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