]>
Commit | Line | Data |
---|---|---|
8a9562d2 | 1 | /* |
12d0d124 | 2 | * Copyright (c) 2014, 2015, 2016, 2017 Nicira, Inc. |
8a9562d2 PS |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
01961bbd | 18 | #include "netdev-dpdk.h" |
8a9562d2 | 19 | |
6ebc4b09 | 20 | #include <errno.h> |
8a9562d2 PS |
21 | #include <signal.h> |
22 | #include <stdlib.h> | |
6ebc4b09 | 23 | #include <string.h> |
8a9562d2 | 24 | #include <unistd.h> |
f3e7ec25 MW |
25 | #include <linux/virtio_net.h> |
26 | #include <sys/socket.h> | |
27 | #include <linux/if.h> | |
01961bbd | 28 | |
5e925ccc | 29 | #include <rte_bus_pci.h> |
01961bbd DDP |
30 | #include <rte_config.h> |
31 | #include <rte_cycles.h> | |
32 | #include <rte_errno.h> | |
33 | #include <rte_eth_ring.h> | |
34 | #include <rte_ethdev.h> | |
6ebc4b09 | 35 | #include <rte_flow.h> |
01961bbd DDP |
36 | #include <rte_malloc.h> |
37 | #include <rte_mbuf.h> | |
38 | #include <rte_meter.h> | |
fc56f5e0 | 39 | #include <rte_pci.h> |
3eb8d4fa | 40 | #include <rte_version.h> |
6ebc4b09 | 41 | #include <rte_vhost.h> |
8a9562d2 | 42 | |
e8a2b5bf | 43 | #include "cmap.h" |
7d1ced01 | 44 | #include "dirs.h" |
e14deea0 | 45 | #include "dp-packet.h" |
01961bbd | 46 | #include "dpdk.h" |
8a9562d2 | 47 | #include "dpif-netdev.h" |
e5c0f5a4 | 48 | #include "fatal-signal.h" |
8a9562d2 PS |
49 | #include "netdev-provider.h" |
50 | #include "netdev-vport.h" | |
51 | #include "odp-util.h" | |
eac84432 | 52 | #include "openvswitch/dynamic-string.h" |
25d436fb | 53 | #include "openvswitch/list.h" |
6ebc4b09 | 54 | #include "openvswitch/match.h" |
25d436fb | 55 | #include "openvswitch/ofp-print.h" |
6ebc4b09 | 56 | #include "openvswitch/shash.h" |
25d436fb | 57 | #include "openvswitch/vlog.h" |
94143fc4 | 58 | #include "ovs-numa.h" |
8a9562d2 | 59 | #include "ovs-rcu.h" |
6ebc4b09 | 60 | #include "ovs-thread.h" |
8a9562d2 | 61 | #include "packets.h" |
0bf765f7 | 62 | #include "smap.h" |
8a9562d2 | 63 | #include "sset.h" |
8a9562d2 | 64 | #include "timeval.h" |
6ebc4b09 | 65 | #include "unaligned.h" |
8a9562d2 | 66 | #include "unixctl.h" |
6ebc4b09 IM |
67 | #include "util.h" |
68 | #include "uuid.h" | |
8a9562d2 | 69 | |
f3e7ec25 MW |
70 | enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM}; |
71 | ||
05b49df6 | 72 | VLOG_DEFINE_THIS_MODULE(netdev_dpdk); |
8a9562d2 PS |
73 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); |
74 | ||
75 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
76 | ||
77 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
78 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
79 | ||
80 | /* | |
81 | * need to reserve tons of extra space in the mbufs so we can align the | |
82 | * DMA addresses to 4KB. | |
18f777b2 TP |
83 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
84 | * performance for standard Ethernet MTU. | |
8a9562d2 | 85 | */ |
58be5c0e MK |
86 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN \ |
87 | + (2 * VLAN_HEADER_LEN)) | |
4be4d22c MK |
88 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) |
89 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
58be5c0e MK |
90 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len) \ |
91 | - ETHER_HDR_LEN - ETHER_CRC_LEN) | |
4be4d22c | 92 | #define NETDEV_DPDK_MBUF_ALIGN 1024 |
0072e931 | 93 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 94 | |
43307ad0 IS |
95 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
96 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
97 | * enough hugepages) we keep halving the number until the allocation succeeds | |
98 | * or we reach MIN_NB_MBUF */ | |
99 | ||
100 | #define MAX_NB_MBUF (4096 * 64) | |
da79ce2b DDP |
101 | #define MIN_NB_MBUF (4096 * 4) |
102 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
103 | ||
43307ad0 IS |
104 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ |
105 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF / MIN_NB_MBUF) | |
106 | == 0); | |
107 | ||
108 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
109 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
110 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF / MIN_NB_MBUF)) | |
111 | % MP_CACHE_SZ == 0); | |
112 | ||
8a9562d2 PS |
113 | #define SOCKET0 0 |
114 | ||
b685696b CL |
115 | /* Default size of Physical NIC RXQ */ |
116 | #define NIC_PORT_DEFAULT_RXQ_SIZE 2048 | |
117 | /* Default size of Physical NIC TXQ */ | |
118 | #define NIC_PORT_DEFAULT_TXQ_SIZE 2048 | |
119 | /* Maximum size of Physical NIC Queues */ | |
120 | #define NIC_PORT_MAX_Q_SIZE 4096 | |
79f5354c | 121 | |
585a5bea | 122 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
123 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
124 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
125 | * yet mapped to another queue. */ | |
585a5bea | 126 | |
bb37956a IM |
127 | #define DPDK_ETH_PORT_ID_INVALID RTE_MAX_ETHPORTS |
128 | ||
5e925ccc MK |
129 | /* DPDK library uses uint16_t for port_id. */ |
130 | typedef uint16_t dpdk_port_t; | |
fa9f4eeb | 131 | #define DPDK_PORT_ID_FMT "%"PRIu16 |
bb37956a | 132 | |
080f080c KT |
133 | /* Minimum amount of vhost tx retries, effectively a disable. */ |
134 | #define VHOST_ENQ_RETRY_MIN 0 | |
135 | /* Maximum amount of vhost tx retries. */ | |
136 | #define VHOST_ENQ_RETRY_MAX 32 | |
137 | /* Legacy default value for vhost tx retries. */ | |
138 | #define VHOST_ENQ_RETRY_DEF 8 | |
139 | ||
0a0f39df | 140 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 141 | |
8a9562d2 | 142 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
143 | .rxmode = { |
144 | .mq_mode = ETH_MQ_RX_RSS, | |
145 | .split_hdr_size = 0, | |
03f3f9c0 | 146 | .offloads = 0, |
a28ddd11 DDP |
147 | }, |
148 | .rx_adv_conf = { | |
149 | .rss_conf = { | |
150 | .rss_key = NULL, | |
543342a4 | 151 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 152 | }, |
a28ddd11 DDP |
153 | }, |
154 | .txmode = { | |
155 | .mq_mode = ETH_MQ_TX_NONE, | |
156 | }, | |
8a9562d2 PS |
157 | }; |
158 | ||
f3e7ec25 MW |
159 | /* |
160 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
161 | * configuration has been fully completed. | |
162 | */ | |
163 | static int new_device(int vid); | |
164 | static void destroy_device(int vid); | |
165 | static int vring_state_changed(int vid, uint16_t queue_id, int enable); | |
61473a0e | 166 | static void destroy_connection(int vid); |
f3e7ec25 MW |
167 | static const struct vhost_device_ops virtio_net_device_ops = |
168 | { | |
169 | .new_device = new_device, | |
170 | .destroy_device = destroy_device, | |
171 | .vring_state_changed = vring_state_changed, | |
61473a0e DM |
172 | .features_changed = NULL, |
173 | .new_connection = NULL, | |
174 | .destroy_connection = destroy_connection, | |
f3e7ec25 MW |
175 | }; |
176 | ||
58f7c37b DDP |
177 | enum { DPDK_RING_SIZE = 256 }; |
178 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
179 | enum { DRAIN_TSC = 200000ULL }; |
180 | ||
58397e6c KT |
181 | enum dpdk_dev_type { |
182 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 183 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
184 | }; |
185 | ||
0bf765f7 IS |
186 | /* Quality of Service */ |
187 | ||
188 | /* An instance of a QoS configuration. Always associated with a particular | |
189 | * network device. | |
190 | * | |
191 | * Each QoS implementation subclasses this with whatever additional data it | |
192 | * needs. | |
193 | */ | |
194 | struct qos_conf { | |
195 | const struct dpdk_qos_ops *ops; | |
78bd47cf | 196 | rte_spinlock_t lock; |
0bf765f7 IS |
197 | }; |
198 | ||
199 | /* A particular implementation of dpdk QoS operations. | |
200 | * | |
201 | * The functions below return 0 if successful or a positive errno value on | |
202 | * failure, except where otherwise noted. All of them must be provided, except | |
203 | * where otherwise noted. | |
204 | */ | |
205 | struct dpdk_qos_ops { | |
206 | ||
207 | /* Name of the QoS type */ | |
208 | const char *qos_name; | |
209 | ||
78bd47cf DDP |
210 | /* Called to construct a qos_conf object. The implementation should make |
211 | * the appropriate calls to configure QoS according to 'details'. | |
0bf765f7 IS |
212 | * |
213 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
214 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
215 | * (which is built as ovs-vswitchd.conf.db(8)). | |
216 | * | |
78bd47cf DDP |
217 | * This function must return 0 if and only if it sets '*conf' to an |
218 | * initialized 'struct qos_conf'. | |
0bf765f7 IS |
219 | * |
220 | * For all QoS implementations it should always be non-null. | |
221 | */ | |
78bd47cf | 222 | int (*qos_construct)(const struct smap *details, struct qos_conf **conf); |
0bf765f7 IS |
223 | |
224 | /* Destroys the data structures allocated by the implementation as part of | |
78bd47cf | 225 | * 'qos_conf'. |
0bf765f7 IS |
226 | * |
227 | * For all QoS implementations it should always be non-null. | |
228 | */ | |
78bd47cf | 229 | void (*qos_destruct)(struct qos_conf *conf); |
0bf765f7 | 230 | |
78bd47cf | 231 | /* Retrieves details of 'conf' configuration into 'details'. |
0bf765f7 IS |
232 | * |
233 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
234 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
235 | * (which is built as ovs-vswitchd.conf.db(8)). | |
236 | */ | |
78bd47cf | 237 | int (*qos_get)(const struct qos_conf *conf, struct smap *details); |
0bf765f7 | 238 | |
78bd47cf | 239 | /* Returns true if 'conf' is already configured according to 'details'. |
0bf765f7 IS |
240 | * |
241 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
242 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
243 | * (which is built as ovs-vswitchd.conf.db(8)). | |
244 | * | |
78bd47cf | 245 | * For all QoS implementations it should always be non-null. |
0bf765f7 | 246 | */ |
78bd47cf DDP |
247 | bool (*qos_is_equal)(const struct qos_conf *conf, |
248 | const struct smap *details); | |
0bf765f7 IS |
249 | |
250 | /* Modify an array of rte_mbufs. The modification is specific to | |
251 | * each qos implementation. | |
252 | * | |
253 | * The function should take and array of mbufs and an int representing | |
254 | * the current number of mbufs present in the array. | |
255 | * | |
256 | * After the function has performed a qos modification to the array of | |
257 | * mbufs it returns an int representing the number of mbufs now present in | |
258 | * the array. This value is can then be passed to the port send function | |
259 | * along with the modified array for transmission. | |
260 | * | |
261 | * For all QoS implementations it should always be non-null. | |
262 | */ | |
78bd47cf | 263 | int (*qos_run)(struct qos_conf *qos_conf, struct rte_mbuf **pkts, |
7d7ded7a | 264 | int pkt_cnt, bool should_steal); |
0bf765f7 IS |
265 | }; |
266 | ||
267 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
268 | static const struct dpdk_qos_ops egress_policer_ops; | |
269 | ||
270 | /* | |
271 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
272 | * operations. | |
273 | */ | |
274 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
275 | &egress_policer_ops, | |
276 | NULL | |
277 | }; | |
278 | ||
c2adb102 IM |
279 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; |
280 | ||
8a9562d2 | 281 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 282 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 283 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 284 | |
c2adb102 IM |
285 | static struct ovs_mutex dpdk_mp_mutex OVS_ACQ_AFTER(dpdk_mutex) |
286 | = OVS_MUTEX_INITIALIZER; | |
287 | ||
91fccdad | 288 | /* Contains all 'struct dpdk_mp's. */ |
43307ad0 IS |
289 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mp_mutex) |
290 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); | |
91fccdad | 291 | |
91fccdad KT |
292 | struct dpdk_mp { |
293 | struct rte_mempool *mp; | |
43307ad0 IS |
294 | int mtu; |
295 | int socket_id; | |
296 | int refcount; | |
91fccdad KT |
297 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mp_mutex); |
298 | }; | |
299 | ||
5a034064 | 300 | /* There should be one 'struct dpdk_tx_queue' created for |
15ba075d | 301 | * each netdev tx queue. */ |
8a9562d2 | 302 | struct dpdk_tx_queue { |
15ba075d IM |
303 | /* Padding to make dpdk_tx_queue exactly one cache line long. */ |
304 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
305 | /* Protects the members and the NIC queue from concurrent access. | |
306 | * It is used only if the queue is shared among different pmd threads | |
307 | * (see 'concurrent_txq'). */ | |
308 | rte_spinlock_t tx_lock; | |
309 | /* Mapping of configured vhost-user queue to enabled by guest. */ | |
310 | int map; | |
311 | ); | |
8a9562d2 PS |
312 | }; |
313 | ||
95fb793a | 314 | /* dpdk has no way to remove dpdk ring ethernet devices |
315 | so we have to keep them around once they've been created | |
316 | */ | |
317 | ||
ca6ba700 | 318 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 319 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 320 | |
321 | struct dpdk_ring { | |
322 | /* For the client rings */ | |
323 | struct rte_ring *cring_tx; | |
324 | struct rte_ring *cring_rx; | |
b83a2df1 | 325 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
bb37956a | 326 | dpdk_port_t eth_port_id; /* ethernet device port id */ |
ca6ba700 | 327 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 328 | }; |
329 | ||
9509913a IS |
330 | struct ingress_policer { |
331 | struct rte_meter_srtcm_params app_srtcm_params; | |
332 | struct rte_meter_srtcm in_policer; | |
03f3f9c0 | 333 | struct rte_meter_srtcm_profile in_prof; |
9509913a IS |
334 | rte_spinlock_t policer_lock; |
335 | }; | |
336 | ||
1a2bb118 SC |
337 | enum dpdk_hw_ol_features { |
338 | NETDEV_RX_CHECKSUM_OFFLOAD = 1 << 0, | |
e10ca8b9 | 339 | NETDEV_RX_HW_CRC_STRIP = 1 << 1, |
03f3f9c0 | 340 | NETDEV_RX_HW_SCATTER = 1 << 2 |
1a2bb118 SC |
341 | }; |
342 | ||
b2e72a9c IM |
343 | /* |
344 | * In order to avoid confusion in variables names, following naming convention | |
345 | * should be used, if possible: | |
346 | * | |
347 | * 'struct netdev' : 'netdev' | |
348 | * 'struct netdev_dpdk' : 'dev' | |
349 | * 'struct netdev_rxq' : 'rxq' | |
350 | * 'struct netdev_rxq_dpdk' : 'rx' | |
351 | * | |
352 | * Example: | |
353 | * struct netdev *netdev = netdev_from_name(name); | |
354 | * struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
355 | * | |
356 | * Also, 'netdev' should be used instead of 'dev->up', where 'netdev' was | |
357 | * already defined. | |
358 | */ | |
359 | ||
8a9562d2 | 360 | struct netdev_dpdk { |
23d4d53f BB |
361 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline0, |
362 | dpdk_port_t port_id; | |
363 | ||
364 | /* If true, device was attached by rte_eth_dev_attach(). */ | |
365 | bool attached; | |
606f6650 EC |
366 | /* If true, rte_eth_dev_start() was successfully called */ |
367 | bool started; | |
23d4d53f BB |
368 | struct eth_addr hwaddr; |
369 | int mtu; | |
370 | int socket_id; | |
371 | int buf_size; | |
372 | int max_packet_len; | |
373 | enum dpdk_dev_type type; | |
374 | enum netdev_flags flags; | |
eaa43581 | 375 | int link_reset_cnt; |
bb9d2623 IM |
376 | union { |
377 | /* Device arguments for dpdk ports. */ | |
378 | char *devargs; | |
379 | /* Identifier used to distinguish vhost devices from each other. */ | |
380 | char *vhost_id; | |
381 | }; | |
23d4d53f BB |
382 | struct dpdk_tx_queue *tx_q; |
383 | struct rte_eth_link link; | |
23d4d53f BB |
384 | ); |
385 | ||
386 | PADDED_MEMBERS_CACHELINE_MARKER(CACHE_LINE_SIZE, cacheline1, | |
387 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
43307ad0 | 388 | struct dpdk_mp *dpdk_mp; |
23d4d53f BB |
389 | |
390 | /* virtio identifier for vhost devices */ | |
391 | ovsrcu_index vid; | |
392 | ||
393 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
394 | bool vhost_reconfigured; | |
080f080c KT |
395 | |
396 | atomic_uint8_t vhost_tx_retries_max; | |
397 | /* 2 pad bytes here. */ | |
23d4d53f BB |
398 | ); |
399 | ||
23d4d53f BB |
400 | PADDED_MEMBERS(CACHE_LINE_SIZE, |
401 | struct netdev up; | |
402 | /* In dpdk_list. */ | |
403 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); | |
404 | ||
405 | /* QoS configuration and lock for the device */ | |
406 | OVSRCU_TYPE(struct qos_conf *) qos_conf; | |
407 | ||
408 | /* Ingress Policer */ | |
409 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
410 | uint32_t policer_rate; | |
411 | uint32_t policer_burst; | |
35c91567 DM |
412 | |
413 | /* Array of vhost rxq states, see vring_state_changed. */ | |
414 | bool *vhost_rxq_enabled; | |
23d4d53f BB |
415 | ); |
416 | ||
417 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
418 | struct netdev_stats stats; | |
c161357d KT |
419 | /* Custom stat for retries when unable to transmit. */ |
420 | uint64_t tx_retries; | |
23d4d53f BB |
421 | /* Protects stats */ |
422 | rte_spinlock_t stats_lock; | |
c161357d | 423 | /* 4 pad bytes here. */ |
23d4d53f BB |
424 | ); |
425 | ||
426 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
427 | /* The following properties cannot be changed when a device is running, | |
428 | * so we remember the request and update them next time | |
429 | * netdev_dpdk*_reconfigure() is called */ | |
430 | int requested_mtu; | |
431 | int requested_n_txq; | |
432 | int requested_n_rxq; | |
433 | int requested_rxq_size; | |
434 | int requested_txq_size; | |
435 | ||
436 | /* Number of rx/tx descriptors for physical devices */ | |
437 | int rxq_size; | |
438 | int txq_size; | |
439 | ||
440 | /* Socket ID detected when vHost device is brought up */ | |
441 | int requested_socket_id; | |
442 | ||
443 | /* Denotes whether vHost port is client/server mode */ | |
444 | uint64_t vhost_driver_flags; | |
445 | ||
446 | /* DPDK-ETH Flow control */ | |
447 | struct rte_eth_fc_conf fc_conf; | |
448 | ||
449 | /* DPDK-ETH hardware offload features, | |
450 | * from the enum set 'dpdk_hw_ol_features' */ | |
451 | uint32_t hw_ol_features; | |
f8b64a61 RM |
452 | |
453 | /* Properties for link state change detection mode. | |
454 | * If lsc_interrupt_mode is set to false, poll mode is used, | |
455 | * otherwise interrupt mode is used. */ | |
456 | bool requested_lsc_interrupt_mode; | |
457 | bool lsc_interrupt_mode; | |
23d4d53f | 458 | ); |
971f4b39 MW |
459 | |
460 | PADDED_MEMBERS(CACHE_LINE_SIZE, | |
461 | /* Names of all XSTATS counters */ | |
462 | struct rte_eth_xstat_name *rte_xstats_names; | |
463 | int rte_xstats_names_size; | |
464 | int rte_xstats_ids_size; | |
465 | uint64_t *rte_xstats_ids; | |
466 | ); | |
8a9562d2 PS |
467 | }; |
468 | ||
469 | struct netdev_rxq_dpdk { | |
470 | struct netdev_rxq up; | |
bb37956a | 471 | dpdk_port_t port_id; |
8a9562d2 PS |
472 | }; |
473 | ||
f3e7ec25 MW |
474 | static void netdev_dpdk_destruct(struct netdev *netdev); |
475 | static void netdev_dpdk_vhost_destruct(struct netdev *netdev); | |
8a9562d2 | 476 | |
ac1a9bb9 IM |
477 | static void netdev_dpdk_clear_xstats(struct netdev_dpdk *dev); |
478 | ||
0a0f39df | 479 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 480 | |
9509913a IS |
481 | struct ingress_policer * |
482 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
483 | ||
8a9562d2 PS |
484 | static bool |
485 | is_dpdk_class(const struct netdev_class *class) | |
486 | { | |
f3e7ec25 MW |
487 | return class->destruct == netdev_dpdk_destruct |
488 | || class->destruct == netdev_dpdk_vhost_destruct; | |
8a9562d2 PS |
489 | } |
490 | ||
4be4d22c MK |
491 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
492 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
493 | * value, insufficient buffers are allocated to accomodate the packet in its | |
494 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
495 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
496 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
58be5c0e MK |
497 | * behaviour, which reduces performance. To prevent this, use a buffer size |
498 | * that is closest to 'mtu', but which satisfies the aforementioned criteria. | |
4be4d22c MK |
499 | */ |
500 | static uint32_t | |
501 | dpdk_buf_size(int mtu) | |
502 | { | |
a32bab26 TL |
503 | return ROUND_UP(MTU_TO_MAX_FRAME_LEN(mtu), NETDEV_DPDK_MBUF_ALIGN) |
504 | + RTE_PKTMBUF_HEADROOM; | |
4be4d22c MK |
505 | } |
506 | ||
eff23640 DDP |
507 | /* Allocates an area of 'sz' bytes from DPDK. The memory is zero'ed. |
508 | * | |
509 | * Unlike xmalloc(), this function can return NULL on failure. */ | |
8a9562d2 PS |
510 | static void * |
511 | dpdk_rte_mzalloc(size_t sz) | |
512 | { | |
eff23640 | 513 | return rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); |
8a9562d2 PS |
514 | } |
515 | ||
516 | void | |
e14deea0 | 517 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 518 | { |
db73f716 | 519 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 520 | |
b00b4a81 | 521 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
522 | } |
523 | ||
b3cd9f9d | 524 | static void |
401b70d6 | 525 | ovs_rte_pktmbuf_init(struct rte_mempool *mp OVS_UNUSED, |
b3cd9f9d | 526 | void *opaque_arg OVS_UNUSED, |
2391135c | 527 | void *_p, |
b3cd9f9d PS |
528 | unsigned i OVS_UNUSED) |
529 | { | |
2391135c | 530 | struct rte_mbuf *pkt = _p; |
b3cd9f9d | 531 | |
3aaa6201 | 532 | dp_packet_init_dpdk((struct dp_packet *) pkt); |
b3cd9f9d PS |
533 | } |
534 | ||
91fccdad KT |
535 | static int |
536 | dpdk_mp_full(const struct rte_mempool *mp) OVS_REQUIRES(dpdk_mp_mutex) | |
537 | { | |
1f84a2d5 KT |
538 | /* At this point we want to know if all the mbufs are back |
539 | * in the mempool. rte_mempool_full() is not atomic but it's | |
540 | * the best available and as we are no longer requesting mbufs | |
541 | * from the mempool, it means mbufs will not move from | |
542 | * 'mempool ring' --> 'mempool cache'. In rte_mempool_full() | |
543 | * the ring is counted before caches, so we won't get false | |
544 | * positives in this use case and we handle false negatives. | |
545 | * | |
546 | * If future implementations of rte_mempool_full() were to change | |
547 | * it could be possible for a false positive. Even that would | |
548 | * likely be ok, as there are additional checks during mempool | |
549 | * freeing but it would make things racey. | |
91fccdad | 550 | */ |
1f84a2d5 | 551 | return rte_mempool_full(mp); |
91fccdad KT |
552 | } |
553 | ||
554 | /* Free unused mempools. */ | |
555 | static void | |
43307ad0 | 556 | dpdk_mp_sweep(void) OVS_REQUIRES(dpdk_mp_mutex) |
91fccdad KT |
557 | { |
558 | struct dpdk_mp *dmp, *next; | |
559 | ||
43307ad0 IS |
560 | LIST_FOR_EACH_SAFE (dmp, next, list_node, &dpdk_mp_list) { |
561 | if (!dmp->refcount && dpdk_mp_full(dmp->mp)) { | |
91fccdad KT |
562 | VLOG_DBG("Freeing mempool \"%s\"", dmp->mp->name); |
563 | ovs_list_remove(&dmp->list_node); | |
564 | rte_mempool_free(dmp->mp); | |
565 | rte_free(dmp); | |
566 | } | |
567 | } | |
91fccdad KT |
568 | } |
569 | ||
43307ad0 IS |
570 | /* Calculating the required number of mbufs differs depending on the |
571 | * mempool model being used. Check if per port memory is in use before | |
572 | * calculating. | |
573 | */ | |
574 | static uint32_t | |
575 | dpdk_calculate_mbufs(struct netdev_dpdk *dev, int mtu, bool per_port_mp) | |
91fccdad | 576 | { |
43307ad0 | 577 | uint32_t n_mbufs; |
91fccdad | 578 | |
43307ad0 IS |
579 | if (!per_port_mp) { |
580 | /* Shared memory are being used. | |
581 | * XXX: this is a really rough method of provisioning memory. | |
582 | * It's impossible to determine what the exact memory requirements are | |
583 | * when the number of ports and rxqs that utilize a particular mempool | |
584 | * can change dynamically at runtime. For now, use this rough | |
585 | * heurisitic. | |
586 | */ | |
587 | if (mtu >= ETHER_MTU) { | |
588 | n_mbufs = MAX_NB_MBUF; | |
589 | } else { | |
590 | n_mbufs = MIN_NB_MBUF; | |
91fccdad | 591 | } |
43307ad0 IS |
592 | } else { |
593 | /* Per port memory is being used. | |
594 | * XXX: rough estimation of number of mbufs required for this port: | |
595 | * <packets required to fill the device rxqs> | |
596 | * + <packets that could be stuck on other ports txqs> | |
597 | * + <packets in the pmd threads> | |
598 | * + <additional memory for corner cases> | |
599 | */ | |
600 | n_mbufs = dev->requested_n_rxq * dev->requested_rxq_size | |
601 | + dev->requested_n_txq * dev->requested_txq_size | |
602 | + MIN(RTE_MAX_LCORE, dev->requested_n_rxq) * NETDEV_MAX_BURST | |
603 | + MIN_NB_MBUF; | |
91fccdad | 604 | } |
43307ad0 IS |
605 | |
606 | return n_mbufs; | |
91fccdad KT |
607 | } |
608 | ||
43307ad0 IS |
609 | static struct dpdk_mp * |
610 | dpdk_mp_create(struct netdev_dpdk *dev, int mtu, bool per_port_mp) | |
8a9562d2 | 611 | { |
24e78f93 IM |
612 | char mp_name[RTE_MEMPOOL_NAMESIZE]; |
613 | const char *netdev_name = netdev_get_name(&dev->up); | |
614 | int socket_id = dev->requested_socket_id; | |
dfaf00e8 MK |
615 | uint32_t n_mbufs = 0; |
616 | uint32_t mbuf_size = 0; | |
617 | uint32_t aligned_mbuf_size = 0; | |
618 | uint32_t mbuf_priv_data_len = 0; | |
619 | uint32_t pkt_size = 0; | |
24e78f93 | 620 | uint32_t hash = hash_string(netdev_name, 0); |
43307ad0 IS |
621 | struct dpdk_mp *dmp = NULL; |
622 | int ret; | |
623 | ||
624 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
625 | if (!dmp) { | |
626 | return NULL; | |
627 | } | |
628 | dmp->socket_id = socket_id; | |
629 | dmp->mtu = mtu; | |
630 | dmp->refcount = 1; | |
631 | ||
dfaf00e8 MK |
632 | /* Get the size of each mbuf, based on the MTU */ |
633 | mbuf_size = MTU_TO_FRAME_LEN(mtu); | |
634 | ||
43307ad0 | 635 | n_mbufs = dpdk_calculate_mbufs(dev, mtu, per_port_mp); |
d555d9bd | 636 | |
da79ce2b | 637 | do { |
24e78f93 | 638 | /* Full DPDK memory pool name must be unique and cannot be |
43307ad0 IS |
639 | * longer than RTE_MEMPOOL_NAMESIZE. Note that for the shared |
640 | * mempool case this can result in one device using a mempool | |
641 | * which references a different device in it's name. However as | |
642 | * mempool names are hashed, the device name will not be readable | |
643 | * so this is not an issue for tasks such as debugging. | |
644 | */ | |
645 | ret = snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, | |
dfaf00e8 MK |
646 | "ovs%08x%02d%05d%07u", |
647 | hash, socket_id, mtu, n_mbufs); | |
24e78f93 IM |
648 | if (ret < 0 || ret >= RTE_MEMPOOL_NAMESIZE) { |
649 | VLOG_DBG("snprintf returned %d. " | |
650 | "Failed to generate a mempool name for \"%s\". " | |
651 | "Hash:0x%x, socket_id: %d, mtu:%d, mbufs:%u.", | |
652 | ret, netdev_name, hash, socket_id, mtu, n_mbufs); | |
653 | break; | |
65056fd7 | 654 | } |
95fb793a | 655 | |
dfaf00e8 MK |
656 | VLOG_DBG("Port %s: Requesting a mempool of %u mbufs of size %u " |
657 | "on socket %d for %d Rx and %d Tx queues, " | |
658 | "cache line size of %u", | |
659 | netdev_name, n_mbufs, mbuf_size, socket_id, | |
660 | dev->requested_n_rxq, dev->requested_n_txq, | |
661 | RTE_CACHE_LINE_SIZE); | |
662 | ||
a32bab26 TL |
663 | /* The size of the mbuf's private area (i.e. area that holds OvS' |
664 | * dp_packet data)*/ | |
dfaf00e8 MK |
665 | mbuf_priv_data_len = sizeof(struct dp_packet) - |
666 | sizeof(struct rte_mbuf); | |
667 | /* The size of the entire dp_packet. */ | |
668 | pkt_size = sizeof(struct dp_packet) + mbuf_size; | |
669 | /* mbuf size, rounded up to cacheline size. */ | |
670 | aligned_mbuf_size = ROUND_UP(pkt_size, RTE_CACHE_LINE_SIZE); | |
671 | /* If there is a size discrepancy, add padding to mbuf_priv_data_len. | |
672 | * This maintains mbuf size cache alignment, while also honoring RX | |
673 | * buffer alignment in the data portion of the mbuf. If this adjustment | |
674 | * is not made, there is a possiblity later on that for an element of | |
675 | * the mempool, buf, buf->data_len < (buf->buf_len - buf->data_off). | |
676 | * This is problematic in the case of multi-segment mbufs, particularly | |
677 | * when an mbuf segment needs to be resized (when [push|popp]ing a VLAN | |
678 | * header, for example. | |
679 | */ | |
680 | mbuf_priv_data_len += (aligned_mbuf_size - pkt_size); | |
681 | ||
682 | dmp->mp = rte_pktmbuf_pool_create(mp_name, n_mbufs, MP_CACHE_SZ, | |
683 | mbuf_priv_data_len, | |
684 | mbuf_size, | |
43307ad0 | 685 | socket_id); |
24e78f93 | 686 | |
43307ad0 | 687 | if (dmp->mp) { |
24e78f93 IM |
688 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", |
689 | mp_name, n_mbufs); | |
837c1761 | 690 | /* rte_pktmbuf_pool_create has done some initialization of the |
43307ad0 IS |
691 | * rte_mbuf part of each dp_packet, while ovs_rte_pktmbuf_init |
692 | * initializes some OVS specific fields of dp_packet. | |
693 | */ | |
694 | rte_mempool_obj_iter(dmp->mp, ovs_rte_pktmbuf_init, NULL); | |
695 | return dmp; | |
d555d9bd RW |
696 | } else if (rte_errno == EEXIST) { |
697 | /* A mempool with the same name already exists. We just | |
698 | * retrieve its pointer to be returned to the caller. */ | |
43307ad0 | 699 | dmp->mp = rte_mempool_lookup(mp_name); |
d555d9bd RW |
700 | /* As the mempool create returned EEXIST we can expect the |
701 | * lookup has returned a valid pointer. If for some reason | |
702 | * that's not the case we keep track of it. */ | |
24e78f93 | 703 | VLOG_DBG("A mempool with name \"%s\" already exists at %p.", |
43307ad0 IS |
704 | mp_name, dmp->mp); |
705 | return dmp; | |
d555d9bd | 706 | } else { |
43307ad0 IS |
707 | VLOG_DBG("Failed to create mempool \"%s\" with a request of " |
708 | "%u mbufs, retrying with %u mbufs", | |
709 | mp_name, n_mbufs, n_mbufs / 2); | |
0c6f39e5 | 710 | } |
43307ad0 | 711 | } while (!dmp->mp && rte_errno == ENOMEM && (n_mbufs /= 2) >= MIN_NB_MBUF); |
2ae3d542 | 712 | |
43307ad0 IS |
713 | VLOG_ERR("Failed to create mempool \"%s\" with a request of %u mbufs", |
714 | mp_name, n_mbufs); | |
715 | ||
716 | rte_free(dmp); | |
717 | return NULL; | |
8a9562d2 PS |
718 | } |
719 | ||
43307ad0 IS |
720 | static struct dpdk_mp * |
721 | dpdk_mp_get(struct netdev_dpdk *dev, int mtu, bool per_port_mp) | |
8a9562d2 | 722 | { |
43307ad0 IS |
723 | struct dpdk_mp *dmp, *next; |
724 | bool reuse = false; | |
8a9562d2 | 725 | |
c2adb102 | 726 | ovs_mutex_lock(&dpdk_mp_mutex); |
43307ad0 IS |
727 | /* Check if shared memory is being used, if so check existing mempools |
728 | * to see if reuse is possible. */ | |
729 | if (!per_port_mp) { | |
730 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
731 | if (dmp->socket_id == dev->requested_socket_id | |
732 | && dmp->mtu == mtu) { | |
733 | VLOG_DBG("Reusing mempool \"%s\"", dmp->mp->name); | |
734 | dmp->refcount++; | |
735 | reuse = true; | |
736 | break; | |
737 | } | |
738 | } | |
739 | } | |
740 | /* Sweep mempools after reuse or before create. */ | |
741 | dpdk_mp_sweep(); | |
91fccdad | 742 | |
43307ad0 IS |
743 | if (!reuse) { |
744 | dmp = dpdk_mp_create(dev, mtu, per_port_mp); | |
91fccdad | 745 | if (dmp) { |
43307ad0 IS |
746 | /* Shared memory will hit the reuse case above so will not |
747 | * request a mempool that already exists but we need to check | |
748 | * for the EEXIST case for per port memory case. Compare the | |
749 | * mempool returned by dmp to each entry in dpdk_mp_list. If a | |
750 | * match is found, free dmp as a new entry is not required, set | |
751 | * dmp to point to the existing entry and increment the refcount | |
752 | * to avoid being freed at a later stage. | |
753 | */ | |
754 | if (per_port_mp && rte_errno == EEXIST) { | |
755 | LIST_FOR_EACH (next, list_node, &dpdk_mp_list) { | |
756 | if (dmp->mp == next->mp) { | |
757 | rte_free(dmp); | |
758 | dmp = next; | |
759 | dmp->refcount++; | |
760 | } | |
761 | } | |
762 | } else { | |
763 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); | |
764 | } | |
91fccdad KT |
765 | } |
766 | } | |
43307ad0 | 767 | |
43307ad0 IS |
768 | ovs_mutex_unlock(&dpdk_mp_mutex); |
769 | ||
770 | return dmp; | |
771 | } | |
772 | ||
773 | /* Decrement reference to a mempool. */ | |
774 | static void | |
775 | dpdk_mp_put(struct dpdk_mp *dmp) | |
776 | { | |
777 | if (!dmp) { | |
778 | return; | |
779 | } | |
780 | ||
781 | ovs_mutex_lock(&dpdk_mp_mutex); | |
782 | ovs_assert(dmp->refcount); | |
783 | dmp->refcount--; | |
c2adb102 | 784 | ovs_mutex_unlock(&dpdk_mp_mutex); |
8a9562d2 PS |
785 | } |
786 | ||
43307ad0 IS |
787 | /* Depending on the memory model being used this function tries to |
788 | * identify and reuse an existing mempool or tries to allocate a new | |
789 | * mempool on requested_socket_id with mbuf size corresponding to the | |
790 | * requested_mtu. On success, a new configuration will be applied. | |
0072e931 MK |
791 | * On error, device will be left unchanged. */ |
792 | static int | |
793 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
0072e931 MK |
794 | OVS_REQUIRES(dev->mutex) |
795 | { | |
796 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
43307ad0 | 797 | struct dpdk_mp *dmp; |
24e78f93 | 798 | int ret = 0; |
43307ad0 | 799 | bool per_port_mp = dpdk_per_port_memory(); |
0072e931 | 800 | |
43307ad0 IS |
801 | /* With shared memory we do not need to configure a mempool if the MTU |
802 | * and socket ID have not changed, the previous configuration is still | |
803 | * valid so return 0 */ | |
804 | if (!per_port_mp && dev->mtu == dev->requested_mtu | |
805 | && dev->socket_id == dev->requested_socket_id) { | |
806 | return ret; | |
807 | } | |
91fccdad | 808 | |
43307ad0 IS |
809 | dmp = dpdk_mp_get(dev, FRAME_LEN_TO_MTU(buf_size), per_port_mp); |
810 | if (!dmp) { | |
c67e46c0 MK |
811 | VLOG_ERR("Failed to create memory pool for netdev " |
812 | "%s, with MTU %d on socket %d: %s\n", | |
813 | dev->up.name, dev->requested_mtu, dev->requested_socket_id, | |
814 | rte_strerror(rte_errno)); | |
24e78f93 | 815 | ret = rte_errno; |
0072e931 | 816 | } else { |
43307ad0 IS |
817 | /* Check for any pre-existing dpdk_mp for the device before accessing |
818 | * the associated mempool. | |
819 | */ | |
820 | if (dev->dpdk_mp != NULL) { | |
821 | /* A new MTU was requested, decrement the reference count for the | |
822 | * devices current dpdk_mp. This is required even if a pointer to | |
823 | * same dpdk_mp is returned by dpdk_mp_get. The refcount for dmp | |
824 | * has already been incremented by dpdk_mp_get at this stage so it | |
825 | * must be decremented to keep an accurate refcount for the | |
826 | * dpdk_mp. | |
827 | */ | |
828 | dpdk_mp_put(dev->dpdk_mp); | |
829 | } | |
830 | dev->dpdk_mp = dmp; | |
0072e931 MK |
831 | dev->mtu = dev->requested_mtu; |
832 | dev->socket_id = dev->requested_socket_id; | |
833 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
834 | } | |
835 | ||
24e78f93 | 836 | return ret; |
0072e931 MK |
837 | } |
838 | ||
8a9562d2 PS |
839 | static void |
840 | check_link_status(struct netdev_dpdk *dev) | |
841 | { | |
842 | struct rte_eth_link link; | |
843 | ||
844 | rte_eth_link_get_nowait(dev->port_id, &link); | |
845 | ||
846 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 847 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
848 | |
849 | dev->link_reset_cnt++; | |
850 | dev->link = link; | |
851 | if (dev->link.link_status) { | |
fa9f4eeb IM |
852 | VLOG_DBG_RL(&rl, |
853 | "Port "DPDK_PORT_ID_FMT" Link Up - speed %u Mbps - %s", | |
58be5c0e | 854 | dev->port_id, (unsigned) dev->link.link_speed, |
fa9f4eeb IM |
855 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) |
856 | ? "full-duplex" : "half-duplex"); | |
8a9562d2 | 857 | } else { |
fa9f4eeb IM |
858 | VLOG_DBG_RL(&rl, "Port "DPDK_PORT_ID_FMT" Link Down", |
859 | dev->port_id); | |
8a9562d2 PS |
860 | } |
861 | } | |
862 | } | |
863 | ||
864 | static void * | |
865 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
866 | { | |
867 | struct netdev_dpdk *dev; | |
868 | ||
869 | pthread_detach(pthread_self()); | |
870 | ||
871 | for (;;) { | |
872 | ovs_mutex_lock(&dpdk_mutex); | |
873 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
874 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
875 | if (dev->type == DPDK_DEV_ETH) { |
876 | check_link_status(dev); | |
877 | } | |
8a9562d2 PS |
878 | ovs_mutex_unlock(&dev->mutex); |
879 | } | |
880 | ovs_mutex_unlock(&dpdk_mutex); | |
881 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
882 | } | |
883 | ||
884 | return NULL; | |
885 | } | |
886 | ||
b98d7669 | 887 | static int |
f8b64a61 | 888 | dpdk_eth_dev_port_config(struct netdev_dpdk *dev, int n_rxq, int n_txq) |
b98d7669 DDP |
889 | { |
890 | int diag = 0; | |
891 | int i; | |
0072e931 | 892 | struct rte_eth_conf conf = port_conf; |
65a87968 | 893 | struct rte_eth_dev_info info; |
4dd16ca0 | 894 | uint16_t conf_mtu; |
65a87968 | 895 | |
03f3f9c0 OM |
896 | rte_eth_dev_info_get(dev->port_id, &info); |
897 | ||
65a87968 | 898 | /* As of DPDK 17.11.1 a few PMDs require to explicitly enable |
03f3f9c0 OM |
899 | * scatter to support jumbo RX. |
900 | * Setting scatter for the device is done after checking for | |
901 | * scatter support in the device capabilites. */ | |
0072e931 | 902 | if (dev->mtu > ETHER_MTU) { |
03f3f9c0 OM |
903 | if (dev->hw_ol_features & NETDEV_RX_HW_SCATTER) { |
904 | conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER; | |
65a87968 | 905 | } |
0072e931 | 906 | } |
67fe6d63 | 907 | |
f8b64a61 | 908 | conf.intr_conf.lsc = dev->lsc_interrupt_mode; |
e10ca8b9 | 909 | |
03f3f9c0 OM |
910 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
911 | conf.rxmode.offloads |= DEV_RX_OFFLOAD_CHECKSUM; | |
912 | } | |
913 | ||
914 | if (!(dev->hw_ol_features & NETDEV_RX_HW_CRC_STRIP) | |
915 | && info.rx_offload_capa & DEV_RX_OFFLOAD_KEEP_CRC) { | |
916 | conf.rxmode.offloads |= DEV_RX_OFFLOAD_KEEP_CRC; | |
e10ca8b9 MW |
917 | } |
918 | ||
03f3f9c0 OM |
919 | /* Limit configured rss hash functions to only those supported |
920 | * by the eth device. */ | |
921 | conf.rx_adv_conf.rss_conf.rss_hf &= info.flow_type_rss_offloads; | |
922 | ||
b98d7669 DDP |
923 | /* A device may report more queues than it makes available (this has |
924 | * been observed for Intel xl710, which reserves some of them for | |
925 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
926 | * available. When this happens we can retry the configuration | |
927 | * and request less queues */ | |
928 | while (n_rxq && n_txq) { | |
929 | if (diag) { | |
930 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
931 | } | |
932 | ||
0072e931 | 933 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 934 | if (diag) { |
0072e931 MK |
935 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
936 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
937 | break; |
938 | } | |
939 | ||
67fe6d63 MK |
940 | diag = rte_eth_dev_set_mtu(dev->port_id, dev->mtu); |
941 | if (diag) { | |
4dd16ca0 IS |
942 | /* A device may not support rte_eth_dev_set_mtu, in this case |
943 | * flag a warning to the user and include the devices configured | |
944 | * MTU value that will be used instead. */ | |
945 | if (-ENOTSUP == diag) { | |
946 | rte_eth_dev_get_mtu(dev->port_id, &conf_mtu); | |
947 | VLOG_WARN("Interface %s does not support MTU configuration, " | |
948 | "max packet size supported is %"PRIu16".", | |
949 | dev->up.name, conf_mtu); | |
950 | } else { | |
951 | VLOG_ERR("Interface %s MTU (%d) setup error: %s", | |
952 | dev->up.name, dev->mtu, rte_strerror(-diag)); | |
953 | break; | |
954 | } | |
67fe6d63 MK |
955 | } |
956 | ||
b98d7669 | 957 | for (i = 0; i < n_txq; i++) { |
b685696b | 958 | diag = rte_eth_tx_queue_setup(dev->port_id, i, dev->txq_size, |
b98d7669 DDP |
959 | dev->socket_id, NULL); |
960 | if (diag) { | |
1dfebee9 | 961 | VLOG_INFO("Interface %s unable to setup txq(%d): %s", |
b98d7669 DDP |
962 | dev->up.name, i, rte_strerror(-diag)); |
963 | break; | |
964 | } | |
965 | } | |
966 | ||
967 | if (i != n_txq) { | |
968 | /* Retry with less tx queues */ | |
969 | n_txq = i; | |
970 | continue; | |
971 | } | |
972 | ||
973 | for (i = 0; i < n_rxq; i++) { | |
b685696b | 974 | diag = rte_eth_rx_queue_setup(dev->port_id, i, dev->rxq_size, |
43307ad0 IS |
975 | dev->socket_id, NULL, |
976 | dev->dpdk_mp->mp); | |
b98d7669 | 977 | if (diag) { |
1dfebee9 | 978 | VLOG_INFO("Interface %s unable to setup rxq(%d): %s", |
b98d7669 DDP |
979 | dev->up.name, i, rte_strerror(-diag)); |
980 | break; | |
981 | } | |
982 | } | |
983 | ||
984 | if (i != n_rxq) { | |
985 | /* Retry with less rx queues */ | |
986 | n_rxq = i; | |
987 | continue; | |
988 | } | |
989 | ||
990 | dev->up.n_rxq = n_rxq; | |
81acebda | 991 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
992 | |
993 | return 0; | |
994 | } | |
995 | ||
996 | return diag; | |
997 | } | |
998 | ||
9fd39370 SC |
999 | static void |
1000 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
1001 | { | |
1002 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
fa9f4eeb | 1003 | VLOG_WARN("Failed to enable flow control on device "DPDK_PORT_ID_FMT, |
bb37956a | 1004 | dev->port_id); |
9fd39370 SC |
1005 | } |
1006 | } | |
b98d7669 | 1007 | |
8a9562d2 | 1008 | static int |
c2adb102 IM |
1009 | dpdk_eth_dev_init(struct netdev_dpdk *dev) |
1010 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
1011 | { |
1012 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 1013 | struct rte_eth_dev_info info; |
8a9562d2 PS |
1014 | struct ether_addr eth_addr; |
1015 | int diag; | |
b98d7669 | 1016 | int n_rxq, n_txq; |
d4f5282c KT |
1017 | uint32_t rx_chksm_offload_capa = DEV_RX_OFFLOAD_UDP_CKSUM | |
1018 | DEV_RX_OFFLOAD_TCP_CKSUM | | |
1019 | DEV_RX_OFFLOAD_IPV4_CKSUM; | |
8a9562d2 | 1020 | |
a0cb2d66 | 1021 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 1022 | |
e10ca8b9 MW |
1023 | if (strstr(info.driver_name, "vf") != NULL) { |
1024 | VLOG_INFO("Virtual function detected, HW_CRC_STRIP will be enabled"); | |
1025 | dev->hw_ol_features |= NETDEV_RX_HW_CRC_STRIP; | |
1026 | } else { | |
1027 | dev->hw_ol_features &= ~NETDEV_RX_HW_CRC_STRIP; | |
1028 | } | |
1029 | ||
d4f5282c KT |
1030 | if ((info.rx_offload_capa & rx_chksm_offload_capa) != |
1031 | rx_chksm_offload_capa) { | |
fa9f4eeb IM |
1032 | VLOG_WARN("Rx checksum offload is not supported on port " |
1033 | DPDK_PORT_ID_FMT, dev->port_id); | |
d4f5282c KT |
1034 | dev->hw_ol_features &= ~NETDEV_RX_CHECKSUM_OFFLOAD; |
1035 | } else { | |
1036 | dev->hw_ol_features |= NETDEV_RX_CHECKSUM_OFFLOAD; | |
1037 | } | |
1038 | ||
03f3f9c0 OM |
1039 | if (info.rx_offload_capa & DEV_RX_OFFLOAD_SCATTER) { |
1040 | dev->hw_ol_features |= NETDEV_RX_HW_SCATTER; | |
1041 | } else { | |
1042 | /* Do not warn on lack of scatter support */ | |
1043 | dev->hw_ol_features &= ~NETDEV_RX_HW_SCATTER; | |
1044 | } | |
1045 | ||
b98d7669 DDP |
1046 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
1047 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
1048 | ||
f8b64a61 | 1049 | diag = dpdk_eth_dev_port_config(dev, n_rxq, n_txq); |
8a9562d2 | 1050 | if (diag) { |
f8b64a61 RM |
1051 | VLOG_ERR("Interface %s(rxq:%d txq:%d lsc interrupt mode:%s) " |
1052 | "configure error: %s", | |
1053 | dev->up.name, n_rxq, n_txq, | |
1054 | dev->lsc_interrupt_mode ? "true" : "false", | |
1055 | rte_strerror(-diag)); | |
95fb793a | 1056 | return -diag; |
8a9562d2 PS |
1057 | } |
1058 | ||
8a9562d2 PS |
1059 | diag = rte_eth_dev_start(dev->port_id); |
1060 | if (diag) { | |
b98d7669 DDP |
1061 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
1062 | rte_strerror(-diag)); | |
95fb793a | 1063 | return -diag; |
8a9562d2 | 1064 | } |
606f6650 | 1065 | dev->started = true; |
8a9562d2 PS |
1066 | |
1067 | rte_eth_promiscuous_enable(dev->port_id); | |
1068 | rte_eth_allmulticast_enable(dev->port_id); | |
1069 | ||
1070 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
1071 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
fa9f4eeb IM |
1072 | VLOG_INFO_RL(&rl, "Port "DPDK_PORT_ID_FMT": "ETH_ADDR_FMT, |
1073 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); | |
8a9562d2 | 1074 | |
ca92d173 | 1075 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
1076 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
1077 | ||
43307ad0 | 1078 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); |
8a9562d2 | 1079 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; |
8a9562d2 PS |
1080 | return 0; |
1081 | } | |
1082 | ||
1083 | static struct netdev_dpdk * | |
1084 | netdev_dpdk_cast(const struct netdev *netdev) | |
1085 | { | |
1086 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
1087 | } | |
1088 | ||
1089 | static struct netdev * | |
1090 | netdev_dpdk_alloc(void) | |
1091 | { | |
bab69409 AC |
1092 | struct netdev_dpdk *dev; |
1093 | ||
65e19e70 DDP |
1094 | dev = dpdk_rte_mzalloc(sizeof *dev); |
1095 | if (dev) { | |
1096 | return &dev->up; | |
bab69409 | 1097 | } |
65e19e70 | 1098 | |
bab69409 | 1099 | return NULL; |
8a9562d2 PS |
1100 | } |
1101 | ||
eff23640 DDP |
1102 | static struct dpdk_tx_queue * |
1103 | netdev_dpdk_alloc_txq(unsigned int n_txqs) | |
5a034064 | 1104 | { |
eff23640 | 1105 | struct dpdk_tx_queue *txqs; |
bd5131ba | 1106 | unsigned i; |
5a034064 | 1107 | |
eff23640 DDP |
1108 | txqs = dpdk_rte_mzalloc(n_txqs * sizeof *txqs); |
1109 | if (txqs) { | |
1110 | for (i = 0; i < n_txqs; i++) { | |
1111 | /* Initialize map for vhost devices. */ | |
1112 | txqs[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
1113 | rte_spinlock_init(&txqs[i].tx_lock); | |
1114 | } | |
5a034064 | 1115 | } |
eff23640 DDP |
1116 | |
1117 | return txqs; | |
5a034064 AW |
1118 | } |
1119 | ||
8a9562d2 | 1120 | static int |
bb37956a | 1121 | common_construct(struct netdev *netdev, dpdk_port_t port_no, |
1ce30dfd | 1122 | enum dpdk_dev_type type, int socket_id) |
5a034064 | 1123 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 1124 | { |
d46285a2 | 1125 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1126 | |
d46285a2 | 1127 | ovs_mutex_init(&dev->mutex); |
8a9562d2 | 1128 | |
d46285a2 | 1129 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 1130 | |
1b7a04e0 AW |
1131 | /* If the 'sid' is negative, it means that the kernel fails |
1132 | * to obtain the pci numa info. In that situation, always | |
1133 | * use 'SOCKET0'. */ | |
1ce30dfd | 1134 | dev->socket_id = socket_id < 0 ? SOCKET0 : socket_id; |
db8f13b0 | 1135 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
1136 | dev->port_id = port_no; |
1137 | dev->type = type; | |
1138 | dev->flags = 0; | |
7f381c2e | 1139 | dev->requested_mtu = ETHER_MTU; |
d46285a2 | 1140 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
f8b64a61 | 1141 | dev->requested_lsc_interrupt_mode = 0; |
0a0f39df CL |
1142 | ovsrcu_index_init(&dev->vid, -1); |
1143 | dev->vhost_reconfigured = false; | |
5dcde09c | 1144 | dev->attached = false; |
8a9562d2 | 1145 | |
78bd47cf | 1146 | ovsrcu_init(&dev->qos_conf, NULL); |
0bf765f7 | 1147 | |
9509913a IS |
1148 | ovsrcu_init(&dev->ingress_policer, NULL); |
1149 | dev->policer_rate = 0; | |
1150 | dev->policer_burst = 0; | |
1151 | ||
7f381c2e DDP |
1152 | netdev->n_rxq = 0; |
1153 | netdev->n_txq = 0; | |
1154 | dev->requested_n_rxq = NR_QUEUE; | |
1155 | dev->requested_n_txq = NR_QUEUE; | |
1156 | dev->requested_rxq_size = NIC_PORT_DEFAULT_RXQ_SIZE; | |
1157 | dev->requested_txq_size = NIC_PORT_DEFAULT_TXQ_SIZE; | |
58397e6c | 1158 | |
9fd39370 SC |
1159 | /* Initialize the flow control to NULL */ |
1160 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
1a2bb118 SC |
1161 | |
1162 | /* Initilize the hardware offload flags to 0 */ | |
1163 | dev->hw_ol_features = 0; | |
3b1fb077 DDP |
1164 | |
1165 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
1166 | ||
d46285a2 | 1167 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 1168 | |
7f381c2e DDP |
1169 | netdev_request_reconfigure(netdev); |
1170 | ||
971f4b39 MW |
1171 | dev->rte_xstats_names = NULL; |
1172 | dev->rte_xstats_names_size = 0; | |
1173 | ||
1174 | dev->rte_xstats_ids = NULL; | |
1175 | dev->rte_xstats_ids_size = 0; | |
1176 | ||
c161357d KT |
1177 | dev->tx_retries = 0; |
1178 | ||
1ce30dfd | 1179 | return 0; |
95fb793a | 1180 | } |
1181 | ||
b83a2df1 MV |
1182 | /* dev_name must be the prefix followed by a positive decimal number. |
1183 | * (no leading + or - signs are allowed) */ | |
95fb793a | 1184 | static int |
1185 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
1186 | unsigned int *port_no) | |
1187 | { | |
1188 | const char *cport; | |
1189 | ||
1190 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
1191 | return ENODEV; | |
1192 | } | |
1193 | ||
1194 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
1195 | |
1196 | if (str_to_uint(cport, 10, port_no)) { | |
1197 | return 0; | |
1198 | } else { | |
1199 | return ENODEV; | |
1200 | } | |
95fb793a | 1201 | } |
1202 | ||
40e940e4 OM |
1203 | /* Get the number of OVS interfaces which have the same DPDK |
1204 | * rte device (e.g. same pci bus address). | |
1205 | * FIXME: avoid direct access to DPDK internal array rte_eth_devices. | |
1206 | */ | |
1207 | static int | |
1208 | netdev_dpdk_get_num_ports(struct rte_device *device) | |
1209 | OVS_REQUIRES(dpdk_mutex) | |
1210 | { | |
1211 | struct netdev_dpdk *dev; | |
1212 | int count = 0; | |
1213 | ||
1214 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1215 | if (rte_eth_devices[dev->port_id].device == device | |
1216 | && rte_eth_devices[dev->port_id].state != RTE_ETH_DEV_UNUSED) { | |
1217 | count++; | |
1218 | } | |
1219 | } | |
1220 | return count; | |
1221 | } | |
1222 | ||
1ce30dfd DDP |
1223 | static int |
1224 | vhost_common_construct(struct netdev *netdev) | |
1225 | OVS_REQUIRES(dpdk_mutex) | |
1226 | { | |
1227 | int socket_id = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
1228 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1229 | ||
35c91567 DM |
1230 | dev->vhost_rxq_enabled = dpdk_rte_mzalloc(OVS_VHOST_MAX_QUEUE_NUM * |
1231 | sizeof *dev->vhost_rxq_enabled); | |
1232 | if (!dev->vhost_rxq_enabled) { | |
1233 | return ENOMEM; | |
1234 | } | |
1ce30dfd DDP |
1235 | dev->tx_q = netdev_dpdk_alloc_txq(OVS_VHOST_MAX_QUEUE_NUM); |
1236 | if (!dev->tx_q) { | |
35c91567 | 1237 | rte_free(dev->vhost_rxq_enabled); |
1ce30dfd DDP |
1238 | return ENOMEM; |
1239 | } | |
1240 | ||
080f080c KT |
1241 | atomic_init(&dev->vhost_tx_retries_max, VHOST_ENQ_RETRY_DEF); |
1242 | ||
bb37956a IM |
1243 | return common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1244 | DPDK_DEV_VHOST, socket_id); | |
1ce30dfd DDP |
1245 | } |
1246 | ||
7d1ced01 | 1247 | static int |
53f50d24 | 1248 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 1249 | { |
d46285a2 DDP |
1250 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1251 | const char *name = netdev->name; | |
7d1ced01 | 1252 | int err; |
a0cb2d66 | 1253 | |
1af27e8a DDP |
1254 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
1255 | * the file system. '/' or '\' would traverse directories, so they're not | |
1256 | * acceptable in 'name'. */ | |
1257 | if (strchr(name, '/') || strchr(name, '\\')) { | |
1258 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
1259 | "A valid name must not include '/' or '\\'", | |
1260 | name); | |
1261 | return EINVAL; | |
1262 | } | |
1263 | ||
7d1ced01 CL |
1264 | ovs_mutex_lock(&dpdk_mutex); |
1265 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 1266 | * the socket is to be created, then register the socket. |
7d1ced01 | 1267 | */ |
bb9d2623 | 1268 | dev->vhost_id = xasprintf("%s/%s", dpdk_get_vhost_sock_dir(), name); |
1af27e8a | 1269 | |
2d24d165 CL |
1270 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
1271 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
1272 | if (err) { |
1273 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 1274 | dev->vhost_id); |
f3e7ec25 | 1275 | goto out; |
e5c0f5a4 | 1276 | } else { |
2d24d165 CL |
1277 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
1278 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
1279 | dev->vhost_id, name); | |
1280 | } | |
f3e7ec25 MW |
1281 | |
1282 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
1283 | &virtio_net_device_ops); | |
1284 | if (err) { | |
1285 | VLOG_ERR("rte_vhost_driver_callback_register failed for vhost user " | |
1286 | "port: %s\n", name); | |
1287 | goto out; | |
1288 | } | |
1289 | ||
1290 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
1291 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
1292 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
1293 | | 1ULL << VIRTIO_NET_F_CSUM); | |
1294 | if (err) { | |
1295 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
1296 | "port: %s\n", name); | |
1297 | goto out; | |
1298 | } | |
1299 | ||
1300 | err = rte_vhost_driver_start(dev->vhost_id); | |
1301 | if (err) { | |
1302 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
1303 | "port: %s\n", name); | |
1304 | goto out; | |
1305 | } | |
1306 | ||
1ce30dfd | 1307 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1308 | if (err) { |
1309 | VLOG_ERR("vhost_common_construct failed for vhost user " | |
1310 | "port: %s\n", name); | |
1311 | } | |
2d24d165 | 1312 | |
f3e7ec25 | 1313 | out: |
bb9d2623 IM |
1314 | if (err) { |
1315 | free(dev->vhost_id); | |
1316 | dev->vhost_id = NULL; | |
1317 | } | |
1318 | ||
2d24d165 | 1319 | ovs_mutex_unlock(&dpdk_mutex); |
28ca969e AC |
1320 | VLOG_WARN_ONCE("dpdkvhostuser ports are considered deprecated; " |
1321 | "please migrate to dpdkvhostuserclient ports."); | |
2d24d165 CL |
1322 | return err; |
1323 | } | |
1324 | ||
1325 | static int | |
1326 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
1327 | { | |
1328 | int err; | |
1329 | ||
2d24d165 | 1330 | ovs_mutex_lock(&dpdk_mutex); |
1ce30dfd | 1331 | err = vhost_common_construct(netdev); |
f3e7ec25 MW |
1332 | if (err) { |
1333 | VLOG_ERR("vhost_common_construct failed for vhost user client" | |
1334 | "port: %s\n", netdev->name); | |
1335 | } | |
7d1ced01 | 1336 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
1337 | return err; |
1338 | } | |
1339 | ||
95fb793a | 1340 | static int |
1341 | netdev_dpdk_construct(struct netdev *netdev) | |
1342 | { | |
95fb793a | 1343 | int err; |
1344 | ||
95fb793a | 1345 | ovs_mutex_lock(&dpdk_mutex); |
bb37956a IM |
1346 | err = common_construct(netdev, DPDK_ETH_PORT_ID_INVALID, |
1347 | DPDK_DEV_ETH, SOCKET0); | |
8a9562d2 PS |
1348 | ovs_mutex_unlock(&dpdk_mutex); |
1349 | return err; | |
1350 | } | |
1351 | ||
1ce30dfd DDP |
1352 | static void |
1353 | common_destruct(struct netdev_dpdk *dev) | |
1354 | OVS_REQUIRES(dpdk_mutex) | |
1355 | OVS_EXCLUDED(dev->mutex) | |
1356 | { | |
1357 | rte_free(dev->tx_q); | |
43307ad0 | 1358 | dpdk_mp_put(dev->dpdk_mp); |
1ce30dfd DDP |
1359 | |
1360 | ovs_list_remove(&dev->list_node); | |
1361 | free(ovsrcu_get_protected(struct ingress_policer *, | |
1362 | &dev->ingress_policer)); | |
1363 | ovs_mutex_destroy(&dev->mutex); | |
1364 | } | |
1365 | ||
8a9562d2 | 1366 | static void |
d46285a2 | 1367 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 1368 | { |
d46285a2 | 1369 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
40e940e4 OM |
1370 | struct rte_device *rte_dev; |
1371 | struct rte_eth_dev *eth_dev; | |
1372 | bool remove_on_close; | |
8a9562d2 | 1373 | |
8d38823b | 1374 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1375 | |
8a9562d2 | 1376 | rte_eth_dev_stop(dev->port_id); |
606f6650 | 1377 | dev->started = false; |
5dcde09c IM |
1378 | |
1379 | if (dev->attached) { | |
40e940e4 OM |
1380 | /* Retrieve eth device data before closing it. |
1381 | * FIXME: avoid direct access to DPDK internal array rte_eth_devices. | |
1382 | */ | |
1383 | eth_dev = &rte_eth_devices[dev->port_id]; | |
1384 | remove_on_close = | |
1385 | eth_dev->data && | |
1386 | (eth_dev->data->dev_flags & RTE_ETH_DEV_CLOSE_REMOVE); | |
1387 | rte_dev = eth_dev->device; | |
1388 | ||
1389 | /* Remove the eth device. */ | |
5dcde09c | 1390 | rte_eth_dev_close(dev->port_id); |
40e940e4 OM |
1391 | |
1392 | /* Remove this rte device and all its eth devices if flag | |
1393 | * RTE_ETH_DEV_CLOSE_REMOVE is not supported (which means representors | |
1394 | * are not supported), or if all the eth devices belonging to the rte | |
1395 | * device are closed. | |
1396 | */ | |
1397 | if (!remove_on_close || !netdev_dpdk_get_num_ports(rte_dev)) { | |
595ce47c IM |
1398 | int ret = rte_dev_remove(rte_dev); |
1399 | ||
1400 | if (ret < 0) { | |
1401 | VLOG_ERR("Device '%s' can not be detached: %s.", | |
1402 | dev->devargs, rte_strerror(-ret)); | |
40e940e4 OM |
1403 | } else { |
1404 | /* Device was closed and detached. */ | |
1405 | VLOG_INFO("Device '%s' has been removed and detached", | |
1406 | dev->devargs); | |
1407 | } | |
5dcde09c | 1408 | } else { |
40e940e4 OM |
1409 | /* Device was only closed. rte_dev_remove() was not called. */ |
1410 | VLOG_INFO("Device '%s' has been removed", dev->devargs); | |
5dcde09c IM |
1411 | } |
1412 | } | |
1413 | ||
ac1a9bb9 | 1414 | netdev_dpdk_clear_xstats(dev); |
55e075e6 | 1415 | free(dev->devargs); |
1ce30dfd | 1416 | common_destruct(dev); |
8d38823b | 1417 | |
8a9562d2 | 1418 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 1419 | } |
8a9562d2 | 1420 | |
3f891bbe DDP |
1421 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
1422 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
1423 | * deadlock, none of the mutexes must be held while calling this function. */ | |
1424 | static int | |
c1ff66ac CL |
1425 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
1426 | char *vhost_id) | |
3f891bbe DDP |
1427 | OVS_EXCLUDED(dpdk_mutex) |
1428 | OVS_EXCLUDED(dev->mutex) | |
1429 | { | |
c1ff66ac | 1430 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
1431 | } |
1432 | ||
58397e6c | 1433 | static void |
d46285a2 | 1434 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 1435 | { |
d46285a2 | 1436 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1437 | char *vhost_id; |
58397e6c | 1438 | |
8d38823b | 1439 | ovs_mutex_lock(&dpdk_mutex); |
8d38823b | 1440 | |
c62da695 | 1441 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1442 | if (netdev_dpdk_get_vid(dev) >= 0 |
1443 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1444 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1445 | netdev->name); |
58be5c0e MK |
1446 | VLOG_ERR("To restore connectivity after re-adding of port, VM on " |
1447 | "socket '%s' must be restarted.", dev->vhost_id); | |
58397e6c KT |
1448 | } |
1449 | ||
bb9d2623 IM |
1450 | vhost_id = dev->vhost_id; |
1451 | dev->vhost_id = NULL; | |
35c91567 | 1452 | rte_free(dev->vhost_rxq_enabled); |
c1ff66ac | 1453 | |
1ce30dfd DDP |
1454 | common_destruct(dev); |
1455 | ||
58397e6c | 1456 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1457 | |
bb9d2623 | 1458 | if (!vhost_id) { |
821b8664 IM |
1459 | goto out; |
1460 | } | |
1461 | ||
c1ff66ac | 1462 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1463 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1464 | netdev->name, vhost_id); | |
c1ff66ac CL |
1465 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1466 | /* OVS server mode - remove this socket from list for deletion */ | |
1467 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1468 | } |
821b8664 | 1469 | out: |
c1ff66ac | 1470 | free(vhost_id); |
8a9562d2 PS |
1471 | } |
1472 | ||
1473 | static void | |
d46285a2 | 1474 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1475 | { |
d46285a2 | 1476 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1477 | |
d46285a2 | 1478 | rte_free(dev); |
8a9562d2 PS |
1479 | } |
1480 | ||
971f4b39 | 1481 | static void |
ac1a9bb9 | 1482 | netdev_dpdk_clear_xstats(struct netdev_dpdk *dev) |
971f4b39 MW |
1483 | { |
1484 | /* If statistics are already allocated, we have to | |
1485 | * reconfigure, as port_id could have been changed. */ | |
1486 | if (dev->rte_xstats_names) { | |
1487 | free(dev->rte_xstats_names); | |
1488 | dev->rte_xstats_names = NULL; | |
1489 | dev->rte_xstats_names_size = 0; | |
1490 | } | |
1491 | if (dev->rte_xstats_ids) { | |
1492 | free(dev->rte_xstats_ids); | |
1493 | dev->rte_xstats_ids = NULL; | |
1494 | dev->rte_xstats_ids_size = 0; | |
1495 | } | |
1496 | } | |
1497 | ||
1498 | static const char* | |
1499 | netdev_dpdk_get_xstat_name(struct netdev_dpdk *dev, uint64_t id) | |
1500 | { | |
1501 | if (id >= dev->rte_xstats_names_size) { | |
1502 | return "UNKNOWN"; | |
1503 | } | |
1504 | return dev->rte_xstats_names[id].name; | |
1505 | } | |
1506 | ||
1507 | static bool | |
1508 | netdev_dpdk_configure_xstats(struct netdev_dpdk *dev) | |
1509 | OVS_REQUIRES(dev->mutex) | |
1510 | { | |
1511 | int rte_xstats_len; | |
1512 | bool ret; | |
1513 | struct rte_eth_xstat *rte_xstats; | |
1514 | uint64_t id; | |
1515 | int xstats_no; | |
1516 | const char *name; | |
1517 | ||
1518 | /* Retrieving all XSTATS names. If something will go wrong | |
1519 | * or amount of counters will be equal 0, rte_xstats_names | |
1520 | * buffer will be marked as NULL, and any further xstats | |
1521 | * query won't be performed (e.g. during netdev_dpdk_get_stats | |
1522 | * execution). */ | |
1523 | ||
1524 | ret = false; | |
1525 | rte_xstats = NULL; | |
1526 | ||
1527 | if (dev->rte_xstats_names == NULL || dev->rte_xstats_ids == NULL) { | |
1528 | dev->rte_xstats_names_size = | |
1529 | rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
1530 | ||
1531 | if (dev->rte_xstats_names_size < 0) { | |
fa9f4eeb IM |
1532 | VLOG_WARN("Cannot get XSTATS for port: "DPDK_PORT_ID_FMT, |
1533 | dev->port_id); | |
971f4b39 MW |
1534 | dev->rte_xstats_names_size = 0; |
1535 | } else { | |
1536 | /* Reserve memory for xstats names and values */ | |
1537 | dev->rte_xstats_names = xcalloc(dev->rte_xstats_names_size, | |
1538 | sizeof *dev->rte_xstats_names); | |
1539 | ||
1540 | if (dev->rte_xstats_names) { | |
1541 | /* Retreive xstats names */ | |
1542 | rte_xstats_len = | |
1543 | rte_eth_xstats_get_names(dev->port_id, | |
1544 | dev->rte_xstats_names, | |
1545 | dev->rte_xstats_names_size); | |
1546 | ||
1547 | if (rte_xstats_len < 0) { | |
fa9f4eeb IM |
1548 | VLOG_WARN("Cannot get XSTATS names for port: " |
1549 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 MW |
1550 | goto out; |
1551 | } else if (rte_xstats_len != dev->rte_xstats_names_size) { | |
fa9f4eeb IM |
1552 | VLOG_WARN("XSTATS size doesn't match for port: " |
1553 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 MW |
1554 | goto out; |
1555 | } | |
1556 | ||
1557 | dev->rte_xstats_ids = xcalloc(dev->rte_xstats_names_size, | |
1558 | sizeof(uint64_t)); | |
1559 | ||
1560 | /* We have to calculate number of counters */ | |
1561 | rte_xstats = xmalloc(rte_xstats_len * sizeof *rte_xstats); | |
1562 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
1563 | ||
1564 | /* Retreive xstats values */ | |
1565 | if (rte_eth_xstats_get(dev->port_id, rte_xstats, | |
1566 | rte_xstats_len) > 0) { | |
1567 | dev->rte_xstats_ids_size = 0; | |
1568 | xstats_no = 0; | |
1569 | for (uint32_t i = 0; i < rte_xstats_len; i++) { | |
1570 | id = rte_xstats[i].id; | |
1571 | name = netdev_dpdk_get_xstat_name(dev, id); | |
1572 | /* We need to filter out everything except | |
1573 | * dropped, error and management counters */ | |
1574 | if (string_ends_with(name, "_errors") || | |
1575 | strstr(name, "_management_") || | |
1576 | string_ends_with(name, "_dropped")) { | |
1577 | ||
1578 | dev->rte_xstats_ids[xstats_no] = id; | |
1579 | xstats_no++; | |
1580 | } | |
1581 | } | |
1582 | dev->rte_xstats_ids_size = xstats_no; | |
1583 | ret = true; | |
1584 | } else { | |
fa9f4eeb IM |
1585 | VLOG_WARN("Can't get XSTATS IDs for port: " |
1586 | DPDK_PORT_ID_FMT, dev->port_id); | |
971f4b39 | 1587 | } |
34eb0863 IM |
1588 | |
1589 | free(rte_xstats); | |
971f4b39 MW |
1590 | } |
1591 | } | |
1592 | } else { | |
1593 | /* Already configured */ | |
1594 | ret = true; | |
1595 | } | |
1596 | ||
1597 | out: | |
1598 | if (!ret) { | |
1599 | netdev_dpdk_clear_xstats(dev); | |
1600 | } | |
1601 | return ret; | |
1602 | } | |
1603 | ||
8a9562d2 | 1604 | static int |
a14b8947 | 1605 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1606 | { |
a14b8947 | 1607 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1608 | |
1609 | ovs_mutex_lock(&dev->mutex); | |
1610 | ||
050c60bf | 1611 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1612 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1613 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1614 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1615 | smap_add_format(args, "mtu", "%d", dev->mtu); |
451f26fd IM |
1616 | |
1617 | if (dev->type == DPDK_DEV_ETH) { | |
1618 | smap_add_format(args, "requested_rxq_descriptors", "%d", | |
1619 | dev->requested_rxq_size); | |
1620 | smap_add_format(args, "configured_rxq_descriptors", "%d", | |
1621 | dev->rxq_size); | |
1622 | smap_add_format(args, "requested_txq_descriptors", "%d", | |
1623 | dev->requested_txq_size); | |
1624 | smap_add_format(args, "configured_txq_descriptors", "%d", | |
1625 | dev->txq_size); | |
1a2bb118 SC |
1626 | if (dev->hw_ol_features & NETDEV_RX_CHECKSUM_OFFLOAD) { |
1627 | smap_add(args, "rx_csum_offload", "true"); | |
8155ab7e KT |
1628 | } else { |
1629 | smap_add(args, "rx_csum_offload", "false"); | |
1a2bb118 | 1630 | } |
f8b64a61 RM |
1631 | smap_add(args, "lsc_interrupt_mode", |
1632 | dev->lsc_interrupt_mode ? "true" : "false"); | |
451f26fd | 1633 | } |
8a9562d2 PS |
1634 | ovs_mutex_unlock(&dev->mutex); |
1635 | ||
1636 | return 0; | |
1637 | } | |
1638 | ||
55e075e6 | 1639 | static struct netdev_dpdk * |
bb37956a | 1640 | netdev_dpdk_lookup_by_port_id(dpdk_port_t port_id) |
55e075e6 CL |
1641 | OVS_REQUIRES(dpdk_mutex) |
1642 | { | |
1643 | struct netdev_dpdk *dev; | |
1644 | ||
1645 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
1646 | if (dev->port_id == port_id) { | |
1647 | return dev; | |
1648 | } | |
1649 | } | |
1650 | ||
1651 | return NULL; | |
1652 | } | |
1653 | ||
5e758818 YL |
1654 | static dpdk_port_t |
1655 | netdev_dpdk_get_port_by_mac(const char *mac_str) | |
1656 | { | |
1657 | dpdk_port_t port_id; | |
1658 | struct eth_addr mac, port_mac; | |
1659 | ||
1660 | if (!eth_addr_from_string(mac_str, &mac)) { | |
1661 | VLOG_ERR("invalid mac: %s", mac_str); | |
1662 | return DPDK_ETH_PORT_ID_INVALID; | |
1663 | } | |
1664 | ||
1665 | RTE_ETH_FOREACH_DEV (port_id) { | |
1666 | struct ether_addr ea; | |
1667 | ||
1668 | rte_eth_macaddr_get(port_id, &ea); | |
1669 | memcpy(port_mac.ea, ea.addr_bytes, ETH_ADDR_LEN); | |
1670 | if (eth_addr_equals(mac, port_mac)) { | |
1671 | return port_id; | |
1672 | } | |
1673 | } | |
1674 | ||
1675 | return DPDK_ETH_PORT_ID_INVALID; | |
1676 | } | |
1677 | ||
40e940e4 OM |
1678 | /* Return the first DPDK port id matching the devargs pattern. */ |
1679 | static dpdk_port_t netdev_dpdk_get_port_by_devargs(const char *devargs) | |
1680 | OVS_REQUIRES(dpdk_mutex) | |
1681 | { | |
1682 | dpdk_port_t port_id; | |
1683 | struct rte_dev_iterator iterator; | |
1684 | ||
1685 | RTE_ETH_FOREACH_MATCHING_DEV (port_id, devargs, &iterator) { | |
1686 | /* If a break is done - must call rte_eth_iterator_cleanup. */ | |
1687 | rte_eth_iterator_cleanup(&iterator); | |
1688 | break; | |
1689 | } | |
1690 | ||
1691 | return port_id; | |
1692 | } | |
1693 | ||
5e758818 | 1694 | /* |
40e940e4 OM |
1695 | * Normally, a PCI id (optionally followed by a representor number) |
1696 | * is enough for identifying a specific DPDK port. | |
5e758818 YL |
1697 | * However, for some NICs having multiple ports sharing the same PCI |
1698 | * id, using PCI id won't work then. | |
1699 | * | |
1700 | * To fix that, here one more method is introduced: "class=eth,mac=$MAC". | |
1701 | * | |
1702 | * Note that the compatibility is fully kept: user can still use the | |
1703 | * PCI id for adding ports (when it's enough for them). | |
1704 | */ | |
bb37956a | 1705 | static dpdk_port_t |
5dcde09c IM |
1706 | netdev_dpdk_process_devargs(struct netdev_dpdk *dev, |
1707 | const char *devargs, char **errp) | |
40e940e4 | 1708 | OVS_REQUIRES(dpdk_mutex) |
55e075e6 | 1709 | { |
40e940e4 | 1710 | dpdk_port_t new_port_id; |
55e075e6 | 1711 | |
5e758818 YL |
1712 | if (strncmp(devargs, "class=eth,mac=", 14) == 0) { |
1713 | new_port_id = netdev_dpdk_get_port_by_mac(&devargs[14]); | |
1714 | } else { | |
40e940e4 OM |
1715 | new_port_id = netdev_dpdk_get_port_by_devargs(devargs); |
1716 | if (!rte_eth_dev_is_valid_port(new_port_id)) { | |
5e758818 | 1717 | /* Device not found in DPDK, attempt to attach it */ |
40e940e4 | 1718 | if (rte_dev_probe(devargs)) { |
5e758818 | 1719 | new_port_id = DPDK_ETH_PORT_ID_INVALID; |
40e940e4 OM |
1720 | } else { |
1721 | new_port_id = netdev_dpdk_get_port_by_devargs(devargs); | |
1722 | if (rte_eth_dev_is_valid_port(new_port_id)) { | |
1723 | /* Attach successful */ | |
1724 | dev->attached = true; | |
1725 | VLOG_INFO("Device '%s' attached to DPDK", devargs); | |
1726 | } else { | |
1727 | /* Attach unsuccessful */ | |
1728 | new_port_id = DPDK_ETH_PORT_ID_INVALID; | |
1729 | } | |
5e758818 | 1730 | } |
55e075e6 | 1731 | } |
5e758818 YL |
1732 | } |
1733 | ||
1734 | if (new_port_id == DPDK_ETH_PORT_ID_INVALID) { | |
1735 | VLOG_WARN_BUF(errp, "Error attaching device '%s' to DPDK", devargs); | |
55e075e6 CL |
1736 | } |
1737 | ||
1738 | return new_port_id; | |
1739 | } | |
1740 | ||
c3d062a7 CL |
1741 | static void |
1742 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
b614c894 | 1743 | OVS_REQUIRES(dev->mutex) |
a14b8947 | 1744 | { |
050c60bf | 1745 | int new_n_rxq; |
a14b8947 | 1746 | |
2a21e757 | 1747 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", NR_QUEUE), 1); |
050c60bf DDP |
1748 | if (new_n_rxq != dev->requested_n_rxq) { |
1749 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1750 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1751 | } |
c3d062a7 CL |
1752 | } |
1753 | ||
b685696b CL |
1754 | static void |
1755 | dpdk_process_queue_size(struct netdev *netdev, const struct smap *args, | |
1756 | const char *flag, int default_size, int *new_size) | |
1757 | { | |
1758 | int queue_size = smap_get_int(args, flag, default_size); | |
1759 | ||
1760 | if (queue_size <= 0 || queue_size > NIC_PORT_MAX_Q_SIZE | |
1761 | || !is_pow2(queue_size)) { | |
1762 | queue_size = default_size; | |
1763 | } | |
1764 | ||
1765 | if (queue_size != *new_size) { | |
1766 | *new_size = queue_size; | |
1767 | netdev_request_reconfigure(netdev); | |
1768 | } | |
1769 | } | |
1770 | ||
c3d062a7 | 1771 | static int |
9fff138e DDP |
1772 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args, |
1773 | char **errp) | |
c3d062a7 CL |
1774 | { |
1775 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
f8b64a61 | 1776 | bool rx_fc_en, tx_fc_en, autoneg, lsc_interrupt_mode; |
c2c84474 | 1777 | bool flow_control_requested = true; |
b614c894 IM |
1778 | enum rte_eth_fc_mode fc_mode; |
1779 | static const enum rte_eth_fc_mode fc_mode_set[2][2] = { | |
1780 | {RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1781 | {RTE_FC_RX_PAUSE, RTE_FC_FULL } | |
1782 | }; | |
55e075e6 CL |
1783 | const char *new_devargs; |
1784 | int err = 0; | |
c3d062a7 | 1785 | |
55e075e6 | 1786 | ovs_mutex_lock(&dpdk_mutex); |
c3d062a7 CL |
1787 | ovs_mutex_lock(&dev->mutex); |
1788 | ||
1789 | dpdk_set_rxq_config(dev, args); | |
1790 | ||
b685696b CL |
1791 | dpdk_process_queue_size(netdev, args, "n_rxq_desc", |
1792 | NIC_PORT_DEFAULT_RXQ_SIZE, | |
1793 | &dev->requested_rxq_size); | |
1794 | dpdk_process_queue_size(netdev, args, "n_txq_desc", | |
1795 | NIC_PORT_DEFAULT_TXQ_SIZE, | |
1796 | &dev->requested_txq_size); | |
1797 | ||
55e075e6 CL |
1798 | new_devargs = smap_get(args, "dpdk-devargs"); |
1799 | ||
1800 | if (dev->devargs && strcmp(new_devargs, dev->devargs)) { | |
1801 | /* The user requested a new device. If we return error, the caller | |
1802 | * will delete this netdev and try to recreate it. */ | |
1803 | err = EAGAIN; | |
1804 | goto out; | |
1805 | } | |
1806 | ||
1807 | /* dpdk-devargs is required for device configuration */ | |
1808 | if (new_devargs && new_devargs[0]) { | |
1809 | /* Don't process dpdk-devargs if value is unchanged and port id | |
1810 | * is valid */ | |
1811 | if (!(dev->devargs && !strcmp(dev->devargs, new_devargs) | |
1812 | && rte_eth_dev_is_valid_port(dev->port_id))) { | |
bb37956a IM |
1813 | dpdk_port_t new_port_id = netdev_dpdk_process_devargs(dev, |
1814 | new_devargs, | |
1815 | errp); | |
55e075e6 CL |
1816 | if (!rte_eth_dev_is_valid_port(new_port_id)) { |
1817 | err = EINVAL; | |
1818 | } else if (new_port_id == dev->port_id) { | |
1819 | /* Already configured, do not reconfigure again */ | |
1820 | err = 0; | |
1821 | } else { | |
1822 | struct netdev_dpdk *dup_dev; | |
bb37956a | 1823 | |
55e075e6 CL |
1824 | dup_dev = netdev_dpdk_lookup_by_port_id(new_port_id); |
1825 | if (dup_dev) { | |
9fff138e | 1826 | VLOG_WARN_BUF(errp, "'%s' is trying to use device '%s' " |
40e940e4 | 1827 | "which is already in use by '%s'", |
9fff138e DDP |
1828 | netdev_get_name(netdev), new_devargs, |
1829 | netdev_get_name(&dup_dev->up)); | |
55e075e6 CL |
1830 | err = EADDRINUSE; |
1831 | } else { | |
bd4e172b | 1832 | int sid = rte_eth_dev_socket_id(new_port_id); |
bb37956a | 1833 | |
bd4e172b | 1834 | dev->requested_socket_id = sid < 0 ? SOCKET0 : sid; |
55e075e6 CL |
1835 | dev->devargs = xstrdup(new_devargs); |
1836 | dev->port_id = new_port_id; | |
1837 | netdev_request_reconfigure(&dev->up); | |
971f4b39 | 1838 | netdev_dpdk_clear_xstats(dev); |
55e075e6 CL |
1839 | err = 0; |
1840 | } | |
1841 | } | |
1842 | } | |
1843 | } else { | |
9fff138e DDP |
1844 | VLOG_WARN_BUF(errp, "'%s' is missing 'options:dpdk-devargs'. " |
1845 | "The old 'dpdk<port_id>' names are not supported", | |
1846 | netdev_get_name(netdev)); | |
55e075e6 CL |
1847 | err = EINVAL; |
1848 | } | |
1849 | ||
1850 | if (err) { | |
1851 | goto out; | |
1852 | } | |
1853 | ||
f8b64a61 RM |
1854 | lsc_interrupt_mode = smap_get_bool(args, "dpdk-lsc-interrupt", false); |
1855 | if (dev->requested_lsc_interrupt_mode != lsc_interrupt_mode) { | |
1856 | dev->requested_lsc_interrupt_mode = lsc_interrupt_mode; | |
1857 | netdev_request_reconfigure(netdev); | |
1858 | } | |
1859 | ||
c3d062a7 CL |
1860 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); |
1861 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
b614c894 | 1862 | autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); |
c3d062a7 | 1863 | |
b614c894 | 1864 | fc_mode = fc_mode_set[tx_fc_en][rx_fc_en]; |
c2c84474 TK |
1865 | |
1866 | if (!smap_get(args, "rx-flow-ctrl") && !smap_get(args, "tx-flow-ctrl") | |
1867 | && !smap_get(args, "flow-ctrl-autoneg")) { | |
1868 | /* FIXME: User didn't ask for flow control configuration. | |
1869 | * For now we'll not print a warning if flow control is not | |
1870 | * supported by the DPDK port. */ | |
1871 | flow_control_requested = false; | |
1872 | } | |
1873 | ||
1874 | /* Get the Flow control configuration. */ | |
1875 | err = -rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
1876 | if (err) { | |
1877 | if (err == ENOTSUP) { | |
1878 | if (flow_control_requested) { | |
1879 | VLOG_WARN("%s: Flow control is not supported.", | |
1880 | netdev_get_name(netdev)); | |
1881 | } | |
1882 | err = 0; /* Not fatal. */ | |
1883 | } else { | |
1884 | VLOG_WARN("%s: Cannot get flow control parameters: %s", | |
1885 | netdev_get_name(netdev), rte_strerror(err)); | |
1886 | } | |
1887 | goto out; | |
1888 | } | |
1889 | ||
b614c894 IM |
1890 | if (dev->fc_conf.mode != fc_mode || autoneg != dev->fc_conf.autoneg) { |
1891 | dev->fc_conf.mode = fc_mode; | |
1892 | dev->fc_conf.autoneg = autoneg; | |
1893 | dpdk_eth_flow_ctrl_setup(dev); | |
1894 | } | |
9fd39370 | 1895 | |
55e075e6 | 1896 | out: |
c3d062a7 | 1897 | ovs_mutex_unlock(&dev->mutex); |
55e075e6 | 1898 | ovs_mutex_unlock(&dpdk_mutex); |
c3d062a7 | 1899 | |
55e075e6 | 1900 | return err; |
c3d062a7 CL |
1901 | } |
1902 | ||
1903 | static int | |
9fff138e DDP |
1904 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args, |
1905 | char **errp OVS_UNUSED) | |
c3d062a7 CL |
1906 | { |
1907 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1908 | ||
1909 | ovs_mutex_lock(&dev->mutex); | |
1910 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1911 | ovs_mutex_unlock(&dev->mutex); |
1912 | ||
1913 | return 0; | |
1914 | } | |
1915 | ||
c1ff66ac | 1916 | static int |
2d24d165 | 1917 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
9fff138e DDP |
1918 | const struct smap *args, |
1919 | char **errp OVS_UNUSED) | |
c1ff66ac CL |
1920 | { |
1921 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1922 | const char *path; | |
080f080c | 1923 | int max_tx_retries, cur_max_tx_retries; |
c1ff66ac | 1924 | |
6881885a | 1925 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1926 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1927 | path = smap_get(args, "vhost-server-path"); | |
bb9d2623 IM |
1928 | if (!nullable_string_is_equal(path, dev->vhost_id)) { |
1929 | free(dev->vhost_id); | |
1930 | dev->vhost_id = nullable_xstrdup(path); | |
10087cba CL |
1931 | /* check zero copy configuration */ |
1932 | if (smap_get_bool(args, "dq-zero-copy", false)) { | |
1933 | dev->vhost_driver_flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
1934 | } else { | |
1935 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
1936 | } | |
c1ff66ac CL |
1937 | netdev_request_reconfigure(netdev); |
1938 | } | |
1939 | } | |
080f080c KT |
1940 | |
1941 | max_tx_retries = smap_get_int(args, "tx-retries-max", | |
1942 | VHOST_ENQ_RETRY_DEF); | |
1943 | if (max_tx_retries < VHOST_ENQ_RETRY_MIN | |
1944 | || max_tx_retries > VHOST_ENQ_RETRY_MAX) { | |
1945 | max_tx_retries = VHOST_ENQ_RETRY_DEF; | |
1946 | } | |
1947 | atomic_read_relaxed(&dev->vhost_tx_retries_max, &cur_max_tx_retries); | |
1948 | if (max_tx_retries != cur_max_tx_retries) { | |
1949 | atomic_store_relaxed(&dev->vhost_tx_retries_max, max_tx_retries); | |
1950 | VLOG_INFO("Max Tx retries for vhost device '%s' set to %d", | |
1951 | netdev_get_name(netdev), max_tx_retries); | |
1952 | } | |
6881885a | 1953 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1954 | |
1955 | return 0; | |
1956 | } | |
1957 | ||
7dec44fe | 1958 | static int |
d46285a2 | 1959 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1960 | { |
d46285a2 | 1961 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1962 | |
d46285a2 | 1963 | return dev->socket_id; |
7dec44fe AW |
1964 | } |
1965 | ||
050c60bf | 1966 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1967 | static int |
050c60bf | 1968 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1969 | { |
d46285a2 | 1970 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1971 | |
d46285a2 | 1972 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1973 | |
050c60bf DDP |
1974 | if (dev->requested_n_txq == n_txq) { |
1975 | goto out; | |
4573fbd3 FL |
1976 | } |
1977 | ||
050c60bf DDP |
1978 | dev->requested_n_txq = n_txq; |
1979 | netdev_request_reconfigure(netdev); | |
58397e6c | 1980 | |
050c60bf | 1981 | out: |
d46285a2 | 1982 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1983 | return 0; |
58397e6c KT |
1984 | } |
1985 | ||
8a9562d2 PS |
1986 | static struct netdev_rxq * |
1987 | netdev_dpdk_rxq_alloc(void) | |
1988 | { | |
1989 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1990 | ||
eff23640 DDP |
1991 | if (rx) { |
1992 | return &rx->up; | |
1993 | } | |
1994 | ||
1995 | return NULL; | |
8a9562d2 PS |
1996 | } |
1997 | ||
1998 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1999 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 2000 | { |
d46285a2 | 2001 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
2002 | } |
2003 | ||
2004 | static int | |
d46285a2 | 2005 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 2006 | { |
d46285a2 DDP |
2007 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
2008 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 2009 | |
d46285a2 DDP |
2010 | ovs_mutex_lock(&dev->mutex); |
2011 | rx->port_id = dev->port_id; | |
2012 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2013 | |
2014 | return 0; | |
2015 | } | |
2016 | ||
2017 | static void | |
d46285a2 | 2018 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
2019 | { |
2020 | } | |
2021 | ||
2022 | static void | |
d46285a2 | 2023 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 2024 | { |
d46285a2 | 2025 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
2026 | |
2027 | rte_free(rx); | |
2028 | } | |
2029 | ||
819f13bd DDP |
2030 | /* Tries to transmit 'pkts' to txq 'qid' of device 'dev'. Takes ownership of |
2031 | * 'pkts', even in case of failure. | |
2032 | * | |
2033 | * Returns the number of packets that weren't transmitted. */ | |
2034 | static inline int | |
b59cc14e | 2035 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
819f13bd | 2036 | struct rte_mbuf **pkts, int cnt) |
8a9562d2 | 2037 | { |
1304f1f8 DDP |
2038 | uint32_t nb_tx = 0; |
2039 | ||
b59cc14e | 2040 | while (nb_tx != cnt) { |
1304f1f8 DDP |
2041 | uint32_t ret; |
2042 | ||
b59cc14e | 2043 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
2044 | if (!ret) { |
2045 | break; | |
2046 | } | |
2047 | ||
2048 | nb_tx += ret; | |
2049 | } | |
8a9562d2 | 2050 | |
b59cc14e | 2051 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
819f13bd | 2052 | /* Free buffers, which we couldn't transmit, one at a time (each |
db73f716 DDP |
2053 | * packet could come from a different mempool) */ |
2054 | int i; | |
2055 | ||
b59cc14e IM |
2056 | for (i = nb_tx; i < cnt; i++) { |
2057 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 2058 | } |
8a9562d2 | 2059 | } |
819f13bd DDP |
2060 | |
2061 | return cnt - nb_tx; | |
8a9562d2 PS |
2062 | } |
2063 | ||
f3926f29 IS |
2064 | static inline bool |
2065 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
03f3f9c0 | 2066 | struct rte_meter_srtcm_profile *profile, |
f3926f29 IS |
2067 | struct rte_mbuf *pkt, uint64_t time) |
2068 | { | |
2069 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
2070 | ||
03f3f9c0 OM |
2071 | return rte_meter_srtcm_color_blind_check(meter, profile, time, pkt_len) == |
2072 | e_RTE_METER_GREEN; | |
f3926f29 IS |
2073 | } |
2074 | ||
2075 | static int | |
2076 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
03f3f9c0 | 2077 | struct rte_meter_srtcm_profile *profile, |
3e90f7d7 | 2078 | struct rte_mbuf **pkts, int pkt_cnt, |
7d7ded7a | 2079 | bool should_steal) |
f3926f29 IS |
2080 | { |
2081 | int i = 0; | |
2082 | int cnt = 0; | |
2083 | struct rte_mbuf *pkt = NULL; | |
2084 | uint64_t current_time = rte_rdtsc(); | |
2085 | ||
2086 | for (i = 0; i < pkt_cnt; i++) { | |
2087 | pkt = pkts[i]; | |
2088 | /* Handle current packet */ | |
03f3f9c0 OM |
2089 | if (netdev_dpdk_policer_pkt_handle(meter, profile, |
2090 | pkt, current_time)) { | |
f3926f29 IS |
2091 | if (cnt != i) { |
2092 | pkts[cnt] = pkt; | |
2093 | } | |
2094 | cnt++; | |
2095 | } else { | |
7d7ded7a | 2096 | if (should_steal) { |
3e90f7d7 GZ |
2097 | rte_pktmbuf_free(pkt); |
2098 | } | |
f3926f29 IS |
2099 | } |
2100 | } | |
2101 | ||
2102 | return cnt; | |
2103 | } | |
2104 | ||
9509913a IS |
2105 | static int |
2106 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
7d7ded7a | 2107 | int pkt_cnt, bool should_steal) |
9509913a IS |
2108 | { |
2109 | int cnt = 0; | |
2110 | ||
2111 | rte_spinlock_lock(&policer->policer_lock); | |
03f3f9c0 OM |
2112 | cnt = netdev_dpdk_policer_run(&policer->in_policer, &policer->in_prof, |
2113 | pkts, pkt_cnt, should_steal); | |
9509913a IS |
2114 | rte_spinlock_unlock(&policer->policer_lock); |
2115 | ||
2116 | return cnt; | |
2117 | } | |
2118 | ||
58397e6c | 2119 | static bool |
0a0f39df | 2120 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 2121 | { |
0a0f39df | 2122 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
2123 | } |
2124 | ||
d6e3feb5 | 2125 | static inline void |
2126 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
2127 | unsigned int packet_size) | |
2128 | { | |
2129 | /* Hard-coded search for the size bucket. */ | |
2130 | if (packet_size < 256) { | |
2131 | if (packet_size >= 128) { | |
2132 | stats->rx_128_to_255_packets++; | |
2133 | } else if (packet_size <= 64) { | |
2134 | stats->rx_1_to_64_packets++; | |
2135 | } else { | |
2136 | stats->rx_65_to_127_packets++; | |
2137 | } | |
2138 | } else { | |
2139 | if (packet_size >= 1523) { | |
2140 | stats->rx_1523_to_max_packets++; | |
2141 | } else if (packet_size >= 1024) { | |
2142 | stats->rx_1024_to_1522_packets++; | |
2143 | } else if (packet_size < 512) { | |
2144 | stats->rx_256_to_511_packets++; | |
2145 | } else { | |
2146 | stats->rx_512_to_1023_packets++; | |
2147 | } | |
2148 | } | |
2149 | } | |
2150 | ||
9e3ddd45 TP |
2151 | static inline void |
2152 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
2153 | struct dp_packet **packets, int count, |
2154 | int dropped) | |
9e3ddd45 TP |
2155 | { |
2156 | int i; | |
d6e3feb5 | 2157 | unsigned int packet_size; |
9e3ddd45 TP |
2158 | struct dp_packet *packet; |
2159 | ||
2160 | stats->rx_packets += count; | |
9509913a | 2161 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
2162 | for (i = 0; i < count; i++) { |
2163 | packet = packets[i]; | |
d6e3feb5 | 2164 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 2165 | |
d6e3feb5 | 2166 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
2167 | /* This only protects the following multicast counting from |
2168 | * too short packets, but it does not stop the packet from | |
2169 | * further processing. */ | |
2170 | stats->rx_errors++; | |
2171 | stats->rx_length_errors++; | |
2172 | continue; | |
2173 | } | |
2174 | ||
d6e3feb5 | 2175 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
2176 | ||
9e3ddd45 TP |
2177 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
2178 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
2179 | stats->multicast++; | |
2180 | } | |
2181 | ||
d6e3feb5 | 2182 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
2183 | } |
2184 | } | |
2185 | ||
58397e6c KT |
2186 | /* |
2187 | * The receive path for the vhost port is the TX path out from guest. | |
2188 | */ | |
2189 | static int | |
d46285a2 | 2190 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
8492adc2 | 2191 | struct dp_packet_batch *batch, int *qfill) |
58397e6c | 2192 | { |
d46285a2 | 2193 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
9509913a | 2194 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 2195 | uint16_t nb_rx = 0; |
9509913a | 2196 | uint16_t dropped = 0; |
8492adc2 | 2197 | int qid = rxq->queue_id * VIRTIO_QNUM + VIRTIO_TXQ; |
daf22bf7 | 2198 | int vid = netdev_dpdk_get_vid(dev); |
58397e6c | 2199 | |
daf22bf7 | 2200 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured |
e543851d | 2201 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
2202 | return EAGAIN; |
2203 | } | |
2204 | ||
43307ad0 | 2205 | nb_rx = rte_vhost_dequeue_burst(vid, qid, dev->dpdk_mp->mp, |
64839cf4 | 2206 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 2207 | NETDEV_MAX_BURST); |
58397e6c KT |
2208 | if (!nb_rx) { |
2209 | return EAGAIN; | |
2210 | } | |
2211 | ||
8492adc2 JS |
2212 | if (qfill) { |
2213 | if (nb_rx == NETDEV_MAX_BURST) { | |
2214 | /* The DPDK API returns a uint32_t which often has invalid bits in | |
2215 | * the upper 16-bits. Need to restrict the value to uint16_t. */ | |
2216 | *qfill = rte_vhost_rx_queue_count(vid, qid) & UINT16_MAX; | |
2217 | } else { | |
2218 | *qfill = 0; | |
2219 | } | |
2220 | } | |
2221 | ||
9509913a IS |
2222 | if (policer) { |
2223 | dropped = nb_rx; | |
64839cf4 WT |
2224 | nb_rx = ingress_policer_run(policer, |
2225 | (struct rte_mbuf **) batch->packets, | |
3e90f7d7 | 2226 | nb_rx, true); |
9509913a IS |
2227 | dropped -= nb_rx; |
2228 | } | |
2229 | ||
d46285a2 | 2230 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
2231 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
2232 | nb_rx, dropped); | |
d46285a2 | 2233 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 2234 | |
75fb9148 ZB |
2235 | batch->count = nb_rx; |
2236 | dp_packet_batch_init_packet_fields(batch); | |
2237 | ||
58397e6c KT |
2238 | return 0; |
2239 | } | |
2240 | ||
35c91567 DM |
2241 | static bool |
2242 | netdev_dpdk_vhost_rxq_enabled(struct netdev_rxq *rxq) | |
2243 | { | |
2244 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
2245 | ||
2246 | return dev->vhost_rxq_enabled[rxq->queue_id]; | |
2247 | } | |
2248 | ||
8a9562d2 | 2249 | static int |
8492adc2 JS |
2250 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch, |
2251 | int *qfill) | |
8a9562d2 | 2252 | { |
d46285a2 DDP |
2253 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
2254 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 2255 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 2256 | int nb_rx; |
9509913a | 2257 | int dropped = 0; |
8a9562d2 | 2258 | |
3b1fb077 DDP |
2259 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
2260 | return EAGAIN; | |
2261 | } | |
2262 | ||
d46285a2 | 2263 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 2264 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 2265 | NETDEV_MAX_BURST); |
8a9562d2 PS |
2266 | if (!nb_rx) { |
2267 | return EAGAIN; | |
2268 | } | |
2269 | ||
9509913a IS |
2270 | if (policer) { |
2271 | dropped = nb_rx; | |
64839cf4 | 2272 | nb_rx = ingress_policer_run(policer, |
58be5c0e | 2273 | (struct rte_mbuf **) batch->packets, |
3e90f7d7 | 2274 | nb_rx, true); |
9509913a IS |
2275 | dropped -= nb_rx; |
2276 | } | |
2277 | ||
2278 | /* Update stats to reflect dropped packets */ | |
2279 | if (OVS_UNLIKELY(dropped)) { | |
2280 | rte_spinlock_lock(&dev->stats_lock); | |
2281 | dev->stats.rx_dropped += dropped; | |
2282 | rte_spinlock_unlock(&dev->stats_lock); | |
2283 | } | |
2284 | ||
64839cf4 | 2285 | batch->count = nb_rx; |
75fb9148 | 2286 | dp_packet_batch_init_packet_fields(batch); |
8a9562d2 | 2287 | |
8492adc2 JS |
2288 | if (qfill) { |
2289 | if (nb_rx == NETDEV_MAX_BURST) { | |
2290 | *qfill = rte_eth_rx_queue_count(rx->port_id, rxq->queue_id); | |
2291 | } else { | |
2292 | *qfill = 0; | |
2293 | } | |
2294 | } | |
2295 | ||
8a9562d2 PS |
2296 | return 0; |
2297 | } | |
2298 | ||
0bf765f7 | 2299 | static inline int |
78bd47cf | 2300 | netdev_dpdk_qos_run(struct netdev_dpdk *dev, struct rte_mbuf **pkts, |
7d7ded7a | 2301 | int cnt, bool should_steal) |
0bf765f7 | 2302 | { |
78bd47cf | 2303 | struct qos_conf *qos_conf = ovsrcu_get(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 2304 | |
78bd47cf DDP |
2305 | if (qos_conf) { |
2306 | rte_spinlock_lock(&qos_conf->lock); | |
7d7ded7a | 2307 | cnt = qos_conf->ops->qos_run(qos_conf, pkts, cnt, should_steal); |
78bd47cf | 2308 | rte_spinlock_unlock(&qos_conf->lock); |
0bf765f7 IS |
2309 | } |
2310 | ||
2311 | return cnt; | |
2312 | } | |
2313 | ||
c6ec9d17 IM |
2314 | static int |
2315 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
2316 | int pkt_cnt) | |
2317 | { | |
2318 | int i = 0; | |
2319 | int cnt = 0; | |
2320 | struct rte_mbuf *pkt; | |
2321 | ||
2322 | for (i = 0; i < pkt_cnt; i++) { | |
2323 | pkt = pkts[i]; | |
2324 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
2325 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
2326 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
2327 | rte_pktmbuf_free(pkt); | |
2328 | continue; | |
2329 | } | |
2330 | ||
2331 | if (OVS_UNLIKELY(i != cnt)) { | |
2332 | pkts[cnt] = pkt; | |
2333 | } | |
2334 | cnt++; | |
2335 | } | |
2336 | ||
2337 | return cnt; | |
2338 | } | |
2339 | ||
9e3ddd45 TP |
2340 | static inline void |
2341 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
2342 | struct dp_packet **packets, | |
2343 | int attempted, | |
2344 | int dropped) | |
2345 | { | |
2346 | int i; | |
2347 | int sent = attempted - dropped; | |
2348 | ||
2349 | stats->tx_packets += sent; | |
2350 | stats->tx_dropped += dropped; | |
2351 | ||
2352 | for (i = 0; i < sent; i++) { | |
2353 | stats->tx_bytes += dp_packet_size(packets[i]); | |
2354 | } | |
2355 | } | |
2356 | ||
58397e6c | 2357 | static void |
4573fbd3 | 2358 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 2359 | struct dp_packet **pkts, int cnt) |
58397e6c | 2360 | { |
d46285a2 | 2361 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
2362 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
2363 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 2364 | unsigned int dropped = 0; |
dd52de45 | 2365 | int i, retries = 0; |
080f080c | 2366 | int max_retries = VHOST_ENQ_RETRY_MIN; |
daf22bf7 | 2367 | int vid = netdev_dpdk_get_vid(dev); |
58397e6c | 2368 | |
81acebda | 2369 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 2370 | |
daf22bf7 | 2371 | if (OVS_UNLIKELY(vid < 0 || !dev->vhost_reconfigured || qid < 0 |
e543851d | 2372 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
2373 | rte_spinlock_lock(&dev->stats_lock); |
2374 | dev->stats.tx_dropped+= cnt; | |
2375 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 2376 | goto out; |
58397e6c KT |
2377 | } |
2378 | ||
d46285a2 | 2379 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 2380 | |
c6ec9d17 | 2381 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 2382 | /* Check has QoS has been configured for the netdev */ |
3e90f7d7 | 2383 | cnt = netdev_dpdk_qos_run(dev, cur_pkts, cnt, true); |
c6ec9d17 | 2384 | dropped = total_pkts - cnt; |
0bf765f7 | 2385 | |
95e9881f | 2386 | do { |
4573fbd3 | 2387 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
2388 | unsigned int tx_pkts; |
2389 | ||
daf22bf7 | 2390 | tx_pkts = rte_vhost_enqueue_burst(vid, vhost_qid, cur_pkts, cnt); |
95e9881f KT |
2391 | if (OVS_LIKELY(tx_pkts)) { |
2392 | /* Packets have been sent.*/ | |
2393 | cnt -= tx_pkts; | |
31871ee3 | 2394 | /* Prepare for possible retry.*/ |
95e9881f | 2395 | cur_pkts = &cur_pkts[tx_pkts]; |
080f080c KT |
2396 | if (OVS_UNLIKELY(cnt && !retries)) { |
2397 | /* | |
2398 | * Read max retries as there are packets not sent | |
2399 | * and no retries have already occurred. | |
2400 | */ | |
2401 | atomic_read_relaxed(&dev->vhost_tx_retries_max, &max_retries); | |
2402 | } | |
95e9881f | 2403 | } else { |
31871ee3 KT |
2404 | /* No packets sent - do not retry.*/ |
2405 | break; | |
95e9881f | 2406 | } |
080f080c | 2407 | } while (cnt && (retries++ < max_retries)); |
4573fbd3 | 2408 | |
d46285a2 | 2409 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 2410 | |
d46285a2 | 2411 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 2412 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 2413 | cnt + dropped); |
080f080c | 2414 | dev->tx_retries += MIN(retries, max_retries); |
d46285a2 | 2415 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
2416 | |
2417 | out: | |
c6ec9d17 | 2418 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 2419 | dp_packet_delete(pkts[i]); |
58397e6c KT |
2420 | } |
2421 | } | |
2422 | ||
8a9562d2 PS |
2423 | /* Tx function. Transmit packets indefinitely */ |
2424 | static void | |
64839cf4 | 2425 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 2426 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 2427 | { |
8a14bd7b | 2428 | const size_t batch_cnt = dp_packet_batch_size(batch); |
bce01e3a | 2429 | #if !defined(__CHECKER__) && !defined(_WIN32) |
8a14bd7b | 2430 | const size_t PKT_ARRAY_SIZE = batch_cnt; |
bce01e3a EJ |
2431 | #else |
2432 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 2433 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 2434 | #endif |
8a9562d2 | 2435 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2391135c | 2436 | struct rte_mbuf *pkts[PKT_ARRAY_SIZE]; |
8a14bd7b | 2437 | uint32_t cnt = batch_cnt; |
3e90f7d7 GZ |
2438 | uint32_t dropped = 0; |
2439 | ||
2440 | if (dev->type != DPDK_DEV_VHOST) { | |
2441 | /* Check if QoS has been configured for this netdev. */ | |
2442 | cnt = netdev_dpdk_qos_run(dev, (struct rte_mbuf **) batch->packets, | |
8a14bd7b BB |
2443 | batch_cnt, false); |
2444 | dropped += batch_cnt - cnt; | |
3e90f7d7 | 2445 | } |
8a9562d2 | 2446 | |
3e90f7d7 GZ |
2447 | uint32_t txcnt = 0; |
2448 | ||
2449 | for (uint32_t i = 0; i < cnt; i++) { | |
8a14bd7b BB |
2450 | struct dp_packet *packet = batch->packets[i]; |
2451 | uint32_t size = dp_packet_size(packet); | |
95fb793a | 2452 | |
f98d7864 | 2453 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
3e90f7d7 GZ |
2454 | VLOG_WARN_RL(&rl, "Too big size %u max_packet_len %d", |
2455 | size, dev->max_packet_len); | |
f4fd623c | 2456 | |
175cf4de | 2457 | dropped++; |
f4fd623c DDP |
2458 | continue; |
2459 | } | |
8a9562d2 | 2460 | |
43307ad0 | 2461 | pkts[txcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a14bd7b | 2462 | if (OVS_UNLIKELY(!pkts[txcnt])) { |
3e90f7d7 | 2463 | dropped += cnt - i; |
175cf4de | 2464 | break; |
f4fd623c DDP |
2465 | } |
2466 | ||
2467 | /* We have to do a copy for now */ | |
3e90f7d7 | 2468 | memcpy(rte_pktmbuf_mtod(pkts[txcnt], void *), |
8a14bd7b BB |
2469 | dp_packet_data(packet), size); |
2470 | dp_packet_set_size((struct dp_packet *)pkts[txcnt], size); | |
f4fd623c | 2471 | |
3e90f7d7 | 2472 | txcnt++; |
f4fd623c | 2473 | } |
8a9562d2 | 2474 | |
3e90f7d7 GZ |
2475 | if (OVS_LIKELY(txcnt)) { |
2476 | if (dev->type == DPDK_DEV_VHOST) { | |
2477 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) pkts, | |
2478 | txcnt); | |
2479 | } else { | |
2480 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, txcnt); | |
2481 | } | |
58397e6c | 2482 | } |
db73f716 | 2483 | |
0bf765f7 IS |
2484 | if (OVS_UNLIKELY(dropped)) { |
2485 | rte_spinlock_lock(&dev->stats_lock); | |
2486 | dev->stats.tx_dropped += dropped; | |
2487 | rte_spinlock_unlock(&dev->stats_lock); | |
2488 | } | |
8a9562d2 PS |
2489 | } |
2490 | ||
58397e6c | 2491 | static int |
64839cf4 WT |
2492 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
2493 | struct dp_packet_batch *batch, | |
b30896c9 | 2494 | bool concurrent_txq OVS_UNUSED) |
58397e6c | 2495 | { |
58397e6c | 2496 | |
b30896c9 | 2497 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 | 2498 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2499 | dp_packet_delete_batch(batch, true); |
58397e6c | 2500 | } else { |
940ac2ce PC |
2501 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, |
2502 | dp_packet_batch_size(batch)); | |
58397e6c KT |
2503 | } |
2504 | return 0; | |
2505 | } | |
2506 | ||
7251515e DV |
2507 | static inline void |
2508 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
b30896c9 | 2509 | struct dp_packet_batch *batch, |
324c8374 | 2510 | bool concurrent_txq) |
8a9562d2 | 2511 | { |
3b1fb077 | 2512 | if (OVS_UNLIKELY(!(dev->flags & NETDEV_UP))) { |
b30896c9 | 2513 | dp_packet_delete_batch(batch, true); |
3b1fb077 DDP |
2514 | return; |
2515 | } | |
2516 | ||
324c8374 | 2517 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 2518 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
2519 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
2520 | } | |
2521 | ||
b30896c9 | 2522 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
2523 | struct netdev *netdev = &dev->up; |
2524 | ||
64839cf4 | 2525 | dpdk_do_tx_copy(netdev, qid, batch); |
b30896c9 | 2526 | dp_packet_delete_batch(batch, true); |
8a9562d2 | 2527 | } else { |
fd57eeba BB |
2528 | int tx_cnt, dropped; |
2529 | int batch_cnt = dp_packet_batch_size(batch); | |
2391135c | 2530 | struct rte_mbuf **pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 2531 | |
fd57eeba BB |
2532 | tx_cnt = netdev_dpdk_filter_packet_len(dev, pkts, batch_cnt); |
2533 | tx_cnt = netdev_dpdk_qos_run(dev, pkts, tx_cnt, true); | |
2534 | dropped = batch_cnt - tx_cnt; | |
1b99bb05 | 2535 | |
fd57eeba | 2536 | dropped += netdev_dpdk_eth_tx_burst(dev, qid, pkts, tx_cnt); |
8a9562d2 | 2537 | |
f4fd623c | 2538 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 2539 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 2540 | dev->stats.tx_dropped += dropped; |
45d947c4 | 2541 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 2542 | } |
8a9562d2 | 2543 | } |
a0cb2d66 | 2544 | |
324c8374 | 2545 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
2546 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
2547 | } | |
7251515e DV |
2548 | } |
2549 | ||
2550 | static int | |
2551 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
b30896c9 | 2552 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e DV |
2553 | { |
2554 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 2555 | |
b30896c9 | 2556 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e | 2557 | return 0; |
8a9562d2 PS |
2558 | } |
2559 | ||
2560 | static int | |
74ff3298 | 2561 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
2562 | { |
2563 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2564 | ||
2565 | ovs_mutex_lock(&dev->mutex); | |
2566 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 2567 | dev->hwaddr = mac; |
045c0d1a | 2568 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
2569 | } |
2570 | ovs_mutex_unlock(&dev->mutex); | |
2571 | ||
2572 | return 0; | |
2573 | } | |
2574 | ||
2575 | static int | |
74ff3298 | 2576 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
2577 | { |
2578 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2579 | ||
2580 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 2581 | *mac = dev->hwaddr; |
8a9562d2 PS |
2582 | ovs_mutex_unlock(&dev->mutex); |
2583 | ||
2584 | return 0; | |
2585 | } | |
2586 | ||
2587 | static int | |
2588 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
2589 | { | |
2590 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2591 | ||
2592 | ovs_mutex_lock(&dev->mutex); | |
2593 | *mtup = dev->mtu; | |
2594 | ovs_mutex_unlock(&dev->mutex); | |
2595 | ||
2596 | return 0; | |
2597 | } | |
2598 | ||
0072e931 MK |
2599 | static int |
2600 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
2601 | { | |
2602 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2603 | ||
f6f50552 IS |
2604 | /* XXX: Ensure that the overall frame length of the requested MTU does not |
2605 | * surpass the NETDEV_DPDK_MAX_PKT_LEN. DPDK device drivers differ in how | |
2606 | * the L2 frame length is calculated for a given MTU when | |
2607 | * rte_eth_dev_set_mtu(mtu) is called e.g. i40e driver includes 2 x vlan | |
2608 | * headers, the em driver includes 1 x vlan header, the ixgbe driver does | |
2609 | * not include vlan headers. As such we should use | |
2610 | * MTU_TO_MAX_FRAME_LEN(mtu) which includes an additional 2 x vlan headers | |
2611 | * (8 bytes) for comparison. This avoids a failure later with | |
2612 | * rte_eth_dev_set_mtu(). This approach should be used until DPDK provides | |
2613 | * a method to retrieve the upper bound MTU for a given device. | |
2614 | */ | |
2615 | if (MTU_TO_MAX_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
0072e931 MK |
2616 | || mtu < ETHER_MIN_MTU) { |
2617 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
2618 | return EINVAL; | |
2619 | } | |
2620 | ||
2621 | ovs_mutex_lock(&dev->mutex); | |
2622 | if (dev->requested_mtu != mtu) { | |
2623 | dev->requested_mtu = mtu; | |
2624 | netdev_request_reconfigure(netdev); | |
2625 | } | |
2626 | ovs_mutex_unlock(&dev->mutex); | |
2627 | ||
2628 | return 0; | |
2629 | } | |
2630 | ||
8a9562d2 | 2631 | static int |
d46285a2 | 2632 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 2633 | |
58397e6c KT |
2634 | static int |
2635 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
2636 | struct netdev_stats *stats) | |
2637 | { | |
2638 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2639 | ||
2640 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 2641 | |
45d947c4 | 2642 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c | 2643 | /* Supported Stats */ |
50986e78 | 2644 | stats->rx_packets = dev->stats.rx_packets; |
2645 | stats->tx_packets = dev->stats.tx_packets; | |
9509913a | 2646 | stats->rx_dropped = dev->stats.rx_dropped; |
50986e78 | 2647 | stats->tx_dropped = dev->stats.tx_dropped; |
9e3ddd45 TP |
2648 | stats->multicast = dev->stats.multicast; |
2649 | stats->rx_bytes = dev->stats.rx_bytes; | |
2650 | stats->tx_bytes = dev->stats.tx_bytes; | |
2651 | stats->rx_errors = dev->stats.rx_errors; | |
2652 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 2653 | |
2654 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
2655 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
2656 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
2657 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
2658 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
2659 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
2660 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
2661 | ||
45d947c4 | 2662 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 2663 | |
58397e6c KT |
2664 | ovs_mutex_unlock(&dev->mutex); |
2665 | ||
2666 | return 0; | |
2667 | } | |
2668 | ||
d6e3feb5 | 2669 | static void |
2670 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
2671 | const struct rte_eth_xstat *xstats, |
2672 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 2673 | const unsigned int size) |
2674 | { | |
18366d16 IM |
2675 | /* DPDK XSTATS Counter names definition. */ |
2676 | #define DPDK_XSTATS \ | |
2677 | DPDK_XSTAT(multicast, "rx_multicast_packets" ) \ | |
2678 | DPDK_XSTAT(tx_multicast_packets, "tx_multicast_packets" ) \ | |
2679 | DPDK_XSTAT(rx_broadcast_packets, "rx_broadcast_packets" ) \ | |
2680 | DPDK_XSTAT(tx_broadcast_packets, "tx_broadcast_packets" ) \ | |
2681 | DPDK_XSTAT(rx_undersized_errors, "rx_undersized_errors" ) \ | |
2682 | DPDK_XSTAT(rx_oversize_errors, "rx_oversize_errors" ) \ | |
2683 | DPDK_XSTAT(rx_fragmented_errors, "rx_fragmented_errors" ) \ | |
2684 | DPDK_XSTAT(rx_jabber_errors, "rx_jabber_errors" ) \ | |
2685 | DPDK_XSTAT(rx_1_to_64_packets, "rx_size_64_packets" ) \ | |
2686 | DPDK_XSTAT(rx_65_to_127_packets, "rx_size_65_to_127_packets" ) \ | |
2687 | DPDK_XSTAT(rx_128_to_255_packets, "rx_size_128_to_255_packets" ) \ | |
2688 | DPDK_XSTAT(rx_256_to_511_packets, "rx_size_256_to_511_packets" ) \ | |
2689 | DPDK_XSTAT(rx_512_to_1023_packets, "rx_size_512_to_1023_packets" ) \ | |
2690 | DPDK_XSTAT(rx_1024_to_1522_packets, "rx_size_1024_to_1522_packets" ) \ | |
2691 | DPDK_XSTAT(rx_1523_to_max_packets, "rx_size_1523_to_max_packets" ) \ | |
2692 | DPDK_XSTAT(tx_1_to_64_packets, "tx_size_64_packets" ) \ | |
2693 | DPDK_XSTAT(tx_65_to_127_packets, "tx_size_65_to_127_packets" ) \ | |
2694 | DPDK_XSTAT(tx_128_to_255_packets, "tx_size_128_to_255_packets" ) \ | |
2695 | DPDK_XSTAT(tx_256_to_511_packets, "tx_size_256_to_511_packets" ) \ | |
2696 | DPDK_XSTAT(tx_512_to_1023_packets, "tx_size_512_to_1023_packets" ) \ | |
2697 | DPDK_XSTAT(tx_1024_to_1522_packets, "tx_size_1024_to_1522_packets" ) \ | |
2698 | DPDK_XSTAT(tx_1523_to_max_packets, "tx_size_1523_to_max_packets" ) | |
2699 | ||
d6e3feb5 | 2700 | for (unsigned int i = 0; i < size; i++) { |
18366d16 IM |
2701 | #define DPDK_XSTAT(MEMBER, NAME) \ |
2702 | if (strcmp(NAME, names[i].name) == 0) { \ | |
2703 | stats->MEMBER = xstats[i].value; \ | |
2704 | continue; \ | |
d6e3feb5 | 2705 | } |
18366d16 IM |
2706 | DPDK_XSTATS; |
2707 | #undef DPDK_XSTAT | |
d6e3feb5 | 2708 | } |
18366d16 | 2709 | #undef DPDK_XSTATS |
d6e3feb5 | 2710 | } |
2711 | ||
8a9562d2 PS |
2712 | static int |
2713 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
2714 | { | |
2715 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2716 | struct rte_eth_stats rte_stats; | |
2717 | bool gg; | |
2718 | ||
2719 | netdev_dpdk_get_carrier(netdev, &gg); | |
2720 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 2721 | |
0a0f39df CL |
2722 | struct rte_eth_xstat *rte_xstats = NULL; |
2723 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
2724 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 2725 | |
2726 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
fa9f4eeb IM |
2727 | VLOG_ERR("Can't get ETH statistics for port: "DPDK_PORT_ID_FMT, |
2728 | dev->port_id); | |
f9256822 | 2729 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 2730 | return EPROTO; |
2731 | } | |
2732 | ||
0a0f39df CL |
2733 | /* Get length of statistics */ |
2734 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
2735 | if (rte_xstats_len < 0) { | |
fa9f4eeb IM |
2736 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
2737 | dev->port_id); | |
0a0f39df CL |
2738 | goto out; |
2739 | } | |
2740 | /* Reserve memory for xstats names and values */ | |
2741 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
2742 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
2743 | ||
2744 | /* Retreive xstats names */ | |
2745 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
2746 | rte_xstats_names, | |
2747 | rte_xstats_len); | |
2748 | if (rte_xstats_new_len != rte_xstats_len) { | |
fa9f4eeb IM |
2749 | VLOG_WARN("Cannot get XSTATS names for port: "DPDK_PORT_ID_FMT, |
2750 | dev->port_id); | |
0a0f39df CL |
2751 | goto out; |
2752 | } | |
2753 | /* Retreive xstats values */ | |
2754 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
2755 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
2756 | rte_xstats_len); | |
2757 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
2758 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
2759 | rte_xstats_len); | |
d6e3feb5 | 2760 | } else { |
fa9f4eeb IM |
2761 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
2762 | dev->port_id); | |
d6e3feb5 | 2763 | } |
8a9562d2 | 2764 | |
0a0f39df CL |
2765 | out: |
2766 | free(rte_xstats); | |
2767 | free(rte_xstats_names); | |
2768 | ||
2f9dd77f PS |
2769 | stats->rx_packets = rte_stats.ipackets; |
2770 | stats->tx_packets = rte_stats.opackets; | |
2771 | stats->rx_bytes = rte_stats.ibytes; | |
2772 | stats->tx_bytes = rte_stats.obytes; | |
21e9844c | 2773 | stats->rx_errors = rte_stats.ierrors; |
2f9dd77f | 2774 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 2775 | |
45d947c4 | 2776 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 2777 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 2778 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 2779 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
2780 | |
2781 | /* These are the available DPDK counters for packets not received due to | |
2782 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 2783 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
2784 | stats->rx_missed_errors = rte_stats.imissed; |
2785 | ||
8a9562d2 PS |
2786 | ovs_mutex_unlock(&dev->mutex); |
2787 | ||
2788 | return 0; | |
2789 | } | |
2790 | ||
971f4b39 MW |
2791 | static int |
2792 | netdev_dpdk_get_custom_stats(const struct netdev *netdev, | |
2793 | struct netdev_custom_stats *custom_stats) | |
2794 | { | |
2795 | ||
2796 | uint32_t i; | |
2797 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2798 | int rte_xstats_ret; | |
2799 | ||
2800 | ovs_mutex_lock(&dev->mutex); | |
2801 | ||
2802 | if (netdev_dpdk_configure_xstats(dev)) { | |
2803 | uint64_t *values = xcalloc(dev->rte_xstats_ids_size, | |
2804 | sizeof(uint64_t)); | |
2805 | ||
2806 | rte_xstats_ret = | |
2807 | rte_eth_xstats_get_by_id(dev->port_id, dev->rte_xstats_ids, | |
2808 | values, dev->rte_xstats_ids_size); | |
2809 | ||
2810 | if (rte_xstats_ret > 0 && | |
2811 | rte_xstats_ret <= dev->rte_xstats_ids_size) { | |
2812 | ||
2813 | custom_stats->size = rte_xstats_ret; | |
2814 | custom_stats->counters = | |
2815 | (struct netdev_custom_counter *) xcalloc(rte_xstats_ret, | |
2816 | sizeof(struct netdev_custom_counter)); | |
2817 | ||
2818 | for (i = 0; i < rte_xstats_ret; i++) { | |
2819 | ovs_strlcpy(custom_stats->counters[i].name, | |
2820 | netdev_dpdk_get_xstat_name(dev, | |
2821 | dev->rte_xstats_ids[i]), | |
2822 | NETDEV_CUSTOM_STATS_NAME_SIZE); | |
2823 | custom_stats->counters[i].value = values[i]; | |
2824 | } | |
2825 | } else { | |
fa9f4eeb | 2826 | VLOG_WARN("Cannot get XSTATS values for port: "DPDK_PORT_ID_FMT, |
971f4b39 MW |
2827 | dev->port_id); |
2828 | custom_stats->counters = NULL; | |
2829 | custom_stats->size = 0; | |
2830 | /* Let's clear statistics cache, so it will be | |
2831 | * reconfigured */ | |
2832 | netdev_dpdk_clear_xstats(dev); | |
2833 | } | |
526259f2 IM |
2834 | |
2835 | free(values); | |
971f4b39 MW |
2836 | } |
2837 | ||
2838 | ovs_mutex_unlock(&dev->mutex); | |
2839 | ||
2840 | return 0; | |
2841 | } | |
2842 | ||
c161357d KT |
2843 | static int |
2844 | netdev_dpdk_vhost_get_custom_stats(const struct netdev *netdev, | |
2845 | struct netdev_custom_stats *custom_stats) | |
2846 | { | |
2847 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
5c7ba90d | 2848 | int i; |
c161357d | 2849 | |
5c7ba90d IM |
2850 | #define VHOST_CSTATS \ |
2851 | VHOST_CSTAT(tx_retries) | |
c161357d | 2852 | |
5c7ba90d IM |
2853 | #define VHOST_CSTAT(NAME) + 1 |
2854 | custom_stats->size = VHOST_CSTATS; | |
2855 | #undef VHOST_CSTAT | |
c161357d KT |
2856 | custom_stats->counters = xcalloc(custom_stats->size, |
2857 | sizeof *custom_stats->counters); | |
5c7ba90d IM |
2858 | i = 0; |
2859 | #define VHOST_CSTAT(NAME) \ | |
2860 | ovs_strlcpy(custom_stats->counters[i++].name, #NAME, \ | |
c161357d | 2861 | NETDEV_CUSTOM_STATS_NAME_SIZE); |
5c7ba90d IM |
2862 | VHOST_CSTATS; |
2863 | #undef VHOST_CSTAT | |
2864 | ||
2865 | ovs_mutex_lock(&dev->mutex); | |
c161357d KT |
2866 | |
2867 | rte_spinlock_lock(&dev->stats_lock); | |
5c7ba90d IM |
2868 | i = 0; |
2869 | #define VHOST_CSTAT(NAME) \ | |
2870 | custom_stats->counters[i++].value = dev->NAME; | |
2871 | VHOST_CSTATS; | |
2872 | #undef VHOST_CSTAT | |
c161357d KT |
2873 | rte_spinlock_unlock(&dev->stats_lock); |
2874 | ||
2875 | ovs_mutex_unlock(&dev->mutex); | |
2876 | ||
2877 | return 0; | |
2878 | } | |
2879 | ||
8a9562d2 | 2880 | static int |
d46285a2 | 2881 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 | 2882 | enum netdev_features *current, |
ca3d4f55 BX |
2883 | enum netdev_features *advertised, |
2884 | enum netdev_features *supported, | |
2885 | enum netdev_features *peer) | |
8a9562d2 | 2886 | { |
d46285a2 | 2887 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 2888 | struct rte_eth_link link; |
dfcb5b8a | 2889 | uint32_t feature = 0; |
8a9562d2 PS |
2890 | |
2891 | ovs_mutex_lock(&dev->mutex); | |
2892 | link = dev->link; | |
2893 | ovs_mutex_unlock(&dev->mutex); | |
2894 | ||
dfcb5b8a IS |
2895 | /* Match against OpenFlow defined link speed values. */ |
2896 | if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
2897 | switch (link.link_speed) { | |
2898 | case ETH_SPEED_NUM_10M: | |
2899 | feature |= NETDEV_F_10MB_FD; | |
2900 | break; | |
2901 | case ETH_SPEED_NUM_100M: | |
2902 | feature |= NETDEV_F_100MB_FD; | |
2903 | break; | |
2904 | case ETH_SPEED_NUM_1G: | |
2905 | feature |= NETDEV_F_1GB_FD; | |
2906 | break; | |
2907 | case ETH_SPEED_NUM_10G: | |
2908 | feature |= NETDEV_F_10GB_FD; | |
2909 | break; | |
2910 | case ETH_SPEED_NUM_40G: | |
2911 | feature |= NETDEV_F_40GB_FD; | |
2912 | break; | |
2913 | case ETH_SPEED_NUM_100G: | |
2914 | feature |= NETDEV_F_100GB_FD; | |
2915 | break; | |
2916 | default: | |
2917 | feature |= NETDEV_F_OTHER; | |
8a9562d2 | 2918 | } |
dfcb5b8a IS |
2919 | } else if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
2920 | switch (link.link_speed) { | |
2921 | case ETH_SPEED_NUM_10M: | |
2922 | feature |= NETDEV_F_10MB_HD; | |
2923 | break; | |
2924 | case ETH_SPEED_NUM_100M: | |
2925 | feature |= NETDEV_F_100MB_HD; | |
2926 | break; | |
2927 | case ETH_SPEED_NUM_1G: | |
2928 | feature |= NETDEV_F_1GB_HD; | |
2929 | break; | |
2930 | default: | |
2931 | feature |= NETDEV_F_OTHER; | |
74cd69a4 | 2932 | } |
8a9562d2 PS |
2933 | } |
2934 | ||
362ca396 | 2935 | if (link.link_autoneg) { |
dfcb5b8a | 2936 | feature |= NETDEV_F_AUTONEG; |
362ca396 | 2937 | } |
2938 | ||
dfcb5b8a | 2939 | *current = feature; |
ca3d4f55 BX |
2940 | *advertised = *supported = *peer = 0; |
2941 | ||
8a9562d2 PS |
2942 | return 0; |
2943 | } | |
2944 | ||
9509913a IS |
2945 | static struct ingress_policer * |
2946 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
2947 | { | |
2948 | struct ingress_policer *policer = NULL; | |
2949 | uint64_t rate_bytes; | |
2950 | uint64_t burst_bytes; | |
2951 | int err = 0; | |
2952 | ||
2953 | policer = xmalloc(sizeof *policer); | |
2954 | rte_spinlock_init(&policer->policer_lock); | |
2955 | ||
2956 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
602c8668 LR |
2957 | rate_bytes = rate * 1000ULL / 8; |
2958 | burst_bytes = burst * 1000ULL / 8; | |
9509913a IS |
2959 | |
2960 | policer->app_srtcm_params.cir = rate_bytes; | |
2961 | policer->app_srtcm_params.cbs = burst_bytes; | |
2962 | policer->app_srtcm_params.ebs = 0; | |
03f3f9c0 OM |
2963 | err = rte_meter_srtcm_profile_config(&policer->in_prof, |
2964 | &policer->app_srtcm_params); | |
2965 | if (!err) { | |
2966 | err = rte_meter_srtcm_config(&policer->in_policer, | |
2967 | &policer->in_prof); | |
2968 | } | |
58be5c0e | 2969 | if (err) { |
9509913a | 2970 | VLOG_ERR("Could not create rte meter for ingress policer"); |
4c47ddde | 2971 | free(policer); |
9509913a IS |
2972 | return NULL; |
2973 | } | |
2974 | ||
2975 | return policer; | |
2976 | } | |
2977 | ||
2978 | static int | |
2979 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2980 | uint32_t policer_burst) | |
2981 | { | |
2982 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2983 | struct ingress_policer *policer; | |
2984 | ||
2985 | /* Force to 0 if no rate specified, | |
2986 | * default to 8000 kbits if burst is 0, | |
2987 | * else stick with user-specified value. | |
2988 | */ | |
2989 | policer_burst = (!policer_rate ? 0 | |
2990 | : !policer_burst ? 8000 | |
2991 | : policer_burst); | |
2992 | ||
2993 | ovs_mutex_lock(&dev->mutex); | |
2994 | ||
2995 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2996 | &dev->ingress_policer); | |
2997 | ||
2998 | if (dev->policer_rate == policer_rate && | |
2999 | dev->policer_burst == policer_burst) { | |
3000 | /* Assume that settings haven't changed since we last set them. */ | |
3001 | ovs_mutex_unlock(&dev->mutex); | |
3002 | return 0; | |
3003 | } | |
3004 | ||
3005 | /* Destroy any existing ingress policer for the device if one exists */ | |
3006 | if (policer) { | |
3007 | ovsrcu_postpone(free, policer); | |
3008 | } | |
3009 | ||
3010 | if (policer_rate != 0) { | |
3011 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
3012 | } else { | |
3013 | policer = NULL; | |
3014 | } | |
3015 | ovsrcu_set(&dev->ingress_policer, policer); | |
3016 | dev->policer_rate = policer_rate; | |
3017 | dev->policer_burst = policer_burst; | |
3018 | ovs_mutex_unlock(&dev->mutex); | |
3019 | ||
3020 | return 0; | |
3021 | } | |
3022 | ||
8a9562d2 PS |
3023 | static int |
3024 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
3025 | { | |
3026 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 PS |
3027 | |
3028 | ovs_mutex_lock(&dev->mutex); | |
12d0d124 PL |
3029 | /* Calculate hash from the netdev name. Ensure that ifindex is a 24-bit |
3030 | * postive integer to meet RFC 2863 recommendations. | |
3031 | */ | |
3032 | int ifindex = hash_string(netdev->name, 0) % 0xfffffe + 1; | |
8a9562d2 PS |
3033 | ovs_mutex_unlock(&dev->mutex); |
3034 | ||
3035 | return ifindex; | |
3036 | } | |
3037 | ||
3038 | static int | |
d46285a2 | 3039 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 3040 | { |
d46285a2 | 3041 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
3042 | |
3043 | ovs_mutex_lock(&dev->mutex); | |
3044 | check_link_status(dev); | |
3045 | *carrier = dev->link.link_status; | |
58397e6c KT |
3046 | |
3047 | ovs_mutex_unlock(&dev->mutex); | |
3048 | ||
3049 | return 0; | |
3050 | } | |
3051 | ||
3052 | static int | |
d46285a2 | 3053 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 3054 | { |
d46285a2 | 3055 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
3056 | |
3057 | ovs_mutex_lock(&dev->mutex); | |
3058 | ||
0a0f39df | 3059 | if (is_vhost_running(dev)) { |
58397e6c KT |
3060 | *carrier = 1; |
3061 | } else { | |
3062 | *carrier = 0; | |
3063 | } | |
3064 | ||
8a9562d2 PS |
3065 | ovs_mutex_unlock(&dev->mutex); |
3066 | ||
3067 | return 0; | |
3068 | } | |
3069 | ||
3070 | static long long int | |
d46285a2 | 3071 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 3072 | { |
d46285a2 | 3073 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
3074 | long long int carrier_resets; |
3075 | ||
3076 | ovs_mutex_lock(&dev->mutex); | |
3077 | carrier_resets = dev->link_reset_cnt; | |
3078 | ovs_mutex_unlock(&dev->mutex); | |
3079 | ||
3080 | return carrier_resets; | |
3081 | } | |
3082 | ||
3083 | static int | |
d46285a2 | 3084 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
3085 | long long int interval OVS_UNUSED) |
3086 | { | |
ee32150e | 3087 | return EOPNOTSUPP; |
8a9562d2 PS |
3088 | } |
3089 | ||
3090 | static int | |
3091 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
3092 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
3093 | enum netdev_flags *old_flagsp) |
3094 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 | 3095 | { |
8a9562d2 PS |
3096 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { |
3097 | return EINVAL; | |
3098 | } | |
3099 | ||
3100 | *old_flagsp = dev->flags; | |
3101 | dev->flags |= on; | |
3102 | dev->flags &= ~off; | |
3103 | ||
3104 | if (dev->flags == *old_flagsp) { | |
3105 | return 0; | |
3106 | } | |
3107 | ||
58397e6c | 3108 | if (dev->type == DPDK_DEV_ETH) { |
2d37de73 EC |
3109 | |
3110 | if ((dev->flags ^ *old_flagsp) & NETDEV_UP) { | |
3111 | int err; | |
3112 | ||
3113 | if (dev->flags & NETDEV_UP) { | |
3114 | err = rte_eth_dev_set_link_up(dev->port_id); | |
3115 | } else { | |
3116 | err = rte_eth_dev_set_link_down(dev->port_id); | |
3117 | } | |
3118 | if (err == -ENOTSUP) { | |
3119 | VLOG_INFO("Interface %s does not support link state " | |
3120 | "configuration", netdev_get_name(&dev->up)); | |
3121 | } else if (err < 0) { | |
3122 | VLOG_ERR("Interface %s link change error: %s", | |
3123 | netdev_get_name(&dev->up), rte_strerror(-err)); | |
3124 | dev->flags = *old_flagsp; | |
3125 | return -err; | |
3126 | } | |
3127 | } | |
3128 | ||
58397e6c KT |
3129 | if (dev->flags & NETDEV_PROMISC) { |
3130 | rte_eth_promiscuous_enable(dev->port_id); | |
3131 | } | |
8a9562d2 | 3132 | |
314fb5ad | 3133 | netdev_change_seq_changed(&dev->up); |
e543851d ZB |
3134 | } else { |
3135 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
3136 | * running then change netdev's change_seq to trigger link state | |
3137 | * update. */ | |
e543851d ZB |
3138 | |
3139 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 3140 | && is_vhost_running(dev)) { |
e543851d ZB |
3141 | netdev_change_seq_changed(&dev->up); |
3142 | ||
3143 | /* Clear statistics if device is getting up. */ | |
3144 | if (NETDEV_UP & on) { | |
3145 | rte_spinlock_lock(&dev->stats_lock); | |
58be5c0e | 3146 | memset(&dev->stats, 0, sizeof dev->stats); |
e543851d ZB |
3147 | rte_spinlock_unlock(&dev->stats_lock); |
3148 | } | |
3149 | } | |
8a9562d2 PS |
3150 | } |
3151 | ||
3152 | return 0; | |
3153 | } | |
3154 | ||
3155 | static int | |
d46285a2 | 3156 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
3157 | enum netdev_flags off, enum netdev_flags on, |
3158 | enum netdev_flags *old_flagsp) | |
3159 | { | |
d46285a2 | 3160 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
3161 | int error; |
3162 | ||
d46285a2 DDP |
3163 | ovs_mutex_lock(&dev->mutex); |
3164 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
3165 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
3166 | |
3167 | return error; | |
3168 | } | |
3169 | ||
b2e8b12f FL |
3170 | static int |
3171 | netdev_dpdk_vhost_user_get_status(const struct netdev *netdev, | |
3172 | struct smap *args) | |
3173 | { | |
3174 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3175 | ||
3176 | ovs_mutex_lock(&dev->mutex); | |
3177 | ||
3178 | bool client_mode = dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT; | |
3179 | smap_add_format(args, "mode", "%s", client_mode ? "client" : "server"); | |
3180 | ||
3181 | int vid = netdev_dpdk_get_vid(dev); | |
3182 | if (vid < 0) { | |
3183 | smap_add_format(args, "status", "disconnected"); | |
3184 | ovs_mutex_unlock(&dev->mutex); | |
3185 | return 0; | |
3186 | } else { | |
3187 | smap_add_format(args, "status", "connected"); | |
3188 | } | |
3189 | ||
3190 | char socket_name[PATH_MAX]; | |
3191 | if (!rte_vhost_get_ifname(vid, socket_name, PATH_MAX)) { | |
3192 | smap_add_format(args, "socket", "%s", socket_name); | |
3193 | } | |
3194 | ||
3195 | uint64_t features; | |
3196 | if (!rte_vhost_get_negotiated_features(vid, &features)) { | |
3197 | smap_add_format(args, "features", "0x%016"PRIx64, features); | |
3198 | } | |
3199 | ||
3200 | uint16_t mtu; | |
3201 | if (!rte_vhost_get_mtu(vid, &mtu)) { | |
3202 | smap_add_format(args, "mtu", "%d", mtu); | |
3203 | } | |
3204 | ||
3205 | int numa = rte_vhost_get_numa_node(vid); | |
3206 | if (numa >= 0) { | |
3207 | smap_add_format(args, "numa", "%d", numa); | |
3208 | } | |
3209 | ||
3210 | uint16_t vring_num = rte_vhost_get_vring_num(vid); | |
3211 | if (vring_num) { | |
3212 | smap_add_format(args, "num_of_vrings", "%d", vring_num); | |
3213 | } | |
3214 | ||
3215 | for (int i = 0; i < vring_num; i++) { | |
3216 | struct rte_vhost_vring vring; | |
b2e8b12f FL |
3217 | |
3218 | rte_vhost_get_vhost_vring(vid, i, &vring); | |
b9a3183d AC |
3219 | smap_add_nocopy(args, xasprintf("vring_%d_size", i), |
3220 | xasprintf("%d", vring.size)); | |
b2e8b12f FL |
3221 | } |
3222 | ||
3223 | ovs_mutex_unlock(&dev->mutex); | |
3224 | return 0; | |
3225 | } | |
3226 | ||
31154f95 IS |
3227 | /* |
3228 | * Convert a given uint32_t link speed defined in DPDK to a string | |
3229 | * equivalent. | |
3230 | */ | |
3231 | static const char * | |
3232 | netdev_dpdk_link_speed_to_str__(uint32_t link_speed) | |
3233 | { | |
3234 | switch (link_speed) { | |
3235 | case ETH_SPEED_NUM_10M: return "10Mbps"; | |
3236 | case ETH_SPEED_NUM_100M: return "100Mbps"; | |
3237 | case ETH_SPEED_NUM_1G: return "1Gbps"; | |
3238 | case ETH_SPEED_NUM_2_5G: return "2.5Gbps"; | |
3239 | case ETH_SPEED_NUM_5G: return "5Gbps"; | |
3240 | case ETH_SPEED_NUM_10G: return "10Gbps"; | |
3241 | case ETH_SPEED_NUM_20G: return "20Gbps"; | |
3242 | case ETH_SPEED_NUM_25G: return "25Gbps"; | |
3243 | case ETH_SPEED_NUM_40G: return "40Gbps"; | |
3244 | case ETH_SPEED_NUM_50G: return "50Gbps"; | |
3245 | case ETH_SPEED_NUM_56G: return "56Gbps"; | |
3246 | case ETH_SPEED_NUM_100G: return "100Gbps"; | |
3247 | default: return "Not Defined"; | |
3248 | } | |
3249 | } | |
3250 | ||
8a9562d2 | 3251 | static int |
d46285a2 | 3252 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 3253 | { |
d46285a2 | 3254 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 3255 | struct rte_eth_dev_info dev_info; |
31154f95 | 3256 | uint32_t link_speed; |
8a9562d2 | 3257 | |
7cd1261d | 3258 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 3259 | return ENODEV; |
7cd1261d | 3260 | } |
8a9562d2 | 3261 | |
03f3f9c0 | 3262 | ovs_mutex_lock(&dpdk_mutex); |
8a9562d2 PS |
3263 | ovs_mutex_lock(&dev->mutex); |
3264 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
31154f95 | 3265 | link_speed = dev->link.link_speed; |
8a9562d2 | 3266 | ovs_mutex_unlock(&dev->mutex); |
03f3f9c0 OM |
3267 | const struct rte_bus *bus; |
3268 | const struct rte_pci_device *pci_dev; | |
3269 | uint16_t vendor_id = PCI_ANY_ID; | |
3270 | uint16_t device_id = PCI_ANY_ID; | |
3271 | bus = rte_bus_find_by_device(dev_info.device); | |
3272 | if (bus && !strcmp(bus->name, "pci")) { | |
3273 | pci_dev = RTE_DEV_TO_PCI(dev_info.device); | |
3274 | if (pci_dev) { | |
3275 | vendor_id = pci_dev->id.vendor_id; | |
3276 | device_id = pci_dev->id.device_id; | |
3277 | } | |
3278 | } | |
3279 | ovs_mutex_unlock(&dpdk_mutex); | |
8a9562d2 | 3280 | |
fa9f4eeb | 3281 | smap_add_format(args, "port_no", DPDK_PORT_ID_FMT, dev->port_id); |
58be5c0e MK |
3282 | smap_add_format(args, "numa_id", "%d", |
3283 | rte_eth_dev_socket_id(dev->port_id)); | |
8a9562d2 PS |
3284 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); |
3285 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 3286 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
3287 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
3288 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
3289 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
58be5c0e MK |
3290 | smap_add_format(args, "max_hash_mac_addrs", "%u", |
3291 | dev_info.max_hash_mac_addrs); | |
8a9562d2 PS |
3292 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); |
3293 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
3294 | ||
3eb8d4fa MW |
3295 | /* Querying the DPDK library for iftype may be done in future, pending |
3296 | * support; cf. RFC 3635 Section 3.2.4. */ | |
3297 | enum { IF_TYPE_ETHERNETCSMACD = 6 }; | |
3298 | ||
3299 | smap_add_format(args, "if_type", "%"PRIu32, IF_TYPE_ETHERNETCSMACD); | |
3300 | smap_add_format(args, "if_descr", "%s %s", rte_version(), | |
3301 | dev_info.driver_name); | |
03f3f9c0 OM |
3302 | smap_add_format(args, "pci-vendor_id", "0x%x", vendor_id); |
3303 | smap_add_format(args, "pci-device_id", "0x%x", device_id); | |
8a9562d2 | 3304 | |
31154f95 IS |
3305 | /* Not all link speeds are defined in the OpenFlow specs e.g. 25 Gbps. |
3306 | * In that case the speed will not be reported as part of the usual | |
3307 | * call to get_features(). Get the link speed of the device and add it | |
3308 | * to the device status in an easy to read string format. | |
3309 | */ | |
3310 | smap_add(args, "link_speed", | |
3311 | netdev_dpdk_link_speed_to_str__(link_speed)); | |
3312 | ||
8a9562d2 PS |
3313 | return 0; |
3314 | } | |
3315 | ||
3316 | static void | |
3317 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
3318 | OVS_REQUIRES(dev->mutex) | |
3319 | { | |
3320 | enum netdev_flags old_flags; | |
3321 | ||
3322 | if (admin_state) { | |
3323 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
3324 | } else { | |
3325 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
3326 | } | |
3327 | } | |
3328 | ||
3329 | static void | |
3330 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
3331 | const char *argv[], void *aux OVS_UNUSED) | |
3332 | { | |
3333 | bool up; | |
3334 | ||
3335 | if (!strcasecmp(argv[argc - 1], "up")) { | |
3336 | up = true; | |
3337 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
3338 | up = false; | |
3339 | } else { | |
3340 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
3341 | return; | |
3342 | } | |
3343 | ||
3344 | if (argc > 2) { | |
3345 | struct netdev *netdev = netdev_from_name(argv[1]); | |
3d0d5ab1 | 3346 | |
8a9562d2 | 3347 | if (netdev && is_dpdk_class(netdev->netdev_class)) { |
3d0d5ab1 | 3348 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 3349 | |
3d0d5ab1 IM |
3350 | ovs_mutex_lock(&dev->mutex); |
3351 | netdev_dpdk_set_admin_state__(dev, up); | |
3352 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
3353 | |
3354 | netdev_close(netdev); | |
3355 | } else { | |
3356 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
3357 | netdev_close(netdev); | |
3358 | return; | |
3359 | } | |
3360 | } else { | |
3d0d5ab1 | 3361 | struct netdev_dpdk *dev; |
8a9562d2 PS |
3362 | |
3363 | ovs_mutex_lock(&dpdk_mutex); | |
3d0d5ab1 IM |
3364 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
3365 | ovs_mutex_lock(&dev->mutex); | |
3366 | netdev_dpdk_set_admin_state__(dev, up); | |
3367 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
3368 | } |
3369 | ovs_mutex_unlock(&dpdk_mutex); | |
3370 | } | |
3371 | unixctl_command_reply(conn, "OK"); | |
3372 | } | |
3373 | ||
0ee821c2 DB |
3374 | static void |
3375 | netdev_dpdk_detach(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
3376 | const char *argv[], void *aux OVS_UNUSED) | |
3377 | { | |
0ee821c2 | 3378 | char *response; |
7ee94cba | 3379 | dpdk_port_t port_id; |
0ee821c2 | 3380 | struct netdev_dpdk *dev; |
40e940e4 OM |
3381 | struct rte_device *rte_dev; |
3382 | struct ds used_interfaces = DS_EMPTY_INITIALIZER; | |
3383 | bool used = false; | |
0ee821c2 DB |
3384 | |
3385 | ovs_mutex_lock(&dpdk_mutex); | |
3386 | ||
40e940e4 OM |
3387 | port_id = netdev_dpdk_get_port_by_devargs(argv[1]); |
3388 | if (!rte_eth_dev_is_valid_port(port_id)) { | |
0ee821c2 DB |
3389 | response = xasprintf("Device '%s' not found in DPDK", argv[1]); |
3390 | goto error; | |
3391 | } | |
3392 | ||
40e940e4 OM |
3393 | rte_dev = rte_eth_devices[port_id].device; |
3394 | ds_put_format(&used_interfaces, | |
3395 | "Device '%s' is being used by the following interfaces:", | |
3396 | argv[1]); | |
3397 | ||
3398 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
3399 | /* FIXME: avoid direct access to DPDK array rte_eth_devices. */ | |
3400 | if (rte_eth_devices[dev->port_id].device == rte_dev | |
3401 | && rte_eth_devices[dev->port_id].state != RTE_ETH_DEV_UNUSED) { | |
3402 | used = true; | |
3403 | ds_put_format(&used_interfaces, " %s", | |
3404 | netdev_get_name(&dev->up)); | |
3405 | } | |
3406 | } | |
3407 | ||
3408 | if (used) { | |
3409 | ds_put_cstr(&used_interfaces, ". Remove them before detaching."); | |
3410 | response = ds_steal_cstr(&used_interfaces); | |
3411 | ds_destroy(&used_interfaces); | |
0ee821c2 DB |
3412 | goto error; |
3413 | } | |
40e940e4 | 3414 | ds_destroy(&used_interfaces); |
0ee821c2 DB |
3415 | |
3416 | rte_eth_dev_close(port_id); | |
40e940e4 | 3417 | if (rte_dev_remove(rte_dev) < 0) { |
0ee821c2 DB |
3418 | response = xasprintf("Device '%s' can not be detached", argv[1]); |
3419 | goto error; | |
3420 | } | |
3421 | ||
40e940e4 OM |
3422 | response = xasprintf("All devices shared with device '%s' " |
3423 | "have been detached", argv[1]); | |
0ee821c2 DB |
3424 | |
3425 | ovs_mutex_unlock(&dpdk_mutex); | |
3426 | unixctl_command_reply(conn, response); | |
3427 | free(response); | |
3428 | return; | |
3429 | ||
3430 | error: | |
3431 | ovs_mutex_unlock(&dpdk_mutex); | |
3432 | unixctl_command_reply_error(conn, response); | |
3433 | free(response); | |
3434 | } | |
3435 | ||
be481733 IM |
3436 | static void |
3437 | netdev_dpdk_get_mempool_info(struct unixctl_conn *conn, | |
3438 | int argc, const char *argv[], | |
3439 | void *aux OVS_UNUSED) | |
3440 | { | |
3441 | size_t size; | |
3442 | FILE *stream; | |
3443 | char *response = NULL; | |
3444 | struct netdev *netdev = NULL; | |
3445 | ||
3446 | if (argc == 2) { | |
3447 | netdev = netdev_from_name(argv[1]); | |
3448 | if (!netdev || !is_dpdk_class(netdev->netdev_class)) { | |
3449 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
3450 | goto out; | |
3451 | } | |
3452 | } | |
3453 | ||
3454 | stream = open_memstream(&response, &size); | |
3455 | if (!stream) { | |
3456 | response = xasprintf("Unable to open memstream: %s.", | |
3457 | ovs_strerror(errno)); | |
3458 | unixctl_command_reply_error(conn, response); | |
3459 | goto out; | |
3460 | } | |
3461 | ||
3462 | if (netdev) { | |
3463 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3464 | ||
3465 | ovs_mutex_lock(&dev->mutex); | |
3466 | ovs_mutex_lock(&dpdk_mp_mutex); | |
3467 | ||
43307ad0 | 3468 | rte_mempool_dump(stream, dev->dpdk_mp->mp); |
be481733 IM |
3469 | |
3470 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
3471 | ovs_mutex_unlock(&dev->mutex); | |
3472 | } else { | |
3473 | ovs_mutex_lock(&dpdk_mp_mutex); | |
3474 | rte_mempool_list_dump(stream); | |
3475 | ovs_mutex_unlock(&dpdk_mp_mutex); | |
3476 | } | |
3477 | ||
3478 | fclose(stream); | |
3479 | ||
3480 | unixctl_command_reply(conn, response); | |
3481 | out: | |
3482 | free(response); | |
3483 | netdev_close(netdev); | |
3484 | } | |
3485 | ||
58397e6c KT |
3486 | /* |
3487 | * Set virtqueue flags so that we do not receive interrupts. | |
3488 | */ | |
3489 | static void | |
0a0f39df | 3490 | set_irq_status(int vid) |
58397e6c | 3491 | { |
4573fbd3 | 3492 | uint32_t i; |
4573fbd3 | 3493 | |
f3e7ec25 MW |
3494 | for (i = 0; i < rte_vhost_get_vring_num(vid); i++) { |
3495 | rte_vhost_enable_guest_notification(vid, i, 0); | |
4573fbd3 FL |
3496 | } |
3497 | } | |
3498 | ||
585a5bea IM |
3499 | /* |
3500 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 3501 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
3502 | */ |
3503 | static void | |
d46285a2 DDP |
3504 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
3505 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
3506 | { |
3507 | int *enabled_queues, n_enabled = 0; | |
81acebda | 3508 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea | 3509 | |
eff23640 | 3510 | enabled_queues = xcalloc(total_txqs, sizeof *enabled_queues); |
585a5bea IM |
3511 | |
3512 | for (i = 0; i < total_txqs; i++) { | |
3513 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 3514 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
3515 | enabled_queues[n_enabled++] = i; |
3516 | } | |
3517 | } | |
3518 | ||
3519 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 3520 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
3521 | n_enabled = 1; |
3522 | } | |
3523 | ||
3524 | k = 0; | |
3525 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
3526 | if (dev->tx_q[i].map != i) { |
3527 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
3528 | k = (k + 1) % n_enabled; |
3529 | } | |
3530 | } | |
3531 | ||
170ef726 IM |
3532 | if (VLOG_IS_DBG_ENABLED()) { |
3533 | struct ds mapping = DS_EMPTY_INITIALIZER; | |
3534 | ||
3535 | ds_put_format(&mapping, "TX queue mapping for port '%s':\n", | |
3536 | netdev_get_name(&dev->up)); | |
3537 | for (i = 0; i < total_txqs; i++) { | |
3538 | ds_put_format(&mapping, "%2d --> %2d\n", i, dev->tx_q[i].map); | |
3539 | } | |
3540 | ||
3541 | VLOG_DBG("%s", ds_cstr(&mapping)); | |
3542 | ds_destroy(&mapping); | |
585a5bea IM |
3543 | } |
3544 | ||
eff23640 | 3545 | free(enabled_queues); |
585a5bea | 3546 | } |
4573fbd3 | 3547 | |
58397e6c KT |
3548 | /* |
3549 | * A new virtio-net device is added to a vhost port. | |
3550 | */ | |
3551 | static int | |
0a0f39df | 3552 | new_device(int vid) |
58397e6c | 3553 | { |
d46285a2 | 3554 | struct netdev_dpdk *dev; |
58397e6c | 3555 | bool exists = false; |
db8f13b0 | 3556 | int newnode = 0; |
0a0f39df CL |
3557 | char ifname[IF_NAME_SZ]; |
3558 | ||
58be5c0e | 3559 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
3560 | |
3561 | ovs_mutex_lock(&dpdk_mutex); | |
3562 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 3563 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 3564 | ovs_mutex_lock(&dev->mutex); |
bb9d2623 IM |
3565 | if (nullable_string_is_equal(ifname, dev->vhost_id)) { |
3566 | uint32_t qp_num = rte_vhost_get_vring_num(vid) / VIRTIO_QNUM; | |
db8f13b0 CL |
3567 | |
3568 | /* Get NUMA information */ | |
0a0f39df CL |
3569 | newnode = rte_vhost_get_numa_node(vid); |
3570 | if (newnode == -1) { | |
5b9bf9e0 | 3571 | #ifdef VHOST_NUMA |
db8f13b0 | 3572 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 3573 | ifname); |
5b9bf9e0 | 3574 | #endif |
db8f13b0 | 3575 | newnode = dev->socket_id; |
db8f13b0 CL |
3576 | } |
3577 | ||
7235cd20 DM |
3578 | if (dev->requested_n_txq < qp_num |
3579 | || dev->requested_n_rxq < qp_num | |
7f5f2bd0 IM |
3580 | || dev->requested_socket_id != newnode) { |
3581 | dev->requested_socket_id = newnode; | |
3582 | dev->requested_n_rxq = qp_num; | |
3583 | dev->requested_n_txq = qp_num; | |
3584 | netdev_request_reconfigure(&dev->up); | |
3585 | } else { | |
3586 | /* Reconfiguration not required. */ | |
3587 | dev->vhost_reconfigured = true; | |
3588 | } | |
81acebda | 3589 | |
0a0f39df | 3590 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
3591 | exists = true; |
3592 | ||
58397e6c | 3593 | /* Disable notifications. */ |
0a0f39df | 3594 | set_irq_status(vid); |
e543851d | 3595 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 3596 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
3597 | break; |
3598 | } | |
c1ff66ac | 3599 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
3600 | } |
3601 | ovs_mutex_unlock(&dpdk_mutex); | |
3602 | ||
3603 | if (!exists) { | |
0a0f39df | 3604 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
3605 | |
3606 | return -1; | |
3607 | } | |
3608 | ||
0a0f39df CL |
3609 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
3610 | ifname, newnode); | |
3611 | ||
58397e6c KT |
3612 | return 0; |
3613 | } | |
3614 | ||
f3ea2ad2 IM |
3615 | /* Clears mapping for all available queues of vhost interface. */ |
3616 | static void | |
3617 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
3618 | OVS_REQUIRES(dev->mutex) | |
3619 | { | |
3620 | int i; | |
3621 | ||
81acebda | 3622 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
3623 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
3624 | } | |
3625 | } | |
3626 | ||
58397e6c KT |
3627 | /* |
3628 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
3629 | * flag to stop any more packets from being sent or received to/from a VM and | |
3630 | * ensure all currently queued packets have been sent/received before removing | |
3631 | * the device. | |
3632 | */ | |
3633 | static void | |
0a0f39df | 3634 | destroy_device(int vid) |
58397e6c | 3635 | { |
d46285a2 | 3636 | struct netdev_dpdk *dev; |
afee281f | 3637 | bool exists = false; |
0a0f39df CL |
3638 | char ifname[IF_NAME_SZ]; |
3639 | ||
58be5c0e | 3640 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
58397e6c KT |
3641 | |
3642 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 3643 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 3644 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 3645 | |
d46285a2 | 3646 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
3647 | dev->vhost_reconfigured = false; |
3648 | ovsrcu_index_set(&dev->vid, -1); | |
35c91567 DM |
3649 | memset(dev->vhost_rxq_enabled, 0, |
3650 | dev->up.n_rxq * sizeof *dev->vhost_rxq_enabled); | |
d46285a2 | 3651 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 3652 | |
e543851d | 3653 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 3654 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 3655 | exists = true; |
afee281f | 3656 | break; |
58397e6c KT |
3657 | } |
3658 | } | |
afee281f | 3659 | |
58397e6c KT |
3660 | ovs_mutex_unlock(&dpdk_mutex); |
3661 | ||
0a0f39df | 3662 | if (exists) { |
afee281f KT |
3663 | /* |
3664 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
3665 | * to NULL, before returning. | |
3666 | */ | |
3667 | ovsrcu_synchronize(); | |
3668 | /* | |
3669 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
3670 | * put thread back into quiescent state before returning. | |
3671 | */ | |
3672 | ovsrcu_quiesce_start(); | |
0a0f39df | 3673 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 3674 | } else { |
0a0f39df | 3675 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 3676 | } |
58397e6c KT |
3677 | } |
3678 | ||
585a5bea | 3679 | static int |
0a0f39df | 3680 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 3681 | { |
d46285a2 | 3682 | struct netdev_dpdk *dev; |
585a5bea IM |
3683 | bool exists = false; |
3684 | int qid = queue_id / VIRTIO_QNUM; | |
35c91567 | 3685 | bool is_rx = (queue_id % VIRTIO_QNUM) == VIRTIO_TXQ; |
0a0f39df CL |
3686 | char ifname[IF_NAME_SZ]; |
3687 | ||
58be5c0e | 3688 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); |
585a5bea | 3689 | |
585a5bea | 3690 | ovs_mutex_lock(&dpdk_mutex); |
d46285a2 | 3691 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 3692 | ovs_mutex_lock(&dev->mutex); |
bb9d2623 | 3693 | if (nullable_string_is_equal(ifname, dev->vhost_id)) { |
35c91567 DM |
3694 | if (is_rx) { |
3695 | bool old_state = dev->vhost_rxq_enabled[qid]; | |
3696 | ||
3697 | dev->vhost_rxq_enabled[qid] = enable != 0; | |
3698 | if (old_state != dev->vhost_rxq_enabled[qid]) { | |
3699 | netdev_change_seq_changed(&dev->up); | |
3700 | } | |
585a5bea | 3701 | } else { |
35c91567 DM |
3702 | if (enable) { |
3703 | dev->tx_q[qid].map = qid; | |
3704 | } else { | |
3705 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; | |
3706 | } | |
3707 | netdev_dpdk_remap_txqs(dev); | |
585a5bea | 3708 | } |
585a5bea | 3709 | exists = true; |
d46285a2 | 3710 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3711 | break; |
3712 | } | |
c1ff66ac | 3713 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
3714 | } |
3715 | ovs_mutex_unlock(&dpdk_mutex); | |
3716 | ||
3717 | if (exists) { | |
35c91567 DM |
3718 | VLOG_INFO("State of queue %d ( %s_qid %d ) of vhost device '%s' " |
3719 | "changed to \'%s\'", queue_id, is_rx == true ? "rx" : "tx", | |
3720 | qid, ifname, (enable == 1) ? "enabled" : "disabled"); | |
585a5bea | 3721 | } else { |
0a0f39df | 3722 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
3723 | return -1; |
3724 | } | |
3725 | ||
3726 | return 0; | |
3727 | } | |
3728 | ||
61473a0e DM |
3729 | static void |
3730 | destroy_connection(int vid) | |
3731 | { | |
3732 | struct netdev_dpdk *dev; | |
3733 | char ifname[IF_NAME_SZ]; | |
3734 | bool exists = false; | |
3735 | ||
3736 | rte_vhost_get_ifname(vid, ifname, sizeof ifname); | |
3737 | ||
3738 | ovs_mutex_lock(&dpdk_mutex); | |
3739 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
3740 | ovs_mutex_lock(&dev->mutex); | |
3741 | if (nullable_string_is_equal(ifname, dev->vhost_id)) { | |
3742 | uint32_t qp_num = NR_QUEUE; | |
3743 | ||
3744 | if (netdev_dpdk_get_vid(dev) >= 0) { | |
3745 | VLOG_ERR("Connection on socket '%s' destroyed while vhost " | |
3746 | "device still attached.", dev->vhost_id); | |
3747 | } | |
3748 | ||
3749 | /* Restore the number of queue pairs to default. */ | |
3750 | if (dev->requested_n_txq != qp_num | |
3751 | || dev->requested_n_rxq != qp_num) { | |
3752 | dev->requested_n_rxq = qp_num; | |
3753 | dev->requested_n_txq = qp_num; | |
3754 | netdev_request_reconfigure(&dev->up); | |
3755 | } | |
3756 | ovs_mutex_unlock(&dev->mutex); | |
3757 | exists = true; | |
3758 | break; | |
3759 | } | |
3760 | ovs_mutex_unlock(&dev->mutex); | |
3761 | } | |
3762 | ovs_mutex_unlock(&dpdk_mutex); | |
3763 | ||
3764 | if (exists) { | |
3765 | VLOG_INFO("vHost Device '%s' connection has been destroyed", ifname); | |
3766 | } else { | |
3767 | VLOG_INFO("vHost Device '%s' not found", ifname); | |
3768 | } | |
3769 | } | |
3770 | ||
8492adc2 JS |
3771 | /* |
3772 | * Retrieve the DPDK virtio device ID (vid) associated with a vhostuser | |
3773 | * or vhostuserclient netdev. | |
3774 | * | |
3775 | * Returns a value greater or equal to zero for a valid vid or '-1' if | |
3776 | * there is no valid vid associated. A vid of '-1' must not be used in | |
3777 | * rte_vhost_ APi calls. | |
3778 | * | |
3779 | * Once obtained and validated, a vid can be used by a PMD for multiple | |
3780 | * subsequent rte_vhost API calls until the PMD quiesces. A PMD should | |
3781 | * not fetch the vid again for each of a series of API calls. | |
3782 | */ | |
3783 | ||
0a0f39df CL |
3784 | int |
3785 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 3786 | { |
0a0f39df | 3787 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
3788 | } |
3789 | ||
9509913a IS |
3790 | struct ingress_policer * |
3791 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
3792 | { | |
3793 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
3794 | } | |
3795 | ||
58397e6c | 3796 | static int |
ecc1a34e | 3797 | netdev_dpdk_class_init(void) |
7d1ced01 | 3798 | { |
ecc1a34e DDP |
3799 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; |
3800 | ||
3801 | /* This function can be called for different classes. The initialization | |
3802 | * needs to be done only once */ | |
3803 | if (ovsthread_once_start(&once)) { | |
3804 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); | |
3805 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
3806 | "[netdev] up|down", 1, 2, | |
3807 | netdev_dpdk_set_admin_state, NULL); | |
3808 | ||
0ee821c2 DB |
3809 | unixctl_command_register("netdev-dpdk/detach", |
3810 | "pci address of device", 1, 1, | |
3811 | netdev_dpdk_detach, NULL); | |
3812 | ||
be481733 IM |
3813 | unixctl_command_register("netdev-dpdk/get-mempool-info", |
3814 | "[netdev]", 0, 1, | |
3815 | netdev_dpdk_get_mempool_info, NULL); | |
3816 | ||
ecc1a34e DDP |
3817 | ovsthread_once_done(&once); |
3818 | } | |
362ca396 | 3819 | |
7d1ced01 CL |
3820 | return 0; |
3821 | } | |
3822 | ||
95fb793a | 3823 | /* Client Rings */ |
3824 | ||
95fb793a | 3825 | static int |
3826 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
bb37956a | 3827 | dpdk_port_t *eth_port_id) |
95fb793a | 3828 | { |
48fffdee | 3829 | struct dpdk_ring *ring_pair; |
0c6f39e5 | 3830 | char *ring_name; |
b8374d0d | 3831 | int port_id; |
95fb793a | 3832 | |
48fffdee KT |
3833 | ring_pair = dpdk_rte_mzalloc(sizeof *ring_pair); |
3834 | if (!ring_pair) { | |
95fb793a | 3835 | return ENOMEM; |
3836 | } | |
3837 | ||
7251515e | 3838 | /* XXX: Add support for multiquque ring. */ |
0c6f39e5 | 3839 | ring_name = xasprintf("%s_tx", dev_name); |
95fb793a | 3840 | |
8f0a76c9 | 3841 | /* Create single producer tx ring, netdev does explicit locking. */ |
48fffdee | 3842 | ring_pair->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3843 | RING_F_SP_ENQ); |
0c6f39e5 | 3844 | free(ring_name); |
48fffdee KT |
3845 | if (ring_pair->cring_tx == NULL) { |
3846 | rte_free(ring_pair); | |
95fb793a | 3847 | return ENOMEM; |
3848 | } | |
3849 | ||
0c6f39e5 | 3850 | ring_name = xasprintf("%s_rx", dev_name); |
95fb793a | 3851 | |
8f0a76c9 | 3852 | /* Create single consumer rx ring, netdev does explicit locking. */ |
48fffdee | 3853 | ring_pair->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 3854 | RING_F_SC_DEQ); |
0c6f39e5 | 3855 | free(ring_name); |
48fffdee KT |
3856 | if (ring_pair->cring_rx == NULL) { |
3857 | rte_free(ring_pair); | |
95fb793a | 3858 | return ENOMEM; |
3859 | } | |
3860 | ||
b8374d0d MV |
3861 | port_id = rte_eth_from_rings(dev_name, &ring_pair->cring_rx, 1, |
3862 | &ring_pair->cring_tx, 1, SOCKET0); | |
d7310583 | 3863 | |
b8374d0d | 3864 | if (port_id < 0) { |
48fffdee | 3865 | rte_free(ring_pair); |
95fb793a | 3866 | return ENODEV; |
3867 | } | |
3868 | ||
48fffdee | 3869 | ring_pair->user_port_id = port_no; |
b8374d0d MV |
3870 | ring_pair->eth_port_id = port_id; |
3871 | *eth_port_id = port_id; | |
3872 | ||
48fffdee | 3873 | ovs_list_push_back(&dpdk_ring_list, &ring_pair->list_node); |
95fb793a | 3874 | |
95fb793a | 3875 | return 0; |
3876 | } | |
3877 | ||
3878 | static int | |
bb37956a | 3879 | dpdk_ring_open(const char dev_name[], dpdk_port_t *eth_port_id) |
64839cf4 | 3880 | OVS_REQUIRES(dpdk_mutex) |
95fb793a | 3881 | { |
48fffdee | 3882 | struct dpdk_ring *ring_pair; |
95fb793a | 3883 | unsigned int port_no; |
3884 | int err = 0; | |
3885 | ||
3886 | /* Names always start with "dpdkr" */ | |
3887 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
3888 | if (err) { | |
3889 | return err; | |
3890 | } | |
3891 | ||
58be5c0e | 3892 | /* Look through our list to find the device */ |
48fffdee KT |
3893 | LIST_FOR_EACH (ring_pair, list_node, &dpdk_ring_list) { |
3894 | if (ring_pair->user_port_id == port_no) { | |
58397e6c | 3895 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
58be5c0e | 3896 | /* Really all that is needed */ |
48fffdee | 3897 | *eth_port_id = ring_pair->eth_port_id; |
95fb793a | 3898 | return 0; |
3899 | } | |
3900 | } | |
3901 | /* Need to create the device rings */ | |
3902 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
3903 | } | |
3904 | ||
7251515e | 3905 | static int |
d46285a2 | 3906 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
b30896c9 | 3907 | struct dp_packet_batch *batch, bool concurrent_txq) |
7251515e | 3908 | { |
d46285a2 | 3909 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a543eb0 | 3910 | struct dp_packet *packet; |
1b99bb05 | 3911 | |
58be5c0e | 3912 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that |
a47e2db2 | 3913 | * the offload fields are clear. This is because the same mbuf may be |
58be5c0e | 3914 | * modified by the consumer of the ring and return into the datapath |
a47e2db2 | 3915 | * without recalculating the RSS hash or revalidating the checksums. */ |
e883448e | 3916 | DP_PACKET_BATCH_FOR_EACH (i, packet, batch) { |
a47e2db2 | 3917 | dp_packet_reset_offload(packet); |
1b99bb05 | 3918 | } |
7251515e | 3919 | |
b30896c9 | 3920 | netdev_dpdk_send__(dev, qid, batch, concurrent_txq); |
7251515e DV |
3921 | return 0; |
3922 | } | |
3923 | ||
95fb793a | 3924 | static int |
3925 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
3926 | { | |
bb37956a | 3927 | dpdk_port_t port_no = 0; |
95fb793a | 3928 | int err = 0; |
3929 | ||
95fb793a | 3930 | ovs_mutex_lock(&dpdk_mutex); |
3931 | ||
3932 | err = dpdk_ring_open(netdev->name, &port_no); | |
3933 | if (err) { | |
3934 | goto unlock_dpdk; | |
3935 | } | |
3936 | ||
1ce30dfd DDP |
3937 | err = common_construct(netdev, port_no, DPDK_DEV_ETH, |
3938 | rte_eth_dev_socket_id(port_no)); | |
95fb793a | 3939 | unlock_dpdk: |
3940 | ovs_mutex_unlock(&dpdk_mutex); | |
3941 | return err; | |
3942 | } | |
3943 | ||
0bf765f7 IS |
3944 | /* QoS Functions */ |
3945 | ||
3946 | /* | |
3947 | * Initialize QoS configuration operations. | |
3948 | */ | |
3949 | static void | |
3950 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
3951 | { | |
3952 | conf->ops = ops; | |
78bd47cf | 3953 | rte_spinlock_init(&conf->lock); |
0bf765f7 IS |
3954 | } |
3955 | ||
3956 | /* | |
3957 | * Search existing QoS operations in qos_ops and compare each set of | |
3958 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
3959 | * else return NULL | |
3960 | */ | |
3961 | static const struct dpdk_qos_ops * | |
3962 | qos_lookup_name(const char *name) | |
3963 | { | |
3964 | const struct dpdk_qos_ops *const *opsp; | |
3965 | ||
3966 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3967 | const struct dpdk_qos_ops *ops = *opsp; | |
3968 | if (!strcmp(name, ops->qos_name)) { | |
3969 | return ops; | |
3970 | } | |
3971 | } | |
3972 | return NULL; | |
3973 | } | |
3974 | ||
0bf765f7 IS |
3975 | static int |
3976 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
3977 | struct sset *types) | |
3978 | { | |
3979 | const struct dpdk_qos_ops *const *opsp; | |
3980 | ||
3981 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
3982 | const struct dpdk_qos_ops *ops = *opsp; | |
3983 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
3984 | sset_add(types, ops->qos_name); | |
3985 | } | |
3986 | } | |
3987 | return 0; | |
3988 | } | |
3989 | ||
3990 | static int | |
d46285a2 | 3991 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
3992 | const char **typep, struct smap *details) |
3993 | { | |
d46285a2 | 3994 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
78bd47cf | 3995 | struct qos_conf *qos_conf; |
0bf765f7 IS |
3996 | int error = 0; |
3997 | ||
d46285a2 | 3998 | ovs_mutex_lock(&dev->mutex); |
78bd47cf DDP |
3999 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
4000 | if (qos_conf) { | |
4001 | *typep = qos_conf->ops->qos_name; | |
4002 | error = (qos_conf->ops->qos_get | |
4003 | ? qos_conf->ops->qos_get(qos_conf, details): 0); | |
d03603c4 MC |
4004 | } else { |
4005 | /* No QoS configuration set, return an empty string */ | |
4006 | *typep = ""; | |
0bf765f7 | 4007 | } |
d46285a2 | 4008 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
4009 | |
4010 | return error; | |
4011 | } | |
4012 | ||
4013 | static int | |
78bd47cf DDP |
4014 | netdev_dpdk_set_qos(struct netdev *netdev, const char *type, |
4015 | const struct smap *details) | |
0bf765f7 | 4016 | { |
d46285a2 | 4017 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 4018 | const struct dpdk_qos_ops *new_ops = NULL; |
78bd47cf | 4019 | struct qos_conf *qos_conf, *new_qos_conf = NULL; |
0bf765f7 IS |
4020 | int error = 0; |
4021 | ||
d46285a2 | 4022 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 4023 | |
78bd47cf | 4024 | qos_conf = ovsrcu_get_protected(struct qos_conf *, &dev->qos_conf); |
0bf765f7 | 4025 | |
78bd47cf DDP |
4026 | new_ops = qos_lookup_name(type); |
4027 | ||
4028 | if (!new_ops || !new_ops->qos_construct) { | |
4029 | new_qos_conf = NULL; | |
4030 | if (type && type[0]) { | |
4031 | error = EOPNOTSUPP; | |
0bf765f7 | 4032 | } |
44975bb0 | 4033 | } else if (qos_conf && qos_conf->ops == new_ops |
78bd47cf DDP |
4034 | && qos_conf->ops->qos_is_equal(qos_conf, details)) { |
4035 | new_qos_conf = qos_conf; | |
0bf765f7 | 4036 | } else { |
78bd47cf | 4037 | error = new_ops->qos_construct(details, &new_qos_conf); |
7ea266e9 IS |
4038 | } |
4039 | ||
7ea266e9 | 4040 | if (error) { |
78bd47cf DDP |
4041 | VLOG_ERR("Failed to set QoS type %s on port %s: %s", |
4042 | type, netdev->name, rte_strerror(error)); | |
4043 | } | |
4044 | ||
4045 | if (new_qos_conf != qos_conf) { | |
4046 | ovsrcu_set(&dev->qos_conf, new_qos_conf); | |
4047 | if (qos_conf) { | |
4048 | ovsrcu_postpone(qos_conf->ops->qos_destruct, qos_conf); | |
4049 | } | |
0bf765f7 IS |
4050 | } |
4051 | ||
d46285a2 | 4052 | ovs_mutex_unlock(&dev->mutex); |
78bd47cf | 4053 | |
0bf765f7 IS |
4054 | return error; |
4055 | } | |
4056 | ||
4057 | /* egress-policer details */ | |
4058 | ||
4059 | struct egress_policer { | |
4060 | struct qos_conf qos_conf; | |
4061 | struct rte_meter_srtcm_params app_srtcm_params; | |
4062 | struct rte_meter_srtcm egress_meter; | |
03f3f9c0 | 4063 | struct rte_meter_srtcm_profile egress_prof; |
0bf765f7 IS |
4064 | }; |
4065 | ||
78bd47cf DDP |
4066 | static void |
4067 | egress_policer_details_to_param(const struct smap *details, | |
4068 | struct rte_meter_srtcm_params *params) | |
0bf765f7 | 4069 | { |
78bd47cf DDP |
4070 | memset(params, 0, sizeof *params); |
4071 | params->cir = smap_get_ullong(details, "cir", 0); | |
4072 | params->cbs = smap_get_ullong(details, "cbs", 0); | |
4073 | params->ebs = 0; | |
0bf765f7 IS |
4074 | } |
4075 | ||
4076 | static int | |
78bd47cf DDP |
4077 | egress_policer_qos_construct(const struct smap *details, |
4078 | struct qos_conf **conf) | |
0bf765f7 | 4079 | { |
0bf765f7 | 4080 | struct egress_policer *policer; |
0bf765f7 IS |
4081 | int err = 0; |
4082 | ||
0bf765f7 IS |
4083 | policer = xmalloc(sizeof *policer); |
4084 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
78bd47cf | 4085 | egress_policer_details_to_param(details, &policer->app_srtcm_params); |
03f3f9c0 OM |
4086 | err = rte_meter_srtcm_profile_config(&policer->egress_prof, |
4087 | &policer->app_srtcm_params); | |
4088 | if (!err) { | |
4089 | err = rte_meter_srtcm_config(&policer->egress_meter, | |
4090 | &policer->egress_prof); | |
4091 | } | |
4092 | ||
78bd47cf DDP |
4093 | if (!err) { |
4094 | *conf = &policer->qos_conf; | |
4095 | } else { | |
03f3f9c0 | 4096 | VLOG_ERR("Could not create rte meter for egress policer"); |
7ea266e9 | 4097 | free(policer); |
78bd47cf | 4098 | *conf = NULL; |
7ea266e9 IS |
4099 | err = -err; |
4100 | } | |
0bf765f7 IS |
4101 | |
4102 | return err; | |
4103 | } | |
4104 | ||
4105 | static void | |
78bd47cf | 4106 | egress_policer_qos_destruct(struct qos_conf *conf) |
0bf765f7 IS |
4107 | { |
4108 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
78bd47cf | 4109 | qos_conf); |
0bf765f7 IS |
4110 | free(policer); |
4111 | } | |
4112 | ||
4113 | static int | |
78bd47cf | 4114 | egress_policer_qos_get(const struct qos_conf *conf, struct smap *details) |
0bf765f7 | 4115 | { |
78bd47cf DDP |
4116 | struct egress_policer *policer = |
4117 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
4118 | ||
4119 | smap_add_format(details, "cir", "%"PRIu64, policer->app_srtcm_params.cir); | |
4120 | smap_add_format(details, "cbs", "%"PRIu64, policer->app_srtcm_params.cbs); | |
050c60bf | 4121 | |
0bf765f7 IS |
4122 | return 0; |
4123 | } | |
4124 | ||
78bd47cf | 4125 | static bool |
47a45d86 KT |
4126 | egress_policer_qos_is_equal(const struct qos_conf *conf, |
4127 | const struct smap *details) | |
0bf765f7 | 4128 | { |
78bd47cf DDP |
4129 | struct egress_policer *policer = |
4130 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
4131 | struct rte_meter_srtcm_params params; | |
0bf765f7 | 4132 | |
78bd47cf | 4133 | egress_policer_details_to_param(details, ¶ms); |
7ea266e9 | 4134 | |
78bd47cf | 4135 | return !memcmp(¶ms, &policer->app_srtcm_params, sizeof params); |
0bf765f7 IS |
4136 | } |
4137 | ||
0bf765f7 | 4138 | static int |
3e90f7d7 | 4139 | egress_policer_run(struct qos_conf *conf, struct rte_mbuf **pkts, int pkt_cnt, |
7d7ded7a | 4140 | bool should_steal) |
0bf765f7 | 4141 | { |
0bf765f7 | 4142 | int cnt = 0; |
78bd47cf DDP |
4143 | struct egress_policer *policer = |
4144 | CONTAINER_OF(conf, struct egress_policer, qos_conf); | |
0bf765f7 | 4145 | |
03f3f9c0 OM |
4146 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, |
4147 | &policer->egress_prof, pkts, | |
7d7ded7a | 4148 | pkt_cnt, should_steal); |
0bf765f7 IS |
4149 | |
4150 | return cnt; | |
4151 | } | |
4152 | ||
4153 | static const struct dpdk_qos_ops egress_policer_ops = { | |
4154 | "egress-policer", /* qos_name */ | |
4155 | egress_policer_qos_construct, | |
4156 | egress_policer_qos_destruct, | |
4157 | egress_policer_qos_get, | |
78bd47cf | 4158 | egress_policer_qos_is_equal, |
0bf765f7 IS |
4159 | egress_policer_run |
4160 | }; | |
4161 | ||
050c60bf DDP |
4162 | static int |
4163 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
4164 | { | |
4165 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
4166 | int err = 0; | |
4167 | ||
050c60bf DDP |
4168 | ovs_mutex_lock(&dev->mutex); |
4169 | ||
4170 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 | 4171 | && netdev->n_rxq == dev->requested_n_rxq |
b685696b | 4172 | && dev->mtu == dev->requested_mtu |
f8b64a61 | 4173 | && dev->lsc_interrupt_mode == dev->requested_lsc_interrupt_mode |
b685696b | 4174 | && dev->rxq_size == dev->requested_rxq_size |
bd4e172b | 4175 | && dev->txq_size == dev->requested_txq_size |
606f6650 EC |
4176 | && dev->socket_id == dev->requested_socket_id |
4177 | && dev->started) { | |
050c60bf DDP |
4178 | /* Reconfiguration is unnecessary */ |
4179 | ||
4180 | goto out; | |
4181 | } | |
4182 | ||
4183 | rte_eth_dev_stop(dev->port_id); | |
606f6650 | 4184 | dev->started = false; |
050c60bf | 4185 | |
d555d9bd | 4186 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 | 4187 | if (err && err != EEXIST) { |
d555d9bd | 4188 | goto out; |
0072e931 MK |
4189 | } |
4190 | ||
f8b64a61 RM |
4191 | dev->lsc_interrupt_mode = dev->requested_lsc_interrupt_mode; |
4192 | ||
050c60bf DDP |
4193 | netdev->n_txq = dev->requested_n_txq; |
4194 | netdev->n_rxq = dev->requested_n_rxq; | |
4195 | ||
b685696b CL |
4196 | dev->rxq_size = dev->requested_rxq_size; |
4197 | dev->txq_size = dev->requested_txq_size; | |
4198 | ||
050c60bf DDP |
4199 | rte_free(dev->tx_q); |
4200 | err = dpdk_eth_dev_init(dev); | |
eff23640 DDP |
4201 | dev->tx_q = netdev_dpdk_alloc_txq(netdev->n_txq); |
4202 | if (!dev->tx_q) { | |
4203 | err = ENOMEM; | |
4204 | } | |
050c60bf | 4205 | |
0072e931 MK |
4206 | netdev_change_seq_changed(netdev); |
4207 | ||
050c60bf | 4208 | out: |
050c60bf | 4209 | ovs_mutex_unlock(&dev->mutex); |
050c60bf DDP |
4210 | return err; |
4211 | } | |
4212 | ||
7f381c2e | 4213 | static int |
2d24d165 | 4214 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) |
2d24d165 | 4215 | OVS_REQUIRES(dev->mutex) |
050c60bf | 4216 | { |
2d24d165 CL |
4217 | dev->up.n_txq = dev->requested_n_txq; |
4218 | dev->up.n_rxq = dev->requested_n_rxq; | |
96e9b168 | 4219 | int err; |
050c60bf | 4220 | |
35c91567 DM |
4221 | /* Always keep RX queue 0 enabled for implementations that won't |
4222 | * report vring states. */ | |
4223 | dev->vhost_rxq_enabled[0] = true; | |
4224 | ||
81acebda IM |
4225 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
4226 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
4227 | dev->tx_q[0].map = 0; | |
4228 | } | |
4229 | ||
4230 | netdev_dpdk_remap_txqs(dev); | |
4231 | ||
d555d9bd | 4232 | err = netdev_dpdk_mempool_configure(dev); |
b6b26021 | 4233 | if (!err) { |
43307ad0 | 4234 | /* A new mempool was created or re-used. */ |
d555d9bd | 4235 | netdev_change_seq_changed(&dev->up); |
03f3f9c0 | 4236 | } else if (err != EEXIST) { |
b6b26021 | 4237 | return err; |
db8f13b0 | 4238 | } |
0a0f39df | 4239 | if (netdev_dpdk_get_vid(dev) >= 0) { |
894af647 | 4240 | if (dev->vhost_reconfigured == false) { |
4241 | dev->vhost_reconfigured = true; | |
4242 | /* Carrier status may need updating. */ | |
4243 | netdev_change_seq_changed(&dev->up); | |
4244 | } | |
81acebda | 4245 | } |
7f381c2e DDP |
4246 | |
4247 | return 0; | |
2d24d165 CL |
4248 | } |
4249 | ||
4250 | static int | |
4251 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
4252 | { | |
4253 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 4254 | int err; |
2d24d165 | 4255 | |
2d24d165 | 4256 | ovs_mutex_lock(&dev->mutex); |
7f381c2e | 4257 | err = dpdk_vhost_reconfigure_helper(dev); |
2d24d165 | 4258 | ovs_mutex_unlock(&dev->mutex); |
7f381c2e DDP |
4259 | |
4260 | return err; | |
2d24d165 CL |
4261 | } |
4262 | ||
4263 | static int | |
4264 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
4265 | { | |
4266 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
7f381c2e | 4267 | int err; |
a14d1cc8 | 4268 | uint64_t vhost_flags = 0; |
10087cba | 4269 | bool zc_enabled; |
2d24d165 | 4270 | |
2d24d165 CL |
4271 | ovs_mutex_lock(&dev->mutex); |
4272 | ||
c1ff66ac CL |
4273 | /* Configure vHost client mode if requested and if the following criteria |
4274 | * are met: | |
2d24d165 CL |
4275 | * 1. Device hasn't been registered yet. |
4276 | * 2. A path has been specified. | |
c1ff66ac | 4277 | */ |
bb9d2623 | 4278 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) && dev->vhost_id) { |
a14d1cc8 MK |
4279 | /* Register client-mode device. */ |
4280 | vhost_flags |= RTE_VHOST_USER_CLIENT; | |
4281 | ||
4282 | /* Enable IOMMU support, if explicitly requested. */ | |
4283 | if (dpdk_vhost_iommu_enabled()) { | |
4284 | vhost_flags |= RTE_VHOST_USER_IOMMU_SUPPORT; | |
4285 | } | |
10087cba | 4286 | |
30e834dc LB |
4287 | /* Enable POSTCOPY support, if explicitly requested. */ |
4288 | if (dpdk_vhost_postcopy_enabled()) { | |
4289 | vhost_flags |= RTE_VHOST_USER_POSTCOPY_SUPPORT; | |
4290 | } | |
4291 | ||
10087cba CL |
4292 | zc_enabled = dev->vhost_driver_flags |
4293 | & RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
4294 | /* Enable zero copy flag, if requested */ | |
4295 | if (zc_enabled) { | |
4296 | vhost_flags |= RTE_VHOST_USER_DEQUEUE_ZERO_COPY; | |
4297 | } | |
4298 | ||
a14d1cc8 | 4299 | err = rte_vhost_driver_register(dev->vhost_id, vhost_flags); |
c1ff66ac | 4300 | if (err) { |
2d24d165 CL |
4301 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
4302 | dev->vhost_id); | |
7f381c2e | 4303 | goto unlock; |
c1ff66ac | 4304 | } else { |
2d24d165 | 4305 | /* Configuration successful */ |
a14d1cc8 | 4306 | dev->vhost_driver_flags |= vhost_flags; |
2d24d165 CL |
4307 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " |
4308 | "using client socket '%s'", | |
4309 | dev->up.name, dev->vhost_id); | |
10087cba CL |
4310 | if (zc_enabled) { |
4311 | VLOG_INFO("Zero copy enabled for vHost port %s", dev->up.name); | |
4312 | } | |
c1ff66ac | 4313 | } |
f3e7ec25 MW |
4314 | |
4315 | err = rte_vhost_driver_callback_register(dev->vhost_id, | |
4316 | &virtio_net_device_ops); | |
4317 | if (err) { | |
4318 | VLOG_ERR("rte_vhost_driver_callback_register failed for " | |
4319 | "vhost user client port: %s\n", dev->up.name); | |
4320 | goto unlock; | |
4321 | } | |
4322 | ||
4323 | err = rte_vhost_driver_disable_features(dev->vhost_id, | |
4324 | 1ULL << VIRTIO_NET_F_HOST_TSO4 | |
4325 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
4326 | | 1ULL << VIRTIO_NET_F_CSUM); | |
4327 | if (err) { | |
4328 | VLOG_ERR("rte_vhost_driver_disable_features failed for vhost user " | |
4329 | "client port: %s\n", dev->up.name); | |
4330 | goto unlock; | |
4331 | } | |
4332 | ||
4333 | err = rte_vhost_driver_start(dev->vhost_id); | |
4334 | if (err) { | |
4335 | VLOG_ERR("rte_vhost_driver_start failed for vhost user " | |
4336 | "client port: %s\n", dev->up.name); | |
4337 | goto unlock; | |
4338 | } | |
c1ff66ac CL |
4339 | } |
4340 | ||
7f381c2e DDP |
4341 | err = dpdk_vhost_reconfigure_helper(dev); |
4342 | ||
4343 | unlock: | |
050c60bf | 4344 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 4345 | |
7f381c2e | 4346 | return err; |
050c60bf DDP |
4347 | } |
4348 | ||
5fc5c50f IM |
4349 | bool |
4350 | netdev_dpdk_flow_api_supported(struct netdev *netdev) | |
4351 | { | |
4352 | struct netdev_dpdk *dev; | |
4353 | bool ret = false; | |
4354 | ||
4355 | if (!is_dpdk_class(netdev->netdev_class)) { | |
4356 | goto out; | |
4357 | } | |
4358 | ||
4359 | dev = netdev_dpdk_cast(netdev); | |
4360 | ovs_mutex_lock(&dev->mutex); | |
4361 | if (dev->type == DPDK_DEV_ETH) { | |
4362 | /* TODO: Check if we able to offload some minimal flow. */ | |
4363 | ret = true; | |
4364 | } | |
4365 | ovs_mutex_unlock(&dev->mutex); | |
4366 | out: | |
4367 | return ret; | |
4368 | } | |
4369 | ||
6775bdfc RBY |
4370 | int |
4371 | netdev_dpdk_rte_flow_destroy(struct netdev *netdev, | |
4372 | struct rte_flow *rte_flow, | |
4373 | struct rte_flow_error *error) | |
4374 | { | |
4375 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
4376 | int ret; | |
4377 | ||
4378 | ovs_mutex_lock(&dev->mutex); | |
4379 | ret = rte_flow_destroy(dev->port_id, rte_flow, error); | |
4380 | ovs_mutex_unlock(&dev->mutex); | |
4381 | return ret; | |
4382 | } | |
4383 | ||
4384 | struct rte_flow * | |
4385 | netdev_dpdk_rte_flow_create(struct netdev *netdev, | |
4386 | const struct rte_flow_attr *attr, | |
4387 | const struct rte_flow_item *items, | |
4388 | const struct rte_flow_action *actions, | |
4389 | struct rte_flow_error *error) | |
4390 | { | |
4391 | struct rte_flow *flow; | |
4392 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
4393 | ||
4394 | ovs_mutex_lock(&dev->mutex); | |
4395 | flow = rte_flow_create(dev->port_id, attr, items, actions, error); | |
4396 | ovs_mutex_unlock(&dev->mutex); | |
4397 | return flow; | |
4398 | } | |
e8a2b5bf | 4399 | |
89c09c1c BP |
4400 | #define NETDEV_DPDK_CLASS_COMMON \ |
4401 | .is_pmd = true, \ | |
4402 | .alloc = netdev_dpdk_alloc, \ | |
4403 | .dealloc = netdev_dpdk_dealloc, \ | |
4404 | .get_config = netdev_dpdk_get_config, \ | |
4405 | .get_numa_id = netdev_dpdk_get_numa_id, \ | |
4406 | .set_etheraddr = netdev_dpdk_set_etheraddr, \ | |
4407 | .get_etheraddr = netdev_dpdk_get_etheraddr, \ | |
4408 | .get_mtu = netdev_dpdk_get_mtu, \ | |
4409 | .set_mtu = netdev_dpdk_set_mtu, \ | |
4410 | .get_ifindex = netdev_dpdk_get_ifindex, \ | |
4411 | .get_carrier_resets = netdev_dpdk_get_carrier_resets, \ | |
4412 | .set_miimon_interval = netdev_dpdk_set_miimon, \ | |
4413 | .set_policing = netdev_dpdk_set_policing, \ | |
4414 | .get_qos_types = netdev_dpdk_get_qos_types, \ | |
4415 | .get_qos = netdev_dpdk_get_qos, \ | |
4416 | .set_qos = netdev_dpdk_set_qos, \ | |
4417 | .update_flags = netdev_dpdk_update_flags, \ | |
4418 | .rxq_alloc = netdev_dpdk_rxq_alloc, \ | |
4419 | .rxq_construct = netdev_dpdk_rxq_construct, \ | |
4420 | .rxq_destruct = netdev_dpdk_rxq_destruct, \ | |
c0af6425 | 4421 | .rxq_dealloc = netdev_dpdk_rxq_dealloc |
89c09c1c BP |
4422 | |
4423 | #define NETDEV_DPDK_CLASS_BASE \ | |
4424 | NETDEV_DPDK_CLASS_COMMON, \ | |
4425 | .init = netdev_dpdk_class_init, \ | |
4426 | .destruct = netdev_dpdk_destruct, \ | |
4427 | .set_tx_multiq = netdev_dpdk_set_tx_multiq, \ | |
4428 | .get_carrier = netdev_dpdk_get_carrier, \ | |
4429 | .get_stats = netdev_dpdk_get_stats, \ | |
4430 | .get_custom_stats = netdev_dpdk_get_custom_stats, \ | |
4431 | .get_features = netdev_dpdk_get_features, \ | |
4432 | .get_status = netdev_dpdk_get_status, \ | |
4433 | .reconfigure = netdev_dpdk_reconfigure, \ | |
5fc5c50f | 4434 | .rxq_recv = netdev_dpdk_rxq_recv |
89c09c1c BP |
4435 | |
4436 | static const struct netdev_class dpdk_class = { | |
4437 | .type = "dpdk", | |
4438 | NETDEV_DPDK_CLASS_BASE, | |
4439 | .construct = netdev_dpdk_construct, | |
4440 | .set_config = netdev_dpdk_set_config, | |
4441 | .send = netdev_dpdk_eth_send, | |
4442 | }; | |
4443 | ||
4444 | static const struct netdev_class dpdk_ring_class = { | |
4445 | .type = "dpdkr", | |
4446 | NETDEV_DPDK_CLASS_BASE, | |
4447 | .construct = netdev_dpdk_ring_construct, | |
4448 | .set_config = netdev_dpdk_ring_set_config, | |
4449 | .send = netdev_dpdk_ring_send, | |
4450 | }; | |
4451 | ||
4452 | static const struct netdev_class dpdk_vhost_class = { | |
4453 | .type = "dpdkvhostuser", | |
4454 | NETDEV_DPDK_CLASS_COMMON, | |
4455 | .construct = netdev_dpdk_vhost_construct, | |
4456 | .destruct = netdev_dpdk_vhost_destruct, | |
4457 | .send = netdev_dpdk_vhost_send, | |
4458 | .get_carrier = netdev_dpdk_vhost_get_carrier, | |
4459 | .get_stats = netdev_dpdk_vhost_get_stats, | |
c161357d | 4460 | .get_custom_stats = netdev_dpdk_vhost_get_custom_stats, |
89c09c1c BP |
4461 | .get_status = netdev_dpdk_vhost_user_get_status, |
4462 | .reconfigure = netdev_dpdk_vhost_reconfigure, | |
35c91567 DM |
4463 | .rxq_recv = netdev_dpdk_vhost_rxq_recv, |
4464 | .rxq_enabled = netdev_dpdk_vhost_rxq_enabled, | |
89c09c1c BP |
4465 | }; |
4466 | ||
4467 | static const struct netdev_class dpdk_vhost_client_class = { | |
4468 | .type = "dpdkvhostuserclient", | |
4469 | NETDEV_DPDK_CLASS_COMMON, | |
4470 | .construct = netdev_dpdk_vhost_client_construct, | |
4471 | .destruct = netdev_dpdk_vhost_destruct, | |
4472 | .set_config = netdev_dpdk_vhost_client_set_config, | |
4473 | .send = netdev_dpdk_vhost_send, | |
4474 | .get_carrier = netdev_dpdk_vhost_get_carrier, | |
4475 | .get_stats = netdev_dpdk_vhost_get_stats, | |
c161357d | 4476 | .get_custom_stats = netdev_dpdk_vhost_get_custom_stats, |
89c09c1c BP |
4477 | .get_status = netdev_dpdk_vhost_user_get_status, |
4478 | .reconfigure = netdev_dpdk_vhost_client_reconfigure, | |
35c91567 DM |
4479 | .rxq_recv = netdev_dpdk_vhost_rxq_recv, |
4480 | .rxq_enabled = netdev_dpdk_vhost_rxq_enabled, | |
89c09c1c | 4481 | }; |
95fb793a | 4482 | |
8a9562d2 PS |
4483 | void |
4484 | netdev_dpdk_register(void) | |
4485 | { | |
bab69409 AC |
4486 | netdev_register_provider(&dpdk_class); |
4487 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 4488 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 4489 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 4490 | } |