk_configure_xstats(struct netdev_dpdk *dev) | |
1530 | OVS_REQUIRES(dev->mutex) | |
1531 | { | |
1532 | int rte_xstats_len; | |
1533 | bool ret; | |
1534 | struct rte_eth_xstat *rte_xstats; | |
1535 | uint64_t id; | |
1536 | int xstats_no; | |
1537 | const char *name; | |
1538 | ||
1539 | /* Retrieving all XSTATS names. If something will go wrong | |
1540 | * or amount of counters will be equal 0, rte_xstats_names | |
1541 | * buffer will be marked as NULL, and any further xstats | |
1542 | * query won't be performed (e.g. during netdev_dpdk_get_stats | |
1543 | * execution). */ | |
1544 | ||
1545 | ret = false; | |
1546 | rte_xstats = NULL; | |
1547 | ||
1548 | if (dev->rte_xstats_names == NULL || dev->rte_xstats_ids == NULL) { | |
1549 | dev->rte_xstats_names_size = | |
1550 | rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
1551 | ||
1552 | if (dev->rte_xstats_names_size < 0) { | |
fa9f4eeb IM |
1553 | VLOG_WARN("Cannot get XSTATS for port: "DPDK_PORT_ID_FMT, |
1554 | dev->port_id); | |
971f4b39 MW |
1555 | dev->rte_xstats_names_size = 0; |
1556 | } else { | |
1557 | /* Reserve memory for xstats names and values */ | |
1558 | dev->rte_xstats_names = xcalloc(dev->rte_xstats_names_size, | |
1559 | sizeof *dev->rte_xstats_names); | |
1560 | ||
1561 | if (dev->rte_xstats_names) { | |
1562 | /* Retreive xstats names */ | |
1563 | rte_xstats_len = | |
1564 | rte_eth_xstats_get_names(dev->port_id, | |
1565 | dev->rte_xstats_names, | |
1566 | dev->rte_xstats_names_size); | |
1567 | ||
1568 | if (rte_xstats_len < 0) { | |
fa9f4eeb IM |
1569 | VLOG_WARN("Cannot get XSTATS names for port: " |
1570 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 MW |
1571 | goto out; |
1572 | } else if (rte_xstats_len != dev->rte_xstats_names_size) { | |
fa9f4eeb IM |
1573 | VLOG_WARN("XSTATS size doesn't match for port: " |
1574 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 MW |
1575 | goto out; |
1576 | } | |
1577 | ||
1578 | dev->rte_xstats_ids = xcalloc(dev->rte_xstats_names_size, | |
1579 | sizeof(uint64_t)); | |
1580 | ||
1581 | /* We have to calculate number of counters */ | |
1582 | rte_xstats = xmalloc(rte_xstats_len * sizeof *rte_xstats); | |
1583 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
1584 | ||
1585 | /* Retreive xstats values */ | |
1586 | if (rte_eth_xstats_get(dev->port_id, rte_xstats, | |
1587 | rte_xstats_len) > 0) { | |
1588 | dev->rte_xstats_ids_size = 0; | |
1589 | xstats_no = 0; | |
1590 | for (uint32_t i = 0; i < rte_xstats_len; i++) { | |
1591 | id = rte_xstats[i].id; | |
1592 | name = netdev_dpdk_get_xstat_name(dev, id); | |
1593 | /* We need to filter out everything except | |
1594 | * dropped, error and management counters */ | |
1595 | if (string_ends_with(name, "_errors") || | |
1596 | strstr(name, "_management_") || | |
1597 | string_ends_with(name, "_dropped")) { | |
1598 | ||
1599 | dev->rte_xstats_ids[xstats_no] = id; | |
1600 | xstats_no++; | |
1601 | } | |
1602 | } | |
1603 | dev->rte_xstats_ids_size = xstats_no; | |
1604 | ret = true; | |
1605 | } else { | |
fa9f4eeb IM |
1606 | VLOG_WARN("Can't get XSTATS IDs for port: " |
1607 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 | 1608 | } |
34eb0863 IM |
1609 | |
1610 | free(rte_xstats); | |
971f4b39 MW |
1611 | } |
1612 | } | |
1613 | } else { | |
1614 | /* Already configured */ | |
1615 | ret = true; | |
1616 | } | |
1617 | ||
1618 | out: | |
1619 | if (!ret) { | |
1620 | netdev_dpdk_clear_xstats(dev); | |
1621 | } | |
1622 | return ret; | |
1623 | } | |
1624 | ||
8a9562d2 | 1625 | static int |
a14b8947 | 1626 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1627 | { |
a14b8947 | 1628 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1629 | |
1630 | ovs_mutex_lock(&dev->mutex); | |
1631 | ||
050c60bf | 1632 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1633 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1634 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1635 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1636 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1637 | |
1638 | if (dev->type == DPDK_DEV_ETH) { | |
1639 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1640 | dev->requested_rxq_size); | |
1641 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1642 | dev->rxq_size); | |
1643 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1644 | dev->requested_txq_size); | |
1645 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1646 | dev->txq_size); | |
1a2bb118 SC |
1647 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1648 | smap_add(args, "rx_csum_offload", "true"); | |
8155ab7e KT |
1649 | } else { |
1650 | smap_add(args, "rx_csum_offload", "false"); | |
1a2bb118 | 1651 | } |
f8b64a61 RM |
1652 | smap_add(args, "lsc_interrupt_mode", |
1653 | dev->lsc_interrupt_mode ? "true" : "false"); | |
451f26fd | 1654 | } |
8a9562d2 PS |
1655 | ovs_mutex_unlock(&dev->mutex); |
1656 | ||
1657 | return 0; | |
1658 | } | |
1659 | ||
55e075e6 | 1660 | static struct netdev_dpdk * |
bb37956a | 1661 | netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id) |
55e075e6 CL |
1662 | OVS_REQUIRES(dpdk_mutex) |
1663 | { | |
1664 | struct netdev_dpdk *dev; | |
1665 | ||
1666 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1667 | if (dev->port_id == port_id) { | |
1668 | return dev; | |
1669 | } | |
1670 | } | |
1671 | ||
1672 | return NULL; | |
1673 | } | |
1674 | ||
5e758818 YL |
1675 | static dpdk_port_t |
1676 | netdev_dpdk_get_port_by_mac(const char *mac_str) | |
1677 | { | |
1678 | dpdk_port_t port_id; | |
1679 | struct eth_addr mac, port_mac; | |
1680 | ||
1681 | if (!eth_addr_from_string(mac_str, &mac)) { | |
1682 | VLOG_ERR("invalid mac: %s", mac_str); | |
1683 | return DPDK_ETH_PORT_ID_INVALID; | |
1684 | } | |
1685 | ||
1686 | RTE_ETH_FOREACH_DEV (port_id) { | |
1687 | struct ether_addr ea; | |
1688 | ||
1689 | rte_eth_macaddr_get(port_id, &ea); | |
1690 | memcpy(port_mac.ea, ea.addr_bytes, ETH_ADDR_LEN); | |
1691 | if (eth_addr_equals(mac, port_mac)) { | |
1692 | return port_id; | |
1693 | } | |
1694 | } | |
1695 | ||
1696 | return DPDK_ETH_PORT_ID_INVALID; | |
1697 | } | |
1698 | ||
40e940e4 OM |
1699 | /* Return the first DPDK port id matching the devargs pattern. */ |
1700 | static dpdk_port_t netdev_dpdk_get_port_by_devargs(const char *devargs) | |
1701 | OVS_REQUIRES(dpdk_mutex) | |
1702 | { | |
1703 | dpdk_port_t port_id; | |
1704 | struct rte_dev_iterator iterator; | |
1705 | ||
1706 | RTE_ETH_FOREACH_MATCHING_DEV (port_id, devargs, &iterator) { | |
1707 | /* If a break is done - must call rte_eth_iterator_cleanup. */ | |
1708 | rte_eth_iterator_cleanup(&iterator); | |
1709 | break; | |
1710 | } | |
1711 | ||
1712 | return port_id; | |
1713 | } | |
1714 | ||
5e758818 | 1715 | /* |
40e940e4 OM |
1716 | * Normally, a PCI id (optionally followed by a representor number) |
1717 | * is enough for identifying a specific DPDK port. | |
5e758818 YL |
1718 | * However, for some NICs having multiple ports sharing the same PCI |
1719 | * id, using PCI id won't work then. | |
1720 | * | |
1721 | * To fix that, here one more method is introduced: "class=eth,mac=$MAC". | |
1722 | * | |
1723 | * Note that the compatibility is fully kept: user can still use the | |
1724 | * PCI id for adding ports (when it's enough for them). | |
1725 | */ | |
bb37956a | 1726 | static dpdk_port_t |
5dcde09c IM |
1727 | netdev_dpdk_process_devargs(struct netdev_dpdk *dev, |
1728 | const char *devargs, char **errp) | |
40e940e4 | 1729 | OVS_REQUIRES(dpdk_mutex) |
55e075e6 | 1730 | { |
40e940e4 | 1731 | dpdk_port_t new_port_id; |
55e075e6 | 1732 | |
5e758818 YL |
1733 | if (strncmp(devargs, "class=eth,mac=", 14) == 0) { |
1734 | new_port_id = netdev_dpdk_get_port_by_mac(&devargs[14]); | |
1735 | } else { | |
40e940e4 OM |
1736 | new_port_id = netdev_dpdk_get_port_by_devargs(devargs); |
1737 | if (!rte_eth_dev_is_valid_port(new_port_id)) { | |
5e758818 | 1738 | /* Device not found in DPDK, attempt to attach it */ |
40e940e4 | 1739 | if (rte_dev_probe(devargs)) { |
5e758818 | 1740 | new_port_id = DPDK_ETH_PORT_ID_INVALID; |
40e940e4 OM |
1741 | } else { |
1742 | new_port_id = netdev_dpdk_get_port_by_devargs(devargs); | |
1743 | if (rte_eth_dev_is_valid_port(new_port_id)) { | |
1744 | /* Attach successful */ | |
1745 | dev->attached = true; | |
1746 | VLOG_INFO("Device '%s' attached to DPDK", devargs); | |
1747 | } else { | |
1748 | /* Attach unsuccessful */ | |
1749 | new_port_id = DPDK_ETH_PORT_ID_INVALID; | |
1750 | } | |
5e758818 | 1751 | } |
55e075e6 | 1752 | } |
5e758818 YL |
1753 | } |
1754 | ||
1755 | if (new_port_id == DPDK_ETH_PORT_ID_INVALID) { | |
1756 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", devargs); | |
55e075e6 CL |
1757 | } |
1758 | ||
1759 | return new_port_id; | |
1760 | } | |
1761 | ||
c3d062a7 CL |
1762 | static void |
1763 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1764 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1765 | { |
050c60bf | 1766 | int new_n_rxq; |
a14b8947 | 1767 | |
2a21e757 | 1768 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1769 | if (new_n_rxq != dev->requested_n_rxq) { |
1770 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1771 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1772 | } |
c3d062a7 CL |
1773 | } |
1774 | ||
b685696b CL |
1775 | static void |
1776 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1777 | const char *flag, int default_size, int *new_size) | |
1778 | { | |
1779 | int queue_size = smap_get_int(args, flag, default_size); | |
1780 | ||
1781 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1782 | || !is_pow2(queue_size)) { | |
1783 | queue_size = default_size; | |
1784 | } | |
1785 | ||
1786 | if (queue_size != *new_size) { | |
1787 | *new_size = queue_size; | |
1788 | netdev_request_reconfigure(netdev); | |
1789 | } | |
1790 | } | |
1791 | ||
c3d062a7 | 1792 | static int |
9fff138e DDP |
1793 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1794 | char **errp) | |
c3d062a7 CL |
1795 | { |
1796 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
f8b64a61 | 1797 | bool rx_fc_en, tx_fc_en, autoneg, lsc_interrupt_mode; |
b614c894 IM |
1798 | enum rte_eth_fc_mode fc_mode; |
1799 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1800 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1801 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1802 | }; | |
55e075e6 CL |
1803 | const char *new_devargs; |
1804 | int err = 0; | |
c3d062a7 | 1805 | |
55e075e6 | 1806 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1807 | ovs_mutex_lock(&dev->mutex); |
1808 | ||
1809 | dpdk_set_rxq_config(dev, args); | |
1810 | ||
b685696b CL |
1811 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1812 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1813 | &dev->requested_rxq_size); | |
1814 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1815 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1816 | &dev->requested_txq_size); | |
1817 | ||
55e075e6 CL |
1818 | new_devargs = smap_get(args, "dpdk-devargs"); |
1819 | ||
1820 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1821 | /* The user requested a new device. If we return error, the caller | |
1822 | * will delete this netdev and try to recreate it. */ | |
1823 | err = EAGAIN; | |
1824 | goto out; | |
1825 | } | |
1826 | ||
1827 | /* dpdk-devargs is required for device configuration */ | |
1828 | if (new_devargs && new_devargs[0]) { | |
1829 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1830 | * is valid */ | |
1831 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1832 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
bb37956a IM |
1833 | dpdk_port_t new_port_id = netdev_dpdk_process_devargs(dev, |
1834 | new_devargs, | |
1835 | errp); | |
55e075e6 CL |
1836 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1837 | err = EINVAL; | |
1838 | } else if (new_port_id == dev->port_id) { | |
1839 | /* Already configured, do not reconfigure again */ | |
1840 | err = 0; | |
1841 | } else { | |
1842 | struct netdev_dpdk *dup_dev; | |
bb37956a | 1843 | |
55e075e6 CL |
1844 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); |
1845 | if (dup_dev) { | |
9fff138e | 1846 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
40e940e4 | 1847 | "which is already in use by '%s'", |
9fff138e DDP |
1848 | netdev_get_name(netdev), new_devargs, |
1849 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1850 | err = EADDRINUSE; |
1851 | } else { | |
bd4e172b | 1852 | int sid = rte_eth_dev_socket_id(new_port_id); |
bb37956a | 1853 | |
bd4e172b | 1854 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; |
55e075e6 CL |
1855 | dev->devargs = xstrdup(new_devargs); |
1856 | dev->port_id = new_port_id; | |
1857 | netdev_request_reconfigure(&dev->up); | |
971f4b39 | 1858 | netdev_dpdk_clear_xstats(dev); |
55e075e6 CL |
1859 | err = 0; |
1860 | } | |
1861 | } | |
1862 | } | |
1863 | } else { | |
9fff138e DDP |
1864 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1865 | "The old 'dpdk<port_id>' names are not supported", | |
1866 | netdev_get_name(netdev)); | |
55e075e6 CL |
1867 | err = EINVAL; |
1868 | } | |
1869 | ||
1870 | if (err) { | |
1871 | goto out; | |
1872 | } | |
1873 | ||
f8b64a61 RM |
1874 | lsc_interrupt_mode = smap_get_bool(args, "dpdk-lsc-interrupt", false); |
1875 | if (dev->requested_lsc_interrupt_mode != lsc_interrupt_mode) { | |
1876 | dev->requested_lsc_interrupt_mode = lsc_interrupt_mode; | |
1877 | netdev_request_reconfigure(netdev); | |
1878 | } | |
1879 | ||
c3d062a7 CL |
1880 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1881 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1882 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1883 | |
b614c894 IM |
1884 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1885 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1886 | dev->fc_conf.mode = fc_mode; | |
1887 | dev->fc_conf.autoneg = autoneg; | |
7e1de65e SC |
1888 | /* Get the Flow control configuration for DPDK-ETH */ |
1889 | err = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
1890 | if (err) { | |
1891 | VLOG_WARN("Cannot get flow control parameters on port " | |
1892 | DPDK_PORT_ID_FMT", err=%d", dev->port_id, err); | |
1893 | } | |
b614c894 IM |
1894 | dpdk_eth_flow_ctrl_setup(dev); |
1895 | } | |
9fd39370 | 1896 | |
55e075e6 | 1897 | out: |
c3d062a7 | 1898 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1899 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1900 | |
55e075e6 | 1901 | return err; |
c3d062a7 CL |
1902 | } |
1903 | ||
1904 | static int | |
9fff138e DDP |
1905 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1906 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1907 | { |
1908 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1909 | ||
1910 | ovs_mutex_lock(&dev->mutex); | |
1911 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1912 | ovs_mutex_unlock(&dev->mutex); |
1913 | ||
1914 | return 0; | |
1915 | } | |
1916 | ||
c1ff66ac | 1917 | static int |
2d24d165 | 1918 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1919 | const struct smap *args, |
1920 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1921 | { |
1922 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1923 | const char *path; | |
1924 | ||
6881885a | 1925 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1926 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1927 | path = smap_get(args, "vhost-server-path"); | |
bb9d2623 IM |
1928 | if (!nullable_string_is_equal(path, dev->vhost_id)) { |
1929 | free(dev->vhost_id); | |
1930 | dev->vhost_id = nullable_xstrdup(path); | |
10087cba CL |
1931 | /* check zero copy configuration */ |
1932 | if (smap_get_bool(args, "dq-zero-copy", false)) { | |
1933 | dev->vhost_driver_flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
1934 | } else { | |
1935 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
1936 | } | |
c1ff66ac CL |
1937 | netdev_request_reconfigure(netdev); |
1938 | } | |
1939 | } | |
6881885a | 1940 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1941 | |
1942 | return 0; | |
1943 | } | |
1944 | ||
7dec44fe | 1945 | static int |
d46285a2 | 1946 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1947 | { |
d46285a2 | 1948 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1949 | |
d46285a2 | 1950 | return dev->socket_id; |
7dec44fe AW |
1951 | } |
1952 | ||
050c60bf | 1953 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1954 | static int |
050c60bf | 1955 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1956 | { |
d46285a2 | 1957 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1958 | |
d46285a2 | 1959 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1960 | |
050c60bf DDP |
1961 | if (dev->requested_n_txq == n_txq) { |
1962 | goto out; | |
4573fbd3 FL |
1963 | } |
1964 | ||
050c60bf DDP |
1965 | dev->requested_n_txq = n_txq; |
1966 | netdev_request_reconfigure(netdev); | |
58397e6c | 1967 | |
050c60bf | 1968 | out: |
d46285a2 | 1969 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1970 | return 0; |
58397e6c KT |
1971 | } |
1972 | ||
8a9562d2 PS |
1973 | static struct netdev_rxq * |
1974 | netdev_dpdk_rxq_alloc(void) | |
1975 | { | |
1976 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1977 | ||
eff23640 DDP |
1978 | if (rx) { |
1979 | return &rx->up; | |
1980 | } | |
1981 | ||
1982 | return NULL; | |
8a9562d2 PS |
1983 | } |
1984 | ||
1985 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1986 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1987 | { |
d46285a2 | 1988 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1989 | } |
1990 | ||
1991 | static int | |
d46285a2 | 1992 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1993 | { |
d46285a2 DDP |
1994 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1995 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1996 | |
d46285a2 DDP |
1997 | ovs_mutex_lock(&dev->mutex); |
1998 | rx->port_id = dev->port_id; | |
1999 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2000 | |
2001 | return 0; | |
2002 | } | |
2003 | ||
2004 | static void | |
d46285a2 | 2005 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
2006 | { |
2007 | } | |
2008 | ||
2009 | static void | |
d46285a2 | 2010 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 2011 | { |
d46285a2 | 2012 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
2013 | |
2014 | rte_free(rx); | |
2015 | } | |
2016 | ||
819f13bd DDP |
2017 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
2018 | * 'pkts', even in case of failure. | |
2019 | * | |
2020 | * Returns the number of packets that weren't transmitted. */ | |
2021 | static inline int | |
b59cc14e | 2022 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 2023 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 2024 | { |
1304f1f8 DDP |
2025 | uint32_t nb_tx = 0; |
2026 | ||
b59cc14e | 2027 | while (nb_tx != cnt) { |
1304f1f8 DDP |
2028 | uint32_t ret; |
2029 | ||
b59cc14e | 2030 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
2031 | if (!ret) { |
2032 | break; | |
2033 | } | |
2034 | ||
2035 | nb_tx += ret; | |
2036 | } | |
8a9562d2 | 2037 | |
b59cc14e | 2038 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 2039 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
2040 | * packet could come from a different mempool) */ |
2041 | int i; | |
2042 | ||
b59cc14e IM |
2043 | for (i = nb_tx; i < cnt; i++) { |
2044 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 2045 | } |
8a9562d2 | 2046 | } |
819f13bd DDP |
2047 | |
2048 | return cnt - nb_tx; | |
8a9562d2 PS |
2049 | } |
2050 | ||
f3926f29 IS |
2051 | static inline bool |
2052 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
03f3f9c0 | 2053 | struct rte_meter_srtcm_profile *profile, |
f3926f29 IS |
2054 | struct rte_mbuf *pkt, uint64_t time) |
2055 | { | |
2056 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
2057 | ||
03f3f9c0 OM |
2058 | return rte_meter_srtcm_color_blind_check(meter, profile, time, pkt_len) == |
2059 | e_RTE_METER_GREEN; | |
f3926f29 IS |
2060 | } |
2061 | ||
2062 | static int | |
2063 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
03f3f9c0 | 2064 | struct rte_meter_srtcm_profile *profile, |
3e90f7d7 | 2065 | struct rte_mbuf **pkts, int pkt_cnt, |
7d7ded7a | 2066 | bool should_steal) |
f3926f29 IS |
2067 | { |
2068 | int i = 0; | |
2069 | int cnt = 0; | |
2070 | struct rte_mbuf *pkt = NULL; | |
2071 | uint64_t current_time = rte_rdtsc(); | |
2072 | ||
2073 | for (i = 0; i < pkt_cnt; i++) { | |
2074 | pkt = pkts[i]; | |
2075 | /* Handle current packet */ | |
03f3f9c0 OM |
2076 | if (netdev_dpdk_policer_pkt_handle(meter, profile, |
2077 | pkt, current_time)) { | |
f3926f29 IS |
2078 | if (cnt != i) { |
2079 | pkts[cnt] = pkt; | |
2080 | } | |
2081 | cnt++; | |
2082 | } else { | |
7d7ded7a | 2083 | if (should_steal) { |
3e90f7d7 GZ |
2084 | rte_pktmbuf_free(pkt); |
2085 | } | |
f3926f29 IS |
2086 | } |
2087 | } | |
2088 | ||
2089 | return cnt; | |
2090 | } | |
2091 | ||
9509913a IS |
2092 | static int |
2093 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
7d7ded7a | 2094 | int pkt_cnt, bool should_steal) |
9509913a IS |
2095 | { |
2096 | int cnt = 0; | |
2097 | ||
2098 | rte_spinlock_lock(&policer->policer_lock); | |
03f3f9c0 OM |
2099 | cnt = netdev_dpdk_policer_run(&policer->in_policer, &policer->in_prof, |
2100 | pkts, pkt_cnt, should_steal); | |
9509913a IS |
2101 | rte_spinlock_unlock(&policer->policer_lock); |
2102 | ||
2103 | return cnt; | |
2104 | } | |
2105 | ||
58397e6c | 2106 | static bool |
0a0f39df | 2107 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 2108 | { |
0a0f39df | 2109 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
2110 | } |
2111 | ||
d6e3feb5 | 2112 | static inline void |
2113 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
2114 | unsigned int packet_size) | |
2115 | { | |
2116 | /* Hard-coded search for the size bucket. */ | |
2117 | if (packet_size < 256) { | |
2118 | if (packet_size >= 128) { | |
2119 | stats->rx_128_to_255_packets++; | |
2120 | } else if (packet_size <= 64) { | |
2121 | stats->rx_1_to_64_packets++; | |
2122 | } else { | |
2123 | stats->rx_65_to_127_packets++; | |
2124 | } | |
2125 | } else { | |
2126 | if (packet_size >= 1523) { | |
2127 | stats->rx_1523_to_max_packets++; | |
2128 | } else if (packet_size >= 1024) { | |
2129 | stats->rx_1024_to_1522_packets++; | |
2130 | } else if (packet_size < 512) { | |
2131 | stats->rx_256_to_511_packets++; | |
2132 | } else { | |
2133 | stats->rx_512_to_1023_packets++; | |
2134 | } | |
2135 | } | |
2136 | } | |
2137 | ||
9e3ddd45 TP |
2138 | static inline void |
2139 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
2140 | struct dp_packet **packets, int count, |
2141 | int dropped) | |
9e3ddd45 TP |
2142 | { |
2143 | int i; | |
d6e3feb5 | 2144 | unsigned int packet_size; |
9e3ddd45 TP |
2145 | struct dp_packet *packet; |
2146 | ||
2147 | stats->rx_packets += count; | |
9509913a | 2148 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
2149 | for (i = 0; i < count; i++) { |
2150 | packet = packets[i]; | |
d6e3feb5 | 2151 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 2152 | |
d6e3feb5 | 2153 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
2154 | /* This only protects the following multicast counting from |
2155 | * too short packets, but it does not stop the packet from | |
2156 | * further processing. */ | |
2157 | stats->rx_errors++; | |
2158 | stats->rx_length_errors++; | |
2159 | continue; | |
2160 | } | |
2161 | ||
d6e3feb5 | 2162 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
2163 | ||
9e3ddd45 TP |
2164 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
2165 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
2166 | stats->multicast++; | |
2167 | } | |
2168 | ||
d6e3feb5 | 2169 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
2170 | } |
2171 | } | |
2172 | ||
58397e6c KT |
2173 | /* |
2174 | * The receive path for the vhost port is the TX path out from guest. | |
2175 | */ | |
2176 | static int | |
d46285a2 | 2177 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
8492adc2 | 2178 | struct dp_packet_batch *batch, int *qfill) |
58397e6c | 2179 | { |
d46285a2 | 2180 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
9509913a | 2181 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 2182 | uint16_t nb_rx = 0; |
9509913a | 2183 | uint16_t dropped = 0; |
8492adc2 | 2184 | int qid = rxq->queue_id * VIRTIO_QNUM + VIRTIO_TXQ; |
daf22bf7 | 2185 | int vid = netdev_dpdk_get_vid(dev); |
58397e6c | 2186 | |
daf22bf7 | 2187 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured |
e543851d | 2188 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
2189 | return EAGAIN; |
2190 | } | |
2191 | ||
43307ad0 | 2192 | nb_rx = rte_vhost_dequeue_burst(vid, qid, dev->dpdk_mp->mp, |
64839cf4 | 2193 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 2194 | NETDEV_MAX_BURST); |
58397e6c KT |
2195 | if (!nb_rx) { |
2196 | return EAGAIN; | |
2197 | } | |
2198 | ||
8492adc2 JS |
2199 | if (qfill) { |
2200 | if (nb_rx == NETDEV_MAX_BURST) { | |
2201 | /* The DPDK API returns a uint32_t which often has invalid bits in | |
2202 | * the upper 16-bits. Need to restrict the value to uint16_t. */ | |
2203 | *qfill = rte_vhost_rx_queue_count(vid, qid) & UINT16_MAX; | |
2204 | } else { | |
2205 | *qfill = 0; | |
2206 | } | |
2207 | } | |
2208 | ||
9509913a IS |
2209 | if (policer) { |
2210 | dropped = nb_rx; | |
64839cf4 WT |
2211 | nb_rx = ingress_policer_run(policer, |
2212 | (struct rte_mbuf **) batch->packets, | |
3e90f7d7 | 2213 | nb_rx, true); |
9509913a IS |
2214 | dropped -= nb_rx; |
2215 | } | |
2216 | ||
d46285a2 | 2217 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
2218 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
2219 | nb_rx, dropped); | |
d46285a2 | 2220 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 2221 | |
75fb9148 ZB |
2222 | batch->count = nb_rx; |
2223 | dp_packet_batch_init_packet_fields(batch); | |
2224 | ||
58397e6c KT |
2225 | return 0; |
2226 | } | |
2227 | ||
35c91567 DM |
2228 | static bool |
2229 | netdev_dpdk_vhost_rxq_enabled(struct netdev_rxq *rxq) | |
2230 | { | |
2231 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
2232 | ||
2233 | return dev->vhost_rxq_enabled[rxq->queue_id]; | |
2234 | } | |
2235 | ||
8a9562d2 | 2236 | static int |
8492adc2 JS |
2237 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch, |
2238 | int *qfill) | |
8a9562d2 | 2239 | { |
d46285a2 DDP |
2240 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
2241 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 2242 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 2243 | int nb_rx; |
9509913a | 2244 | int dropped = 0; |
8a9562d2 | 2245 | |
3b1fb077 DDP |
2246 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
2247 | return EAGAIN; | |
2248 | } | |
2249 | ||
d46285a2 | 2250 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 2251 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 2252 | NETDEV_MAX_BURST); |
8a9562d2 PS |
2253 | if (!nb_rx) { |
2254 | return EAGAIN; | |
2255 | } | |
2256 | ||
9509913a IS |
2257 | if (policer) { |
2258 | dropped = nb_rx; | |
64839cf4 | 2259 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 2260 | (struct rte_mbuf **) batch->packets, |
3e90f7d7 | 2261 | nb_rx, true); |
9509913a IS |
2262 | dropped -= nb_rx; |
2263 | } | |
2264 | ||
2265 | /* Update stats to reflect dropped packets */ | |
2266 | if (OVS_UNLIKELY(dropped)) { | |
2267 | rte_spinlock_lock(&dev->stats_lock); | |
2268 | dev->stats.rx_dropped += dropped; | |
2269 | rte_spinlock_unlock(&dev->stats_lock); | |
2270 | } | |
2271 | ||
64839cf4 | 2272 | batch->count = nb_rx; |
75fb9148 | 2273 | dp_packet_batch_init_packet_fields(batch); |
8a9562d2 | 2274 | |
8492adc2 JS |
2275 | if (qfill) { |
2276 | if (nb_rx == NETDEV_MAX_BURST) { | |
2277 | *qfill = rte_eth_rx_queue_count(rx->port_id, rxq->queue_id); | |
2278 | } else { | |
2279 | *qfill = 0; | |
2280 | } | |
2281 | } | |
2282 | ||
8a9562d2 PS |
2283 | return 0; |
2284 | } | |
2285 | ||
0bf765f7 | 2286 | static inline int |
78bd47cf | 2287 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
7d7ded7a | 2288 | int cnt, bool should_steal) |
0bf765f7 | 2289 | { |
78bd47cf | 2290 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 2291 | |
78bd47cf DDP |
2292 | if (qos_conf) { |
2293 | rte_spinlock_lock(&qos_conf->lock); | |
7d7ded7a | 2294 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt, should_steal); |
78bd47cf | 2295 | rte_spinlock_unlock(&qos_conf->lock); |
0bf765f7 IS |
2296 | } |
2297 | ||
2298 | return cnt; | |
2299 | } | |
2300 | ||
c6ec9d17 IM |
2301 | static int |
2302 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
2303 | int pkt_cnt) | |
2304 | { | |
2305 | int i = 0; | |
2306 | int cnt = 0; | |
2307 | struct rte_mbuf *pkt; | |
2308 | ||
2309 | for (i = 0; i < pkt_cnt; i++) { | |
2310 | pkt = pkts[i]; | |
2311 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
2312 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
2313 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
2314 | rte_pktmbuf_free(pkt); | |
2315 | continue; | |
2316 | } | |
2317 | ||
2318 | if (OVS_UNLIKELY(i != cnt)) { | |
2319 | pkts[cnt] = pkt; | |
2320 | } | |
2321 | cnt++; | |
2322 | } | |
2323 | ||
2324 | return cnt; | |
2325 | } | |
2326 | ||
9e3ddd45 TP |
2327 | static inline void |
2328 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
2329 | struct dp_packet **packets, | |
2330 | int attempted, | |
2331 | int dropped) | |
2332 | { | |
2333 | int i; | |
2334 | int sent = attempted - dropped; | |
2335 | ||
2336 | stats->tx_packets += sent; | |
2337 | stats->tx_dropped += dropped; | |
2338 | ||
2339 | for (i = 0; i < sent; i++) { | |
2340 | stats->tx_bytes += dp_packet_size(packets[i]); | |
2341 | } | |
2342 | } | |
2343 | ||
58397e6c | 2344 | static void |
4573fbd3 | 2345 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 2346 | struct dp_packet **pkts, int cnt) |
58397e6c | 2347 | { |
d46285a2 | 2348 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
2349 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
2350 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 2351 | unsigned int dropped = 0; |
dd52de45 | 2352 | int i, retries = 0; |
daf22bf7 | 2353 | int vid = netdev_dpdk_get_vid(dev); |
58397e6c | 2354 | |
81acebda | 2355 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 2356 | |
daf22bf7 | 2357 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured || qid < 0 |
e543851d | 2358 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
2359 | rte_spinlock_lock(&dev->stats_lock); |
2360 | dev->stats.tx_dropped+= cnt; | |
2361 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 2362 | goto out; |
58397e6c KT |
2363 | } |
2364 | ||
d46285a2 | 2365 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 2366 | |
c6ec9d17 | 2367 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 2368 | /* Check has QoS has been configured for the netdev */ |
3e90f7d7 | 2369 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt, true); |
c6ec9d17 | 2370 | dropped = total_pkts - cnt; |
0bf765f7 | 2371 | |
95e9881f | 2372 | do { |
4573fbd3 | 2373 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
2374 | unsigned int tx_pkts; |
2375 | ||
daf22bf7 | 2376 | tx_pkts = rte_vhost_enqueue_burst(vid, vhost_qid, cur_pkts, cnt); |
95e9881f KT |
2377 | if (OVS_LIKELY(tx_pkts)) { |
2378 | /* Packets have been sent.*/ | |
2379 | cnt -= tx_pkts; | |
31871ee3 | 2380 | /* Prepare for possible retry.*/ |
95e9881f KT |
2381 | cur_pkts = &cur_pkts[tx_pkts]; |
2382 | } else { | |
31871ee3 KT |
2383 | /* No packets sent - do not retry.*/ |
2384 | break; | |
95e9881f | 2385 | } |
730b3485 | 2386 | } while (cnt && (retries++ < VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 2387 | |
d46285a2 | 2388 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 2389 | |
d46285a2 | 2390 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 2391 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 2392 | cnt + dropped); |
c161357d | 2393 | dev->tx_retries += MIN(retries, VHOST_ENQ_RETRY_NUM); |
d46285a2 | 2394 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
2395 | |
2396 | out: | |
c6ec9d17 | 2397 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 2398 | dp_packet_delete(pkts[i]); |
58397e6c KT |
2399 | } |
2400 | } | |
2401 | ||
8a9562d2 PS |
2402 | /* Tx function. Transmit packets indefinitely */ |
2403 | static void | |
64839cf4 | 2404 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 2405 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 2406 | { |
8a14bd7b | 2407 | const size_t batch_cnt = dp_packet_batch_size(batch); |
bce01e3a | 2408 | #if !defined(__CHECKER__) && !defined(_WIN32) |
8a14bd7b | 2409 | const size_t PKT_ARRAY_SIZE = batch_cnt; |
bce01e3a EJ |
2410 | #else |
2411 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 2412 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 2413 | #endif |
8a9562d2 | 2414 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 2415 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
8a14bd7b | 2416 | uint32_t cnt = batch_cnt; |
3e90f7d7 GZ |
2417 | uint32_t dropped = 0; |
2418 | ||
2419 | if (dev->type != DPDK_DEV_VHOST) { | |
2420 | /* Check if QoS has been configured for this netdev. */ | |
2421 | cnt = netdev_dpdk_qos_run(dev, (struct rte_mbuf **) batch->packets, | |
8a14bd7b BB |
2422 | batch_cnt, false); |
2423 | dropped += batch_cnt - cnt; | |
3e90f7d7 | 2424 | } |
8a9562d2 | 2425 | |
3e90f7d7 GZ |
2426 | uint32_t txcnt = 0; |
2427 | ||
2428 | for (uint32_t i = 0; i < cnt; i++) { | |
8a14bd7b BB |
2429 | struct dp_packet *packet = batch->packets[i]; |
2430 | uint32_t size = dp_packet_size(packet); | |
95fb793a | 2431 | |
f98d7864 | 2432 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
3e90f7d7 GZ |
2433 | VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d", |
2434 | size, dev->max_packet_len); | |
f4fd623c | 2435 | |
175cf4de | 2436 | dropped++; |
f4fd623c DDP |
2437 | continue; |
2438 | } | |
8a9562d2 | 2439 | |
43307ad0 | 2440 | pkts[txcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a14bd7b | 2441 | if (OVS_UNLIKELY(!pkts[txcnt])) { |
3e90f7d7 | 2442 | dropped += cnt - i; |
175cf4de | 2443 | break; |
f4fd623c DDP |
2444 | } |
2445 | ||
2446 | /* We have to do a copy for now */ | |
3e90f7d7 | 2447 | memcpy(rte_pktmbuf_mtod(pkts[txcnt], void *), |
8a14bd7b BB |
2448 | dp_packet_data(packet), size); |
2449 | dp_packet_set_size((struct dp_packet *)pkts[txcnt], size); | |
f4fd623c | 2450 | |
3e90f7d7 | 2451 | txcnt++; |
f4fd623c | 2452 | } |
8a9562d2 | 2453 | |
3e90f7d7 GZ |
2454 | if (OVS_LIKELY(txcnt)) { |
2455 | if (dev->type == DPDK_DEV_VHOST) { | |
2456 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, | |
2457 | txcnt); | |
2458 | } else { | |
2459 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, txcnt); | |
2460 | } | |
58397e6c | 2461 | } |
db73f716 | 2462 | |
0bf765f7 IS |
2463 | if (OVS_UNLIKELY(dropped)) { |
2464 | rte_spinlock_lock(&dev->stats_lock); | |
2465 | dev->stats.tx_dropped += dropped; | |
2466 | rte_spinlock_unlock(&dev->stats_lock); | |
2467 | } | |
8a9562d2 PS |
2468 | } |
2469 | ||
58397e6c | 2470 | static int |
64839cf4 WT |
2471 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
2472 | struct dp_packet_batch *batch, | |
b30896c9 | 2473 | bool concurrent_txq OVS_UNUSED) |
58397e6c | 2474 | { |
58397e6c | 2475 | |
b30896c9 | 2476 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 | 2477 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2478 | dp_packet_delete_batch(batch, true); |
58397e6c | 2479 | } else { |
dd52de45 | 2480 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
2481 | } |
2482 | return 0; | |
2483 | } | |
2484 | ||
7251515e DV |
2485 | static inline void |
2486 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
b30896c9 | 2487 | struct dp_packet_batch *batch, |
324c8374 | 2488 | bool concurrent_txq) |
8a9562d2 | 2489 | { |
3b1fb077 | 2490 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
b30896c9 | 2491 | dp_packet_delete_batch(batch, true); |
3b1fb077 DDP |
2492 | return; |
2493 | } | |
2494 | ||
324c8374 | 2495 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 2496 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
2497 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
2498 | } | |
2499 | ||
b30896c9 | 2500 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
2501 | struct netdev *netdev = &dev->up; |
2502 | ||
64839cf4 | 2503 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2504 | dp_packet_delete_batch(batch, true); |
8a9562d2 | 2505 | } else { |
fd57eeba BB |
2506 | int tx_cnt, dropped; |
2507 | int batch_cnt = dp_packet_batch_size(batch); | |
2391135c | 2508 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 2509 | |
fd57eeba BB |
2510 | tx_cnt = netdev_dpdk_filter_packet_len(dev, pkts, batch_cnt); |
2511 | tx_cnt = netdev_dpdk_qos_run(dev, pkts, tx_cnt, true); | |
2512 | dropped = batch_cnt - tx_cnt; | |
1b99bb05 | 2513 | |
fd57eeba | 2514 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, tx_cnt); |
8a9562d2 | 2515 | |
f4fd623c | 2516 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 2517 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 2518 | dev->stats.tx_dropped += dropped; |
45d947c4 | 2519 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 2520 | } |
8a9562d2 | 2521 | } |
a0cb2d66 | 2522 | |
324c8374 | 2523 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
2524 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
2525 | } | |
7251515e DV |
2526 | } |
2527 | ||
2528 | static int | |
2529 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
b30896c9 | 2530 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e DV |
2531 | { |
2532 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 2533 | |
b30896c9 | 2534 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e | 2535 | return 0; |
8a9562d2 PS |
2536 | } |
2537 | ||
2538 | static int | |
74ff3298 | 2539 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
2540 | { |
2541 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2542 | ||
2543 | ovs_mutex_lock(&dev->mutex); | |
2544 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 2545 | dev->hwaddr = mac; |
045c0d1a | 2546 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
2547 | } |
2548 | ovs_mutex_unlock(&dev->mutex); | |
2549 | ||
2550 | return 0; | |
2551 | } | |
2552 | ||
2553 | static int | |
74ff3298 | 2554 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
2555 | { |
2556 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2557 | ||
2558 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 2559 | *mac = dev->hwaddr; |
8a9562d2 PS |
2560 | ovs_mutex_unlock(&dev->mutex); |
2561 | ||
2562 | return 0; | |
2563 | } | |
2564 | ||
2565 | static int | |
2566 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
2567 | { | |
2568 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2569 | ||
2570 | ovs_mutex_lock(&dev->mutex); | |
2571 | *mtup = dev->mtu; | |
2572 | ovs_mutex_unlock(&dev->mutex); | |
2573 | ||
2574 | return 0; | |
2575 | } | |
2576 | ||
0072e931 MK |
2577 | static int |
2578 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
2579 | { | |
2580 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2581 | ||
f6f50552 IS |
2582 | /* XXX: Ensure that the overall frame length of the requested MTU does not |
2583 | * surpass the NETDEV_DPDK_MAX_PKT_LEN. DPDK device drivers differ in how | |
2584 | * the L2 frame length is calculated for a given MTU when | |
2585 | * rte_eth_dev_set_mtu(mtu) is called e.g. i40e driver includes 2 x vlan | |
2586 | * headers, the em driver includes 1 x vlan header, the ixgbe driver does | |
2587 | * not include vlan headers. As such we should use | |
2588 | * MTU_TO_MAX_FRAME_LEN(mtu) which includes an additional 2 x vlan headers | |
2589 | * (8 bytes) for comparison. This avoids a failure later with | |
2590 | * rte_eth_dev_set_mtu(). This approach should be used until DPDK provides | |
2591 | * a method to retrieve the upper bound MTU for a given device. | |
2592 | */ | |
2593 | if (MTU_TO_MAX_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
0072e931 MK |
2594 | || mtu < ETHER_MIN_MTU) { |
2595 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
2596 | return EINVAL; | |
2597 | } | |
2598 | ||
2599 | ovs_mutex_lock(&dev->mutex); | |
2600 | if (dev->requested_mtu != mtu) { | |
2601 | dev->requested_mtu = mtu; | |
2602 | netdev_request_reconfigure(netdev); | |
2603 | } | |
2604 | ovs_mutex_unlock(&dev->mutex); | |
2605 | ||
2606 | return 0; | |
2607 | } | |
2608 | ||
8a9562d2 | 2609 | static int |
d46285a2 | 2610 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 2611 | |
58397e6c KT |
2612 | static int |
2613 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
2614 | struct netdev_stats *stats) | |
2615 | { | |
2616 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2617 | ||
2618 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 2619 | |
45d947c4 | 2620 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c | 2621 | /* Supported Stats */ |
50986e78 | 2622 | stats->rx_packets = dev->stats.rx_packets; |
2623 | stats->tx_packets = dev->stats.tx_packets; | |
9509913a | 2624 | stats->rx_dropped = dev->stats.rx_dropped; |
50986e78 | 2625 | stats->tx_dropped = dev->stats.tx_dropped; |
9e3ddd45 TP |
2626 | stats->multicast = dev->stats.multicast; |
2627 | stats->rx_bytes = dev->stats.rx_bytes; | |
2628 | stats->tx_bytes = dev->stats.tx_bytes; | |
2629 | stats->rx_errors = dev->stats.rx_errors; | |
2630 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 2631 | |
2632 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
2633 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
2634 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
2635 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
2636 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
2637 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
2638 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
2639 | ||
45d947c4 | 2640 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 2641 | |
58397e6c KT |
2642 | ovs_mutex_unlock(&dev->mutex); |
2643 | ||
2644 | return 0; | |
2645 | } | |
2646 | ||
d6e3feb5 | 2647 | static void |
2648 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
2649 | const struct rte_eth_xstat *xstats, |
2650 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 2651 | const unsigned int size) |
2652 | { | |
d6e3feb5 | 2653 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 2654 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2655 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2656 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2657 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2658 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2659 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2660 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2661 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2662 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2663 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2664 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2665 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2666 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2667 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 2668 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2669 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 2670 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2671 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 2672 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2673 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 2674 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2675 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 2676 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2677 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 2678 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2679 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 2680 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2681 | stats->tx_1523_to_max_packets = xstats[i].value; |
d57f777f PS |
2682 | } else if (strcmp(XSTAT_RX_MULTICAST_PACKETS, names[i].name) == 0) { |
2683 | stats->multicast = xstats[i].value; | |
0a0f39df | 2684 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2685 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 2686 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2687 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2688 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2689 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2690 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2691 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2692 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2693 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2694 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2695 | stats->rx_jabber_errors = xstats[i].value; |
2696 | } | |
2697 | } | |
2698 | } | |
2699 | ||
8a9562d2 PS |
2700 | static int |
2701 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2702 | { | |
2703 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2704 | struct rte_eth_stats rte_stats; | |
2705 | bool gg; | |
2706 | ||
2707 | netdev_dpdk_get_carrier(netdev, &gg); | |
2708 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2709 | |
0a0f39df CL |
2710 | struct rte_eth_xstat *rte_xstats = NULL; |
2711 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2712 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2713 | |
2714 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
fa9f4eeb IM |
2715 | VLOG_ERR("Can't get ETH statistics for port: "DPDK_PORT_ID_FMT, |
2716 | dev->port_id); | |
f9256822 | 2717 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2718 | return EPROTO; |
2719 | } | |
2720 | ||
0a0f39df CL |
2721 | /* Get length of statistics */ |
2722 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2723 | if (rte_xstats_len < 0) { | |
fa9f4eeb IM |
2724 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
2725 | dev->port_id); | |
0a0f39df CL |
2726 | goto out; |
2727 | } | |
2728 | /* Reserve memory for xstats names and values */ | |
2729 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2730 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2731 | ||
2732 | /* Retreive xstats names */ | |
2733 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2734 | rte_xstats_names, | |
2735 | rte_xstats_len); | |
2736 | if (rte_xstats_new_len != rte_xstats_len) { | |
fa9f4eeb IM |
2737 | VLOG_WARN("Cannot get XSTATS names for port: "DPDK_PORT_ID_FMT, |
2738 | dev->port_id); | |
0a0f39df CL |
2739 | goto out; |
2740 | } | |
2741 | /* Retreive xstats values */ | |
2742 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2743 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2744 | rte_xstats_len); | |
2745 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2746 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2747 | rte_xstats_len); | |
d6e3feb5 | 2748 | } else { |
fa9f4eeb IM |
2749 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
2750 | dev->port_id); | |
d6e3feb5 | 2751 | } |
8a9562d2 | 2752 | |
0a0f39df CL |
2753 | out: |
2754 | free(rte_xstats); | |
2755 | free(rte_xstats_names); | |
2756 | ||
2f9dd77f PS |
2757 | stats->rx_packets = rte_stats.ipackets; |
2758 | stats->tx_packets = rte_stats.opackets; | |
2759 | stats->rx_bytes = rte_stats.ibytes; | |
2760 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2761 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2762 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2763 | |
45d947c4 | 2764 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2765 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2766 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2767 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2768 | |
2769 | /* These are the available DPDK counters for packets not received due to | |
2770 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2771 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2772 | stats->rx_missed_errors = rte_stats.imissed; |
2773 | ||
8a9562d2 PS |
2774 | ovs_mutex_unlock(&dev->mutex); |
2775 | ||
2776 | return 0; | |
2777 | } | |
2778 | ||
971f4b39 MW |
2779 | static int |
2780 | netdev_dpdk_get_custom_stats(const struct netdev *netdev, | |
2781 | struct netdev_custom_stats *custom_stats) | |
2782 | { | |
2783 | ||
2784 | uint32_t i; | |
2785 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2786 | int rte_xstats_ret; | |
2787 | ||
2788 | ovs_mutex_lock(&dev->mutex); | |
2789 | ||
2790 | if (netdev_dpdk_configure_xstats(dev)) { | |
2791 | uint64_t *values = xcalloc(dev->rte_xstats_ids_size, | |
2792 | sizeof(uint64_t)); | |
2793 | ||
2794 | rte_xstats_ret = | |
2795 | rte_eth_xstats_get_by_id(dev->port_id, dev->rte_xstats_ids, | |
2796 | values, dev->rte_xstats_ids_size); | |
2797 | ||
2798 | if (rte_xstats_ret > 0 && | |
2799 | rte_xstats_ret <= dev->rte_xstats_ids_size) { | |
2800 | ||
2801 | custom_stats->size = rte_xstats_ret; | |
2802 | custom_stats->counters = | |
2803 | (struct netdev_custom_counter *) xcalloc(rte_xstats_ret, | |
2804 | sizeof(struct netdev_custom_counter)); | |
2805 | ||
2806 | for (i = 0; i < rte_xstats_ret; i++) { | |
2807 | ovs_strlcpy(custom_stats->counters[i].name, | |
2808 | netdev_dpdk_get_xstat_name(dev, | |
2809 | dev->rte_xstats_ids[i]), | |
2810 | NETDEV_CUSTOM_STATS_NAME_SIZE); | |
2811 | custom_stats->counters[i].value = values[i]; | |
2812 | } | |
2813 | } else { | |
fa9f4eeb | 2814 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
971f4b39 MW |
2815 | dev->port_id); |
2816 | custom_stats->counters = NULL; | |
2817 | custom_stats->size = 0; | |
2818 | /* Let's clear statistics cache, so it will be | |
2819 | * reconfigured */ | |
2820 | netdev_dpdk_clear_xstats(dev); | |
2821 | } | |
526259f2 IM |
2822 | |
2823 | free(values); | |
971f4b39 MW |
2824 | } |
2825 | ||
2826 | ovs_mutex_unlock(&dev->mutex); | |
2827 | ||
2828 | return 0; | |
2829 | } | |
2830 | ||
c161357d KT |
2831 | static int |
2832 | netdev_dpdk_vhost_get_custom_stats(const struct netdev *netdev, | |
2833 | struct netdev_custom_stats *custom_stats) | |
2834 | { | |
2835 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2836 | ||
2837 | ovs_mutex_lock(&dev->mutex); | |
2838 | ||
2839 | custom_stats->size = VHOST_CUSTOM_STATS_SIZE; | |
2840 | custom_stats->counters = xcalloc(custom_stats->size, | |
2841 | sizeof *custom_stats->counters); | |
2842 | ovs_strlcpy(custom_stats->counters[0].name, VHOST_STAT_TX_RETRIES, | |
2843 | NETDEV_CUSTOM_STATS_NAME_SIZE); | |
2844 | ||
2845 | rte_spinlock_lock(&dev->stats_lock); | |
2846 | custom_stats->counters[0].value = dev->tx_retries; | |
2847 | rte_spinlock_unlock(&dev->stats_lock); | |
2848 | ||
2849 | ovs_mutex_unlock(&dev->mutex); | |
2850 | ||
2851 | return 0; | |
2852 | } | |
2853 | ||
8a9562d2 | 2854 | static int |
d46285a2 | 2855 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2856 | enum netdev_features *current, |
ca3d4f55 BX |
2857 | enum netdev_features *advertised, |
2858 | enum netdev_features *supported, | |
2859 | enum netdev_features *peer) | |
8a9562d2 | 2860 | { |
d46285a2 | 2861 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 2862 | struct rte_eth_link link; |
dfcb5b8a | 2863 | uint32_t feature = 0; |
8a9562d2 PS |
2864 | |
2865 | ovs_mutex_lock(&dev->mutex); | |
2866 | link = dev->link; | |
2867 | ovs_mutex_unlock(&dev->mutex); | |
2868 | ||
dfcb5b8a IS |
2869 | /* Match against OpenFlow defined link speed values. */ |
2870 | if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
2871 | switch (link.link_speed) { | |
2872 | case ETH_SPEED_NUM_10M: | |
2873 | feature |= NETDEV_F_10MB_FD; | |
2874 | break; | |
2875 | case ETH_SPEED_NUM_100M: | |
2876 | feature |= NETDEV_F_100MB_FD; | |
2877 | break; | |
2878 | case ETH_SPEED_NUM_1G: | |
2879 | feature |= NETDEV_F_1GB_FD; | |
2880 | break; | |
2881 | case ETH_SPEED_NUM_10G: | |
2882 | feature |= NETDEV_F_10GB_FD; | |
2883 | break; | |
2884 | case ETH_SPEED_NUM_40G: | |
2885 | feature |= NETDEV_F_40GB_FD; | |
2886 | break; | |
2887 | case ETH_SPEED_NUM_100G: | |
2888 | feature |= NETDEV_F_100GB_FD; | |
2889 | break; | |
2890 | default: | |
2891 | feature |= NETDEV_F_OTHER; | |
8a9562d2 | 2892 | } |
dfcb5b8a IS |
2893 | } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2894 | switch (link.link_speed) { | |
2895 | case ETH_SPEED_NUM_10M: | |
2896 | feature |= NETDEV_F_10MB_HD; | |
2897 | break; | |
2898 | case ETH_SPEED_NUM_100M: | |
2899 | feature |= NETDEV_F_100MB_HD; | |
2900 | break; | |
2901 | case ETH_SPEED_NUM_1G: | |
2902 | feature |= NETDEV_F_1GB_HD; | |
2903 | break; | |
2904 | default: | |
2905 | feature |= NETDEV_F_OTHER; | |
74cd69a4 | 2906 | } |
8a9562d2 PS |
2907 | } |
2908 | ||
362ca396 | 2909 | if (link.link_autoneg) { |
dfcb5b8a | 2910 | feature |= NETDEV_F_AUTONEG; |
362ca396 | 2911 | } |
2912 | ||
dfcb5b8a | 2913 | *current = feature; |
ca3d4f55 BX |
2914 | *advertised = *supported = *peer = 0; |
2915 | ||
8a9562d2 PS |
2916 | return 0; |
2917 | } | |
2918 | ||
9509913a IS |
2919 | static struct ingress_policer * |
2920 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2921 | { | |
2922 | struct ingress_policer *policer = NULL; | |
2923 | uint64_t rate_bytes; | |
2924 | uint64_t burst_bytes; | |
2925 | int err = 0; | |
2926 | ||
2927 | policer = xmalloc(sizeof *policer); | |
2928 | rte_spinlock_init(&policer->policer_lock); | |
2929 | ||
2930 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
602c8668 LR |
2931 | rate_bytes = rate * 1000ULL / 8; |
2932 | burst_bytes = burst * 1000ULL / 8; | |
9509913a IS |
2933 | |
2934 | policer->app_srtcm_params.cir = rate_bytes; | |
2935 | policer->app_srtcm_params.cbs = burst_bytes; | |
2936 | policer->app_srtcm_params.ebs = 0; | |
03f3f9c0 OM |
2937 | err = rte_meter_srtcm_profile_config(&policer->in_prof, |
2938 | &policer->app_srtcm_params); | |
2939 | if (!err) { | |
2940 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2941 | &policer->in_prof); | |
2942 | } | |
58be5c0e | 2943 | if (err) { |
9509913a | 2944 | VLOG_ERR("Could not create rte meter for ingress policer"); |
4c47ddde | 2945 | free(policer); |
9509913a IS |
2946 | return NULL; |
2947 | } | |
2948 | ||
2949 | return policer; | |
2950 | } | |
2951 | ||
2952 | static int | |
2953 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2954 | uint32_t policer_burst) | |
2955 | { | |
2956 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2957 | struct ingress_policer *policer; | |
2958 | ||
2959 | /* Force to 0 if no rate specified, | |
2960 | * default to 8000 kbits if burst is 0, | |
2961 | * else stick with user-specified value. | |
2962 | */ | |
2963 | policer_burst = (!policer_rate ? 0 | |
2964 | : !policer_burst ? 8000 | |
2965 | : policer_burst); | |
2966 | ||
2967 | ovs_mutex_lock(&dev->mutex); | |
2968 | ||
2969 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2970 | &dev->ingress_policer); | |
2971 | ||
2972 | if (dev->policer_rate == policer_rate && | |
2973 | dev->policer_burst == policer_burst) { | |
2974 | /* Assume that settings haven't changed since we last set them. */ | |
2975 | ovs_mutex_unlock(&dev->mutex); | |
2976 | return 0; | |
2977 | } | |
2978 | ||
2979 | /* Destroy any existing ingress policer for the device if one exists */ | |
2980 | if (policer) { | |
2981 | ovsrcu_postpone(free, policer); | |
2982 | } | |
2983 | ||
2984 | if (policer_rate != 0) { | |
2985 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2986 | } else { | |
2987 | policer = NULL; | |
2988 | } | |
2989 | ovsrcu_set(&dev->ingress_policer, policer); | |
2990 | dev->policer_rate = policer_rate; | |
2991 | dev->policer_burst = policer_burst; | |
2992 | ovs_mutex_unlock(&dev->mutex); | |
2993 | ||
2994 | return 0; | |
2995 | } | |
2996 | ||
8a9562d2 PS |
2997 | static int |
2998 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2999 | { | |
3000 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
3001 | |
3002 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
3003 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
3004 | * postive integer to meet RFC 2863 recommendations. | |
3005 | */ | |
3006 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
3007 | ovs_mutex_unlock(&dev->mutex); |
3008 | ||
3009 | return ifindex; | |
3010 | } | |
3011 | ||
3012 | static int | |
d46285a2 | 3013 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 3014 | { |
d46285a2 | 3015 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
3016 | |
3017 | ovs_mutex_lock(&dev->mutex); | |
3018 | check_link_status(dev); | |
3019 | *carrier = dev->link.link_status; | |
58397e6c KT |
3020 | |
3021 | ovs_mutex_unlock(&dev->mutex); | |
3022 | ||
3023 | return 0; | |
3024 | } | |
3025 | ||
3026 | static int | |
d46285a2 | 3027 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 3028 | { |
d46285a2 | 3029 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
3030 | |
3031 | ovs_mutex_lock(&dev->mutex); | |
3032 | ||
0a0f39df | 3033 | if (is_vhost_running(dev)) { |
58397e6c KT |
3034 | *carrier = 1; |
3035 | } else { | |
3036 | *carrier = 0; | |
3037 | } | |
3038 | ||
8a9562d2 PS |
3039 | ovs_mutex_unlock(&dev->mutex); |
3040 | ||
3041 | return 0; | |
3042 | } | |
3043 | ||
3044 | static long long int | |
d46285a2 | 3045 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 3046 | { |
d46285a2 | 3047 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
3048 | long long int carrier_resets; |
3049 | ||
3050 | ovs_mutex_lock(&dev->mutex); | |
3051 | carrier_resets = dev->link_reset_cnt; | |
3052 | ovs_mutex_unlock(&dev->mutex); | |
3053 | ||
3054 | return carrier_resets; | |
3055 | } | |
3056 | ||
3057 | static int | |
d46285a2 | 3058 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
3059 | long long int interval OVS_UNUSED) |
3060 | { | |
ee32150e | 3061 | return EOPNOTSUPP; |
8a9562d2 PS |
3062 | } |
3063 | ||
3064 | static int | |
3065 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
3066 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
3067 | enum netdev_flags *old_flagsp) |
3068 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 3069 | { |
8a9562d2 PS |
3070 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
3071 | return EINVAL; | |
3072 | } | |
3073 | ||
3074 | *old_flagsp = dev->flags; | |
3075 | dev->flags |= on; | |
3076 | dev->flags &= ~off; | |
3077 | ||
3078 | if (dev->flags == *old_flagsp) { | |
3079 | return 0; | |
3080 | } | |
3081 | ||
58397e6c | 3082 | if (dev->type == DPDK_DEV_ETH) { |
2d37de73 EC |
3083 | |
3084 | if ((dev->flags ^ *old_flagsp) & NETDEV_UP) { | |
3085 | int err; | |
3086 | ||
3087 | if (dev->flags & NETDEV_UP) { | |
3088 | err = rte_eth_dev_set_link_up(dev->port_id); | |
3089 | } else { | |
3090 | err = rte_eth_dev_set_link_down(dev->port_id); | |
3091 | } | |
3092 | if (err == -ENOTSUP) { | |
3093 | VLOG_INFO("Interface %s does not support link state " | |
3094 | "configuration", netdev_get_name(&dev->up)); | |
3095 | } else if (err < 0) { | |
3096 | VLOG_ERR("Interface %s link change error: %s", | |
3097 | netdev_get_name(&dev->up), rte_strerror(-err)); | |
3098 | dev->flags = *old_flagsp; | |
3099 | return -err; | |
3100 | } | |
3101 | } | |
3102 | ||
58397e6c KT |
3103 | if (dev->flags & NETDEV_PROMISC) { |
3104 | rte_eth_promiscuous_enable(dev->port_id); | |
3105 | } | |
8a9562d2 | 3106 | |
314fb5ad | 3107 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
3108 | } else { |
3109 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
3110 | * running then change netdev's change_seq to trigger link state | |
3111 | * update. */ | |
e543851d ZB |
3112 | |
3113 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 3114 | && is_vhost_running(dev)) { |
e543851d ZB |
3115 | netdev_change_seq_changed(&dev->up); |
3116 | ||
3117 | /* Clear statistics if device is getting up. */ | |
3118 | if (NETDEV_UP & on) { | |
3119 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 3120 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
3121 | rte_spinlock_unlock(&dev->stats_lock); |
3122 | } | |
3123 | } | |
8a9562d2 PS |
3124 | } |
3125 | ||
3126 | return 0; | |
3127 | } | |
3128 | ||
3129 | static int | |
d46285a2 | 3130 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
3131 | enum netdev_flags off, enum netdev_flags on, |
3132 | enum netdev_flags *old_flagsp) | |
3133 | { | |
d46285a2 | 3134 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
3135 | int error; |
3136 | ||
d46285a2 DDP |
3137 | ovs_mutex_lock(&dev->mutex); |
3138 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
3139 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
3140 | |
3141 | return error; | |
3142 | } | |
3143 | ||
b2e8b12f FL |
3144 | static int |
3145 | netdev_dpdk_vhost_user_get_status(const struct netdev *netdev, | |
3146 | struct smap *args) | |
3147 | { | |
3148 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3149 | ||
3150 | ovs_mutex_lock(&dev->mutex); | |
3151 | ||
3152 | bool client_mode = dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT; | |
3153 | smap_add_format(args, "mode", "%s", client_mode ? "client" : "server"); | |
3154 | ||
3155 | int vid = netdev_dpdk_get_vid(dev); | |
3156 | if (vid < 0) { | |
3157 | smap_add_format(args, "status", "disconnected"); | |
3158 | ovs_mutex_unlock(&dev->mutex); | |
3159 | return 0; | |
3160 | } else { | |
3161 | smap_add_format(args, "status", "connected"); | |
3162 | } | |
3163 | ||
3164 | char socket_name[PATH_MAX]; | |
3165 | if (!rte_vhost_get_ifname(vid, socket_name, PATH_MAX)) { | |
3166 | smap_add_format(args, "socket", "%s", socket_name); | |
3167 | } | |
3168 | ||
3169 | uint64_t features; | |
3170 | if (!rte_vhost_get_negotiated_features(vid, &features)) { | |
3171 | smap_add_format(args, "features", "0x%016"PRIx64, features); | |
3172 | } | |
3173 | ||
3174 | uint16_t mtu; | |
3175 | if (!rte_vhost_get_mtu(vid, &mtu)) { | |
3176 | smap_add_format(args, "mtu", "%d", mtu); | |
3177 | } | |
3178 | ||
3179 | int numa = rte_vhost_get_numa_node(vid); | |
3180 | if (numa >= 0) { | |
3181 | smap_add_format(args, "numa", "%d", numa); | |
3182 | } | |
3183 | ||
3184 | uint16_t vring_num = rte_vhost_get_vring_num(vid); | |
3185 | if (vring_num) { | |
3186 | smap_add_format(args, "num_of_vrings", "%d", vring_num); | |
3187 | } | |
3188 | ||
3189 | for (int i = 0; i < vring_num; i++) { | |
3190 | struct rte_vhost_vring vring; | |
b2e8b12f FL |
3191 | |
3192 | rte_vhost_get_vhost_vring(vid, i, &vring); | |
b9a3183d AC |
3193 | smap_add_nocopy(args, xasprintf("vring_%d_size", i), |
3194 | xasprintf("%d", vring.size)); | |
b2e8b12f FL |
3195 | } |
3196 | ||
3197 | ovs_mutex_unlock(&dev->mutex); | |
3198 | return 0; | |
3199 | } | |
3200 | ||
31154f95 IS |
3201 | /* |
3202 | * Convert a given uint32_t link speed defined in DPDK to a string | |
3203 | * equivalent. | |
3204 | */ | |
3205 | static const char * | |
3206 | netdev_dpdk_link_speed_to_str__(uint32_t link_speed) | |
3207 | { | |
3208 | switch (link_speed) { | |
3209 | case ETH_SPEED_NUM_10M: return "10Mbps"; | |
3210 | case ETH_SPEED_NUM_100M: return "100Mbps"; | |
3211 | case ETH_SPEED_NUM_1G: return "1Gbps"; | |
3212 | case ETH_SPEED_NUM_2_5G: return "2.5Gbps"; | |
3213 | case ETH_SPEED_NUM_5G: return "5Gbps"; | |
3214 | case ETH_SPEED_NUM_10G: return "10Gbps"; | |
3215 | case ETH_SPEED_NUM_20G: return "20Gbps"; | |
3216 | case ETH_SPEED_NUM_25G: return "25Gbps"; | |
3217 | case ETH_SPEED_NUM_40G: return "40Gbps"; | |
3218 | case ETH_SPEED_NUM_50G: return "50Gbps"; | |
3219 | case ETH_SPEED_NUM_56G: return "56Gbps"; | |
3220 | case ETH_SPEED_NUM_100G: return "100Gbps"; | |
3221 | default: return "Not Defined"; | |
3222 | } | |
3223 | } | |
3224 | ||
8a9562d2 | 3225 | static int |
d46285a2 | 3226 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 3227 | { |
d46285a2 | 3228 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 3229 | struct rte_eth_dev_info dev_info; |
31154f95 | 3230 | uint32_t link_speed; |
8a9562d2 | 3231 | |
7cd1261d | 3232 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 3233 | return ENODEV; |
7cd1261d | 3234 | } |
8a9562d2 | 3235 | |
03f3f9c0 | 3236 | ovs_mutex_lock(&dpdk_mutex); |
8a9562d2 PS |
3237 | ovs_mutex_lock(&dev->mutex); |
3238 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
31154f95 | 3239 | link_speed = dev->link.link_speed; |
8a9562d2 | 3240 | ovs_mutex_unlock(&dev->mutex); |
03f3f9c0 OM |
3241 | const struct rte_bus *bus; |
3242 | const struct rte_pci_device *pci_dev; | |
3243 | uint16_t vendor_id = PCI_ANY_ID; | |
3244 | uint16_t device_id = PCI_ANY_ID; | |
3245 | bus = rte_bus_find_by_device(dev_info.device); | |
3246 | if (bus && !strcmp(bus->name, "pci")) { | |
3247 | pci_dev = RTE_DEV_TO_PCI(dev_info.device); | |
3248 | if (pci_dev) { | |
3249 | vendor_id = pci_dev->id.vendor_id; | |
3250 | device_id = pci_dev->id.device_id; | |
3251 | } | |
3252 | } | |
3253 | ovs_mutex_unlock(&dpdk_mutex); | |
8a9562d2 | 3254 | |
fa9f4eeb | 3255 | smap_add_format(args, "port_no", DPDK_PORT_ID_FMT, dev->port_id); |
58be5c0e MK |
3256 | smap_add_format(args, "numa_id", "%d", |
3257 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
3258 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
3259 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 3260 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
3261 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
3262 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
3263 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
3264 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
3265 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
3266 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
3267 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
3268 | ||
3eb8d4fa MW |
3269 | /* Querying the DPDK library for iftype may be done in future, pending |
3270 | * support; cf. RFC 3635 Section 3.2.4. */ | |
3271 | enum { IF_TYPE_ETHERNETCSMACD = 6 }; | |
3272 | ||
3273 | smap_add_format(args, "if_type", "%"PRIu32, IF_TYPE_ETHERNETCSMACD); | |
3274 | smap_add_format(args, "if_descr", "%s %s", rte_version(), | |
3275 | dev_info.driver_name); | |
03f3f9c0 OM |
3276 | smap_add_format(args, "pci-vendor_id", "0x%x", vendor_id); |
3277 | smap_add_format(args, "pci-device_id", "0x%x", device_id); | |
8a9562d2 | 3278 | |
31154f95 IS |
3279 | /* Not all link speeds are defined in the OpenFlow specs e.g. 25 Gbps. |
3280 | * In that case the speed will not be reported as part of the usual | |
3281 | * call to get_features(). Get the link speed of the device and add it | |
3282 | * to the device status in an easy to read string format. | |
3283 | */ | |
3284 | smap_add(args, "link_speed", | |
3285 | netdev_dpdk_link_speed_to_str__(link_speed)); | |
3286 | ||
8a9562d2 PS |
3287 | return 0; |
3288 | } | |
3289 | ||
3290 | static void | |
3291 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
3292 | OVS_REQUIRES(dev->mutex) | |
3293 | { | |
3294 | enum netdev_flags old_flags; | |
3295 | ||
3296 | if (admin_state) { | |
3297 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
3298 | } else { | |
3299 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
3300 | } | |
3301 | } | |
3302 | ||
3303 | static void | |
3304 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
3305 | const char *argv[], void *aux OVS_UNUSED) | |
3306 | { | |
3307 | bool up; | |
3308 | ||
3309 | if (!strcasecmp(argv[argc - 1], "up")) { | |
3310 | up = true; | |
3311 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
3312 | up = false; | |
3313 | } else { | |
3314 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
3315 | return; | |
3316 | } | |
3317 | ||
3318 | if (argc > 2) { | |
3319 | struct netdev *netdev = netdev_from_name(argv[1]); | |
3d0d5ab1 | 3320 | |
8a9562d2 | 3321 | if (netdev && is_dpdk_class(netdev->netdev_class)) { |
3d0d5ab1 | 3322 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 3323 | |
3d0d5ab1 IM |
3324 | ovs_mutex_lock(&dev->mutex); |
3325 | netdev_dpdk_set_admin_state__(dev, up); | |
3326 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
3327 | |
3328 | netdev_close(netdev); | |
3329 | } else { | |
3330 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
3331 | netdev_close(netdev); | |
3332 | return; | |
3333 | } | |
3334 | } else { | |
3d0d5ab1 | 3335 | struct netdev_dpdk *dev; |
8a9562d2 PS |
3336 | |
3337 | ovs_mutex_lock(&dpdk_mutex); | |
3d0d5ab1 IM |
3338 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
3339 | ovs_mutex_lock(&dev->mutex); | |
3340 | netdev_dpdk_set_admin_state__(dev, up); | |
3341 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
3342 | } |
3343 | ovs_mutex_unlock(&dpdk_mutex); | |
3344 | } | |
3345 | unixctl_command_reply(conn, "OK"); | |
3346 | } | |
3347 | ||
0ee821c2 DB |
3348 | static void |
3349 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
3350 | const char *argv[], void *aux OVS_UNUSED) | |
3351 | { | |
0ee821c2 | 3352 | char *response; |
7ee94cba | 3353 | dpdk_port_t port_id; |
0ee821c2 | 3354 | struct netdev_dpdk *dev; |
40e940e4 OM |
3355 | struct rte_device *rte_dev; |
3356 | struct ds used_interfaces = DS_EMPTY_INITIALIZER; | |
3357 | bool used = false; | |
0ee821c2 DB |
3358 | |
3359 | ovs_mutex_lock(&dpdk_mutex); | |
3360 | ||
40e940e4 OM |
3361 | port_id = netdev_dpdk_get_port_by_devargs(argv[1]); |
3362 | if (!rte_eth_dev_is_valid_port(port_id)) { | |
0ee821c2 DB |
3363 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); |
3364 | goto error; | |
3365 | } | |
3366 | ||
40e940e4 OM |
3367 | rte_dev = rte_eth_devices[port_id].device; |
3368 | ds_put_format(&used_interfaces, | |
3369 | "Device '%s' is being used by the following interfaces:", | |
3370 | argv[1]); | |
3371 | ||
3372 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
3373 | /* FIXME: avoid direct access to DPDK array rte_eth_devices. */ | |
3374 | if (rte_eth_devices[dev->port_id].device == rte_dev | |
3375 | && rte_eth_devices[dev->port_id].state != RTE_ETH_DEV_UNUSED) { | |
3376 | used = true; | |
3377 | ds_put_format(&used_interfaces, " %s", | |
3378 | netdev_get_name(&dev->up)); | |
3379 | } | |
3380 | } | |
3381 | ||
3382 | if (used) { | |
3383 | ds_put_cstr(&used_interfaces, ". Remove them before detaching."); | |
3384 | response = ds_steal_cstr(&used_interfaces); | |
3385 | ds_destroy(&used_interfaces); | |
0ee821c2 DB |
3386 | goto error; |
3387 | } | |
40e940e4 | 3388 | ds_destroy(&used_interfaces); |
0ee821c2 DB |
3389 | |
3390 | rte_eth_dev_close(port_id); | |
40e940e4 | 3391 | if (rte_dev_remove(rte_dev) < 0) { |
0ee821c2 DB |
3392 | response = xasprintf("Device '%s' can not be detached", argv[1]); |
3393 | goto error; | |
3394 | } | |
3395 | ||
40e940e4 OM |
3396 | response = xasprintf("All devices shared with device '%s' " |
3397 | "have been detached", argv[1]); | |
0ee821c2 DB |
3398 | |
3399 | ovs_mutex_unlock(&dpdk_mutex); | |
3400 | unixctl_command_reply(conn, response); | |
3401 | free(response); | |
3402 | return; | |
3403 | ||
3404 | error: | |
3405 | ovs_mutex_unlock(&dpdk_mutex); | |
3406 | unixctl_command_reply_error(conn, response); | |
3407 | free(response); | |
3408 | } | |
3409 | ||
be481733 IM |
3410 | static void |
3411 | netdev_dpdk_get_mempool_info(struct unixctl_conn *conn, | |
3412 | int argc, const char *argv[], | |
3413 | void *aux OVS_UNUSED) | |
3414 | { | |
3415 | size_t size; | |
3416 | FILE *stream; | |
3417 | char *response = NULL; | |
3418 | struct netdev *netdev = NULL; | |
3419 | ||
3420 | if (argc == 2) { | |
3421 | netdev = netdev_from_name(argv[1]); | |
3422 | if (!netdev || !is_dpdk_class(netdev->netdev_class)) { | |
3423 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
3424 | goto out; | |
3425 | } | |
3426 | } | |
3427 | ||
3428 | stream = open_memstream(&response, &size); | |
3429 | if (!stream) { | |
3430 | response = xasprintf("Unable to open memstream: %s.", | |
3431 | ovs_strerror(errno)); | |
3432 | unixctl_command_reply_error(conn, response); | |
3433 | goto out; | |
3434 | } | |
3435 | ||
3436 | if (netdev) { | |
3437 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3438 | ||
3439 | ovs_mutex_lock(&dev->mutex); | |
3440 | ovs_mutex_lock(&dpdk_mp_mutex); | |
3441 | ||
43307ad0 | 3442 | rte_mempool_dump(stream, dev->dpdk_mp->mp); |
be481733 IM |
3443 | |
3444 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
3445 | ovs_mutex_unlock(&dev->mutex); | |
3446 | } else { | |
3447 | ovs_mutex_lock(&dpdk_mp_mutex); | |
3448 | rte_mempool_list_dump(stream); | |
3449 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
3450 | } | |
3451 | ||
3452 | fclose(stream); | |
3453 | ||
3454 | unixctl_command_reply(conn, response); | |
3455 | out: | |
3456 | free(response); | |
3457 | netdev_close(netdev); | |
3458 | } | |
3459 | ||
58397e6c KT |
3460 | /* |
3461 | * Set virtqueue flags so that we do not receive interrupts. | |
3462 | */ | |
3463 | static void | |
0a0f39df | 3464 | set_irq_status(int vid) |
58397e6c | 3465 | { |
4573fbd3 | 3466 | uint32_t i; |
4573fbd3 | 3467 | |
f3e7ec25 MW |
3468 | for (i = 0; i < rte_vhost_get_vring_num(vid); i++) { |
3469 | rte_vhost_enable_guest_notification(vid, i, 0); | |
4573fbd3 FL |
3470 | } |
3471 | } | |
3472 | ||
585a5bea IM |
3473 | /* |
3474 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 3475 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
3476 | */ |
3477 | static void | |
d46285a2 DDP |
3478 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
3479 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
3480 | { |
3481 | int *enabled_queues, n_enabled = 0; | |
81acebda | 3482 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 3483 | |
eff23640 | 3484 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
3485 | |
3486 | for (i = 0; i < total_txqs; i++) { | |
3487 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 3488 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
3489 | enabled_queues[n_enabled++] = i; |
3490 | } | |
3491 | } | |
3492 | ||
3493 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 3494 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
3495 | n_enabled = 1; |
3496 | } | |
3497 | ||
3498 | k = 0; | |
3499 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
3500 | if (dev->tx_q[i].map != i) { |
3501 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
3502 | k = (k + 1) % n_enabled; |
3503 | } | |
3504 | } | |
3505 | ||
170ef726 IM |
3506 | if (VLOG_IS_DBG_ENABLED()) { |
3507 | struct ds mapping = DS_EMPTY_INITIALIZER; | |
3508 | ||
3509 | ds_put_format(&mapping, "TX queue mapping for port '%s':\n", | |
3510 | netdev_get_name(&dev->up)); | |
3511 | for (i = 0; i < total_txqs; i++) { | |
3512 | ds_put_format(&mapping, "%2d --> %2d\n", i, dev->tx_q[i].map); | |
3513 | } | |
3514 | ||
3515 | VLOG_DBG("%s", ds_cstr(&mapping)); | |
3516 | ds_destroy(&mapping); | |
585a5bea IM |
3517 | } |
3518 | ||
eff23640 | 3519 | free(enabled_queues); |
585a5bea | 3520 | } |
4573fbd3 | 3521 | |
58397e6c KT |
3522 | /* |
3523 | * A new virtio-net device is added to a vhost port. | |
3524 | */ | |
3525 | static int | |
0a0f39df | 3526 | new_device(int vid) |
58397e6c | 3527 | { |
d46285a2 | 3528 | struct netdev_dpdk *dev; |
58397e6c | 3529 | bool exists = false; |
db8f13b0 | 3530 | int newnode = 0; |
0a0f39df CL |
3531 | char ifname[IF_NAME_SZ]; |
3532 | ||
58be5c0e | 3533 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
3534 | |
3535 | ovs_mutex_lock(&dpdk_mutex); | |
3536 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 3537 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 3538 | ovs_mutex_lock(&dev->mutex); |
bb9d2623 IM |
3539 | if (nullable_string_is_equal(ifname, dev->vhost_id)) { |
3540 | uint32_t qp_num = rte_vhost_get_vring_num(vid) / VIRTIO_QNUM; | |
db8f13b0 CL |
3541 | |
3542 | /* Get NUMA information */ | |
0a0f39df CL |
3543 | newnode = rte_vhost_get_numa_node(vid); |
3544 | if (newnode == -1) { | |
5b9bf9e0 | 3545 | #ifdef VHOST_NUMA |
db8f13b0 | 3546 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 3547 | ifname); |
5b9bf9e0 | 3548 | #endif |
db8f13b0 | 3549 | newnode = dev->socket_id; |
db8f13b0 CL |
3550 | } |
3551 | ||
7235cd20 DM |
3552 | if (dev->requested_n_txq < qp_num |
3553 | || dev->requested_n_rxq < qp_num | |
7f5f2bd0 IM |
3554 | || dev->requested_socket_id != newnode) { |
3555 | dev->requested_socket_id = newnode; | |
3556 | dev->requested_n_rxq = qp_num; | |
3557 | dev->requested_n_txq = qp_num; | |
3558 | netdev_request_reconfigure(&dev->up); | |
3559 | } else { | |
3560 | /* Reconfiguration not required. */ | |
3561 | dev->vhost_reconfigured = true; | |
3562 | } | |
81acebda | 3563 | |
0a0f39df | 3564 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
3565 | exists = true; |
3566 | ||
58397e6c | 3567 | /* Disable notifications. */ |
0a0f39df | 3568 | set_irq_status(vid); |
e543851d | 3569 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 3570 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
3571 | break; |
3572 | } | |
c1ff66ac | 3573 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
3574 | } |
3575 | ovs_mutex_unlock(&dpdk_mutex); | |
3576 | ||
3577 | if (!exists) { | |
0a0f39df | 3578 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
3579 | |
3580 | return -1; | |
3581 | } | |
3582 | ||
0a0f39df CL |
3583 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
3584 | ifname, newnode); | |
3585 | ||
58397e6c KT |
3586 | return 0; |
3587 | } | |
3588 | ||
f3ea2ad2 IM |
3589 | /* Clears mapping for all available queues of vhost interface. */ |
3590 | static void | |
3591 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
3592 | OVS_REQUIRES(dev->mutex) | |
3593 | { | |
3594 | int i; | |
3595 | ||
81acebda | 3596 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
3597 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
3598 | } | |
3599 | } | |
3600 | ||
58397e6c KT |
3601 | /* |
3602 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
3603 | * flag to stop any more packets from being sent or received to/from a VM and | |
3604 | * ensure all currently queued packets have been sent/received before removing | |
3605 | * the device. | |
3606 | */ | |
3607 | static void | |
0a0f39df | 3608 | destroy_device(int vid) |
58397e6c | 3609 | { |
d46285a2 | 3610 | struct netdev_dpdk *dev; |
afee281f | 3611 | bool exists = false; |
0a0f39df CL |
3612 | char ifname[IF_NAME_SZ]; |
3613 | ||
58be5c0e | 3614 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
3615 | |
3616 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 3617 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 3618 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 3619 | |
d46285a2 | 3620 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
3621 | dev->vhost_reconfigured = false; |
3622 | ovsrcu_index_set(&dev->vid, -1); | |
35c91567 DM |
3623 | memset(dev->vhost_rxq_enabled, 0, |
3624 | dev->up.n_rxq * sizeof *dev->vhost_rxq_enabled); | |
d46285a2 | 3625 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 3626 | |
e543851d | 3627 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 3628 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 3629 | exists = true; |
afee281f | 3630 | break; |
58397e6c KT |
3631 | } |
3632 | } | |
afee281f | 3633 | |
58397e6c KT |
3634 | ovs_mutex_unlock(&dpdk_mutex); |
3635 | ||
0a0f39df | 3636 | if (exists) { |
afee281f KT |
3637 | /* |
3638 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
3639 | * to NULL, before returning. | |
3640 | */ | |
3641 | ovsrcu_synchronize(); | |
3642 | /* | |
3643 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
3644 | * put thread back into quiescent state before returning. | |
3645 | */ | |
3646 | ovsrcu_quiesce_start(); | |
0a0f39df | 3647 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 3648 | } else { |
0a0f39df | 3649 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 3650 | } |
58397e6c KT |
3651 | } |
3652 | ||
585a5bea | 3653 | static int |
0a0f39df | 3654 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 3655 | { |
d46285a2 | 3656 | struct netdev_dpdk *dev; |
585a5bea IM |
3657 | bool exists = false; |
3658 | int qid = queue_id / VIRTIO_QNUM; | |
35c91567 | 3659 | bool is_rx = (queue_id % VIRTIO_QNUM) == VIRTIO_TXQ; |
0a0f39df CL |
3660 | char ifname[IF_NAME_SZ]; |
3661 | ||
58be5c0e | 3662 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea | 3663 | |
585a5bea | 3664 | ovs_mutex_lock(&dpdk_mutex); |
d46285a2 | 3665 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 3666 | ovs_mutex_lock(&dev->mutex); |
bb9d2623 | 3667 | if (nullable_string_is_equal(ifname, dev->vhost_id)) { |
35c91567 DM |
3668 | if (is_rx) { |
3669 | bool old_state = dev->vhost_rxq_enabled[qid]; | |
3670 | ||
3671 | dev->vhost_rxq_enabled[qid] = enable != 0; | |
3672 | if (old_state != dev->vhost_rxq_enabled[qid]) { | |
3673 | netdev_change_seq_changed(&dev->up); | |
3674 | } | |
585a5bea | 3675 | } else { |
35c91567 DM |
3676 | if (enable) { |
3677 | dev->tx_q[qid].map = qid; | |
3678 | } else { | |
3679 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; | |
3680 | } | |
3681 | netdev_dpdk_remap_txqs(dev); | |
585a5bea | 3682 | } |
585a5bea | 3683 | exists = true; |
d46285a2 | 3684 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3685 | break; |
3686 | } | |
c1ff66ac | 3687 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3688 | } |
3689 | ovs_mutex_unlock(&dpdk_mutex); | |
3690 | ||
3691 | if (exists) { | |
35c91567 DM |
3692 | VLOG_INFO("State of queue %d ( %s_qid %d ) of vhost device '%s' " |
3693 | "changed to \'%s\'", queue_id, is_rx == true ? "rx" : "tx", | |
3694 | qid, ifname, (enable == 1) ? "enabled" : "disabled"); | |
585a5bea | 3695 | } else { |
0a0f39df | 3696 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
3697 | return -1; |
3698 | } | |
3699 | ||
3700 | return 0; | |
3701 | } | |
3702 | ||
61473a0e DM |
3703 | static void |
3704 | destroy_connection(int vid) | |
3705 | { | |
3706 | struct netdev_dpdk *dev; | |
3707 | char ifname[IF_NAME_SZ]; | |
3708 | bool exists = false; | |
3709 | ||
3710 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); | |
3711 | ||
3712 | ovs_mutex_lock(&dpdk_mutex); | |
3713 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
3714 | ovs_mutex_lock(&dev->mutex); | |
3715 | if (nullable_string_is_equal(ifname, dev->vhost_id)) { | |
3716 | uint32_t qp_num = NR_QUEUE; | |
3717 | ||
3718 | if (netdev_dpdk_get_vid(dev) >= 0) { | |
3719 | VLOG_ERR("Connection on socket '%s' destroyed while vhost " | |
3720 | "device still attached.", dev->vhost_id); | |
3721 | } | |
3722 | ||
3723 | /* Restore the number of queue pairs to default. */ | |
3724 | if (dev->requested_n_txq != qp_num | |
3725 | || dev->requested_n_rxq != qp_num) { | |
3726 | dev->requested_n_rxq = qp_num; | |
3727 | dev->requested_n_txq = qp_num; | |
3728 | netdev_request_reconfigure(&dev->up); | |
3729 | } | |
3730 | ovs_mutex_unlock(&dev->mutex); | |
3731 | exists = true; | |
3732 | break; | |
3733 | } | |
3734 | ovs_mutex_unlock(&dev->mutex); | |
3735 | } | |
3736 | ovs_mutex_unlock(&dpdk_mutex); | |
3737 | ||
3738 | if (exists) { | |
3739 | VLOG_INFO("vHost Device '%s' connection has been destroyed", ifname); | |
3740 | } else { | |
3741 | VLOG_INFO("vHost Device '%s' not found", ifname); | |
3742 | } | |
3743 | } | |
3744 | ||
8492adc2 JS |
3745 | /* |
3746 | * Retrieve the DPDK virtio device ID (vid) associated with a vhostuser | |
3747 | * or vhostuserclient netdev. | |
3748 | * | |
3749 | * Returns a value greater or equal to zero for a valid vid or '-1' if | |
3750 | * there is no valid vid associated. A vid of '-1' must not be used in | |
3751 | * rte_vhost_ APi calls. | |
3752 | * | |
3753 | * Once obtained and validated, a vid can be used by a PMD for multiple | |
3754 | * subsequent rte_vhost API calls until the PMD quiesces. A PMD should | |
3755 | * not fetch the vid again for each of a series of API calls. | |
3756 | */ | |
3757 | ||
0a0f39df CL |
3758 | int |
3759 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 3760 | { |
0a0f39df | 3761 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
3762 | } |
3763 | ||
9509913a IS |
3764 | struct ingress_policer * |
3765 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
3766 | { | |
3767 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
3768 | } | |
3769 | ||
58397e6c | 3770 | static int |
ecc1a34e | 3771 | netdev_dpdk_class_init(void) |
7d1ced01 | 3772 | { |
ecc1a34e DDP |
3773 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
3774 | ||
3775 | /* This function can be called for different classes. The initialization | |
3776 | * needs to be done only once */ | |
3777 | if (ovsthread_once_start(&once)) { | |
3778 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
3779 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
3780 | "[netdev] up|down", 1, 2, | |
3781 | netdev_dpdk_set_admin_state, NULL); | |
3782 | ||
0ee821c2 DB |
3783 | unixctl_command_register("netdev-dpdk/detach", |
3784 | "pci address of device", 1, 1, | |
3785 | netdev_dpdk_detach, NULL); | |
3786 | ||
be481733 IM |
3787 | unixctl_command_register("netdev-dpdk/get-mempool-info", |
3788 | "[netdev]", 0, 1, | |
3789 | netdev_dpdk_get_mempool_info, NULL); | |
3790 | ||
ecc1a34e DDP |
3791 | ovsthread_once_done(&once); |
3792 | } | |
362ca396 | 3793 | |
7d1ced01 CL |
3794 | return 0; |
3795 | } | |
3796 | ||
95fb793a | 3797 | /* Client Rings */ |
3798 | ||
95fb793a | 3799 | static int |
3800 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
bb37956a | 3801 | dpdk_port_t *eth_port_id) |
95fb793a | 3802 | { |
48fffdee | 3803 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 3804 | char *ring_name; |
b8374d0d | 3805 | int port_id; |
95fb793a | 3806 | |
48fffdee KT |
3807 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
3808 | if (!ring_pair) { | |
95fb793a | 3809 | return ENOMEM; |
3810 | } | |
3811 | ||
7251515e | 3812 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 3813 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 3814 | |
8f0a76c9 | 3815 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 3816 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3817 | RING_F_SP_ENQ); |
0c6f39e5 | 3818 | free(ring_name); |
48fffdee KT |
3819 | if (ring_pair->cring_tx == NULL) { |
3820 | rte_free(ring_pair); | |
95fb793a | 3821 | return ENOMEM; |
3822 | } | |
3823 | ||
0c6f39e5 | 3824 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 3825 | |
8f0a76c9 | 3826 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 3827 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3828 | RING_F_SC_DEQ); |
0c6f39e5 | 3829 | free(ring_name); |
48fffdee KT |
3830 | if (ring_pair->cring_rx == NULL) { |
3831 | rte_free(ring_pair); | |
95fb793a | 3832 | return ENOMEM; |
3833 | } | |
3834 | ||
b8374d0d MV |
3835 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
3836 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 3837 | |
b8374d0d | 3838 | if (port_id < 0) { |
48fffdee | 3839 | rte_free(ring_pair); |
95fb793a | 3840 | return ENODEV; |
3841 | } | |
3842 | ||
48fffdee | 3843 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
3844 | ring_pair->eth_port_id = port_id; |
3845 | *eth_port_id = port_id; | |
3846 | ||
48fffdee | 3847 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 3848 | |
95fb793a | 3849 | return 0; |
3850 | } | |
3851 | ||
3852 | static int | |
bb37956a | 3853 | dpdk_ring_open(const char dev_name[], dpdk_port_t *eth_port_id) |
64839cf4 | 3854 | OVS_REQUIRES(dpdk_mutex) |
95fb793a | 3855 | { |
48fffdee | 3856 | struct dpdk_ring *ring_pair; |
95fb793a | 3857 | unsigned int port_no; |
3858 | int err = 0; | |
3859 | ||
3860 | /* Names always start with "dpdkr" */ | |
3861 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
3862 | if (err) { | |
3863 | return err; | |
3864 | } | |
3865 | ||
58be5c0e | 3866 | /* Look through our list to find the device */ |
48fffdee KT |
3867 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
3868 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 3869 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 3870 | /* Really all that is needed */ |
48fffdee | 3871 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 3872 | return 0; |
3873 | } | |
3874 | } | |
3875 | /* Need to create the device rings */ | |
3876 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
3877 | } | |
3878 | ||
7251515e | 3879 | static int |
d46285a2 | 3880 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
b30896c9 | 3881 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e | 3882 | { |
d46285a2 | 3883 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a543eb0 | 3884 | struct dp_packet *packet; |
1b99bb05 | 3885 | |
58be5c0e | 3886 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
a47e2db2 | 3887 | * the offload fields are clear. This is because the same mbuf may be |
58be5c0e | 3888 | * modified by the consumer of the ring and return into the datapath |
a47e2db2 | 3889 | * without recalculating the RSS hash or revalidating the checksums. */ |
e883448e | 3890 | DP_PACKET_BATCH_FOR_EACH (i, packet, batch) { |
a47e2db2 | 3891 | dp_packet_reset_offload(packet); |
1b99bb05 | 3892 | } |
7251515e | 3893 | |
b30896c9 | 3894 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e DV |
3895 | return 0; |
3896 | } | |
3897 | ||
95fb793a | 3898 | static int |
3899 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
3900 | { | |
bb37956a | 3901 | dpdk_port_t port_no = 0; |
95fb793a | 3902 | int err = 0; |
3903 | ||
95fb793a | 3904 | ovs_mutex_lock(&dpdk_mutex); |
3905 | ||
3906 | err = dpdk_ring_open(netdev->name, &port_no); | |
3907 | if (err) { | |
3908 | goto unlock_dpdk; | |
3909 | } | |
3910 | ||
1ce30dfd DDP |
3911 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
3912 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 3913 | unlock_dpdk: |
3914 | ovs_mutex_unlock(&dpdk_mutex); | |
3915 | return err; | |
3916 | } | |
3917 | ||
0bf765f7 IS |
3918 | /* QoS Functions */ |
3919 | ||
3920 | /* | |
3921 | * Initialize QoS configuration operations. | |
3922 | */ | |
3923 | static void | |
3924 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
3925 | { | |
3926 | conf->ops = ops; | |
78bd47cf | 3927 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
3928 | } |
3929 | ||
3930 | /* | |
3931 | * Search existing QoS operations in qos_ops and compare each set of | |
3932 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
3933 | * else return NULL | |
3934 | */ | |
3935 | static const struct dpdk_qos_ops * | |
3936 | qos_lookup_name(const char *name) | |
3937 | { | |
3938 | const struct dpdk_qos_ops *const *opsp; | |
3939 | ||
3940 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3941 | const struct dpdk_qos_ops *ops = *opsp; | |
3942 | if (!strcmp(name, ops->qos_name)) { | |
3943 | return ops; | |
3944 | } | |
3945 | } | |
3946 | return NULL; | |
3947 | } | |
3948 | ||
0bf765f7 IS |
3949 | static int |
3950 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
3951 | struct sset *types) | |
3952 | { | |
3953 | const struct dpdk_qos_ops *const *opsp; | |
3954 | ||
3955 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3956 | const struct dpdk_qos_ops *ops = *opsp; | |
3957 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
3958 | sset_add(types, ops->qos_name); | |
3959 | } | |
3960 | } | |
3961 | return 0; | |
3962 | } | |
3963 | ||
3964 | static int | |
d46285a2 | 3965 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
3966 | const char **typep, struct smap *details) |
3967 | { | |
d46285a2 | 3968 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 3969 | struct qos_conf *qos_conf; |
0bf765f7 IS |
3970 | int error = 0; |
3971 | ||
d46285a2 | 3972 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
3973 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
3974 | if (qos_conf) { | |
3975 | *typep = qos_conf->ops->qos_name; | |
3976 | error = (qos_conf->ops->qos_get | |
3977 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
3978 | } else { |
3979 | /* No QoS configuration set, return an empty string */ | |
3980 | *typep = ""; | |
0bf765f7 | 3981 | } |
d46285a2 | 3982 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
3983 | |
3984 | return error; | |
3985 | } | |
3986 | ||
3987 | static int | |
78bd47cf DDP |
3988 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
3989 | const struct smap *details) | |
0bf765f7 | 3990 | { |
d46285a2 | 3991 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 3992 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 3993 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
3994 | int error = 0; |
3995 | ||
d46285a2 | 3996 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 3997 | |
78bd47cf | 3998 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 3999 | |
78bd47cf DDP |
4000 | new_ops = qos_lookup_name(type); |
4001 | ||
4002 | if (!new_ops || !new_ops->qos_construct) { | |
4003 | new_qos_conf = NULL; | |
4004 | if (type && type[0]) { | |
4005 | error = EOPNOTSUPP; | |
0bf765f7 | 4006 | } |
44975bb0 | 4007 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
4008 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
4009 | new_qos_conf = qos_conf; | |
0bf765f7 | 4010 | } else { |
78bd47cf | 4011 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
4012 | } |
4013 | ||
7ea266e9 | 4014 | if (error) { |
78bd47cf DDP |
4015 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
4016 | type, netdev->name, rte_strerror(error)); | |
4017 | } | |
4018 | ||
4019 | if (new_qos_conf != qos_conf) { | |
4020 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
4021 | if (qos_conf) { | |
4022 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
4023 | } | |
0bf765f7 IS |
4024 | } |
4025 | ||
d46285a2 | 4026 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 4027 | |
0bf765f7 IS |
4028 | return error; |
4029 | } | |
4030 | ||
4031 | /* egress-policer details */ | |
4032 | ||
4033 | struct egress_policer { | |
4034 | struct qos_conf qos_conf; | |
4035 | struct rte_meter_srtcm_params app_srtcm_params; | |
4036 | struct rte_meter_srtcm egress_meter; | |
03f3f9c0 | 4037 | struct rte_meter_srtcm_profile egress_prof; |
0bf765f7 IS |
4038 | }; |
4039 | ||
78bd47cf DDP |
4040 | static void |
4041 | egress_policer_details_to_param(const struct smap *details, | |
4042 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 4043 | { |
78bd47cf DDP |
4044 | memset(params, 0, sizeof *params); |
4045 | params->cir = smap_get_ullong(details, "cir", 0); | |
4046 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
4047 | params->ebs = 0; | |
0bf765f7 IS |
4048 | } |
4049 | ||
4050 | static int | |
78bd47cf DDP |
4051 | egress_policer_qos_construct(const struct smap *details, |
4052 | struct qos_conf **conf) | |
0bf765f7 | 4053 | { |
0bf765f7 | 4054 | struct egress_policer *policer; |
0bf765f7 IS |
4055 | int err = 0; |
4056 | ||
0bf765f7 IS |
4057 | policer = xmalloc(sizeof *policer); |
4058 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 4059 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
03f3f9c0 OM |
4060 | err = rte_meter_srtcm_profile_config(&policer->egress_prof, |
4061 | &policer->app_srtcm_params); | |
4062 | if (!err) { | |
4063 | err = rte_meter_srtcm_config(&policer->egress_meter, | |
4064 | &policer->egress_prof); | |
4065 | } | |
4066 | ||
78bd47cf DDP |
4067 | if (!err) { |
4068 | *conf = &policer->qos_conf; | |
4069 | } else { | |
03f3f9c0 | 4070 | VLOG_ERR("Could not create rte meter for egress policer"); |
7ea266e9 | 4071 | free(policer); |
78bd47cf | 4072 | *conf = NULL; |
7ea266e9 IS |
4073 | err = -err; |
4074 | } | |
0bf765f7 IS |
4075 | |
4076 | return err; | |
4077 | } | |
4078 | ||
4079 | static void | |
78bd47cf | 4080 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
4081 | { |
4082 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 4083 | qos_conf); |
0bf765f7 IS |
4084 | free(policer); |
4085 | } | |
4086 | ||
4087 | static int | |
78bd47cf | 4088 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 4089 | { |
78bd47cf DDP |
4090 | struct egress_policer *policer = |
4091 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
4092 | ||
4093 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
4094 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 4095 | |
0bf765f7 IS |
4096 | return 0; |
4097 | } | |
4098 | ||
78bd47cf | 4099 | static bool |
47a45d86 KT |
4100 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
4101 | const struct smap *details) | |
0bf765f7 | 4102 | { |
78bd47cf DDP |
4103 | struct egress_policer *policer = |
4104 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
4105 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 4106 | |
78bd47cf | 4107 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 4108 | |
78bd47cf | 4109 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
4110 | } |
4111 | ||
0bf765f7 | 4112 | static int |
3e90f7d7 | 4113 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt, |
7d7ded7a | 4114 | bool should_steal) |
0bf765f7 | 4115 | { |
0bf765f7 | 4116 | int cnt = 0; |
78bd47cf DDP |
4117 | struct egress_policer *policer = |
4118 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 4119 | |
03f3f9c0 OM |
4120 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, |
4121 | &policer->egress_prof, pkts, | |
7d7ded7a | 4122 | pkt_cnt, should_steal); |
0bf765f7 IS |
4123 | |
4124 | return cnt; | |
4125 | } | |
4126 | ||
4127 | static const struct dpdk_qos_ops egress_policer_ops = { | |
4128 | "egress-policer", /* qos_name */ | |
4129 | egress_policer_qos_construct, | |
4130 | egress_policer_qos_destruct, | |
4131 | egress_policer_qos_get, | |
78bd47cf | 4132 | egress_policer_qos_is_equal, |
0bf765f7 IS |
4133 | egress_policer_run |
4134 | }; | |
4135 | ||
050c60bf DDP |
4136 | static int |
4137 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
4138 | { | |
4139 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
4140 | int err = 0; | |
4141 | ||
050c60bf DDP |
4142 | ovs_mutex_lock(&dev->mutex); |
4143 | ||
4144 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 4145 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b | 4146 | && dev->mtu == dev->requested_mtu |
f8b64a61 | 4147 | && dev->lsc_interrupt_mode == dev->requested_lsc_interrupt_mode |
b685696b | 4148 | && dev->rxq_size == dev->requested_rxq_size |
bd4e172b | 4149 | && dev->txq_size == dev->requested_txq_size |
606f6650 EC |
4150 | && dev->socket_id == dev->requested_socket_id |
4151 | && dev->started) { | |
050c60bf DDP |
4152 | /* Reconfiguration is unnecessary */ |
4153 | ||
4154 | goto out; | |
4155 | } | |
4156 | ||
4157 | rte_eth_dev_stop(dev->port_id); | |
606f6650 | 4158 | dev->started = false; |
050c60bf | 4159 | |
d555d9bd | 4160 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 | 4161 | if (err && err != EEXIST) { |
d555d9bd | 4162 | goto out; |
0072e931 MK |
4163 | } |
4164 | ||
f8b64a61 RM |
4165 | dev->lsc_interrupt_mode = dev->requested_lsc_interrupt_mode; |
4166 | ||
050c60bf DDP |
4167 | netdev->n_txq = dev->requested_n_txq; |
4168 | netdev->n_rxq = dev->requested_n_rxq; | |
4169 | ||
b685696b CL |
4170 | dev->rxq_size = dev->requested_rxq_size; |
4171 | dev->txq_size = dev->requested_txq_size; | |
4172 | ||
050c60bf DDP |
4173 | rte_free(dev->tx_q); |
4174 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
4175 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
4176 | if (!dev->tx_q) { | |
4177 | err = ENOMEM; | |
4178 | } | |
050c60bf | 4179 | |
0072e931 MK |
4180 | netdev_change_seq_changed(netdev); |
4181 | ||
050c60bf | 4182 | out: |
050c60bf | 4183 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
4184 | return err; |
4185 | } | |
4186 | ||
7f381c2e | 4187 | static int |
2d24d165 | 4188 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 4189 | OVS_REQUIRES(dev->mutex) |
050c60bf | 4190 | { |
2d24d165 CL |
4191 | dev->up.n_txq = dev->requested_n_txq; |
4192 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 4193 | int err; |
050c60bf | 4194 | |
35c91567 DM |
4195 | /* Always keep RX queue 0 enabled for implementations that won't |
4196 | * report vring states. */ | |
4197 | dev->vhost_rxq_enabled[0] = true; | |
4198 | ||
81acebda IM |
4199 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
4200 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
4201 | dev->tx_q[0].map = 0; | |
4202 | } | |
4203 | ||
4204 | netdev_dpdk_remap_txqs(dev); | |
4205 | ||
d555d9bd | 4206 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 | 4207 | if (!err) { |
43307ad0 | 4208 | /* A new mempool was created or re-used. */ |
d555d9bd | 4209 | netdev_change_seq_changed(&dev->up); |
03f3f9c0 | 4210 | } else if (err != EEXIST) { |
b6b26021 | 4211 | return err; |
db8f13b0 | 4212 | } |
0a0f39df | 4213 | if (netdev_dpdk_get_vid(dev) >= 0) { |
894af647 | 4214 | if (dev->vhost_reconfigured == false) { |
4215 | dev->vhost_reconfigured = true; | |
4216 | /* Carrier status may need updating. */ | |
4217 | netdev_change_seq_changed(&dev->up); | |
4218 | } | |
81acebda | 4219 | } |
7f381c2e DDP |
4220 | |
4221 | return 0; | |
2d24d165 CL |
4222 | } |
4223 | ||
4224 | static int | |
4225 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
4226 | { | |
4227 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 4228 | int err; |
2d24d165 | 4229 | |
2d24d165 | 4230 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 4231 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 4232 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
4233 | |
4234 | return err; | |
2d24d165 CL |
4235 | } |
4236 | ||
4237 | static int | |
4238 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
4239 | { | |
4240 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 4241 | int err; |
a14d1cc8 | 4242 | uint64_t vhost_flags = 0; |
10087cba | 4243 | bool zc_enabled; |
2d24d165 | 4244 | |
2d24d165 CL |
4245 | ovs_mutex_lock(&dev->mutex); |
4246 | ||
c1ff66ac CL |
4247 | /* Configure vHost client mode if requested and if the following criteria |
4248 | * are met: | |
2d24d165 CL |
4249 | * 1. Device hasn't been registered yet. |
4250 | * 2. A path has been specified. | |
c1ff66ac | 4251 | */ |
bb9d2623 | 4252 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) && dev->vhost_id) { |
a14d1cc8 MK |
4253 | /* Register client-mode device. */ |
4254 | vhost_flags |= RTE_VHOST_USER_CLIENT; | |
4255 | ||
4256 | /* Enable IOMMU support, if explicitly requested. */ | |
4257 | if (dpdk_vhost_iommu_enabled()) { | |
4258 | vhost_flags |= RTE_VHOST_USER_IOMMU_SUPPORT; | |
4259 | } | |
10087cba | 4260 | |
30e834dc LB |
4261 | /* Enable POSTCOPY support, if explicitly requested. */ |
4262 | if (dpdk_vhost_postcopy_enabled()) { | |
4263 | vhost_flags |= RTE_VHOST_USER_POSTCOPY_SUPPORT; | |
4264 | } | |
4265 | ||
10087cba CL |
4266 | zc_enabled = dev->vhost_driver_flags |
4267 | & RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
4268 | /* Enable zero copy flag, if requested */ | |
4269 | if (zc_enabled) { | |
4270 | vhost_flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
4271 | } | |
4272 | ||
a14d1cc8 | 4273 | err = rte_vhost_driver_register(dev->vhost_id, vhost_flags); |
c1ff66ac | 4274 | if (err) { |
2d24d165 CL |
4275 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
4276 | dev->vhost_id); | |
7f381c2e | 4277 | goto unlock; |
c1ff66ac | 4278 | } else { |
2d24d165 | 4279 | /* Configuration successful */ |
a14d1cc8 | 4280 | dev->vhost_driver_flags |= vhost_flags; |
2d24d165 CL |
4281 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " |
4282 | "using client socket '%s'", | |
4283 | dev->up.name, dev->vhost_id); | |
10087cba CL |
4284 | if (zc_enabled) { |
4285 | VLOG_INFO("Zero copy enabled for vHost port %s", dev->up.name); | |
4286 | } | |
c1ff66ac | 4287 | } |
f3e7ec25 MW |
4288 | |
4289 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
4290 | &virtio_net_device_ops); | |
4291 | if (err) { | |
4292 | VLOG_ERR("rte_vhost_driver_callback_register failed for " | |
4293 | "vhost user client port: %s\n", dev->up.name); | |
4294 | goto unlock; | |
4295 | } | |
4296 | ||
4297 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
4298 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
4299 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
4300 | | 1ULL << VIRTIO_NET_F_CSUM); | |
4301 | if (err) { | |
4302 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
4303 | "client port: %s\n", dev->up.name); | |
4304 | goto unlock; | |
4305 | } | |
4306 | ||
4307 | err = rte_vhost_driver_start(dev->vhost_id); | |
4308 | if (err) { | |
4309 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
4310 | "client port: %s\n", dev->up.name); | |
4311 | goto unlock; | |
4312 | } | |
c1ff66ac CL |
4313 | } |
4314 | ||
7f381c2e DDP |
4315 | err = dpdk_vhost_reconfigure_helper(dev); |
4316 | ||
4317 | unlock: | |
050c60bf | 4318 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 4319 | |
7f381c2e | 4320 | return err; |
050c60bf DDP |
4321 | } |
4322 | ||
5fc5c50f IM |
4323 | bool |
4324 | netdev_dpdk_flow_api_supported(struct netdev *netdev) | |
4325 | { | |
4326 | struct netdev_dpdk *dev; | |
4327 | bool ret = false; | |
4328 | ||
4329 | if (!is_dpdk_class(netdev->netdev_class)) { | |
4330 | goto out; | |
4331 | } | |
4332 | ||
4333 | dev = netdev_dpdk_cast(netdev); | |
4334 | ovs_mutex_lock(&dev->mutex); | |
4335 | if (dev->type == DPDK_DEV_ETH) { | |
4336 | /* TODO: Check if we able to offload some minimal flow. */ | |
4337 | ret = true; | |
4338 | } | |
4339 | ovs_mutex_unlock(&dev->mutex); | |
4340 | out: | |
4341 | return ret; | |
4342 | } | |
4343 | ||
6775bdfc RBY |
4344 | int |
4345 | netdev_dpdk_rte_flow_destroy(struct netdev *netdev, | |
4346 | struct rte_flow *rte_flow, | |
4347 | struct rte_flow_error *error) | |
4348 | { | |
4349 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
4350 | int ret; | |
4351 | ||
4352 | ovs_mutex_lock(&dev->mutex); | |
4353 | ret = rte_flow_destroy(dev->port_id, rte_flow, error); | |
4354 | ovs_mutex_unlock(&dev->mutex); | |
4355 | return ret; | |
4356 | } | |
4357 | ||
4358 | struct rte_flow * | |
4359 | netdev_dpdk_rte_flow_create(struct netdev *netdev, | |
4360 | const struct rte_flow_attr *attr, | |
4361 | const struct rte_flow_item *items, | |
4362 | const struct rte_flow_action *actions, | |
4363 | struct rte_flow_error *error) | |
4364 | { | |
4365 | struct rte_flow *flow; | |
4366 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
4367 | ||
4368 | ovs_mutex_lock(&dev->mutex); | |
4369 | flow = rte_flow_create(dev->port_id, attr, items, actions, error); | |
4370 | ovs_mutex_unlock(&dev->mutex); | |
4371 | return flow; | |
4372 | } | |
e8a2b5bf | 4373 | |
89c09c1c BP |
4374 | #define NETDEV_DPDK_CLASS_COMMON \ |
4375 | .is_pmd = true, \ | |
4376 | .alloc = netdev_dpdk_alloc, \ | |
4377 | .dealloc = netdev_dpdk_dealloc, \ | |
4378 | .get_config = netdev_dpdk_get_config, \ | |
4379 | .get_numa_id = netdev_dpdk_get_numa_id, \ | |
4380 | .set_etheraddr = netdev_dpdk_set_etheraddr, \ | |
4381 | .get_etheraddr = netdev_dpdk_get_etheraddr, \ | |
4382 | .get_mtu = netdev_dpdk_get_mtu, \ | |
4383 | .set_mtu = netdev_dpdk_set_mtu, \ | |
4384 | .get_ifindex = netdev_dpdk_get_ifindex, \ | |
4385 | .get_carrier_resets = netdev_dpdk_get_carrier_resets, \ | |
4386 | .set_miimon_interval = netdev_dpdk_set_miimon, \ | |
4387 | .set_policing = netdev_dpdk_set_policing, \ | |
4388 | .get_qos_types = netdev_dpdk_get_qos_types, \ | |
4389 | .get_qos = netdev_dpdk_get_qos, \ | |
4390 | .set_qos = netdev_dpdk_set_qos, \ | |
4391 | .update_flags = netdev_dpdk_update_flags, \ | |
4392 | .rxq_alloc = netdev_dpdk_rxq_alloc, \ | |
4393 | .rxq_construct = netdev_dpdk_rxq_construct, \ | |
4394 | .rxq_destruct = netdev_dpdk_rxq_destruct, \ | |
c0af6425 | 4395 | .rxq_dealloc = netdev_dpdk_rxq_dealloc |
89c09c1c BP |
4396 | |
4397 | #define NETDEV_DPDK_CLASS_BASE \ | |
4398 | NETDEV_DPDK_CLASS_COMMON, \ | |
4399 | .init = netdev_dpdk_class_init, \ | |
4400 | .destruct = netdev_dpdk_destruct, \ | |
4401 | .set_tx_multiq = netdev_dpdk_set_tx_multiq, \ | |
4402 | .get_carrier = netdev_dpdk_get_carrier, \ | |
4403 | .get_stats = netdev_dpdk_get_stats, \ | |
4404 | .get_custom_stats = netdev_dpdk_get_custom_stats, \ | |
4405 | .get_features = netdev_dpdk_get_features, \ | |
4406 | .get_status = netdev_dpdk_get_status, \ | |
4407 | .reconfigure = netdev_dpdk_reconfigure, \ | |
5fc5c50f | 4408 | .rxq_recv = netdev_dpdk_rxq_recv |
89c09c1c BP |
4409 | |
4410 | static const struct netdev_class dpdk_class = { | |
4411 | .type = "dpdk", | |
4412 | NETDEV_DPDK_CLASS_BASE, | |
4413 | .construct = netdev_dpdk_construct, | |
4414 | .set_config = netdev_dpdk_set_config, | |
4415 | .send = netdev_dpdk_eth_send, | |
4416 | }; | |
4417 | ||
4418 | static const struct netdev_class dpdk_ring_class = { | |
4419 | .type = "dpdkr", | |
4420 | NETDEV_DPDK_CLASS_BASE, | |
4421 | .construct = netdev_dpdk_ring_construct, | |
4422 | .set_config = netdev_dpdk_ring_set_config, | |
4423 | .send = netdev_dpdk_ring_send, | |
4424 | }; | |
4425 | ||
4426 | static const struct netdev_class dpdk_vhost_class = { | |
4427 | .type = "dpdkvhostuser", | |
4428 | NETDEV_DPDK_CLASS_COMMON, | |
4429 | .construct = netdev_dpdk_vhost_construct, | |
4430 | .destruct = netdev_dpdk_vhost_destruct, | |
4431 | .send = netdev_dpdk_vhost_send, | |
4432 | .get_carrier = netdev_dpdk_vhost_get_carrier, | |
4433 | .get_stats = netdev_dpdk_vhost_get_stats, | |
c161357d | 4434 | .get_custom_stats = netdev_dpdk_vhost_get_custom_stats, |
89c09c1c BP |
4435 | .get_status = netdev_dpdk_vhost_user_get_status, |
4436 | .reconfigure = netdev_dpdk_vhost_reconfigure, | |
35c91567 DM |
4437 | .rxq_recv = netdev_dpdk_vhost_rxq_recv, |
4438 | .rxq_enabled = netdev_dpdk_vhost_rxq_enabled, | |
89c09c1c BP |
4439 | }; |
4440 | ||
4441 | static const struct netdev_class dpdk_vhost_client_class = { | |
4442 | .type = "dpdkvhostuserclient", | |
4443 | NETDEV_DPDK_CLASS_COMMON, | |
4444 | .construct = netdev_dpdk_vhost_client_construct, | |
4445 | .destruct = netdev_dpdk_vhost_destruct, | |
4446 | .set_config = netdev_dpdk_vhost_client_set_config, | |
4447 | .send = netdev_dpdk_vhost_send, | |
4448 | .get_carrier = netdev_dpdk_vhost_get_carrier, | |
4449 | .get_stats = netdev_dpdk_vhost_get_stats, | |
c161357d | 4450 | .get_custom_stats = netdev_dpdk_vhost_get_custom_stats, |
89c09c1c BP |
4451 | .get_status = netdev_dpdk_vhost_user_get_status, |
4452 | .reconfigure = netdev_dpdk_vhost_client_reconfigure, | |
35c91567 DM |
4453 | .rxq_recv = netdev_dpdk_vhost_rxq_recv, |
4454 | .rxq_enabled = netdev_dpdk_vhost_rxq_enabled, | |
89c09c1c | 4455 | }; |
95fb793a | 4456 | |
8a9562d2 PS |
4457 | void |
4458 | netdev_dpdk_register(void) | |
4459 | { | |
bab69409 AC |
4460 | netdev_register_provider(&dpdk_class); |
4461 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 4462 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 4463 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 4464 | } |