]>
Commit | Line | Data |
---|---|---|
8a9562d2 | 1 | /* |
0d4af148 | 2 | * Copyright (c) 2014, 2015, 2016 Nicira, Inc. |
8a9562d2 PS |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
01961bbd | 18 | #include "netdev-dpdk.h" |
8a9562d2 | 19 | |
8a9562d2 PS |
20 | #include <string.h> |
21 | #include <signal.h> | |
22 | #include <stdlib.h> | |
8a9562d2 | 23 | #include <errno.h> |
8a9562d2 | 24 | #include <unistd.h> |
01961bbd DDP |
25 | |
26 | #include <rte_config.h> | |
27 | #include <rte_cycles.h> | |
28 | #include <rte_errno.h> | |
29 | #include <rte_eth_ring.h> | |
30 | #include <rte_ethdev.h> | |
31 | #include <rte_malloc.h> | |
32 | #include <rte_mbuf.h> | |
33 | #include <rte_meter.h> | |
01961bbd | 34 | #include <rte_virtio_net.h> |
8a9562d2 | 35 | |
7d1ced01 | 36 | #include "dirs.h" |
e14deea0 | 37 | #include "dp-packet.h" |
01961bbd | 38 | #include "dpdk.h" |
8a9562d2 | 39 | #include "dpif-netdev.h" |
e5c0f5a4 | 40 | #include "fatal-signal.h" |
8a9562d2 PS |
41 | #include "netdev-provider.h" |
42 | #include "netdev-vport.h" | |
43 | #include "odp-util.h" | |
eac84432 | 44 | #include "openvswitch/dynamic-string.h" |
25d436fb BW |
45 | #include "openvswitch/list.h" |
46 | #include "openvswitch/ofp-print.h" | |
47 | #include "openvswitch/vlog.h" | |
94143fc4 | 48 | #include "ovs-numa.h" |
8a9562d2 PS |
49 | #include "ovs-thread.h" |
50 | #include "ovs-rcu.h" | |
51 | #include "packets.h" | |
ee89ea7b | 52 | #include "openvswitch/shash.h" |
0bf765f7 | 53 | #include "smap.h" |
8a9562d2 PS |
54 | #include "sset.h" |
55 | #include "unaligned.h" | |
56 | #include "timeval.h" | |
57 | #include "unixctl.h" | |
8a9562d2 | 58 | |
05b49df6 | 59 | VLOG_DEFINE_THIS_MODULE(netdev_dpdk); |
8a9562d2 PS |
60 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
61 | ||
62 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
63 | ||
64 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
65 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
66 | ||
67 | /* | |
68 | * need to reserve tons of extra space in the mbufs so we can align the | |
69 | * DMA addresses to 4KB. | |
18f777b2 TP |
70 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
71 | * performance for standard Ethernet MTU. | |
8a9562d2 | 72 | */ |
58be5c0e MK |
73 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN \ |
74 | + (2 * VLAN_HEADER_LEN)) | |
4be4d22c MK |
75 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) |
76 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
58be5c0e MK |
77 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len) \ |
78 | - ETHER_HDR_LEN - ETHER_CRC_LEN) | |
79 | #define MBUF_SIZE(mtu) (MTU_TO_MAX_FRAME_LEN(mtu) \ | |
80 | + sizeof(struct dp_packet) \ | |
81 | + RTE_PKTMBUF_HEADROOM) | |
4be4d22c | 82 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 83 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 84 | |
da79ce2b DDP |
85 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
86 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
87 | * enough hugepages) we keep halving the number until the allocation succeeds | |
88 | * or we reach MIN_NB_MBUF */ | |
89 | ||
90 | #define MAX_NB_MBUF (4096 * 64) | |
91 | #define MIN_NB_MBUF (4096 * 4) | |
92 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
93 | ||
94 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
95 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
96 | ||
97 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
98 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
99 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
100 | % MP_CACHE_SZ == 0); | |
101 | ||
d6e3feb5 | 102 | /* |
103 | * DPDK XSTATS Counter names definition | |
104 | */ | |
105 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
106 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
107 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
108 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
109 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
110 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
111 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
112 | ||
113 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
114 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
115 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
116 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
117 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
118 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
119 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
120 | ||
121 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" | |
122 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
123 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
124 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
125 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
126 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
127 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
128 | ||
8a9562d2 PS |
129 | #define SOCKET0 0 |
130 | ||
b685696b CL |
131 | /* Default size of Physical NIC RXQ */ |
132 | #define NIC_PORT_DEFAULT_RXQ_SIZE 2048 | |
133 | /* Default size of Physical NIC TXQ */ | |
134 | #define NIC_PORT_DEFAULT_TXQ_SIZE 2048 | |
135 | /* Maximum size of Physical NIC Queues */ | |
136 | #define NIC_PORT_MAX_Q_SIZE 4096 | |
79f5354c | 137 | |
585a5bea | 138 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
139 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
140 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
141 | * yet mapped to another queue. */ | |
585a5bea | 142 | |
31871ee3 | 143 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 144 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 145 | |
8a9562d2 | 146 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
147 | .rxmode = { |
148 | .mq_mode = ETH_MQ_RX_RSS, | |
149 | .split_hdr_size = 0, | |
150 | .header_split = 0, /* Header Split disabled */ | |
151 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
152 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
153 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
154 | .hw_strip_crc = 0, | |
155 | }, | |
156 | .rx_adv_conf = { | |
157 | .rss_conf = { | |
158 | .rss_key = NULL, | |
543342a4 | 159 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 160 | }, |
a28ddd11 DDP |
161 | }, |
162 | .txmode = { | |
163 | .mq_mode = ETH_MQ_TX_NONE, | |
164 | }, | |
8a9562d2 PS |
165 | }; |
166 | ||
58f7c37b DDP |
167 | enum { DPDK_RING_SIZE = 256 }; |
168 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
169 | enum { DRAIN_TSC = 200000ULL }; |
170 | ||
58397e6c KT |
171 | enum dpdk_dev_type { |
172 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 173 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
174 | }; |
175 | ||
0bf765f7 IS |
176 | /* Quality of Service */ |
177 | ||
178 | /* An instance of a QoS configuration. Always associated with a particular | |
179 | * network device. | |
180 | * | |
181 | * Each QoS implementation subclasses this with whatever additional data it | |
182 | * needs. | |
183 | */ | |
184 | struct qos_conf { | |
185 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 186 | rte_spinlock_t lock; |
0bf765f7 IS |
187 | }; |
188 | ||
189 | /* A particular implementation of dpdk QoS operations. | |
190 | * | |
191 | * The functions below return 0 if successful or a positive errno value on | |
192 | * failure, except where otherwise noted. All of them must be provided, except | |
193 | * where otherwise noted. | |
194 | */ | |
195 | struct dpdk_qos_ops { | |
196 | ||
197 | /* Name of the QoS type */ | |
198 | const char *qos_name; | |
199 | ||
78bd47cf DDP |
200 | /* Called to construct a qos_conf object. The implementation should make |
201 | * the appropriate calls to configure QoS according to 'details'. | |
0bf765f7 IS |
202 | * |
203 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
204 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
205 | * (which is built as ovs-vswitchd.conf.db(8)). | |
206 | * | |
78bd47cf DDP |
207 | * This function must return 0 if and only if it sets '*conf' to an |
208 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
209 | * |
210 | * For all QoS implementations it should always be non-null. | |
211 | */ | |
78bd47cf | 212 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
213 | |
214 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 215 | * 'qos_conf'. |
0bf765f7 IS |
216 | * |
217 | * For all QoS implementations it should always be non-null. | |
218 | */ | |
78bd47cf | 219 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 220 | |
78bd47cf | 221 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
222 | * |
223 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
224 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
225 | * (which is built as ovs-vswitchd.conf.db(8)). | |
226 | */ | |
78bd47cf | 227 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 228 | |
78bd47cf | 229 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
230 | * |
231 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
232 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
233 | * (which is built as ovs-vswitchd.conf.db(8)). | |
234 | * | |
78bd47cf | 235 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 236 | */ |
78bd47cf DDP |
237 | bool (*qos_is_equal)(const struct qos_conf *conf, |
238 | const struct smap *details); | |
0bf765f7 IS |
239 | |
240 | /* Modify an array of rte_mbufs. The modification is specific to | |
241 | * each qos implementation. | |
242 | * | |
243 | * The function should take and array of mbufs and an int representing | |
244 | * the current number of mbufs present in the array. | |
245 | * | |
246 | * After the function has performed a qos modification to the array of | |
247 | * mbufs it returns an int representing the number of mbufs now present in | |
248 | * the array. This value is can then be passed to the port send function | |
249 | * along with the modified array for transmission. | |
250 | * | |
251 | * For all QoS implementations it should always be non-null. | |
252 | */ | |
78bd47cf DDP |
253 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
254 | int pkt_cnt); | |
0bf765f7 IS |
255 | }; |
256 | ||
257 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
258 | static const struct dpdk_qos_ops egress_policer_ops; | |
259 | ||
260 | /* | |
261 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
262 | * operations. | |
263 | */ | |
264 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
265 | &egress_policer_ops, | |
266 | NULL | |
267 | }; | |
268 | ||
c2adb102 IM |
269 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
270 | ||
8a9562d2 | 271 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 272 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 273 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 274 | |
c2adb102 IM |
275 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
276 | = OVS_MUTEX_INITIALIZER; | |
277 | ||
278 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mp_mutex) | |
55951e15 | 279 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 280 | |
8a9562d2 PS |
281 | struct dpdk_mp { |
282 | struct rte_mempool *mp; | |
283 | int mtu; | |
284 | int socket_id; | |
285 | int refcount; | |
c2adb102 | 286 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mp_mutex); |
8a9562d2 PS |
287 | }; |
288 | ||
5a034064 AW |
289 | /* There should be one 'struct dpdk_tx_queue' created for |
290 | * each cpu core. */ | |
8a9562d2 | 291 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
292 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
293 | * from concurrent access. It is used only | |
294 | * if the queue is shared among different | |
324c8374 | 295 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
296 | int map; /* Mapping of configured vhost-user queues |
297 | * to enabled by guest. */ | |
8a9562d2 PS |
298 | }; |
299 | ||
95fb793a | 300 | /* dpdk has no way to remove dpdk ring ethernet devices |
301 | so we have to keep them around once they've been created | |
302 | */ | |
303 | ||
ca6ba700 | 304 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 305 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 306 | |
307 | struct dpdk_ring { | |
308 | /* For the client rings */ | |
309 | struct rte_ring *cring_tx; | |
310 | struct rte_ring *cring_rx; | |
b83a2df1 | 311 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
95fb793a | 312 | int eth_port_id; /* ethernet device port id */ |
ca6ba700 | 313 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 314 | }; |
315 | ||
9509913a IS |
316 | struct ingress_policer { |
317 | struct rte_meter_srtcm_params app_srtcm_params; | |
318 | struct rte_meter_srtcm in_policer; | |
319 | rte_spinlock_t policer_lock; | |
320 | }; | |
321 | ||
1a2bb118 SC |
322 | enum dpdk_hw_ol_features { |
323 | NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0, | |
324 | }; | |
325 | ||
8a9562d2 PS |
326 | struct netdev_dpdk { |
327 | struct netdev up; | |
328 | int port_id; | |
329 | int max_packet_len; | |
58397e6c | 330 | enum dpdk_dev_type type; |
8a9562d2 | 331 | |
5a034064 | 332 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
333 | |
334 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
335 | ||
336 | struct dpdk_mp *dpdk_mp; | |
337 | int mtu; | |
338 | int socket_id; | |
339 | int buf_size; | |
8a9562d2 | 340 | struct netdev_stats stats; |
45d947c4 DDP |
341 | /* Protects stats */ |
342 | rte_spinlock_t stats_lock; | |
8a9562d2 | 343 | |
74ff3298 | 344 | struct eth_addr hwaddr; |
8a9562d2 PS |
345 | enum netdev_flags flags; |
346 | ||
347 | struct rte_eth_link link; | |
348 | int link_reset_cnt; | |
349 | ||
0a0f39df CL |
350 | /* virtio identifier for vhost devices */ |
351 | ovsrcu_index vid; | |
352 | ||
353 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
354 | bool vhost_reconfigured; | |
58397e6c | 355 | |
2d24d165 CL |
356 | /* Identifier used to distinguish vhost devices from each other. */ |
357 | char vhost_id[PATH_MAX]; | |
7d1ced01 | 358 | |
55e075e6 CL |
359 | /* Device arguments for dpdk ports */ |
360 | char *devargs; | |
361 | ||
8a9562d2 | 362 | /* In dpdk_list. */ |
ca6ba700 | 363 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
0bf765f7 IS |
364 | |
365 | /* QoS configuration and lock for the device */ | |
78bd47cf | 366 | OVSRCU_TYPE(struct qos_conf *) qos_conf; |
0bf765f7 | 367 | |
050c60bf DDP |
368 | /* The following properties cannot be changed when a device is running, |
369 | * so we remember the request and update them next time | |
370 | * netdev_dpdk*_reconfigure() is called */ | |
0072e931 | 371 | int requested_mtu; |
050c60bf DDP |
372 | int requested_n_txq; |
373 | int requested_n_rxq; | |
b685696b CL |
374 | int requested_rxq_size; |
375 | int requested_txq_size; | |
376 | ||
377 | /* Number of rx/tx descriptors for physical devices */ | |
378 | int rxq_size; | |
379 | int txq_size; | |
9509913a | 380 | |
db8f13b0 CL |
381 | /* Socket ID detected when vHost device is brought up */ |
382 | int requested_socket_id; | |
383 | ||
c1ff66ac CL |
384 | /* Denotes whether vHost port is client/server mode */ |
385 | uint64_t vhost_driver_flags; | |
386 | ||
9509913a IS |
387 | /* Ingress Policer */ |
388 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
389 | uint32_t policer_rate; | |
390 | uint32_t policer_burst; | |
9fd39370 SC |
391 | |
392 | /* DPDK-ETH Flow control */ | |
393 | struct rte_eth_fc_conf fc_conf; | |
1a2bb118 SC |
394 | |
395 | /* DPDK-ETH hardware offload features, | |
396 | * from the enum set 'dpdk_hw_ol_features' */ | |
397 | uint32_t hw_ol_features; | |
8a9562d2 PS |
398 | }; |
399 | ||
400 | struct netdev_rxq_dpdk { | |
401 | struct netdev_rxq up; | |
402 | int port_id; | |
403 | }; | |
404 | ||
b3722dca | 405 | static int netdev_dpdk_class_init(void); |
406 | static int netdev_dpdk_vhost_class_init(void); | |
8a9562d2 | 407 | |
0a0f39df | 408 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 409 | |
9509913a IS |
410 | struct ingress_policer * |
411 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
412 | ||
8a9562d2 PS |
413 | static bool |
414 | is_dpdk_class(const struct netdev_class *class) | |
415 | { | |
b3722dca | 416 | return class->init == netdev_dpdk_class_init |
417 | || class->init == netdev_dpdk_vhost_class_init; | |
8a9562d2 PS |
418 | } |
419 | ||
4be4d22c MK |
420 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
421 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
422 | * value, insufficient buffers are allocated to accomodate the packet in its | |
423 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
424 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
425 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
426 | * behaviour, which reduces performance. To prevent this, use a buffer size |
427 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
428 | */ |
429 | static uint32_t | |
430 | dpdk_buf_size(int mtu) | |
431 | { | |
432 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
433 | NETDEV_DPDK_MBUF_ALIGN); | |
434 | } | |
435 | ||
eff23640 DDP |
436 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
437 | * | |
438 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
439 | static void * |
440 | dpdk_rte_mzalloc(size_t sz) | |
441 | { | |
eff23640 | 442 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
443 | } |
444 | ||
445 | void | |
e14deea0 | 446 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 447 | { |
db73f716 | 448 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 449 | |
b00b4a81 | 450 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
451 | } |
452 | ||
b3cd9f9d PS |
453 | static void |
454 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
455 | void *opaque_arg OVS_UNUSED, | |
2391135c | 456 | void *_p, |
b3cd9f9d PS |
457 | unsigned i OVS_UNUSED) |
458 | { | |
2391135c | 459 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 460 | |
2391135c | 461 | rte_pktmbuf_init(mp, opaque_arg, _p, i); |
b3cd9f9d | 462 | |
2391135c | 463 | dp_packet_init_dpdk((struct dp_packet *) pkt, pkt->buf_len); |
b3cd9f9d PS |
464 | } |
465 | ||
8a9562d2 | 466 | static struct dpdk_mp * |
2ae3d542 | 467 | dpdk_mp_create(int socket_id, int mtu) |
8a9562d2 | 468 | { |
2ae3d542 | 469 | struct rte_pktmbuf_pool_private mbp_priv; |
2ae3d542 | 470 | struct dpdk_mp *dmp; |
da79ce2b | 471 | unsigned mp_size; |
0c6f39e5 | 472 | char *mp_name; |
8a9562d2 PS |
473 | |
474 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
eff23640 DDP |
475 | if (!dmp) { |
476 | return NULL; | |
477 | } | |
8a9562d2 PS |
478 | dmp->socket_id = socket_id; |
479 | dmp->mtu = mtu; | |
480 | dmp->refcount = 1; | |
4be4d22c | 481 | mbp_priv.mbuf_data_room_size = MBUF_SIZE(mtu) - sizeof(struct dp_packet); |
58be5c0e MK |
482 | mbp_priv.mbuf_priv_size = sizeof(struct dp_packet) |
483 | - sizeof(struct rte_mbuf); | |
0072e931 | 484 | /* XXX: this is a really rough method of provisioning memory. |
2ae3d542 | 485 | * It's impossible to determine what the exact memory requirements are |
58be5c0e MK |
486 | * when the number of ports and rxqs that utilize a particular mempool can |
487 | * change dynamically at runtime. For now, use this rough heurisitic. | |
0072e931 MK |
488 | */ |
489 | if (mtu >= ETHER_MTU) { | |
490 | mp_size = MAX_NB_MBUF; | |
491 | } else { | |
492 | mp_size = MIN_NB_MBUF; | |
493 | } | |
8a9562d2 | 494 | |
da79ce2b | 495 | do { |
0c6f39e5 DDP |
496 | mp_name = xasprintf("ovs_mp_%d_%d_%u", dmp->mtu, dmp->socket_id, |
497 | mp_size); | |
95fb793a | 498 | |
da79ce2b DDP |
499 | dmp->mp = rte_mempool_create(mp_name, mp_size, MBUF_SIZE(mtu), |
500 | MP_CACHE_SZ, | |
501 | sizeof(struct rte_pktmbuf_pool_private), | |
4be4d22c | 502 | rte_pktmbuf_pool_init, &mbp_priv, |
da79ce2b DDP |
503 | ovs_rte_pktmbuf_init, NULL, |
504 | socket_id, 0); | |
0c6f39e5 DDP |
505 | if (dmp->mp) { |
506 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", | |
507 | mp_name, mp_size); | |
508 | } | |
509 | free(mp_name); | |
510 | if (dmp->mp) { | |
511 | return dmp; | |
512 | } | |
513 | } while (rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
2ae3d542 | 514 | |
2ae3d542 DDP |
515 | rte_free(dmp); |
516 | return NULL; | |
517 | } | |
518 | ||
519 | static struct dpdk_mp * | |
520 | dpdk_mp_get(int socket_id, int mtu) | |
521 | { | |
522 | struct dpdk_mp *dmp; | |
523 | ||
524 | ovs_mutex_lock(&dpdk_mp_mutex); | |
525 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
526 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
527 | dmp->refcount++; | |
528 | goto out; | |
529 | } | |
530 | } | |
531 | ||
532 | dmp = dpdk_mp_create(socket_id, mtu); | |
417e7e66 | 533 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); |
5f88de0d | 534 | |
c2adb102 IM |
535 | out: |
536 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
537 | ||
c2adb102 | 538 | return dmp; |
8a9562d2 PS |
539 | } |
540 | ||
541 | static void | |
c2adb102 | 542 | dpdk_mp_put(struct dpdk_mp *dmp) |
8a9562d2 | 543 | { |
8a9562d2 PS |
544 | if (!dmp) { |
545 | return; | |
546 | } | |
547 | ||
c2adb102 | 548 | ovs_mutex_lock(&dpdk_mp_mutex); |
8d38823b | 549 | ovs_assert(dmp->refcount); |
8a9562d2 | 550 | |
8d38823b MK |
551 | if (!--dmp->refcount) { |
552 | ovs_list_remove(&dmp->list_node); | |
553 | rte_mempool_free(dmp->mp); | |
5f88de0d | 554 | rte_free(dmp); |
8a9562d2 | 555 | } |
c2adb102 | 556 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
557 | } |
558 | ||
0072e931 MK |
559 | /* Tries to allocate new mempool on requested_socket_id with |
560 | * mbuf size corresponding to requested_mtu. | |
561 | * On success new configuration will be applied. | |
562 | * On error, device will be left unchanged. */ | |
563 | static int | |
564 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
565 | OVS_REQUIRES(dev->mutex) |
566 | { | |
567 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
568 | struct dpdk_mp *mp; | |
569 | ||
570 | mp = dpdk_mp_get(dev->requested_socket_id, FRAME_LEN_TO_MTU(buf_size)); | |
571 | if (!mp) { | |
572 | VLOG_ERR("Insufficient memory to create memory pool for netdev " | |
573 | "%s, with MTU %d on socket %d\n", | |
574 | dev->up.name, dev->requested_mtu, dev->requested_socket_id); | |
575 | return ENOMEM; | |
576 | } else { | |
577 | dpdk_mp_put(dev->dpdk_mp); | |
578 | dev->dpdk_mp = mp; | |
579 | dev->mtu = dev->requested_mtu; | |
580 | dev->socket_id = dev->requested_socket_id; | |
581 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
582 | } | |
583 | ||
584 | return 0; | |
585 | } | |
586 | ||
8a9562d2 PS |
587 | static void |
588 | check_link_status(struct netdev_dpdk *dev) | |
589 | { | |
590 | struct rte_eth_link link; | |
591 | ||
592 | rte_eth_link_get_nowait(dev->port_id, &link); | |
593 | ||
594 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 595 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
596 | |
597 | dev->link_reset_cnt++; | |
598 | dev->link = link; | |
599 | if (dev->link.link_status) { | |
600 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
58be5c0e | 601 | dev->port_id, (unsigned) dev->link.link_speed, |
8a9562d2 PS |
602 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? |
603 | ("full-duplex") : ("half-duplex")); | |
604 | } else { | |
605 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
606 | } | |
607 | } | |
608 | } | |
609 | ||
610 | static void * | |
611 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
612 | { | |
613 | struct netdev_dpdk *dev; | |
614 | ||
615 | pthread_detach(pthread_self()); | |
616 | ||
617 | for (;;) { | |
618 | ovs_mutex_lock(&dpdk_mutex); | |
619 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
620 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
621 | if (dev->type == DPDK_DEV_ETH) { |
622 | check_link_status(dev); | |
623 | } | |
8a9562d2 PS |
624 | ovs_mutex_unlock(&dev->mutex); |
625 | } | |
626 | ovs_mutex_unlock(&dpdk_mutex); | |
627 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
628 | } | |
629 | ||
630 | return NULL; | |
631 | } | |
632 | ||
b98d7669 DDP |
633 | static int |
634 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
635 | { | |
636 | int diag = 0; | |
637 | int i; | |
0072e931 | 638 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 639 | |
0072e931 MK |
640 | if (dev->mtu > ETHER_MTU) { |
641 | conf.rxmode.jumbo_frame = 1; | |
642 | conf.rxmode.max_rx_pkt_len = dev->max_packet_len; | |
643 | } else { | |
644 | conf.rxmode.jumbo_frame = 0; | |
645 | conf.rxmode.max_rx_pkt_len = 0; | |
646 | } | |
1a2bb118 SC |
647 | conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & |
648 | NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
b98d7669 DDP |
649 | /* A device may report more queues than it makes available (this has |
650 | * been observed for Intel xl710, which reserves some of them for | |
651 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
652 | * available. When this happens we can retry the configuration | |
653 | * and request less queues */ | |
654 | while (n_rxq && n_txq) { | |
655 | if (diag) { | |
656 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
657 | } | |
658 | ||
0072e931 | 659 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 660 | if (diag) { |
0072e931 MK |
661 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
662 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
663 | break; |
664 | } | |
665 | ||
666 | for (i = 0; i < n_txq; i++) { | |
b685696b | 667 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
668 | dev->socket_id, NULL); |
669 | if (diag) { | |
670 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
671 | dev->up.name, i, rte_strerror(-diag)); | |
672 | break; | |
673 | } | |
674 | } | |
675 | ||
676 | if (i != n_txq) { | |
677 | /* Retry with less tx queues */ | |
678 | n_txq = i; | |
679 | continue; | |
680 | } | |
681 | ||
682 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 683 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
b98d7669 DDP |
684 | dev->socket_id, NULL, |
685 | dev->dpdk_mp->mp); | |
686 | if (diag) { | |
687 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
688 | dev->up.name, i, rte_strerror(-diag)); | |
689 | break; | |
690 | } | |
691 | } | |
692 | ||
693 | if (i != n_rxq) { | |
694 | /* Retry with less rx queues */ | |
695 | n_rxq = i; | |
696 | continue; | |
697 | } | |
698 | ||
699 | dev->up.n_rxq = n_rxq; | |
81acebda | 700 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
701 | |
702 | return 0; | |
703 | } | |
704 | ||
705 | return diag; | |
706 | } | |
707 | ||
1a2bb118 SC |
708 | static void |
709 | dpdk_eth_checksum_offload_configure(struct netdev_dpdk *dev) | |
710 | OVS_REQUIRES(dev->mutex) | |
711 | { | |
712 | struct rte_eth_dev_info info; | |
713 | bool rx_csum_ol_flag = false; | |
714 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | | |
715 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
716 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
717 | rte_eth_dev_info_get(dev->port_id, &info); | |
718 | rx_csum_ol_flag = (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) != 0; | |
719 | ||
720 | if (rx_csum_ol_flag && | |
721 | (info.rx_offload_capa & rx_chksm_offload_capa) != | |
722 | rx_chksm_offload_capa) { | |
723 | VLOG_WARN_ONCE("Rx checksum offload is not supported on device %d", | |
724 | dev->port_id); | |
725 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; | |
726 | return; | |
727 | } | |
728 | netdev_request_reconfigure(&dev->up); | |
729 | } | |
730 | ||
9fd39370 SC |
731 | static void |
732 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
733 | { | |
734 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
735 | VLOG_WARN("Failed to enable flow control on device %d", dev->port_id); | |
736 | } | |
737 | } | |
b98d7669 | 738 | |
8a9562d2 | 739 | static int |
c2adb102 IM |
740 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
741 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
742 | { |
743 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 744 | struct rte_eth_dev_info info; |
8a9562d2 PS |
745 | struct ether_addr eth_addr; |
746 | int diag; | |
b98d7669 | 747 | int n_rxq, n_txq; |
8a9562d2 | 748 | |
a0cb2d66 | 749 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 750 | |
b98d7669 DDP |
751 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
752 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
753 | ||
754 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 755 | if (diag) { |
b98d7669 DDP |
756 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
757 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 758 | return -diag; |
8a9562d2 PS |
759 | } |
760 | ||
8a9562d2 PS |
761 | diag = rte_eth_dev_start(dev->port_id); |
762 | if (diag) { | |
b98d7669 DDP |
763 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
764 | rte_strerror(-diag)); | |
95fb793a | 765 | return -diag; |
8a9562d2 PS |
766 | } |
767 | ||
768 | rte_eth_promiscuous_enable(dev->port_id); | |
769 | rte_eth_allmulticast_enable(dev->port_id); | |
770 | ||
771 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
772 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
182ca1ae | 773 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT, |
ca92d173 | 774 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 775 | |
ca92d173 | 776 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
777 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
778 | ||
779 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
780 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
781 | ||
9fd39370 SC |
782 | /* Get the Flow control configuration for DPDK-ETH */ |
783 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
784 | if (diag) { | |
785 | VLOG_DBG("cannot get flow control parameters on port=%d, err=%d", | |
786 | dev->port_id, diag); | |
787 | } | |
788 | ||
8a9562d2 PS |
789 | return 0; |
790 | } | |
791 | ||
792 | static struct netdev_dpdk * | |
793 | netdev_dpdk_cast(const struct netdev *netdev) | |
794 | { | |
795 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
796 | } | |
797 | ||
798 | static struct netdev * | |
799 | netdev_dpdk_alloc(void) | |
800 | { | |
bab69409 AC |
801 | struct netdev_dpdk *dev; |
802 | ||
65e19e70 DDP |
803 | dev = dpdk_rte_mzalloc(sizeof *dev); |
804 | if (dev) { | |
805 | return &dev->up; | |
bab69409 | 806 | } |
65e19e70 | 807 | |
bab69409 | 808 | return NULL; |
8a9562d2 PS |
809 | } |
810 | ||
eff23640 DDP |
811 | static struct dpdk_tx_queue * |
812 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 813 | { |
eff23640 | 814 | struct dpdk_tx_queue *txqs; |
bd5131ba | 815 | unsigned i; |
5a034064 | 816 | |
eff23640 DDP |
817 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
818 | if (txqs) { | |
819 | for (i = 0; i < n_txqs; i++) { | |
820 | /* Initialize map for vhost devices. */ | |
821 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
822 | rte_spinlock_init(&txqs[i].tx_lock); | |
823 | } | |
5a034064 | 824 | } |
eff23640 DDP |
825 | |
826 | return txqs; | |
5a034064 AW |
827 | } |
828 | ||
8a9562d2 | 829 | static int |
1ce30dfd DDP |
830 | common_construct(struct netdev *netdev, unsigned int port_no, |
831 | enum dpdk_dev_type type, int socket_id) | |
5a034064 | 832 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 833 | { |
d46285a2 | 834 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 835 | |
d46285a2 | 836 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 837 | |
d46285a2 | 838 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 839 | |
1b7a04e0 AW |
840 | /* If the 'sid' is negative, it means that the kernel fails |
841 | * to obtain the pci numa info. In that situation, always | |
842 | * use 'SOCKET0'. */ | |
1ce30dfd | 843 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 844 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
845 | dev->port_id = port_no; |
846 | dev->type = type; | |
847 | dev->flags = 0; | |
7f381c2e | 848 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 849 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
850 | ovsrcu_index_init(&dev->vid, -1); |
851 | dev->vhost_reconfigured = false; | |
8a9562d2 | 852 | |
78bd47cf | 853 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 854 | |
9509913a IS |
855 | ovsrcu_init(&dev->ingress_policer, NULL); |
856 | dev->policer_rate = 0; | |
857 | dev->policer_burst = 0; | |
858 | ||
7f381c2e DDP |
859 | netdev->n_rxq = 0; |
860 | netdev->n_txq = 0; | |
861 | dev->requested_n_rxq = NR_QUEUE; | |
862 | dev->requested_n_txq = NR_QUEUE; | |
863 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
864 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 865 | |
9fd39370 SC |
866 | /* Initialize the flow control to NULL */ |
867 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
868 | |
869 | /* Initilize the hardware offload flags to 0 */ | |
870 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
871 | |
872 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
873 | ||
d46285a2 | 874 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 875 | |
7f381c2e DDP |
876 | netdev_request_reconfigure(netdev); |
877 | ||
1ce30dfd | 878 | return 0; |
95fb793a | 879 | } |
880 | ||
b83a2df1 MV |
881 | /* dev_name must be the prefix followed by a positive decimal number. |
882 | * (no leading + or - signs are allowed) */ | |
95fb793a | 883 | static int |
884 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
885 | unsigned int *port_no) | |
886 | { | |
887 | const char *cport; | |
888 | ||
889 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
890 | return ENODEV; | |
891 | } | |
892 | ||
893 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
894 | |
895 | if (str_to_uint(cport, 10, port_no)) { | |
896 | return 0; | |
897 | } else { | |
898 | return ENODEV; | |
899 | } | |
95fb793a | 900 | } |
901 | ||
1ce30dfd DDP |
902 | static int |
903 | vhost_common_construct(struct netdev *netdev) | |
904 | OVS_REQUIRES(dpdk_mutex) | |
905 | { | |
906 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
907 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
908 | ||
909 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); | |
910 | if (!dev->tx_q) { | |
911 | return ENOMEM; | |
912 | } | |
913 | ||
914 | return common_construct(netdev, -1, DPDK_DEV_VHOST, socket_id); | |
915 | } | |
916 | ||
7d1ced01 | 917 | static int |
53f50d24 | 918 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 919 | { |
d46285a2 DDP |
920 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
921 | const char *name = netdev->name; | |
7d1ced01 | 922 | int err; |
a0cb2d66 | 923 | |
1af27e8a DDP |
924 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
925 | * the file system. '/' or '\' would traverse directories, so they're not | |
926 | * acceptable in 'name'. */ | |
927 | if (strchr(name, '/') || strchr(name, '\\')) { | |
928 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
929 | "A valid name must not include '/' or '\\'", | |
930 | name); | |
931 | return EINVAL; | |
932 | } | |
933 | ||
7d1ced01 CL |
934 | ovs_mutex_lock(&dpdk_mutex); |
935 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 936 | * the socket is to be created, then register the socket. |
7d1ced01 | 937 | */ |
2d24d165 | 938 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
01961bbd | 939 | dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 940 | |
2d24d165 CL |
941 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
942 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
943 | if (err) { |
944 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 945 | dev->vhost_id); |
e5c0f5a4 | 946 | } else { |
2d24d165 CL |
947 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
948 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
949 | dev->vhost_id, name); | |
950 | } | |
1ce30dfd | 951 | err = vhost_common_construct(netdev); |
2d24d165 CL |
952 | |
953 | ovs_mutex_unlock(&dpdk_mutex); | |
954 | return err; | |
955 | } | |
956 | ||
957 | static int | |
958 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
959 | { | |
960 | int err; | |
961 | ||
2d24d165 | 962 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 963 | err = vhost_common_construct(netdev); |
7d1ced01 | 964 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
965 | return err; |
966 | } | |
967 | ||
95fb793a | 968 | static int |
969 | netdev_dpdk_construct(struct netdev *netdev) | |
970 | { | |
95fb793a | 971 | int err; |
972 | ||
95fb793a | 973 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 974 | err = common_construct(netdev, -1, DPDK_DEV_ETH, SOCKET0); |
8a9562d2 PS |
975 | ovs_mutex_unlock(&dpdk_mutex); |
976 | return err; | |
977 | } | |
978 | ||
1ce30dfd DDP |
979 | static void |
980 | common_destruct(struct netdev_dpdk *dev) | |
981 | OVS_REQUIRES(dpdk_mutex) | |
982 | OVS_EXCLUDED(dev->mutex) | |
983 | { | |
984 | rte_free(dev->tx_q); | |
985 | dpdk_mp_put(dev->dpdk_mp); | |
986 | ||
987 | ovs_list_remove(&dev->list_node); | |
988 | free(ovsrcu_get_protected(struct ingress_policer *, | |
989 | &dev->ingress_policer)); | |
990 | ovs_mutex_destroy(&dev->mutex); | |
991 | } | |
992 | ||
8a9562d2 | 993 | static void |
d46285a2 | 994 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 995 | { |
d46285a2 | 996 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 997 | |
8d38823b | 998 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 999 | |
8a9562d2 | 1000 | rte_eth_dev_stop(dev->port_id); |
55e075e6 | 1001 | free(dev->devargs); |
1ce30dfd | 1002 | common_destruct(dev); |
8d38823b | 1003 | |
8a9562d2 | 1004 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1005 | } |
8a9562d2 | 1006 | |
3f891bbe DDP |
1007 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1008 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1009 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1010 | static int | |
c1ff66ac CL |
1011 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1012 | char *vhost_id) | |
3f891bbe DDP |
1013 | OVS_EXCLUDED(dpdk_mutex) |
1014 | OVS_EXCLUDED(dev->mutex) | |
1015 | { | |
c1ff66ac | 1016 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1017 | } |
1018 | ||
58397e6c | 1019 | static void |
d46285a2 | 1020 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1021 | { |
d46285a2 | 1022 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1023 | char *vhost_id; |
58397e6c | 1024 | |
8d38823b | 1025 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1026 | |
c62da695 | 1027 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1028 | if (netdev_dpdk_get_vid(dev) >= 0 |
1029 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1030 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1031 | netdev->name); |
58be5c0e MK |
1032 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1033 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1034 | } |
1035 | ||
2d24d165 | 1036 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1037 | |
1ce30dfd DDP |
1038 | common_destruct(dev); |
1039 | ||
58397e6c | 1040 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1041 | |
569c26da | 1042 | if (!vhost_id[0]) { |
821b8664 IM |
1043 | goto out; |
1044 | } | |
1045 | ||
c1ff66ac | 1046 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1047 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1048 | netdev->name, vhost_id); | |
c1ff66ac CL |
1049 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1050 | /* OVS server mode - remove this socket from list for deletion */ | |
1051 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1052 | } |
821b8664 | 1053 | out: |
c1ff66ac | 1054 | free(vhost_id); |
8a9562d2 PS |
1055 | } |
1056 | ||
1057 | static void | |
d46285a2 | 1058 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1059 | { |
d46285a2 | 1060 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1061 | |
d46285a2 | 1062 | rte_free(dev); |
8a9562d2 PS |
1063 | } |
1064 | ||
1065 | static int | |
a14b8947 | 1066 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1067 | { |
a14b8947 | 1068 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1069 | |
1070 | ovs_mutex_lock(&dev->mutex); | |
1071 | ||
050c60bf | 1072 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1073 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1074 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1075 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1076 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1077 | |
1078 | if (dev->type == DPDK_DEV_ETH) { | |
1079 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1080 | dev->requested_rxq_size); | |
1081 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1082 | dev->rxq_size); | |
1083 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1084 | dev->requested_txq_size); | |
1085 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1086 | dev->txq_size); | |
1a2bb118 SC |
1087 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1088 | smap_add(args, "rx_csum_offload", "true"); | |
1089 | } | |
451f26fd | 1090 | } |
8a9562d2 PS |
1091 | ovs_mutex_unlock(&dev->mutex); |
1092 | ||
1093 | return 0; | |
1094 | } | |
1095 | ||
55e075e6 CL |
1096 | static struct netdev_dpdk * |
1097 | netdev_dpdk_lookup_by_port_id(int port_id) | |
1098 | OVS_REQUIRES(dpdk_mutex) | |
1099 | { | |
1100 | struct netdev_dpdk *dev; | |
1101 | ||
1102 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1103 | if (dev->port_id == port_id) { | |
1104 | return dev; | |
1105 | } | |
1106 | } | |
1107 | ||
1108 | return NULL; | |
1109 | } | |
1110 | ||
1111 | static int | |
9fff138e | 1112 | netdev_dpdk_process_devargs(const char *devargs, char **errp) |
55e075e6 | 1113 | { |
55e075e6 CL |
1114 | uint8_t new_port_id = UINT8_MAX; |
1115 | ||
69876ed7 CL |
1116 | if (!rte_eth_dev_count() |
1117 | || rte_eth_dev_get_port_by_name(devargs, &new_port_id) | |
1118 | || !rte_eth_dev_is_valid_port(new_port_id)) { | |
1119 | /* Device not found in DPDK, attempt to attach it */ | |
1120 | if (!rte_eth_dev_attach(devargs, &new_port_id)) { | |
1121 | /* Attach successful */ | |
1122 | VLOG_INFO("Device '%s' attached to DPDK", devargs); | |
1123 | } else { | |
1124 | /* Attach unsuccessful */ | |
9fff138e | 1125 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", devargs); |
69876ed7 | 1126 | return -1; |
55e075e6 CL |
1127 | } |
1128 | } | |
1129 | ||
1130 | return new_port_id; | |
1131 | } | |
1132 | ||
c3d062a7 CL |
1133 | static void |
1134 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1135 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1136 | { |
050c60bf | 1137 | int new_n_rxq; |
a14b8947 | 1138 | |
2a21e757 | 1139 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1140 | if (new_n_rxq != dev->requested_n_rxq) { |
1141 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1142 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1143 | } |
c3d062a7 CL |
1144 | } |
1145 | ||
b685696b CL |
1146 | static void |
1147 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1148 | const char *flag, int default_size, int *new_size) | |
1149 | { | |
1150 | int queue_size = smap_get_int(args, flag, default_size); | |
1151 | ||
1152 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1153 | || !is_pow2(queue_size)) { | |
1154 | queue_size = default_size; | |
1155 | } | |
1156 | ||
1157 | if (queue_size != *new_size) { | |
1158 | *new_size = queue_size; | |
1159 | netdev_request_reconfigure(netdev); | |
1160 | } | |
1161 | } | |
1162 | ||
c3d062a7 | 1163 | static int |
9fff138e DDP |
1164 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1165 | char **errp) | |
c3d062a7 CL |
1166 | { |
1167 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
b614c894 IM |
1168 | bool rx_fc_en, tx_fc_en, autoneg; |
1169 | enum rte_eth_fc_mode fc_mode; | |
1170 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1171 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1172 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1173 | }; | |
1a2bb118 SC |
1174 | bool rx_chksm_ofld; |
1175 | bool temp_flag; | |
55e075e6 CL |
1176 | const char *new_devargs; |
1177 | int err = 0; | |
c3d062a7 | 1178 | |
55e075e6 | 1179 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1180 | ovs_mutex_lock(&dev->mutex); |
1181 | ||
1182 | dpdk_set_rxq_config(dev, args); | |
1183 | ||
b685696b CL |
1184 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1185 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1186 | &dev->requested_rxq_size); | |
1187 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1188 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1189 | &dev->requested_txq_size); | |
1190 | ||
55e075e6 CL |
1191 | new_devargs = smap_get(args, "dpdk-devargs"); |
1192 | ||
1193 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1194 | /* The user requested a new device. If we return error, the caller | |
1195 | * will delete this netdev and try to recreate it. */ | |
1196 | err = EAGAIN; | |
1197 | goto out; | |
1198 | } | |
1199 | ||
1200 | /* dpdk-devargs is required for device configuration */ | |
1201 | if (new_devargs && new_devargs[0]) { | |
1202 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1203 | * is valid */ | |
1204 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1205 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
9fff138e | 1206 | int new_port_id = netdev_dpdk_process_devargs(new_devargs, errp); |
55e075e6 CL |
1207 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1208 | err = EINVAL; | |
1209 | } else if (new_port_id == dev->port_id) { | |
1210 | /* Already configured, do not reconfigure again */ | |
1211 | err = 0; | |
1212 | } else { | |
1213 | struct netdev_dpdk *dup_dev; | |
1214 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); | |
1215 | if (dup_dev) { | |
9fff138e DDP |
1216 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
1217 | "which is already in use by '%s'", | |
1218 | netdev_get_name(netdev), new_devargs, | |
1219 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1220 | err = EADDRINUSE; |
1221 | } else { | |
bd4e172b | 1222 | int sid = rte_eth_dev_socket_id(new_port_id); |
1223 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; | |
55e075e6 CL |
1224 | dev->devargs = xstrdup(new_devargs); |
1225 | dev->port_id = new_port_id; | |
1226 | netdev_request_reconfigure(&dev->up); | |
1227 | err = 0; | |
1228 | } | |
1229 | } | |
1230 | } | |
1231 | } else { | |
9fff138e DDP |
1232 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1233 | "The old 'dpdk<port_id>' names are not supported", | |
1234 | netdev_get_name(netdev)); | |
55e075e6 CL |
1235 | err = EINVAL; |
1236 | } | |
1237 | ||
1238 | if (err) { | |
1239 | goto out; | |
1240 | } | |
1241 | ||
c3d062a7 CL |
1242 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1243 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1244 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1245 | |
b614c894 IM |
1246 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
1247 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { | |
1248 | dev->fc_conf.mode = fc_mode; | |
1249 | dev->fc_conf.autoneg = autoneg; | |
1250 | dpdk_eth_flow_ctrl_setup(dev); | |
1251 | } | |
9fd39370 | 1252 | |
1a2bb118 SC |
1253 | /* Rx checksum offload configuration */ |
1254 | /* By default the Rx checksum offload is ON */ | |
1255 | rx_chksm_ofld = smap_get_bool(args, "rx-checksum-offload", true); | |
1256 | temp_flag = (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) | |
1257 | != 0; | |
1258 | if (temp_flag != rx_chksm_ofld) { | |
1259 | dev->hw_ol_features ^= NETDEV_RX_CHECKSUM_OFFLOAD; | |
1260 | dpdk_eth_checksum_offload_configure(dev); | |
1261 | } | |
55e075e6 CL |
1262 | |
1263 | out: | |
c3d062a7 | 1264 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1265 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1266 | |
55e075e6 | 1267 | return err; |
c3d062a7 CL |
1268 | } |
1269 | ||
1270 | static int | |
9fff138e DDP |
1271 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1272 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1273 | { |
1274 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1275 | ||
1276 | ovs_mutex_lock(&dev->mutex); | |
1277 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1278 | ovs_mutex_unlock(&dev->mutex); |
1279 | ||
1280 | return 0; | |
1281 | } | |
1282 | ||
c1ff66ac | 1283 | static int |
2d24d165 | 1284 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1285 | const struct smap *args, |
1286 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1287 | { |
1288 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1289 | const char *path; | |
1290 | ||
6881885a | 1291 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1292 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1293 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1294 | if (path && strcmp(path, dev->vhost_id)) { |
1295 | strcpy(dev->vhost_id, path); | |
c1ff66ac CL |
1296 | netdev_request_reconfigure(netdev); |
1297 | } | |
1298 | } | |
6881885a | 1299 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1300 | |
1301 | return 0; | |
1302 | } | |
1303 | ||
7dec44fe | 1304 | static int |
d46285a2 | 1305 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1306 | { |
d46285a2 | 1307 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1308 | |
d46285a2 | 1309 | return dev->socket_id; |
7dec44fe AW |
1310 | } |
1311 | ||
050c60bf | 1312 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1313 | static int |
050c60bf | 1314 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1315 | { |
d46285a2 | 1316 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1317 | |
d46285a2 | 1318 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1319 | |
050c60bf DDP |
1320 | if (dev->requested_n_txq == n_txq) { |
1321 | goto out; | |
4573fbd3 FL |
1322 | } |
1323 | ||
050c60bf DDP |
1324 | dev->requested_n_txq = n_txq; |
1325 | netdev_request_reconfigure(netdev); | |
58397e6c | 1326 | |
050c60bf | 1327 | out: |
d46285a2 | 1328 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1329 | return 0; |
58397e6c KT |
1330 | } |
1331 | ||
8a9562d2 PS |
1332 | static struct netdev_rxq * |
1333 | netdev_dpdk_rxq_alloc(void) | |
1334 | { | |
1335 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1336 | ||
eff23640 DDP |
1337 | if (rx) { |
1338 | return &rx->up; | |
1339 | } | |
1340 | ||
1341 | return NULL; | |
8a9562d2 PS |
1342 | } |
1343 | ||
1344 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1345 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1346 | { |
d46285a2 | 1347 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1348 | } |
1349 | ||
1350 | static int | |
d46285a2 | 1351 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1352 | { |
d46285a2 DDP |
1353 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1354 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1355 | |
d46285a2 DDP |
1356 | ovs_mutex_lock(&dev->mutex); |
1357 | rx->port_id = dev->port_id; | |
1358 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1359 | |
1360 | return 0; | |
1361 | } | |
1362 | ||
1363 | static void | |
d46285a2 | 1364 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1365 | { |
1366 | } | |
1367 | ||
1368 | static void | |
d46285a2 | 1369 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1370 | { |
d46285a2 | 1371 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1372 | |
1373 | rte_free(rx); | |
1374 | } | |
1375 | ||
819f13bd DDP |
1376 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
1377 | * 'pkts', even in case of failure. | |
1378 | * | |
1379 | * Returns the number of packets that weren't transmitted. */ | |
1380 | static inline int | |
b59cc14e | 1381 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 1382 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 1383 | { |
1304f1f8 DDP |
1384 | uint32_t nb_tx = 0; |
1385 | ||
b59cc14e | 1386 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1387 | uint32_t ret; |
1388 | ||
b59cc14e | 1389 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1390 | if (!ret) { |
1391 | break; | |
1392 | } | |
1393 | ||
1394 | nb_tx += ret; | |
1395 | } | |
8a9562d2 | 1396 | |
b59cc14e | 1397 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 1398 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
1399 | * packet could come from a different mempool) */ |
1400 | int i; | |
1401 | ||
b59cc14e IM |
1402 | for (i = nb_tx; i < cnt; i++) { |
1403 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1404 | } |
8a9562d2 | 1405 | } |
819f13bd DDP |
1406 | |
1407 | return cnt - nb_tx; | |
8a9562d2 PS |
1408 | } |
1409 | ||
f3926f29 IS |
1410 | static inline bool |
1411 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1412 | struct rte_mbuf *pkt, uint64_t time) | |
1413 | { | |
1414 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1415 | ||
1416 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1417 | e_RTE_METER_GREEN; | |
1418 | } | |
1419 | ||
1420 | static int | |
1421 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
1422 | struct rte_mbuf **pkts, int pkt_cnt) | |
1423 | { | |
1424 | int i = 0; | |
1425 | int cnt = 0; | |
1426 | struct rte_mbuf *pkt = NULL; | |
1427 | uint64_t current_time = rte_rdtsc(); | |
1428 | ||
1429 | for (i = 0; i < pkt_cnt; i++) { | |
1430 | pkt = pkts[i]; | |
1431 | /* Handle current packet */ | |
1432 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1433 | if (cnt != i) { | |
1434 | pkts[cnt] = pkt; | |
1435 | } | |
1436 | cnt++; | |
1437 | } else { | |
1438 | rte_pktmbuf_free(pkt); | |
1439 | } | |
1440 | } | |
1441 | ||
1442 | return cnt; | |
1443 | } | |
1444 | ||
9509913a IS |
1445 | static int |
1446 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
1447 | int pkt_cnt) | |
1448 | { | |
1449 | int cnt = 0; | |
1450 | ||
1451 | rte_spinlock_lock(&policer->policer_lock); | |
1452 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, pkt_cnt); | |
1453 | rte_spinlock_unlock(&policer->policer_lock); | |
1454 | ||
1455 | return cnt; | |
1456 | } | |
1457 | ||
58397e6c | 1458 | static bool |
0a0f39df | 1459 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1460 | { |
0a0f39df | 1461 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1462 | } |
1463 | ||
d6e3feb5 | 1464 | static inline void |
1465 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1466 | unsigned int packet_size) | |
1467 | { | |
1468 | /* Hard-coded search for the size bucket. */ | |
1469 | if (packet_size < 256) { | |
1470 | if (packet_size >= 128) { | |
1471 | stats->rx_128_to_255_packets++; | |
1472 | } else if (packet_size <= 64) { | |
1473 | stats->rx_1_to_64_packets++; | |
1474 | } else { | |
1475 | stats->rx_65_to_127_packets++; | |
1476 | } | |
1477 | } else { | |
1478 | if (packet_size >= 1523) { | |
1479 | stats->rx_1523_to_max_packets++; | |
1480 | } else if (packet_size >= 1024) { | |
1481 | stats->rx_1024_to_1522_packets++; | |
1482 | } else if (packet_size < 512) { | |
1483 | stats->rx_256_to_511_packets++; | |
1484 | } else { | |
1485 | stats->rx_512_to_1023_packets++; | |
1486 | } | |
1487 | } | |
1488 | } | |
1489 | ||
9e3ddd45 TP |
1490 | static inline void |
1491 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1492 | struct dp_packet **packets, int count, |
1493 | int dropped) | |
9e3ddd45 TP |
1494 | { |
1495 | int i; | |
d6e3feb5 | 1496 | unsigned int packet_size; |
9e3ddd45 TP |
1497 | struct dp_packet *packet; |
1498 | ||
1499 | stats->rx_packets += count; | |
9509913a | 1500 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1501 | for (i = 0; i < count; i++) { |
1502 | packet = packets[i]; | |
d6e3feb5 | 1503 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1504 | |
d6e3feb5 | 1505 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1506 | /* This only protects the following multicast counting from |
1507 | * too short packets, but it does not stop the packet from | |
1508 | * further processing. */ | |
1509 | stats->rx_errors++; | |
1510 | stats->rx_length_errors++; | |
1511 | continue; | |
1512 | } | |
1513 | ||
d6e3feb5 | 1514 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1515 | ||
9e3ddd45 TP |
1516 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1517 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1518 | stats->multicast++; | |
1519 | } | |
1520 | ||
d6e3feb5 | 1521 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1522 | } |
1523 | } | |
1524 | ||
58397e6c KT |
1525 | /* |
1526 | * The receive path for the vhost port is the TX path out from guest. | |
1527 | */ | |
1528 | static int | |
d46285a2 | 1529 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1530 | struct dp_packet_batch *batch) |
58397e6c | 1531 | { |
d46285a2 | 1532 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
d46285a2 | 1533 | int qid = rxq->queue_id; |
9509913a | 1534 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1535 | uint16_t nb_rx = 0; |
9509913a | 1536 | uint16_t dropped = 0; |
58397e6c | 1537 | |
0a0f39df | 1538 | if (OVS_UNLIKELY(!is_vhost_running(dev) |
e543851d | 1539 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1540 | return EAGAIN; |
1541 | } | |
1542 | ||
0a0f39df CL |
1543 | nb_rx = rte_vhost_dequeue_burst(netdev_dpdk_get_vid(dev), |
1544 | qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
d46285a2 | 1545 | dev->dpdk_mp->mp, |
64839cf4 | 1546 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1547 | NETDEV_MAX_BURST); |
58397e6c KT |
1548 | if (!nb_rx) { |
1549 | return EAGAIN; | |
1550 | } | |
1551 | ||
9509913a IS |
1552 | if (policer) { |
1553 | dropped = nb_rx; | |
64839cf4 WT |
1554 | nb_rx = ingress_policer_run(policer, |
1555 | (struct rte_mbuf **) batch->packets, | |
1556 | nb_rx); | |
9509913a IS |
1557 | dropped -= nb_rx; |
1558 | } | |
1559 | ||
d46285a2 | 1560 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1561 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1562 | nb_rx, dropped); | |
d46285a2 | 1563 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1564 | |
64839cf4 | 1565 | batch->count = (int) nb_rx; |
58397e6c KT |
1566 | return 0; |
1567 | } | |
1568 | ||
8a9562d2 | 1569 | static int |
64839cf4 | 1570 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1571 | { |
d46285a2 DDP |
1572 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1573 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1574 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1575 | int nb_rx; |
9509913a | 1576 | int dropped = 0; |
8a9562d2 | 1577 | |
3b1fb077 DDP |
1578 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1579 | return EAGAIN; | |
1580 | } | |
1581 | ||
d46285a2 | 1582 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1583 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1584 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1585 | if (!nb_rx) { |
1586 | return EAGAIN; | |
1587 | } | |
1588 | ||
9509913a IS |
1589 | if (policer) { |
1590 | dropped = nb_rx; | |
64839cf4 | 1591 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 1592 | (struct rte_mbuf **) batch->packets, |
64839cf4 | 1593 | nb_rx); |
9509913a IS |
1594 | dropped -= nb_rx; |
1595 | } | |
1596 | ||
1597 | /* Update stats to reflect dropped packets */ | |
1598 | if (OVS_UNLIKELY(dropped)) { | |
1599 | rte_spinlock_lock(&dev->stats_lock); | |
1600 | dev->stats.rx_dropped += dropped; | |
1601 | rte_spinlock_unlock(&dev->stats_lock); | |
1602 | } | |
1603 | ||
64839cf4 | 1604 | batch->count = nb_rx; |
8a9562d2 PS |
1605 | |
1606 | return 0; | |
1607 | } | |
1608 | ||
0bf765f7 | 1609 | static inline int |
78bd47cf DDP |
1610 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
1611 | int cnt) | |
0bf765f7 | 1612 | { |
78bd47cf | 1613 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 1614 | |
78bd47cf DDP |
1615 | if (qos_conf) { |
1616 | rte_spinlock_lock(&qos_conf->lock); | |
1617 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt); | |
1618 | rte_spinlock_unlock(&qos_conf->lock); | |
0bf765f7 IS |
1619 | } |
1620 | ||
1621 | return cnt; | |
1622 | } | |
1623 | ||
c6ec9d17 IM |
1624 | static int |
1625 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1626 | int pkt_cnt) | |
1627 | { | |
1628 | int i = 0; | |
1629 | int cnt = 0; | |
1630 | struct rte_mbuf *pkt; | |
1631 | ||
1632 | for (i = 0; i < pkt_cnt; i++) { | |
1633 | pkt = pkts[i]; | |
1634 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
1635 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
1636 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
1637 | rte_pktmbuf_free(pkt); | |
1638 | continue; | |
1639 | } | |
1640 | ||
1641 | if (OVS_UNLIKELY(i != cnt)) { | |
1642 | pkts[cnt] = pkt; | |
1643 | } | |
1644 | cnt++; | |
1645 | } | |
1646 | ||
1647 | return cnt; | |
1648 | } | |
1649 | ||
9e3ddd45 TP |
1650 | static inline void |
1651 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1652 | struct dp_packet **packets, | |
1653 | int attempted, | |
1654 | int dropped) | |
1655 | { | |
1656 | int i; | |
1657 | int sent = attempted - dropped; | |
1658 | ||
1659 | stats->tx_packets += sent; | |
1660 | stats->tx_dropped += dropped; | |
1661 | ||
1662 | for (i = 0; i < sent; i++) { | |
1663 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1664 | } | |
1665 | } | |
1666 | ||
58397e6c | 1667 | static void |
4573fbd3 | 1668 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 1669 | struct dp_packet **pkts, int cnt) |
58397e6c | 1670 | { |
d46285a2 | 1671 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1672 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1673 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 1674 | unsigned int dropped = 0; |
dd52de45 | 1675 | int i, retries = 0; |
58397e6c | 1676 | |
81acebda | 1677 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1678 | |
0a0f39df | 1679 | if (OVS_UNLIKELY(!is_vhost_running(dev) || qid < 0 |
e543851d | 1680 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1681 | rte_spinlock_lock(&dev->stats_lock); |
1682 | dev->stats.tx_dropped+= cnt; | |
1683 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1684 | goto out; |
58397e6c KT |
1685 | } |
1686 | ||
d46285a2 | 1687 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1688 | |
c6ec9d17 | 1689 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 1690 | /* Check has QoS has been configured for the netdev */ |
78bd47cf | 1691 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt); |
c6ec9d17 | 1692 | dropped = total_pkts - cnt; |
0bf765f7 | 1693 | |
95e9881f | 1694 | do { |
4573fbd3 | 1695 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1696 | unsigned int tx_pkts; |
1697 | ||
0a0f39df | 1698 | tx_pkts = rte_vhost_enqueue_burst(netdev_dpdk_get_vid(dev), |
c6ec9d17 | 1699 | vhost_qid, cur_pkts, cnt); |
95e9881f KT |
1700 | if (OVS_LIKELY(tx_pkts)) { |
1701 | /* Packets have been sent.*/ | |
1702 | cnt -= tx_pkts; | |
31871ee3 | 1703 | /* Prepare for possible retry.*/ |
95e9881f KT |
1704 | cur_pkts = &cur_pkts[tx_pkts]; |
1705 | } else { | |
31871ee3 KT |
1706 | /* No packets sent - do not retry.*/ |
1707 | break; | |
95e9881f | 1708 | } |
c6ec9d17 | 1709 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1710 | |
d46285a2 | 1711 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1712 | |
d46285a2 | 1713 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 1714 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 1715 | cnt + dropped); |
d46285a2 | 1716 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1717 | |
1718 | out: | |
c6ec9d17 | 1719 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 1720 | dp_packet_delete(pkts[i]); |
58397e6c KT |
1721 | } |
1722 | } | |
1723 | ||
8a9562d2 PS |
1724 | /* Tx function. Transmit packets indefinitely */ |
1725 | static void | |
64839cf4 | 1726 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1727 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1728 | { |
bce01e3a | 1729 | #if !defined(__CHECKER__) && !defined(_WIN32) |
64839cf4 | 1730 | const size_t PKT_ARRAY_SIZE = batch->count; |
bce01e3a EJ |
1731 | #else |
1732 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1733 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1734 | #endif |
8a9562d2 | 1735 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 1736 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
175cf4de RW |
1737 | int dropped = 0; |
1738 | int newcnt = 0; | |
1739 | int i; | |
8a9562d2 | 1740 | |
7d6d1a40 WT |
1741 | dp_packet_batch_apply_cutlen(batch); |
1742 | ||
64839cf4 WT |
1743 | for (i = 0; i < batch->count; i++) { |
1744 | int size = dp_packet_size(batch->packets[i]); | |
95fb793a | 1745 | |
f98d7864 | 1746 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c | 1747 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
58be5c0e | 1748 | (int) size, dev->max_packet_len); |
f4fd623c | 1749 | |
175cf4de | 1750 | dropped++; |
f4fd623c DDP |
1751 | continue; |
1752 | } | |
8a9562d2 | 1753 | |
2391135c | 1754 | pkts[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1755 | |
2391135c | 1756 | if (!pkts[newcnt]) { |
64839cf4 | 1757 | dropped += batch->count - i; |
175cf4de | 1758 | break; |
f4fd623c DDP |
1759 | } |
1760 | ||
1761 | /* We have to do a copy for now */ | |
2391135c | 1762 | memcpy(rte_pktmbuf_mtod(pkts[newcnt], void *), |
64839cf4 | 1763 | dp_packet_data(batch->packets[i]), size); |
f4fd623c | 1764 | |
2391135c MK |
1765 | rte_pktmbuf_data_len(pkts[newcnt]) = size; |
1766 | rte_pktmbuf_pkt_len(pkts[newcnt]) = size; | |
f4fd623c DDP |
1767 | |
1768 | newcnt++; | |
1769 | } | |
8a9562d2 | 1770 | |
58397e6c | 1771 | if (dev->type == DPDK_DEV_VHOST) { |
2391135c | 1772 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, |
dd52de45 | 1773 | newcnt); |
58397e6c | 1774 | } else { |
0bf765f7 IS |
1775 | unsigned int qos_pkts = newcnt; |
1776 | ||
1777 | /* Check if QoS has been configured for this netdev. */ | |
78bd47cf | 1778 | newcnt = netdev_dpdk_qos_run(dev, pkts, newcnt); |
0bf765f7 IS |
1779 | |
1780 | dropped += qos_pkts - newcnt; | |
819f13bd | 1781 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, newcnt); |
58397e6c | 1782 | } |
db73f716 | 1783 | |
0bf765f7 IS |
1784 | if (OVS_UNLIKELY(dropped)) { |
1785 | rte_spinlock_lock(&dev->stats_lock); | |
1786 | dev->stats.tx_dropped += dropped; | |
1787 | rte_spinlock_unlock(&dev->stats_lock); | |
1788 | } | |
8a9562d2 PS |
1789 | } |
1790 | ||
58397e6c | 1791 | static int |
64839cf4 WT |
1792 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1793 | struct dp_packet_batch *batch, | |
324c8374 | 1794 | bool may_steal, bool concurrent_txq OVS_UNUSED) |
58397e6c | 1795 | { |
58397e6c | 1796 | |
dd52de45 | 1797 | if (OVS_UNLIKELY(!may_steal || batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 WT |
1798 | dpdk_do_tx_copy(netdev, qid, batch); |
1799 | dp_packet_delete_batch(batch, may_steal); | |
58397e6c | 1800 | } else { |
64839cf4 | 1801 | dp_packet_batch_apply_cutlen(batch); |
dd52de45 | 1802 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
1803 | } |
1804 | return 0; | |
1805 | } | |
1806 | ||
7251515e DV |
1807 | static inline void |
1808 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
324c8374 IM |
1809 | struct dp_packet_batch *batch, bool may_steal, |
1810 | bool concurrent_txq) | |
8a9562d2 | 1811 | { |
3b1fb077 DDP |
1812 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
1813 | dp_packet_delete_batch(batch, may_steal); | |
1814 | return; | |
1815 | } | |
1816 | ||
324c8374 | 1817 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 1818 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
1819 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
1820 | } | |
1821 | ||
7251515e | 1822 | if (OVS_UNLIKELY(!may_steal || |
64839cf4 | 1823 | batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1824 | struct netdev *netdev = &dev->up; |
1825 | ||
64839cf4 WT |
1826 | dpdk_do_tx_copy(netdev, qid, batch); |
1827 | dp_packet_delete_batch(batch, may_steal); | |
8a9562d2 | 1828 | } else { |
6b094bf4 | 1829 | int dropped; |
64839cf4 | 1830 | int cnt = batch->count; |
2391135c | 1831 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 1832 | |
7d6d1a40 WT |
1833 | dp_packet_batch_apply_cutlen(batch); |
1834 | ||
2391135c | 1835 | cnt = netdev_dpdk_filter_packet_len(dev, pkts, cnt); |
78bd47cf | 1836 | cnt = netdev_dpdk_qos_run(dev, pkts, cnt); |
6b094bf4 | 1837 | dropped = batch->count - cnt; |
1b99bb05 | 1838 | |
819f13bd | 1839 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, cnt); |
8a9562d2 | 1840 | |
f4fd623c | 1841 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1842 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1843 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1844 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1845 | } |
8a9562d2 | 1846 | } |
a0cb2d66 | 1847 | |
324c8374 | 1848 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
1849 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
1850 | } | |
7251515e DV |
1851 | } |
1852 | ||
1853 | static int | |
1854 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
324c8374 IM |
1855 | struct dp_packet_batch *batch, bool may_steal, |
1856 | bool concurrent_txq) | |
7251515e DV |
1857 | { |
1858 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1859 | |
324c8374 | 1860 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e | 1861 | return 0; |
8a9562d2 PS |
1862 | } |
1863 | ||
1864 | static int | |
74ff3298 | 1865 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1866 | { |
1867 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1868 | ||
1869 | ovs_mutex_lock(&dev->mutex); | |
1870 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1871 | dev->hwaddr = mac; |
045c0d1a | 1872 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1873 | } |
1874 | ovs_mutex_unlock(&dev->mutex); | |
1875 | ||
1876 | return 0; | |
1877 | } | |
1878 | ||
1879 | static int | |
74ff3298 | 1880 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1881 | { |
1882 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1883 | ||
1884 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1885 | *mac = dev->hwaddr; |
8a9562d2 PS |
1886 | ovs_mutex_unlock(&dev->mutex); |
1887 | ||
1888 | return 0; | |
1889 | } | |
1890 | ||
1891 | static int | |
1892 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1893 | { | |
1894 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1895 | ||
1896 | ovs_mutex_lock(&dev->mutex); | |
1897 | *mtup = dev->mtu; | |
1898 | ovs_mutex_unlock(&dev->mutex); | |
1899 | ||
1900 | return 0; | |
1901 | } | |
1902 | ||
0072e931 MK |
1903 | static int |
1904 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
1905 | { | |
1906 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1907 | ||
1908 | if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
1909 | || mtu < ETHER_MIN_MTU) { | |
1910 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
1911 | return EINVAL; | |
1912 | } | |
1913 | ||
1914 | ovs_mutex_lock(&dev->mutex); | |
1915 | if (dev->requested_mtu != mtu) { | |
1916 | dev->requested_mtu = mtu; | |
1917 | netdev_request_reconfigure(netdev); | |
1918 | } | |
1919 | ovs_mutex_unlock(&dev->mutex); | |
1920 | ||
1921 | return 0; | |
1922 | } | |
1923 | ||
8a9562d2 | 1924 | static int |
d46285a2 | 1925 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 1926 | |
58397e6c KT |
1927 | static int |
1928 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1929 | struct netdev_stats *stats) | |
1930 | { | |
1931 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1932 | ||
1933 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 1934 | |
45d947c4 | 1935 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
1936 | /* Supported Stats */ |
1937 | stats->rx_packets += dev->stats.rx_packets; | |
1938 | stats->tx_packets += dev->stats.tx_packets; | |
9509913a | 1939 | stats->rx_dropped = dev->stats.rx_dropped; |
58397e6c | 1940 | stats->tx_dropped += dev->stats.tx_dropped; |
9e3ddd45 TP |
1941 | stats->multicast = dev->stats.multicast; |
1942 | stats->rx_bytes = dev->stats.rx_bytes; | |
1943 | stats->tx_bytes = dev->stats.tx_bytes; | |
1944 | stats->rx_errors = dev->stats.rx_errors; | |
1945 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 1946 | |
1947 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
1948 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
1949 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
1950 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
1951 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
1952 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
1953 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
1954 | ||
45d947c4 | 1955 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 1956 | |
58397e6c KT |
1957 | ovs_mutex_unlock(&dev->mutex); |
1958 | ||
1959 | return 0; | |
1960 | } | |
1961 | ||
d6e3feb5 | 1962 | static void |
1963 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
1964 | const struct rte_eth_xstat *xstats, |
1965 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 1966 | const unsigned int size) |
1967 | { | |
d6e3feb5 | 1968 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 1969 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1970 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1971 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1972 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1973 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1974 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1975 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1976 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1977 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1978 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1979 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1980 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 1981 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1982 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 1983 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1984 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1985 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1986 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1987 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1988 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1989 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1990 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1991 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1992 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1993 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1994 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 1995 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1996 | stats->tx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 1997 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1998 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 1999 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2000 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 2001 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 2002 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 2003 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2004 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 2005 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2006 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 2007 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 2008 | stats->rx_jabber_errors = xstats[i].value; |
2009 | } | |
2010 | } | |
2011 | } | |
2012 | ||
8a9562d2 PS |
2013 | static int |
2014 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2015 | { | |
2016 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2017 | struct rte_eth_stats rte_stats; | |
2018 | bool gg; | |
2019 | ||
2020 | netdev_dpdk_get_carrier(netdev, &gg); | |
2021 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2022 | |
0a0f39df CL |
2023 | struct rte_eth_xstat *rte_xstats = NULL; |
2024 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2025 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2026 | |
2027 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
2028 | VLOG_ERR("Can't get ETH statistics for port: %i.", dev->port_id); | |
f9256822 | 2029 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2030 | return EPROTO; |
2031 | } | |
2032 | ||
0a0f39df CL |
2033 | /* Get length of statistics */ |
2034 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2035 | if (rte_xstats_len < 0) { | |
2036 | VLOG_WARN("Cannot get XSTATS values for port: %i", dev->port_id); | |
2037 | goto out; | |
2038 | } | |
2039 | /* Reserve memory for xstats names and values */ | |
2040 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2041 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2042 | ||
2043 | /* Retreive xstats names */ | |
2044 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2045 | rte_xstats_names, | |
2046 | rte_xstats_len); | |
2047 | if (rte_xstats_new_len != rte_xstats_len) { | |
2048 | VLOG_WARN("Cannot get XSTATS names for port: %i.", dev->port_id); | |
2049 | goto out; | |
2050 | } | |
2051 | /* Retreive xstats values */ | |
2052 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2053 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2054 | rte_xstats_len); | |
2055 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2056 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2057 | rte_xstats_len); | |
d6e3feb5 | 2058 | } else { |
0a0f39df | 2059 | VLOG_WARN("Cannot get XSTATS values for port: %i.", dev->port_id); |
d6e3feb5 | 2060 | } |
8a9562d2 | 2061 | |
0a0f39df CL |
2062 | out: |
2063 | free(rte_xstats); | |
2064 | free(rte_xstats_names); | |
2065 | ||
2f9dd77f PS |
2066 | stats->rx_packets = rte_stats.ipackets; |
2067 | stats->tx_packets = rte_stats.opackets; | |
2068 | stats->rx_bytes = rte_stats.ibytes; | |
2069 | stats->tx_bytes = rte_stats.obytes; | |
9e3ddd45 TP |
2070 | /* DPDK counts imissed as errors, but count them here as dropped instead */ |
2071 | stats->rx_errors = rte_stats.ierrors - rte_stats.imissed; | |
2f9dd77f | 2072 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2073 | |
45d947c4 | 2074 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2075 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2076 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2077 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2078 | |
2079 | /* These are the available DPDK counters for packets not received due to | |
2080 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2081 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2082 | stats->rx_missed_errors = rte_stats.imissed; |
2083 | ||
8a9562d2 PS |
2084 | ovs_mutex_unlock(&dev->mutex); |
2085 | ||
2086 | return 0; | |
2087 | } | |
2088 | ||
2089 | static int | |
d46285a2 | 2090 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2091 | enum netdev_features *current, |
ca3d4f55 BX |
2092 | enum netdev_features *advertised, |
2093 | enum netdev_features *supported, | |
2094 | enum netdev_features *peer) | |
8a9562d2 | 2095 | { |
d46285a2 | 2096 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2097 | struct rte_eth_link link; |
2098 | ||
2099 | ovs_mutex_lock(&dev->mutex); | |
2100 | link = dev->link; | |
2101 | ovs_mutex_unlock(&dev->mutex); | |
2102 | ||
362ca396 | 2103 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2104 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
2105 | *current = NETDEV_F_10MB_HD; |
2106 | } | |
362ca396 | 2107 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2108 | *current = NETDEV_F_100MB_HD; |
2109 | } | |
362ca396 | 2110 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2111 | *current = NETDEV_F_1GB_HD; |
2112 | } | |
2113 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 2114 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
2115 | *current = NETDEV_F_10MB_FD; |
2116 | } | |
362ca396 | 2117 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
2118 | *current = NETDEV_F_100MB_FD; |
2119 | } | |
362ca396 | 2120 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
2121 | *current = NETDEV_F_1GB_FD; |
2122 | } | |
362ca396 | 2123 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
2124 | *current = NETDEV_F_10GB_FD; |
2125 | } | |
2126 | } | |
2127 | ||
362ca396 | 2128 | if (link.link_autoneg) { |
2129 | *current |= NETDEV_F_AUTONEG; | |
2130 | } | |
2131 | ||
ca3d4f55 BX |
2132 | *advertised = *supported = *peer = 0; |
2133 | ||
8a9562d2 PS |
2134 | return 0; |
2135 | } | |
2136 | ||
9509913a IS |
2137 | static struct ingress_policer * |
2138 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2139 | { | |
2140 | struct ingress_policer *policer = NULL; | |
2141 | uint64_t rate_bytes; | |
2142 | uint64_t burst_bytes; | |
2143 | int err = 0; | |
2144 | ||
2145 | policer = xmalloc(sizeof *policer); | |
2146 | rte_spinlock_init(&policer->policer_lock); | |
2147 | ||
2148 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
2149 | rate_bytes = rate * 1000/8; | |
2150 | burst_bytes = burst * 1000/8; | |
2151 | ||
2152 | policer->app_srtcm_params.cir = rate_bytes; | |
2153 | policer->app_srtcm_params.cbs = burst_bytes; | |
2154 | policer->app_srtcm_params.ebs = 0; | |
2155 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2156 | &policer->app_srtcm_params); | |
58be5c0e | 2157 | if (err) { |
9509913a IS |
2158 | VLOG_ERR("Could not create rte meter for ingress policer"); |
2159 | return NULL; | |
2160 | } | |
2161 | ||
2162 | return policer; | |
2163 | } | |
2164 | ||
2165 | static int | |
2166 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2167 | uint32_t policer_burst) | |
2168 | { | |
2169 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2170 | struct ingress_policer *policer; | |
2171 | ||
2172 | /* Force to 0 if no rate specified, | |
2173 | * default to 8000 kbits if burst is 0, | |
2174 | * else stick with user-specified value. | |
2175 | */ | |
2176 | policer_burst = (!policer_rate ? 0 | |
2177 | : !policer_burst ? 8000 | |
2178 | : policer_burst); | |
2179 | ||
2180 | ovs_mutex_lock(&dev->mutex); | |
2181 | ||
2182 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2183 | &dev->ingress_policer); | |
2184 | ||
2185 | if (dev->policer_rate == policer_rate && | |
2186 | dev->policer_burst == policer_burst) { | |
2187 | /* Assume that settings haven't changed since we last set them. */ | |
2188 | ovs_mutex_unlock(&dev->mutex); | |
2189 | return 0; | |
2190 | } | |
2191 | ||
2192 | /* Destroy any existing ingress policer for the device if one exists */ | |
2193 | if (policer) { | |
2194 | ovsrcu_postpone(free, policer); | |
2195 | } | |
2196 | ||
2197 | if (policer_rate != 0) { | |
2198 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2199 | } else { | |
2200 | policer = NULL; | |
2201 | } | |
2202 | ovsrcu_set(&dev->ingress_policer, policer); | |
2203 | dev->policer_rate = policer_rate; | |
2204 | dev->policer_burst = policer_burst; | |
2205 | ovs_mutex_unlock(&dev->mutex); | |
2206 | ||
2207 | return 0; | |
2208 | } | |
2209 | ||
8a9562d2 PS |
2210 | static int |
2211 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2212 | { | |
2213 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2214 | int ifindex; | |
2215 | ||
2216 | ovs_mutex_lock(&dev->mutex); | |
2217 | ifindex = dev->port_id; | |
2218 | ovs_mutex_unlock(&dev->mutex); | |
2219 | ||
2220 | return ifindex; | |
2221 | } | |
2222 | ||
2223 | static int | |
d46285a2 | 2224 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2225 | { |
d46285a2 | 2226 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2227 | |
2228 | ovs_mutex_lock(&dev->mutex); | |
2229 | check_link_status(dev); | |
2230 | *carrier = dev->link.link_status; | |
58397e6c KT |
2231 | |
2232 | ovs_mutex_unlock(&dev->mutex); | |
2233 | ||
2234 | return 0; | |
2235 | } | |
2236 | ||
2237 | static int | |
d46285a2 | 2238 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2239 | { |
d46285a2 | 2240 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2241 | |
2242 | ovs_mutex_lock(&dev->mutex); | |
2243 | ||
0a0f39df | 2244 | if (is_vhost_running(dev)) { |
58397e6c KT |
2245 | *carrier = 1; |
2246 | } else { | |
2247 | *carrier = 0; | |
2248 | } | |
2249 | ||
8a9562d2 PS |
2250 | ovs_mutex_unlock(&dev->mutex); |
2251 | ||
2252 | return 0; | |
2253 | } | |
2254 | ||
2255 | static long long int | |
d46285a2 | 2256 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2257 | { |
d46285a2 | 2258 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2259 | long long int carrier_resets; |
2260 | ||
2261 | ovs_mutex_lock(&dev->mutex); | |
2262 | carrier_resets = dev->link_reset_cnt; | |
2263 | ovs_mutex_unlock(&dev->mutex); | |
2264 | ||
2265 | return carrier_resets; | |
2266 | } | |
2267 | ||
2268 | static int | |
d46285a2 | 2269 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2270 | long long int interval OVS_UNUSED) |
2271 | { | |
ee32150e | 2272 | return EOPNOTSUPP; |
8a9562d2 PS |
2273 | } |
2274 | ||
2275 | static int | |
2276 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2277 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2278 | enum netdev_flags *old_flagsp) |
2279 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 2280 | { |
8a9562d2 PS |
2281 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
2282 | return EINVAL; | |
2283 | } | |
2284 | ||
2285 | *old_flagsp = dev->flags; | |
2286 | dev->flags |= on; | |
2287 | dev->flags &= ~off; | |
2288 | ||
2289 | if (dev->flags == *old_flagsp) { | |
2290 | return 0; | |
2291 | } | |
2292 | ||
58397e6c | 2293 | if (dev->type == DPDK_DEV_ETH) { |
58397e6c KT |
2294 | if (dev->flags & NETDEV_PROMISC) { |
2295 | rte_eth_promiscuous_enable(dev->port_id); | |
2296 | } | |
8a9562d2 | 2297 | |
314fb5ad | 2298 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
2299 | } else { |
2300 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2301 | * running then change netdev's change_seq to trigger link state | |
2302 | * update. */ | |
e543851d ZB |
2303 | |
2304 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2305 | && is_vhost_running(dev)) { |
e543851d ZB |
2306 | netdev_change_seq_changed(&dev->up); |
2307 | ||
2308 | /* Clear statistics if device is getting up. */ | |
2309 | if (NETDEV_UP & on) { | |
2310 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 2311 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
2312 | rte_spinlock_unlock(&dev->stats_lock); |
2313 | } | |
2314 | } | |
8a9562d2 PS |
2315 | } |
2316 | ||
2317 | return 0; | |
2318 | } | |
2319 | ||
2320 | static int | |
d46285a2 | 2321 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2322 | enum netdev_flags off, enum netdev_flags on, |
2323 | enum netdev_flags *old_flagsp) | |
2324 | { | |
d46285a2 | 2325 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2326 | int error; |
2327 | ||
d46285a2 DDP |
2328 | ovs_mutex_lock(&dev->mutex); |
2329 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2330 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2331 | |
2332 | return error; | |
2333 | } | |
2334 | ||
2335 | static int | |
d46285a2 | 2336 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2337 | { |
d46285a2 | 2338 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2339 | struct rte_eth_dev_info dev_info; |
2340 | ||
7cd1261d | 2341 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2342 | return ENODEV; |
7cd1261d | 2343 | } |
8a9562d2 PS |
2344 | |
2345 | ovs_mutex_lock(&dev->mutex); | |
2346 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2347 | ovs_mutex_unlock(&dev->mutex); | |
2348 | ||
95fb793a | 2349 | smap_add_format(args, "port_no", "%d", dev->port_id); |
58be5c0e MK |
2350 | smap_add_format(args, "numa_id", "%d", |
2351 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
2352 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
2353 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2354 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2355 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2356 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2357 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
2358 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
2359 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
2360 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
2361 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2362 | ||
39c2baa9 | 2363 | if (dev_info.pci_dev) { |
2364 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2365 | dev_info.pci_dev->id.vendor_id); | |
2366 | smap_add_format(args, "pci-device_id", "0x%x", | |
2367 | dev_info.pci_dev->id.device_id); | |
2368 | } | |
8a9562d2 PS |
2369 | |
2370 | return 0; | |
2371 | } | |
2372 | ||
2373 | static void | |
2374 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2375 | OVS_REQUIRES(dev->mutex) | |
2376 | { | |
2377 | enum netdev_flags old_flags; | |
2378 | ||
2379 | if (admin_state) { | |
2380 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2381 | } else { | |
2382 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2383 | } | |
2384 | } | |
2385 | ||
2386 | static void | |
2387 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2388 | const char *argv[], void *aux OVS_UNUSED) | |
2389 | { | |
2390 | bool up; | |
2391 | ||
2392 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2393 | up = true; | |
2394 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2395 | up = false; | |
2396 | } else { | |
2397 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2398 | return; | |
2399 | } | |
2400 | ||
2401 | if (argc > 2) { | |
2402 | struct netdev *netdev = netdev_from_name(argv[1]); | |
2403 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
2404 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
2405 | ||
2406 | ovs_mutex_lock(&dpdk_dev->mutex); | |
2407 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
2408 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
2409 | ||
2410 | netdev_close(netdev); | |
2411 | } else { | |
2412 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2413 | netdev_close(netdev); | |
2414 | return; | |
2415 | } | |
2416 | } else { | |
2417 | struct netdev_dpdk *netdev; | |
2418 | ||
2419 | ovs_mutex_lock(&dpdk_mutex); | |
2420 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
2421 | ovs_mutex_lock(&netdev->mutex); | |
2422 | netdev_dpdk_set_admin_state__(netdev, up); | |
2423 | ovs_mutex_unlock(&netdev->mutex); | |
2424 | } | |
2425 | ovs_mutex_unlock(&dpdk_mutex); | |
2426 | } | |
2427 | unixctl_command_reply(conn, "OK"); | |
2428 | } | |
2429 | ||
b8374d0d MV |
2430 | static void |
2431 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
2432 | const char *argv[], void *aux OVS_UNUSED) | |
2433 | { | |
2434 | int ret; | |
2435 | char *response; | |
b8374d0d MV |
2436 | uint8_t port_id; |
2437 | char devname[RTE_ETH_NAME_MAX_LEN]; | |
55e075e6 | 2438 | struct netdev_dpdk *dev; |
b8374d0d MV |
2439 | |
2440 | ovs_mutex_lock(&dpdk_mutex); | |
2441 | ||
55e075e6 CL |
2442 | if (!rte_eth_dev_count() || rte_eth_dev_get_port_by_name(argv[1], |
2443 | &port_id)) { | |
2444 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); | |
2445 | goto error; | |
2446 | } | |
b8374d0d | 2447 | |
55e075e6 CL |
2448 | dev = netdev_dpdk_lookup_by_port_id(port_id); |
2449 | if (dev) { | |
2450 | response = xasprintf("Device '%s' is being used by interface '%s'. " | |
2451 | "Remove it before detaching", | |
2452 | argv[1], netdev_get_name(&dev->up)); | |
b8374d0d MV |
2453 | goto error; |
2454 | } | |
2455 | ||
2456 | rte_eth_dev_close(port_id); | |
2457 | ||
2458 | ret = rte_eth_dev_detach(port_id, devname); | |
2459 | if (ret < 0) { | |
55e075e6 | 2460 | response = xasprintf("Device '%s' can not be detached", argv[1]); |
b8374d0d MV |
2461 | goto error; |
2462 | } | |
2463 | ||
55e075e6 | 2464 | response = xasprintf("Device '%s' has been detached", argv[1]); |
b8374d0d MV |
2465 | |
2466 | ovs_mutex_unlock(&dpdk_mutex); | |
2467 | unixctl_command_reply(conn, response); | |
2468 | free(response); | |
2469 | return; | |
2470 | ||
2471 | error: | |
2472 | ovs_mutex_unlock(&dpdk_mutex); | |
2473 | unixctl_command_reply_error(conn, response); | |
2474 | free(response); | |
2475 | } | |
2476 | ||
58397e6c KT |
2477 | /* |
2478 | * Set virtqueue flags so that we do not receive interrupts. | |
2479 | */ | |
2480 | static void | |
0a0f39df | 2481 | set_irq_status(int vid) |
58397e6c | 2482 | { |
4573fbd3 FL |
2483 | uint32_t i; |
2484 | uint64_t idx; | |
2485 | ||
0a0f39df | 2486 | for (i = 0; i < rte_vhost_get_queue_num(vid); i++) { |
4573fbd3 | 2487 | idx = i * VIRTIO_QNUM; |
0a0f39df CL |
2488 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_RXQ, 0); |
2489 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_TXQ, 0); | |
4573fbd3 FL |
2490 | } |
2491 | } | |
2492 | ||
585a5bea IM |
2493 | /* |
2494 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2495 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2496 | */ |
2497 | static void | |
d46285a2 DDP |
2498 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2499 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2500 | { |
2501 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2502 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 2503 | |
eff23640 | 2504 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
2505 | |
2506 | for (i = 0; i < total_txqs; i++) { | |
2507 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2508 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2509 | enabled_queues[n_enabled++] = i; |
2510 | } | |
2511 | } | |
2512 | ||
2513 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2514 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2515 | n_enabled = 1; |
2516 | } | |
2517 | ||
2518 | k = 0; | |
2519 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2520 | if (dev->tx_q[i].map != i) { |
2521 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2522 | k = (k + 1) % n_enabled; |
2523 | } | |
2524 | } | |
2525 | ||
2d24d165 | 2526 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2527 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2528 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2529 | } |
2530 | ||
eff23640 | 2531 | free(enabled_queues); |
585a5bea | 2532 | } |
4573fbd3 | 2533 | |
58397e6c KT |
2534 | /* |
2535 | * A new virtio-net device is added to a vhost port. | |
2536 | */ | |
2537 | static int | |
0a0f39df | 2538 | new_device(int vid) |
58397e6c | 2539 | { |
d46285a2 | 2540 | struct netdev_dpdk *dev; |
58397e6c | 2541 | bool exists = false; |
db8f13b0 | 2542 | int newnode = 0; |
0a0f39df CL |
2543 | char ifname[IF_NAME_SZ]; |
2544 | ||
58be5c0e | 2545 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2546 | |
2547 | ovs_mutex_lock(&dpdk_mutex); | |
2548 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2549 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 2550 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2551 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
0a0f39df | 2552 | uint32_t qp_num = rte_vhost_get_queue_num(vid); |
db8f13b0 CL |
2553 | |
2554 | /* Get NUMA information */ | |
0a0f39df CL |
2555 | newnode = rte_vhost_get_numa_node(vid); |
2556 | if (newnode == -1) { | |
5b9bf9e0 | 2557 | #ifdef VHOST_NUMA |
db8f13b0 | 2558 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2559 | ifname); |
5b9bf9e0 | 2560 | #endif |
db8f13b0 | 2561 | newnode = dev->socket_id; |
db8f13b0 CL |
2562 | } |
2563 | ||
7f5f2bd0 IM |
2564 | if (dev->requested_n_txq != qp_num |
2565 | || dev->requested_n_rxq != qp_num | |
2566 | || dev->requested_socket_id != newnode) { | |
2567 | dev->requested_socket_id = newnode; | |
2568 | dev->requested_n_rxq = qp_num; | |
2569 | dev->requested_n_txq = qp_num; | |
2570 | netdev_request_reconfigure(&dev->up); | |
2571 | } else { | |
2572 | /* Reconfiguration not required. */ | |
2573 | dev->vhost_reconfigured = true; | |
2574 | } | |
81acebda | 2575 | |
0a0f39df | 2576 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2577 | exists = true; |
2578 | ||
58397e6c | 2579 | /* Disable notifications. */ |
0a0f39df | 2580 | set_irq_status(vid); |
e543851d | 2581 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2582 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2583 | break; |
2584 | } | |
c1ff66ac | 2585 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2586 | } |
2587 | ovs_mutex_unlock(&dpdk_mutex); | |
2588 | ||
2589 | if (!exists) { | |
0a0f39df | 2590 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2591 | |
2592 | return -1; | |
2593 | } | |
2594 | ||
0a0f39df CL |
2595 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2596 | ifname, newnode); | |
2597 | ||
58397e6c KT |
2598 | return 0; |
2599 | } | |
2600 | ||
f3ea2ad2 IM |
2601 | /* Clears mapping for all available queues of vhost interface. */ |
2602 | static void | |
2603 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2604 | OVS_REQUIRES(dev->mutex) | |
2605 | { | |
2606 | int i; | |
2607 | ||
81acebda | 2608 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
2609 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
2610 | } | |
2611 | } | |
2612 | ||
58397e6c KT |
2613 | /* |
2614 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2615 | * flag to stop any more packets from being sent or received to/from a VM and | |
2616 | * ensure all currently queued packets have been sent/received before removing | |
2617 | * the device. | |
2618 | */ | |
2619 | static void | |
0a0f39df | 2620 | destroy_device(int vid) |
58397e6c | 2621 | { |
d46285a2 | 2622 | struct netdev_dpdk *dev; |
afee281f | 2623 | bool exists = false; |
0a0f39df CL |
2624 | char ifname[IF_NAME_SZ]; |
2625 | ||
58be5c0e | 2626 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
2627 | |
2628 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2629 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2630 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 2631 | |
d46285a2 | 2632 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
2633 | dev->vhost_reconfigured = false; |
2634 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 2635 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 2636 | |
e543851d | 2637 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2638 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 2639 | exists = true; |
afee281f | 2640 | break; |
58397e6c KT |
2641 | } |
2642 | } | |
afee281f | 2643 | |
58397e6c KT |
2644 | ovs_mutex_unlock(&dpdk_mutex); |
2645 | ||
0a0f39df | 2646 | if (exists) { |
afee281f KT |
2647 | /* |
2648 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2649 | * to NULL, before returning. | |
2650 | */ | |
2651 | ovsrcu_synchronize(); | |
2652 | /* | |
2653 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2654 | * put thread back into quiescent state before returning. | |
2655 | */ | |
2656 | ovsrcu_quiesce_start(); | |
0a0f39df | 2657 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 2658 | } else { |
0a0f39df | 2659 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 2660 | } |
58397e6c KT |
2661 | } |
2662 | ||
585a5bea | 2663 | static int |
0a0f39df | 2664 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 2665 | { |
d46285a2 | 2666 | struct netdev_dpdk *dev; |
585a5bea IM |
2667 | bool exists = false; |
2668 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
2669 | char ifname[IF_NAME_SZ]; |
2670 | ||
58be5c0e | 2671 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea IM |
2672 | |
2673 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2674 | return 0; | |
2675 | } | |
2676 | ||
2677 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2678 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 2679 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2680 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 2681 | if (enable) { |
d46285a2 | 2682 | dev->tx_q[qid].map = qid; |
585a5bea | 2683 | } else { |
d46285a2 | 2684 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2685 | } |
d46285a2 | 2686 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2687 | exists = true; |
d46285a2 | 2688 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2689 | break; |
2690 | } | |
c1ff66ac | 2691 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2692 | } |
2693 | ovs_mutex_unlock(&dpdk_mutex); | |
2694 | ||
2695 | if (exists) { | |
0a0f39df CL |
2696 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
2697 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 2698 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 2699 | } else { |
0a0f39df | 2700 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
2701 | return -1; |
2702 | } | |
2703 | ||
2704 | return 0; | |
2705 | } | |
2706 | ||
0a0f39df CL |
2707 | int |
2708 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 2709 | { |
0a0f39df | 2710 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
2711 | } |
2712 | ||
9509913a IS |
2713 | struct ingress_policer * |
2714 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
2715 | { | |
2716 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
2717 | } | |
2718 | ||
58397e6c KT |
2719 | /* |
2720 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
2721 | * configuration has been fully complete. | |
2722 | */ | |
bce01e3a | 2723 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
2724 | { |
2725 | .new_device = new_device, | |
2726 | .destroy_device = destroy_device, | |
585a5bea | 2727 | .vring_state_changed = vring_state_changed |
58397e6c KT |
2728 | }; |
2729 | ||
2730 | static void * | |
7d1ced01 | 2731 | start_vhost_loop(void *dummy OVS_UNUSED) |
58397e6c KT |
2732 | { |
2733 | pthread_detach(pthread_self()); | |
41987644 | 2734 | /* Put the vhost thread into quiescent state. */ |
618f44f7 | 2735 | ovsrcu_quiesce_start(); |
58397e6c KT |
2736 | rte_vhost_driver_session_start(); |
2737 | return NULL; | |
2738 | } | |
2739 | ||
2740 | static int | |
ecc1a34e | 2741 | netdev_dpdk_class_init(void) |
7d1ced01 | 2742 | { |
ecc1a34e DDP |
2743 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
2744 | ||
2745 | /* This function can be called for different classes. The initialization | |
2746 | * needs to be done only once */ | |
2747 | if (ovsthread_once_start(&once)) { | |
2748 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
2749 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2750 | "[netdev] up|down", 1, 2, | |
2751 | netdev_dpdk_set_admin_state, NULL); | |
b8374d0d | 2752 | unixctl_command_register("netdev-dpdk/detach", |
55e075e6 | 2753 | "pci address of device", 1, 1, |
b8374d0d | 2754 | netdev_dpdk_detach, NULL); |
ecc1a34e DDP |
2755 | |
2756 | ovsthread_once_done(&once); | |
2757 | } | |
362ca396 | 2758 | |
7d1ced01 CL |
2759 | return 0; |
2760 | } | |
2761 | ||
ecc1a34e DDP |
2762 | static int |
2763 | netdev_dpdk_vhost_class_init(void) | |
033e9df2 | 2764 | { |
ecc1a34e | 2765 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
033e9df2 | 2766 | |
ecc1a34e DDP |
2767 | /* This function can be called for different classes. The initialization |
2768 | * needs to be done only once */ | |
2769 | if (ovsthread_once_start(&once)) { | |
2770 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
2771 | rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4 | |
2772 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
04de404e CL |
2773 | | 1ULL << VIRTIO_NET_F_CSUM |
2774 | | 1ULL << VIRTIO_RING_F_INDIRECT_DESC); | |
ecc1a34e DDP |
2775 | ovs_thread_create("vhost_thread", start_vhost_loop, NULL); |
2776 | ||
2777 | ovsthread_once_done(&once); | |
2778 | } | |
2779 | ||
2780 | return 0; | |
033e9df2 DDP |
2781 | } |
2782 | ||
95fb793a | 2783 | /* Client Rings */ |
2784 | ||
95fb793a | 2785 | static int |
2786 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
2787 | unsigned int *eth_port_id) | |
2788 | { | |
48fffdee | 2789 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 2790 | char *ring_name; |
b8374d0d | 2791 | int port_id; |
95fb793a | 2792 | |
48fffdee KT |
2793 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
2794 | if (!ring_pair) { | |
95fb793a | 2795 | return ENOMEM; |
2796 | } | |
2797 | ||
7251515e | 2798 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 2799 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 2800 | |
8f0a76c9 | 2801 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 2802 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2803 | RING_F_SP_ENQ); |
0c6f39e5 | 2804 | free(ring_name); |
48fffdee KT |
2805 | if (ring_pair->cring_tx == NULL) { |
2806 | rte_free(ring_pair); | |
95fb793a | 2807 | return ENOMEM; |
2808 | } | |
2809 | ||
0c6f39e5 | 2810 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 2811 | |
8f0a76c9 | 2812 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 2813 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2814 | RING_F_SC_DEQ); |
0c6f39e5 | 2815 | free(ring_name); |
48fffdee KT |
2816 | if (ring_pair->cring_rx == NULL) { |
2817 | rte_free(ring_pair); | |
95fb793a | 2818 | return ENOMEM; |
2819 | } | |
2820 | ||
b8374d0d MV |
2821 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
2822 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 2823 | |
b8374d0d | 2824 | if (port_id < 0) { |
48fffdee | 2825 | rte_free(ring_pair); |
95fb793a | 2826 | return ENODEV; |
2827 | } | |
2828 | ||
48fffdee | 2829 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
2830 | ring_pair->eth_port_id = port_id; |
2831 | *eth_port_id = port_id; | |
2832 | ||
48fffdee | 2833 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 2834 | |
95fb793a | 2835 | return 0; |
2836 | } | |
2837 | ||
2838 | static int | |
64839cf4 WT |
2839 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) |
2840 | OVS_REQUIRES(dpdk_mutex) | |
95fb793a | 2841 | { |
48fffdee | 2842 | struct dpdk_ring *ring_pair; |
95fb793a | 2843 | unsigned int port_no; |
2844 | int err = 0; | |
2845 | ||
2846 | /* Names always start with "dpdkr" */ | |
2847 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2848 | if (err) { | |
2849 | return err; | |
2850 | } | |
2851 | ||
58be5c0e | 2852 | /* Look through our list to find the device */ |
48fffdee KT |
2853 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
2854 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 2855 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 2856 | /* Really all that is needed */ |
48fffdee | 2857 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 2858 | return 0; |
2859 | } | |
2860 | } | |
2861 | /* Need to create the device rings */ | |
2862 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2863 | } | |
2864 | ||
7251515e | 2865 | static int |
d46285a2 | 2866 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
324c8374 IM |
2867 | struct dp_packet_batch *batch, bool may_steal, |
2868 | bool concurrent_txq) | |
7251515e | 2869 | { |
d46285a2 | 2870 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b99bb05 MG |
2871 | unsigned i; |
2872 | ||
58be5c0e MK |
2873 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
2874 | * the rss hash field is clear. This is because the same mbuf may be | |
2875 | * modified by the consumer of the ring and return into the datapath | |
2876 | * without recalculating the RSS hash. */ | |
64839cf4 WT |
2877 | for (i = 0; i < batch->count; i++) { |
2878 | dp_packet_rss_invalidate(batch->packets[i]); | |
1b99bb05 | 2879 | } |
7251515e | 2880 | |
324c8374 | 2881 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e DV |
2882 | return 0; |
2883 | } | |
2884 | ||
95fb793a | 2885 | static int |
2886 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2887 | { | |
2888 | unsigned int port_no = 0; | |
2889 | int err = 0; | |
2890 | ||
95fb793a | 2891 | ovs_mutex_lock(&dpdk_mutex); |
2892 | ||
2893 | err = dpdk_ring_open(netdev->name, &port_no); | |
2894 | if (err) { | |
2895 | goto unlock_dpdk; | |
2896 | } | |
2897 | ||
1ce30dfd DDP |
2898 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
2899 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 2900 | unlock_dpdk: |
2901 | ovs_mutex_unlock(&dpdk_mutex); | |
2902 | return err; | |
2903 | } | |
2904 | ||
0bf765f7 IS |
2905 | /* QoS Functions */ |
2906 | ||
2907 | /* | |
2908 | * Initialize QoS configuration operations. | |
2909 | */ | |
2910 | static void | |
2911 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
2912 | { | |
2913 | conf->ops = ops; | |
78bd47cf | 2914 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
2915 | } |
2916 | ||
2917 | /* | |
2918 | * Search existing QoS operations in qos_ops and compare each set of | |
2919 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
2920 | * else return NULL | |
2921 | */ | |
2922 | static const struct dpdk_qos_ops * | |
2923 | qos_lookup_name(const char *name) | |
2924 | { | |
2925 | const struct dpdk_qos_ops *const *opsp; | |
2926 | ||
2927 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2928 | const struct dpdk_qos_ops *ops = *opsp; | |
2929 | if (!strcmp(name, ops->qos_name)) { | |
2930 | return ops; | |
2931 | } | |
2932 | } | |
2933 | return NULL; | |
2934 | } | |
2935 | ||
0bf765f7 IS |
2936 | static int |
2937 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
2938 | struct sset *types) | |
2939 | { | |
2940 | const struct dpdk_qos_ops *const *opsp; | |
2941 | ||
2942 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2943 | const struct dpdk_qos_ops *ops = *opsp; | |
2944 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
2945 | sset_add(types, ops->qos_name); | |
2946 | } | |
2947 | } | |
2948 | return 0; | |
2949 | } | |
2950 | ||
2951 | static int | |
d46285a2 | 2952 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
2953 | const char **typep, struct smap *details) |
2954 | { | |
d46285a2 | 2955 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 2956 | struct qos_conf *qos_conf; |
0bf765f7 IS |
2957 | int error = 0; |
2958 | ||
d46285a2 | 2959 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
2960 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
2961 | if (qos_conf) { | |
2962 | *typep = qos_conf->ops->qos_name; | |
2963 | error = (qos_conf->ops->qos_get | |
2964 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
2965 | } else { |
2966 | /* No QoS configuration set, return an empty string */ | |
2967 | *typep = ""; | |
0bf765f7 | 2968 | } |
d46285a2 | 2969 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2970 | |
2971 | return error; | |
2972 | } | |
2973 | ||
2974 | static int | |
78bd47cf DDP |
2975 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
2976 | const struct smap *details) | |
0bf765f7 | 2977 | { |
d46285a2 | 2978 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 2979 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 2980 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
2981 | int error = 0; |
2982 | ||
d46285a2 | 2983 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 2984 | |
78bd47cf | 2985 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 2986 | |
78bd47cf DDP |
2987 | new_ops = qos_lookup_name(type); |
2988 | ||
2989 | if (!new_ops || !new_ops->qos_construct) { | |
2990 | new_qos_conf = NULL; | |
2991 | if (type && type[0]) { | |
2992 | error = EOPNOTSUPP; | |
0bf765f7 | 2993 | } |
44975bb0 | 2994 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
2995 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
2996 | new_qos_conf = qos_conf; | |
0bf765f7 | 2997 | } else { |
78bd47cf | 2998 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
2999 | } |
3000 | ||
7ea266e9 | 3001 | if (error) { |
78bd47cf DDP |
3002 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
3003 | type, netdev->name, rte_strerror(error)); | |
3004 | } | |
3005 | ||
3006 | if (new_qos_conf != qos_conf) { | |
3007 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
3008 | if (qos_conf) { | |
3009 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
3010 | } | |
0bf765f7 IS |
3011 | } |
3012 | ||
d46285a2 | 3013 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 3014 | |
0bf765f7 IS |
3015 | return error; |
3016 | } | |
3017 | ||
3018 | /* egress-policer details */ | |
3019 | ||
3020 | struct egress_policer { | |
3021 | struct qos_conf qos_conf; | |
3022 | struct rte_meter_srtcm_params app_srtcm_params; | |
3023 | struct rte_meter_srtcm egress_meter; | |
3024 | }; | |
3025 | ||
78bd47cf DDP |
3026 | static void |
3027 | egress_policer_details_to_param(const struct smap *details, | |
3028 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 3029 | { |
78bd47cf DDP |
3030 | memset(params, 0, sizeof *params); |
3031 | params->cir = smap_get_ullong(details, "cir", 0); | |
3032 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
3033 | params->ebs = 0; | |
0bf765f7 IS |
3034 | } |
3035 | ||
3036 | static int | |
78bd47cf DDP |
3037 | egress_policer_qos_construct(const struct smap *details, |
3038 | struct qos_conf **conf) | |
0bf765f7 | 3039 | { |
0bf765f7 | 3040 | struct egress_policer *policer; |
0bf765f7 IS |
3041 | int err = 0; |
3042 | ||
0bf765f7 IS |
3043 | policer = xmalloc(sizeof *policer); |
3044 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 3045 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
0bf765f7 | 3046 | err = rte_meter_srtcm_config(&policer->egress_meter, |
78bd47cf DDP |
3047 | &policer->app_srtcm_params); |
3048 | if (!err) { | |
3049 | *conf = &policer->qos_conf; | |
3050 | } else { | |
7ea266e9 | 3051 | free(policer); |
78bd47cf | 3052 | *conf = NULL; |
7ea266e9 IS |
3053 | err = -err; |
3054 | } | |
0bf765f7 IS |
3055 | |
3056 | return err; | |
3057 | } | |
3058 | ||
3059 | static void | |
78bd47cf | 3060 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
3061 | { |
3062 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 3063 | qos_conf); |
0bf765f7 IS |
3064 | free(policer); |
3065 | } | |
3066 | ||
3067 | static int | |
78bd47cf | 3068 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 3069 | { |
78bd47cf DDP |
3070 | struct egress_policer *policer = |
3071 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3072 | ||
3073 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
3074 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 3075 | |
0bf765f7 IS |
3076 | return 0; |
3077 | } | |
3078 | ||
78bd47cf DDP |
3079 | static bool |
3080 | egress_policer_qos_is_equal(const struct qos_conf *conf, const struct smap *details) | |
0bf765f7 | 3081 | { |
78bd47cf DDP |
3082 | struct egress_policer *policer = |
3083 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
3084 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 3085 | |
78bd47cf | 3086 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 3087 | |
78bd47cf | 3088 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
3089 | } |
3090 | ||
0bf765f7 | 3091 | static int |
78bd47cf | 3092 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt) |
0bf765f7 | 3093 | { |
0bf765f7 | 3094 | int cnt = 0; |
78bd47cf DDP |
3095 | struct egress_policer *policer = |
3096 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 3097 | |
f3926f29 | 3098 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, pkt_cnt); |
0bf765f7 IS |
3099 | |
3100 | return cnt; | |
3101 | } | |
3102 | ||
3103 | static const struct dpdk_qos_ops egress_policer_ops = { | |
3104 | "egress-policer", /* qos_name */ | |
3105 | egress_policer_qos_construct, | |
3106 | egress_policer_qos_destruct, | |
3107 | egress_policer_qos_get, | |
78bd47cf | 3108 | egress_policer_qos_is_equal, |
0bf765f7 IS |
3109 | egress_policer_run |
3110 | }; | |
3111 | ||
050c60bf DDP |
3112 | static int |
3113 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
3114 | { | |
3115 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3116 | int err = 0; | |
3117 | ||
050c60bf DDP |
3118 | ovs_mutex_lock(&dev->mutex); |
3119 | ||
3120 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 3121 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b CL |
3122 | && dev->mtu == dev->requested_mtu |
3123 | && dev->rxq_size == dev->requested_rxq_size | |
bd4e172b | 3124 | && dev->txq_size == dev->requested_txq_size |
3125 | && dev->socket_id == dev->requested_socket_id) { | |
050c60bf DDP |
3126 | /* Reconfiguration is unnecessary */ |
3127 | ||
3128 | goto out; | |
3129 | } | |
3130 | ||
3131 | rte_eth_dev_stop(dev->port_id); | |
3132 | ||
bd4e172b | 3133 | if (dev->mtu != dev->requested_mtu |
3134 | || dev->socket_id != dev->requested_socket_id) { | |
0072e931 MK |
3135 | netdev_dpdk_mempool_configure(dev); |
3136 | } | |
3137 | ||
050c60bf DDP |
3138 | netdev->n_txq = dev->requested_n_txq; |
3139 | netdev->n_rxq = dev->requested_n_rxq; | |
3140 | ||
b685696b CL |
3141 | dev->rxq_size = dev->requested_rxq_size; |
3142 | dev->txq_size = dev->requested_txq_size; | |
3143 | ||
050c60bf DDP |
3144 | rte_free(dev->tx_q); |
3145 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
3146 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
3147 | if (!dev->tx_q) { | |
3148 | err = ENOMEM; | |
3149 | } | |
050c60bf | 3150 | |
0072e931 MK |
3151 | netdev_change_seq_changed(netdev); |
3152 | ||
050c60bf | 3153 | out: |
050c60bf | 3154 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
3155 | return err; |
3156 | } | |
3157 | ||
7f381c2e | 3158 | static int |
2d24d165 | 3159 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 3160 | OVS_REQUIRES(dev->mutex) |
050c60bf | 3161 | { |
2d24d165 CL |
3162 | dev->up.n_txq = dev->requested_n_txq; |
3163 | dev->up.n_rxq = dev->requested_n_rxq; | |
050c60bf | 3164 | |
81acebda IM |
3165 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
3166 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
3167 | dev->tx_q[0].map = 0; | |
3168 | } | |
3169 | ||
3170 | netdev_dpdk_remap_txqs(dev); | |
3171 | ||
0072e931 MK |
3172 | if (dev->requested_socket_id != dev->socket_id |
3173 | || dev->requested_mtu != dev->mtu) { | |
3174 | if (!netdev_dpdk_mempool_configure(dev)) { | |
2d24d165 | 3175 | netdev_change_seq_changed(&dev->up); |
db8f13b0 CL |
3176 | } |
3177 | } | |
3178 | ||
7f381c2e DDP |
3179 | if (!dev->dpdk_mp) { |
3180 | return ENOMEM; | |
3181 | } | |
3182 | ||
0a0f39df CL |
3183 | if (netdev_dpdk_get_vid(dev) >= 0) { |
3184 | dev->vhost_reconfigured = true; | |
81acebda | 3185 | } |
7f381c2e DDP |
3186 | |
3187 | return 0; | |
2d24d165 CL |
3188 | } |
3189 | ||
3190 | static int | |
3191 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
3192 | { | |
3193 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3194 | int err; |
2d24d165 | 3195 | |
2d24d165 | 3196 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 3197 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 3198 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
3199 | |
3200 | return err; | |
2d24d165 CL |
3201 | } |
3202 | ||
3203 | static int | |
3204 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3205 | { | |
3206 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 3207 | int err; |
2d24d165 | 3208 | |
2d24d165 CL |
3209 | ovs_mutex_lock(&dev->mutex); |
3210 | ||
c1ff66ac CL |
3211 | /* Configure vHost client mode if requested and if the following criteria |
3212 | * are met: | |
2d24d165 CL |
3213 | * 1. Device hasn't been registered yet. |
3214 | * 2. A path has been specified. | |
c1ff66ac CL |
3215 | */ |
3216 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 CL |
3217 | && strlen(dev->vhost_id)) { |
3218 | /* Register client-mode device */ | |
3219 | err = rte_vhost_driver_register(dev->vhost_id, | |
3220 | RTE_VHOST_USER_CLIENT); | |
c1ff66ac | 3221 | if (err) { |
2d24d165 CL |
3222 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3223 | dev->vhost_id); | |
7f381c2e | 3224 | goto unlock; |
c1ff66ac | 3225 | } else { |
2d24d165 CL |
3226 | /* Configuration successful */ |
3227 | dev->vhost_driver_flags |= RTE_VHOST_USER_CLIENT; | |
3228 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " | |
3229 | "using client socket '%s'", | |
3230 | dev->up.name, dev->vhost_id); | |
c1ff66ac CL |
3231 | } |
3232 | } | |
3233 | ||
7f381c2e DDP |
3234 | err = dpdk_vhost_reconfigure_helper(dev); |
3235 | ||
3236 | unlock: | |
050c60bf | 3237 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 3238 | |
7f381c2e | 3239 | return err; |
050c60bf DDP |
3240 | } |
3241 | ||
ecc1a34e | 3242 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
81acebda IM |
3243 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
3244 | GET_CARRIER, GET_STATS, \ | |
3245 | GET_FEATURES, GET_STATUS, \ | |
3246 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3247 | { \ |
3248 | NAME, \ | |
118c77b1 | 3249 | true, /* is_pmd */ \ |
ecc1a34e | 3250 | INIT, /* init */ \ |
95fb793a | 3251 | NULL, /* netdev_dpdk_run */ \ |
3252 | NULL, /* netdev_dpdk_wait */ \ | |
3253 | \ | |
3254 | netdev_dpdk_alloc, \ | |
3255 | CONSTRUCT, \ | |
58397e6c | 3256 | DESTRUCT, \ |
95fb793a | 3257 | netdev_dpdk_dealloc, \ |
3258 | netdev_dpdk_get_config, \ | |
81acebda | 3259 | SET_CONFIG, \ |
95fb793a | 3260 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3261 | NULL, /* build header */ \ |
3262 | NULL, /* push header */ \ | |
3263 | NULL, /* pop header */ \ | |
7dec44fe | 3264 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3265 | SET_TX_MULTIQ, \ |
95fb793a | 3266 | \ |
7251515e | 3267 | SEND, /* send */ \ |
95fb793a | 3268 | NULL, /* send_wait */ \ |
3269 | \ | |
3270 | netdev_dpdk_set_etheraddr, \ | |
3271 | netdev_dpdk_get_etheraddr, \ | |
3272 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3273 | netdev_dpdk_set_mtu, \ |
95fb793a | 3274 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3275 | GET_CARRIER, \ |
95fb793a | 3276 | netdev_dpdk_get_carrier_resets, \ |
3277 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
3278 | GET_STATS, \ |
3279 | GET_FEATURES, \ | |
95fb793a | 3280 | NULL, /* set_advertisements */ \ |
3281 | \ | |
9509913a | 3282 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3283 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3284 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3285 | netdev_dpdk_get_qos, \ |
3286 | netdev_dpdk_set_qos, \ | |
95fb793a | 3287 | NULL, /* get_queue */ \ |
3288 | NULL, /* set_queue */ \ | |
3289 | NULL, /* delete_queue */ \ | |
3290 | NULL, /* get_queue_stats */ \ | |
3291 | NULL, /* queue_dump_start */ \ | |
3292 | NULL, /* queue_dump_next */ \ | |
3293 | NULL, /* queue_dump_done */ \ | |
3294 | NULL, /* dump_queue_stats */ \ | |
3295 | \ | |
95fb793a | 3296 | NULL, /* set_in4 */ \ |
a8704b50 | 3297 | NULL, /* get_addr_list */ \ |
95fb793a | 3298 | NULL, /* add_router */ \ |
3299 | NULL, /* get_next_hop */ \ | |
58397e6c | 3300 | GET_STATUS, \ |
95fb793a | 3301 | NULL, /* arp_lookup */ \ |
3302 | \ | |
3303 | netdev_dpdk_update_flags, \ | |
050c60bf | 3304 | RECONFIGURE, \ |
95fb793a | 3305 | \ |
3306 | netdev_dpdk_rxq_alloc, \ | |
3307 | netdev_dpdk_rxq_construct, \ | |
3308 | netdev_dpdk_rxq_destruct, \ | |
3309 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3310 | RXQ_RECV, \ |
95fb793a | 3311 | NULL, /* rx_wait */ \ |
3312 | NULL, /* rxq_drain */ \ | |
3313 | } | |
8a9562d2 | 3314 | |
bce01e3a | 3315 | static const struct netdev_class dpdk_class = |
95fb793a | 3316 | NETDEV_DPDK_CLASS( |
3317 | "dpdk", | |
ecc1a34e | 3318 | netdev_dpdk_class_init, |
5496878c | 3319 | netdev_dpdk_construct, |
58397e6c | 3320 | netdev_dpdk_destruct, |
81acebda IM |
3321 | netdev_dpdk_set_config, |
3322 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3323 | netdev_dpdk_eth_send, |
3324 | netdev_dpdk_get_carrier, | |
3325 | netdev_dpdk_get_stats, | |
3326 | netdev_dpdk_get_features, | |
3327 | netdev_dpdk_get_status, | |
050c60bf | 3328 | netdev_dpdk_reconfigure, |
58397e6c | 3329 | netdev_dpdk_rxq_recv); |
95fb793a | 3330 | |
bce01e3a | 3331 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3332 | NETDEV_DPDK_CLASS( |
3333 | "dpdkr", | |
ecc1a34e | 3334 | netdev_dpdk_class_init, |
5496878c | 3335 | netdev_dpdk_ring_construct, |
58397e6c | 3336 | netdev_dpdk_destruct, |
c3d062a7 | 3337 | netdev_dpdk_ring_set_config, |
81acebda | 3338 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3339 | netdev_dpdk_ring_send, |
3340 | netdev_dpdk_get_carrier, | |
3341 | netdev_dpdk_get_stats, | |
3342 | netdev_dpdk_get_features, | |
3343 | netdev_dpdk_get_status, | |
050c60bf | 3344 | netdev_dpdk_reconfigure, |
58397e6c KT |
3345 | netdev_dpdk_rxq_recv); |
3346 | ||
53f50d24 | 3347 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3348 | NETDEV_DPDK_CLASS( |
3349 | "dpdkvhostuser", | |
ecc1a34e | 3350 | netdev_dpdk_vhost_class_init, |
53f50d24 | 3351 | netdev_dpdk_vhost_construct, |
58397e6c | 3352 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3353 | NULL, |
81acebda | 3354 | NULL, |
58397e6c KT |
3355 | netdev_dpdk_vhost_send, |
3356 | netdev_dpdk_vhost_get_carrier, | |
3357 | netdev_dpdk_vhost_get_stats, | |
3358 | NULL, | |
7251515e | 3359 | NULL, |
53f50d24 | 3360 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3361 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3362 | static const struct netdev_class dpdk_vhost_client_class = |
3363 | NETDEV_DPDK_CLASS( | |
3364 | "dpdkvhostuserclient", | |
ecc1a34e | 3365 | netdev_dpdk_vhost_class_init, |
2d24d165 CL |
3366 | netdev_dpdk_vhost_client_construct, |
3367 | netdev_dpdk_vhost_destruct, | |
3368 | netdev_dpdk_vhost_client_set_config, | |
3369 | NULL, | |
3370 | netdev_dpdk_vhost_send, | |
3371 | netdev_dpdk_vhost_get_carrier, | |
3372 | netdev_dpdk_vhost_get_stats, | |
3373 | NULL, | |
3374 | NULL, | |
3375 | netdev_dpdk_vhost_client_reconfigure, | |
3376 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 3377 | |
8a9562d2 PS |
3378 | void |
3379 | netdev_dpdk_register(void) | |
3380 | { | |
bab69409 AC |
3381 | netdev_register_provider(&dpdk_class); |
3382 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3383 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 3384 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 3385 | } |