]>
Commit | Line | Data |
---|---|---|
8a9562d2 | 1 | /* |
0d4af148 | 2 | * Copyright (c) 2014, 2015, 2016 Nicira, Inc. |
8a9562d2 PS |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
8a9562d2 PS |
19 | #include <string.h> |
20 | #include <signal.h> | |
21 | #include <stdlib.h> | |
22 | #include <pthread.h> | |
23 | #include <config.h> | |
24 | #include <errno.h> | |
25 | #include <sched.h> | |
26 | #include <stdlib.h> | |
27 | #include <unistd.h> | |
7d1ced01 | 28 | #include <sys/stat.h> |
8a9562d2 | 29 | #include <stdio.h> |
7d1ced01 CL |
30 | #include <sys/types.h> |
31 | #include <sys/stat.h> | |
bab69409 | 32 | #include <getopt.h> |
8a9562d2 | 33 | |
7d1ced01 | 34 | #include "dirs.h" |
e14deea0 | 35 | #include "dp-packet.h" |
8a9562d2 | 36 | #include "dpif-netdev.h" |
e5c0f5a4 | 37 | #include "fatal-signal.h" |
8a9562d2 PS |
38 | #include "netdev-dpdk.h" |
39 | #include "netdev-provider.h" | |
40 | #include "netdev-vport.h" | |
41 | #include "odp-util.h" | |
eac84432 | 42 | #include "openvswitch/dynamic-string.h" |
25d436fb BW |
43 | #include "openvswitch/list.h" |
44 | #include "openvswitch/ofp-print.h" | |
45 | #include "openvswitch/vlog.h" | |
94143fc4 | 46 | #include "ovs-numa.h" |
8a9562d2 PS |
47 | #include "ovs-thread.h" |
48 | #include "ovs-rcu.h" | |
49 | #include "packets.h" | |
ee89ea7b | 50 | #include "openvswitch/shash.h" |
0bf765f7 | 51 | #include "smap.h" |
8a9562d2 PS |
52 | #include "sset.h" |
53 | #include "unaligned.h" | |
54 | #include "timeval.h" | |
55 | #include "unixctl.h" | |
8a9562d2 | 56 | |
b8e57534 MK |
57 | #include "rte_config.h" |
58 | #include "rte_mbuf.h" | |
0bf765f7 | 59 | #include "rte_meter.h" |
58397e6c | 60 | #include "rte_virtio_net.h" |
b8e57534 | 61 | |
8a9562d2 PS |
62 | VLOG_DEFINE_THIS_MODULE(dpdk); |
63 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
64 | ||
65 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
66 | ||
67 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
68 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
69 | ||
70 | /* | |
71 | * need to reserve tons of extra space in the mbufs so we can align the | |
72 | * DMA addresses to 4KB. | |
18f777b2 TP |
73 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
74 | * performance for standard Ethernet MTU. | |
8a9562d2 | 75 | */ |
4be4d22c MK |
76 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN + (2 * VLAN_HEADER_LEN)) |
77 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) | |
78 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
79 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len)- ETHER_HDR_LEN - ETHER_CRC_LEN) | |
80 | #define MBUF_SIZE(mtu) ( MTU_TO_MAX_FRAME_LEN(mtu) \ | |
81 | + sizeof(struct dp_packet) \ | |
82 | + RTE_PKTMBUF_HEADROOM) | |
83 | #define NETDEV_DPDK_MBUF_ALIGN 1024 | |
8a9562d2 | 84 | |
da79ce2b DDP |
85 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
86 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
87 | * enough hugepages) we keep halving the number until the allocation succeeds | |
88 | * or we reach MIN_NB_MBUF */ | |
89 | ||
90 | #define MAX_NB_MBUF (4096 * 64) | |
91 | #define MIN_NB_MBUF (4096 * 4) | |
92 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
93 | ||
94 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
95 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
96 | ||
97 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
98 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
99 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
100 | % MP_CACHE_SZ == 0); | |
101 | ||
d6e3feb5 | 102 | /* |
103 | * DPDK XSTATS Counter names definition | |
104 | */ | |
105 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
106 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
107 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
108 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
109 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
110 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
111 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
112 | ||
113 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
114 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
115 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
116 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
117 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
118 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
119 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
120 | ||
121 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" | |
122 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
123 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
124 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
125 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
126 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
127 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
128 | ||
8a9562d2 PS |
129 | #define SOCKET0 0 |
130 | ||
79f5354c PM |
131 | #define NIC_PORT_RX_Q_SIZE 2048 /* Size of Physical NIC RX Queue, Max (n+32<=4096)*/ |
132 | #define NIC_PORT_TX_Q_SIZE 2048 /* Size of Physical NIC TX Queue, Max (n+32<=4096)*/ | |
133 | ||
585a5bea | 134 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
135 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
136 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
137 | * yet mapped to another queue. */ | |
585a5bea | 138 | |
bab69409 | 139 | #ifdef VHOST_CUSE |
bd8baf47 | 140 | static char *cuse_dev_name = NULL; /* Character device cuse_dev_name. */ |
bab69409 | 141 | #endif |
bd8baf47 | 142 | static char *vhost_sock_dir = NULL; /* Location of vhost-user sockets */ |
58397e6c | 143 | |
31871ee3 | 144 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 145 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 146 | |
8a9562d2 | 147 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
148 | .rxmode = { |
149 | .mq_mode = ETH_MQ_RX_RSS, | |
150 | .split_hdr_size = 0, | |
151 | .header_split = 0, /* Header Split disabled */ | |
152 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
153 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
154 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
155 | .hw_strip_crc = 0, | |
156 | }, | |
157 | .rx_adv_conf = { | |
158 | .rss_conf = { | |
159 | .rss_key = NULL, | |
543342a4 | 160 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 161 | }, |
a28ddd11 DDP |
162 | }, |
163 | .txmode = { | |
164 | .mq_mode = ETH_MQ_TX_NONE, | |
165 | }, | |
8a9562d2 PS |
166 | }; |
167 | ||
58f7c37b DDP |
168 | enum { DPDK_RING_SIZE = 256 }; |
169 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
170 | enum { DRAIN_TSC = 200000ULL }; |
171 | ||
58397e6c KT |
172 | enum dpdk_dev_type { |
173 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 174 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
175 | }; |
176 | ||
8a9562d2 PS |
177 | static int rte_eal_init_ret = ENODEV; |
178 | ||
179 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; | |
180 | ||
0bf765f7 IS |
181 | /* Quality of Service */ |
182 | ||
183 | /* An instance of a QoS configuration. Always associated with a particular | |
184 | * network device. | |
185 | * | |
186 | * Each QoS implementation subclasses this with whatever additional data it | |
187 | * needs. | |
188 | */ | |
189 | struct qos_conf { | |
190 | const struct dpdk_qos_ops *ops; | |
191 | }; | |
192 | ||
193 | /* A particular implementation of dpdk QoS operations. | |
194 | * | |
195 | * The functions below return 0 if successful or a positive errno value on | |
196 | * failure, except where otherwise noted. All of them must be provided, except | |
197 | * where otherwise noted. | |
198 | */ | |
199 | struct dpdk_qos_ops { | |
200 | ||
201 | /* Name of the QoS type */ | |
202 | const char *qos_name; | |
203 | ||
204 | /* Called to construct the QoS implementation on 'netdev'. The | |
205 | * implementation should make the appropriate calls to configure QoS | |
206 | * according to 'details'. The implementation may assume that any current | |
207 | * QoS configuration already installed should be destroyed before | |
208 | * constructing the new configuration. | |
209 | * | |
210 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
211 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
212 | * (which is built as ovs-vswitchd.conf.db(8)). | |
213 | * | |
214 | * This function must return 0 if and only if it sets 'netdev->qos_conf' | |
215 | * to an initialized 'struct qos_conf'. | |
216 | * | |
217 | * For all QoS implementations it should always be non-null. | |
218 | */ | |
219 | int (*qos_construct)(struct netdev *netdev, const struct smap *details); | |
220 | ||
221 | /* Destroys the data structures allocated by the implementation as part of | |
222 | * 'qos_conf. | |
223 | * | |
224 | * For all QoS implementations it should always be non-null. | |
225 | */ | |
226 | void (*qos_destruct)(struct netdev *netdev, struct qos_conf *conf); | |
227 | ||
228 | /* Retrieves details of 'netdev->qos_conf' configuration into 'details'. | |
229 | * | |
230 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
231 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
232 | * (which is built as ovs-vswitchd.conf.db(8)). | |
233 | */ | |
234 | int (*qos_get)(const struct netdev *netdev, struct smap *details); | |
235 | ||
236 | /* Reconfigures 'netdev->qos_conf' according to 'details', performing any | |
237 | * required calls to complete the reconfiguration. | |
238 | * | |
239 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
240 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
241 | * (which is built as ovs-vswitchd.conf.db(8)). | |
242 | * | |
243 | * This function may be null if 'qos_conf' is not configurable. | |
244 | */ | |
245 | int (*qos_set)(struct netdev *netdev, const struct smap *details); | |
246 | ||
247 | /* Modify an array of rte_mbufs. The modification is specific to | |
248 | * each qos implementation. | |
249 | * | |
250 | * The function should take and array of mbufs and an int representing | |
251 | * the current number of mbufs present in the array. | |
252 | * | |
253 | * After the function has performed a qos modification to the array of | |
254 | * mbufs it returns an int representing the number of mbufs now present in | |
255 | * the array. This value is can then be passed to the port send function | |
256 | * along with the modified array for transmission. | |
257 | * | |
258 | * For all QoS implementations it should always be non-null. | |
259 | */ | |
260 | int (*qos_run)(struct netdev *netdev, struct rte_mbuf **pkts, | |
261 | int pkt_cnt); | |
262 | }; | |
263 | ||
264 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
265 | static const struct dpdk_qos_ops egress_policer_ops; | |
266 | ||
267 | /* | |
268 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
269 | * operations. | |
270 | */ | |
271 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
272 | &egress_policer_ops, | |
273 | NULL | |
274 | }; | |
275 | ||
8a9562d2 | 276 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 277 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 278 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 279 | |
ca6ba700 | 280 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 281 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 282 | |
db73f716 | 283 | /* This mutex must be used by non pmd threads when allocating or freeing |
b59cc14e | 284 | * mbufs through mempools. */ |
bce01e3a | 285 | static struct ovs_mutex nonpmd_mempool_mutex = OVS_MUTEX_INITIALIZER; |
db73f716 | 286 | |
8a9562d2 PS |
287 | struct dpdk_mp { |
288 | struct rte_mempool *mp; | |
289 | int mtu; | |
290 | int socket_id; | |
291 | int refcount; | |
ca6ba700 | 292 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
293 | }; |
294 | ||
5a034064 AW |
295 | /* There should be one 'struct dpdk_tx_queue' created for |
296 | * each cpu core. */ | |
8a9562d2 | 297 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
298 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
299 | * from concurrent access. It is used only | |
300 | * if the queue is shared among different | |
324c8374 | 301 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
302 | int map; /* Mapping of configured vhost-user queues |
303 | * to enabled by guest. */ | |
8a9562d2 PS |
304 | }; |
305 | ||
95fb793a | 306 | /* dpdk has no way to remove dpdk ring ethernet devices |
307 | so we have to keep them around once they've been created | |
308 | */ | |
309 | ||
ca6ba700 | 310 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 311 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 312 | |
313 | struct dpdk_ring { | |
314 | /* For the client rings */ | |
315 | struct rte_ring *cring_tx; | |
316 | struct rte_ring *cring_rx; | |
b83a2df1 | 317 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
95fb793a | 318 | int eth_port_id; /* ethernet device port id */ |
ca6ba700 | 319 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 320 | }; |
321 | ||
9509913a IS |
322 | struct ingress_policer { |
323 | struct rte_meter_srtcm_params app_srtcm_params; | |
324 | struct rte_meter_srtcm in_policer; | |
325 | rte_spinlock_t policer_lock; | |
326 | }; | |
327 | ||
8a9562d2 PS |
328 | struct netdev_dpdk { |
329 | struct netdev up; | |
330 | int port_id; | |
331 | int max_packet_len; | |
58397e6c | 332 | enum dpdk_dev_type type; |
8a9562d2 | 333 | |
5a034064 | 334 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
335 | |
336 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
337 | ||
338 | struct dpdk_mp *dpdk_mp; | |
339 | int mtu; | |
340 | int socket_id; | |
341 | int buf_size; | |
8a9562d2 | 342 | struct netdev_stats stats; |
45d947c4 DDP |
343 | /* Protects stats */ |
344 | rte_spinlock_t stats_lock; | |
8a9562d2 | 345 | |
74ff3298 | 346 | struct eth_addr hwaddr; |
8a9562d2 PS |
347 | enum netdev_flags flags; |
348 | ||
349 | struct rte_eth_link link; | |
350 | int link_reset_cnt; | |
351 | ||
0a0f39df CL |
352 | /* virtio identifier for vhost devices */ |
353 | ovsrcu_index vid; | |
354 | ||
355 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
356 | bool vhost_reconfigured; | |
58397e6c | 357 | |
3f891bbe DDP |
358 | /* Identifier used to distinguish vhost devices from each other. It does |
359 | * not change during the lifetime of a struct netdev_dpdk. It can be read | |
360 | * without holding any mutex. */ | |
361 | const char vhost_id[PATH_MAX]; | |
7d1ced01 | 362 | |
8a9562d2 | 363 | /* In dpdk_list. */ |
ca6ba700 | 364 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
0bf765f7 IS |
365 | |
366 | /* QoS configuration and lock for the device */ | |
367 | struct qos_conf *qos_conf; | |
368 | rte_spinlock_t qos_lock; | |
369 | ||
050c60bf DDP |
370 | /* The following properties cannot be changed when a device is running, |
371 | * so we remember the request and update them next time | |
372 | * netdev_dpdk*_reconfigure() is called */ | |
373 | int requested_n_txq; | |
374 | int requested_n_rxq; | |
9509913a | 375 | |
db8f13b0 CL |
376 | /* Socket ID detected when vHost device is brought up */ |
377 | int requested_socket_id; | |
378 | ||
9509913a IS |
379 | /* Ingress Policer */ |
380 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
381 | uint32_t policer_rate; | |
382 | uint32_t policer_burst; | |
9fd39370 SC |
383 | |
384 | /* DPDK-ETH Flow control */ | |
385 | struct rte_eth_fc_conf fc_conf; | |
8a9562d2 PS |
386 | }; |
387 | ||
388 | struct netdev_rxq_dpdk { | |
389 | struct netdev_rxq up; | |
390 | int port_id; | |
391 | }; | |
392 | ||
5f17de68 | 393 | static bool dpdk_thread_is_pmd(void); |
db73f716 | 394 | |
8a9562d2 PS |
395 | static int netdev_dpdk_construct(struct netdev *); |
396 | ||
0a0f39df | 397 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 398 | |
9509913a IS |
399 | struct ingress_policer * |
400 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
401 | ||
8a9562d2 PS |
402 | static bool |
403 | is_dpdk_class(const struct netdev_class *class) | |
404 | { | |
405 | return class->construct == netdev_dpdk_construct; | |
406 | } | |
407 | ||
4be4d22c MK |
408 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
409 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
410 | * value, insufficient buffers are allocated to accomodate the packet in its | |
411 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
412 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
413 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
414 | * behaviour, which reduces performance. To prevent this, use a buffer size that | |
415 | * is closest to 'mtu', but which satisfies the aforementioned criteria. | |
416 | */ | |
417 | static uint32_t | |
418 | dpdk_buf_size(int mtu) | |
419 | { | |
420 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
421 | NETDEV_DPDK_MBUF_ALIGN); | |
422 | } | |
423 | ||
58397e6c KT |
424 | /* XXX: use dpdk malloc for entire OVS. in fact huge page should be used |
425 | * for all other segments data, bss and text. */ | |
8a9562d2 PS |
426 | |
427 | static void * | |
428 | dpdk_rte_mzalloc(size_t sz) | |
429 | { | |
430 | void *ptr; | |
431 | ||
432 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
433 | if (ptr == NULL) { | |
434 | out_of_memory(); | |
435 | } | |
436 | return ptr; | |
437 | } | |
438 | ||
db73f716 DDP |
439 | /* XXX this function should be called only by pmd threads (or by non pmd |
440 | * threads holding the nonpmd_mempool_mutex) */ | |
8a9562d2 | 441 | void |
e14deea0 | 442 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 443 | { |
db73f716 | 444 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 445 | |
b00b4a81 | 446 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
447 | } |
448 | ||
b3cd9f9d PS |
449 | static void |
450 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
451 | void *opaque_arg OVS_UNUSED, | |
452 | void *_m, | |
453 | unsigned i OVS_UNUSED) | |
454 | { | |
455 | struct rte_mbuf *m = _m; | |
456 | ||
4be4d22c | 457 | rte_pktmbuf_init(mp, opaque_arg, _m, i); |
b3cd9f9d | 458 | |
cf62fa4c | 459 | dp_packet_init_dpdk((struct dp_packet *) m, m->buf_len); |
b3cd9f9d PS |
460 | } |
461 | ||
8a9562d2 PS |
462 | static struct dpdk_mp * |
463 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
464 | { | |
465 | struct dpdk_mp *dmp = NULL; | |
466 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
da79ce2b | 467 | unsigned mp_size; |
4be4d22c | 468 | struct rte_pktmbuf_pool_private mbp_priv; |
8a9562d2 PS |
469 | |
470 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
471 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
472 | dmp->refcount++; | |
473 | return dmp; | |
474 | } | |
475 | } | |
476 | ||
477 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
478 | dmp->socket_id = socket_id; | |
479 | dmp->mtu = mtu; | |
480 | dmp->refcount = 1; | |
4be4d22c | 481 | mbp_priv.mbuf_data_room_size = MBUF_SIZE(mtu) - sizeof(struct dp_packet); |
64839cf4 WT |
482 | mbp_priv.mbuf_priv_size = sizeof (struct dp_packet) - |
483 | sizeof (struct rte_mbuf); | |
8a9562d2 | 484 | |
da79ce2b DDP |
485 | mp_size = MAX_NB_MBUF; |
486 | do { | |
487 | if (snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d_%d_%u", | |
488 | dmp->mtu, dmp->socket_id, mp_size) < 0) { | |
489 | return NULL; | |
490 | } | |
95fb793a | 491 | |
da79ce2b DDP |
492 | dmp->mp = rte_mempool_create(mp_name, mp_size, MBUF_SIZE(mtu), |
493 | MP_CACHE_SZ, | |
494 | sizeof(struct rte_pktmbuf_pool_private), | |
4be4d22c | 495 | rte_pktmbuf_pool_init, &mbp_priv, |
da79ce2b DDP |
496 | ovs_rte_pktmbuf_init, NULL, |
497 | socket_id, 0); | |
498 | } while (!dmp->mp && rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
8a9562d2 PS |
499 | |
500 | if (dmp->mp == NULL) { | |
501 | return NULL; | |
da79ce2b DDP |
502 | } else { |
503 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", mp_name, mp_size ); | |
8a9562d2 PS |
504 | } |
505 | ||
417e7e66 | 506 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); |
8a9562d2 PS |
507 | return dmp; |
508 | } | |
509 | ||
510 | static void | |
8d38823b | 511 | dpdk_mp_put(struct dpdk_mp *dmp) OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 512 | { |
8a9562d2 PS |
513 | if (!dmp) { |
514 | return; | |
515 | } | |
516 | ||
8d38823b | 517 | ovs_assert(dmp->refcount); |
8a9562d2 | 518 | |
8d38823b MK |
519 | if (!--dmp->refcount) { |
520 | ovs_list_remove(&dmp->list_node); | |
521 | rte_mempool_free(dmp->mp); | |
8a9562d2 | 522 | } |
8a9562d2 PS |
523 | } |
524 | ||
525 | static void | |
526 | check_link_status(struct netdev_dpdk *dev) | |
527 | { | |
528 | struct rte_eth_link link; | |
529 | ||
530 | rte_eth_link_get_nowait(dev->port_id, &link); | |
531 | ||
532 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 533 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
534 | |
535 | dev->link_reset_cnt++; | |
536 | dev->link = link; | |
537 | if (dev->link.link_status) { | |
538 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
539 | dev->port_id, (unsigned)dev->link.link_speed, | |
540 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
541 | ("full-duplex") : ("half-duplex")); | |
542 | } else { | |
543 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
544 | } | |
545 | } | |
546 | } | |
547 | ||
548 | static void * | |
549 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
550 | { | |
551 | struct netdev_dpdk *dev; | |
552 | ||
553 | pthread_detach(pthread_self()); | |
554 | ||
555 | for (;;) { | |
556 | ovs_mutex_lock(&dpdk_mutex); | |
557 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
558 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
559 | if (dev->type == DPDK_DEV_ETH) { |
560 | check_link_status(dev); | |
561 | } | |
8a9562d2 PS |
562 | ovs_mutex_unlock(&dev->mutex); |
563 | } | |
564 | ovs_mutex_unlock(&dpdk_mutex); | |
565 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
566 | } | |
567 | ||
568 | return NULL; | |
569 | } | |
570 | ||
b98d7669 DDP |
571 | static int |
572 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
573 | { | |
574 | int diag = 0; | |
575 | int i; | |
576 | ||
577 | /* A device may report more queues than it makes available (this has | |
578 | * been observed for Intel xl710, which reserves some of them for | |
579 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
580 | * available. When this happens we can retry the configuration | |
581 | * and request less queues */ | |
582 | while (n_rxq && n_txq) { | |
583 | if (diag) { | |
584 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
585 | } | |
586 | ||
587 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &port_conf); | |
588 | if (diag) { | |
589 | break; | |
590 | } | |
591 | ||
592 | for (i = 0; i < n_txq; i++) { | |
593 | diag = rte_eth_tx_queue_setup(dev->port_id, i, NIC_PORT_TX_Q_SIZE, | |
594 | dev->socket_id, NULL); | |
595 | if (diag) { | |
596 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
597 | dev->up.name, i, rte_strerror(-diag)); | |
598 | break; | |
599 | } | |
600 | } | |
601 | ||
602 | if (i != n_txq) { | |
603 | /* Retry with less tx queues */ | |
604 | n_txq = i; | |
605 | continue; | |
606 | } | |
607 | ||
608 | for (i = 0; i < n_rxq; i++) { | |
609 | diag = rte_eth_rx_queue_setup(dev->port_id, i, NIC_PORT_RX_Q_SIZE, | |
610 | dev->socket_id, NULL, | |
611 | dev->dpdk_mp->mp); | |
612 | if (diag) { | |
613 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
614 | dev->up.name, i, rte_strerror(-diag)); | |
615 | break; | |
616 | } | |
617 | } | |
618 | ||
619 | if (i != n_rxq) { | |
620 | /* Retry with less rx queues */ | |
621 | n_rxq = i; | |
622 | continue; | |
623 | } | |
624 | ||
625 | dev->up.n_rxq = n_rxq; | |
81acebda | 626 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
627 | |
628 | return 0; | |
629 | } | |
630 | ||
631 | return diag; | |
632 | } | |
633 | ||
9fd39370 SC |
634 | static void |
635 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
636 | { | |
637 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
638 | VLOG_WARN("Failed to enable flow control on device %d", dev->port_id); | |
639 | } | |
640 | } | |
b98d7669 | 641 | |
8a9562d2 PS |
642 | static int |
643 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
644 | { | |
645 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 646 | struct rte_eth_dev_info info; |
8a9562d2 PS |
647 | struct ether_addr eth_addr; |
648 | int diag; | |
b98d7669 | 649 | int n_rxq, n_txq; |
8a9562d2 PS |
650 | |
651 | if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) { | |
95fb793a | 652 | return ENODEV; |
8a9562d2 PS |
653 | } |
654 | ||
a0cb2d66 | 655 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 656 | |
b98d7669 DDP |
657 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
658 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
659 | ||
660 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 661 | if (diag) { |
b98d7669 DDP |
662 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
663 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 664 | return -diag; |
8a9562d2 PS |
665 | } |
666 | ||
8a9562d2 PS |
667 | diag = rte_eth_dev_start(dev->port_id); |
668 | if (diag) { | |
b98d7669 DDP |
669 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
670 | rte_strerror(-diag)); | |
95fb793a | 671 | return -diag; |
8a9562d2 PS |
672 | } |
673 | ||
674 | rte_eth_promiscuous_enable(dev->port_id); | |
675 | rte_eth_allmulticast_enable(dev->port_id); | |
676 | ||
677 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
678 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
679 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
ca92d173 | 680 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 681 | |
ca92d173 | 682 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
683 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
684 | ||
685 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
686 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
687 | ||
688 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
9fd39370 SC |
689 | |
690 | /* Get the Flow control configuration for DPDK-ETH */ | |
691 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
692 | if (diag) { | |
693 | VLOG_DBG("cannot get flow control parameters on port=%d, err=%d", | |
694 | dev->port_id, diag); | |
695 | } | |
696 | ||
8a9562d2 PS |
697 | return 0; |
698 | } | |
699 | ||
700 | static struct netdev_dpdk * | |
701 | netdev_dpdk_cast(const struct netdev *netdev) | |
702 | { | |
703 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
704 | } | |
705 | ||
706 | static struct netdev * | |
707 | netdev_dpdk_alloc(void) | |
708 | { | |
bab69409 AC |
709 | struct netdev_dpdk *dev; |
710 | ||
711 | if (!rte_eal_init_ret) { /* Only after successful initialization */ | |
712 | dev = dpdk_rte_mzalloc(sizeof *dev); | |
713 | if (dev) { | |
714 | return &dev->up; | |
715 | } | |
716 | } | |
717 | return NULL; | |
8a9562d2 PS |
718 | } |
719 | ||
5a034064 | 720 | static void |
d46285a2 | 721 | netdev_dpdk_alloc_txq(struct netdev_dpdk *dev, unsigned int n_txqs) |
5a034064 | 722 | { |
bd5131ba | 723 | unsigned i; |
5a034064 | 724 | |
d46285a2 | 725 | dev->tx_q = dpdk_rte_mzalloc(n_txqs * sizeof *dev->tx_q); |
5a034064 | 726 | for (i = 0; i < n_txqs; i++) { |
585a5bea | 727 | /* Initialize map for vhost devices. */ |
d46285a2 DDP |
728 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
729 | rte_spinlock_init(&dev->tx_q[i].tx_lock); | |
5a034064 AW |
730 | } |
731 | } | |
732 | ||
8a9562d2 | 733 | static int |
d46285a2 | 734 | netdev_dpdk_init(struct netdev *netdev, unsigned int port_no, |
58397e6c | 735 | enum dpdk_dev_type type) |
5a034064 | 736 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 737 | { |
d46285a2 | 738 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b7a04e0 | 739 | int sid; |
95fb793a | 740 | int err = 0; |
4be4d22c | 741 | uint32_t buf_size; |
8a9562d2 | 742 | |
d46285a2 DDP |
743 | ovs_mutex_init(&dev->mutex); |
744 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 745 | |
d46285a2 | 746 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 747 | |
1b7a04e0 AW |
748 | /* If the 'sid' is negative, it means that the kernel fails |
749 | * to obtain the pci numa info. In that situation, always | |
750 | * use 'SOCKET0'. */ | |
58397e6c KT |
751 | if (type == DPDK_DEV_ETH) { |
752 | sid = rte_eth_dev_socket_id(port_no); | |
753 | } else { | |
754 | sid = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
755 | } | |
756 | ||
d46285a2 | 757 | dev->socket_id = sid < 0 ? SOCKET0 : sid; |
db8f13b0 | 758 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
759 | dev->port_id = port_no; |
760 | dev->type = type; | |
761 | dev->flags = 0; | |
762 | dev->mtu = ETHER_MTU; | |
763 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
0a0f39df CL |
764 | ovsrcu_index_init(&dev->vid, -1); |
765 | dev->vhost_reconfigured = false; | |
8a9562d2 | 766 | |
d46285a2 DDP |
767 | buf_size = dpdk_buf_size(dev->mtu); |
768 | dev->dpdk_mp = dpdk_mp_get(dev->socket_id, FRAME_LEN_TO_MTU(buf_size)); | |
769 | if (!dev->dpdk_mp) { | |
8a9562d2 | 770 | err = ENOMEM; |
95fb793a | 771 | goto unlock; |
8a9562d2 PS |
772 | } |
773 | ||
0bf765f7 | 774 | /* Initialise QoS configuration to NULL and qos lock to unlocked */ |
d46285a2 DDP |
775 | dev->qos_conf = NULL; |
776 | rte_spinlock_init(&dev->qos_lock); | |
0bf765f7 | 777 | |
9509913a IS |
778 | /* Initialise rcu pointer for ingress policer to NULL */ |
779 | ovsrcu_init(&dev->ingress_policer, NULL); | |
780 | dev->policer_rate = 0; | |
781 | dev->policer_burst = 0; | |
782 | ||
d46285a2 | 783 | netdev->n_rxq = NR_QUEUE; |
81acebda IM |
784 | netdev->n_txq = NR_QUEUE; |
785 | dev->requested_n_rxq = netdev->n_rxq; | |
786 | dev->requested_n_txq = netdev->n_txq; | |
58397e6c | 787 | |
9fd39370 SC |
788 | /* Initialize the flow control to NULL */ |
789 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
58397e6c | 790 | if (type == DPDK_DEV_ETH) { |
d46285a2 | 791 | err = dpdk_eth_dev_init(dev); |
1b99bb05 MG |
792 | if (err) { |
793 | goto unlock; | |
794 | } | |
81acebda | 795 | netdev_dpdk_alloc_txq(dev, netdev->n_txq); |
585a5bea | 796 | } else { |
d46285a2 | 797 | netdev_dpdk_alloc_txq(dev, OVS_VHOST_MAX_QUEUE_NUM); |
e543851d ZB |
798 | /* Enable DPDK_DEV_VHOST device and set promiscuous mode flag. */ |
799 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
8a9562d2 | 800 | } |
8a9562d2 | 801 | |
d46285a2 | 802 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 803 | |
95fb793a | 804 | unlock: |
d46285a2 | 805 | ovs_mutex_unlock(&dev->mutex); |
95fb793a | 806 | return err; |
807 | } | |
808 | ||
b83a2df1 MV |
809 | /* dev_name must be the prefix followed by a positive decimal number. |
810 | * (no leading + or - signs are allowed) */ | |
95fb793a | 811 | static int |
812 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
813 | unsigned int *port_no) | |
814 | { | |
815 | const char *cport; | |
816 | ||
817 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
818 | return ENODEV; | |
819 | } | |
820 | ||
821 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
822 | |
823 | if (str_to_uint(cport, 10, port_no)) { | |
824 | return 0; | |
825 | } else { | |
826 | return ENODEV; | |
827 | } | |
95fb793a | 828 | } |
829 | ||
58397e6c | 830 | static int |
d46285a2 | 831 | vhost_construct_helper(struct netdev *netdev) OVS_REQUIRES(dpdk_mutex) |
58397e6c | 832 | { |
58397e6c KT |
833 | if (rte_eal_init_ret) { |
834 | return rte_eal_init_ret; | |
835 | } | |
836 | ||
d46285a2 | 837 | return netdev_dpdk_init(netdev, -1, DPDK_DEV_VHOST); |
7d1ced01 CL |
838 | } |
839 | ||
840 | static int | |
d46285a2 | 841 | netdev_dpdk_vhost_cuse_construct(struct netdev *netdev) |
7d1ced01 | 842 | { |
d46285a2 | 843 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7d1ced01 CL |
844 | int err; |
845 | ||
bab69409 AC |
846 | if (rte_eal_init_ret) { |
847 | return rte_eal_init_ret; | |
848 | } | |
849 | ||
58397e6c | 850 | ovs_mutex_lock(&dpdk_mutex); |
3f891bbe DDP |
851 | strncpy(CONST_CAST(char *, dev->vhost_id), netdev->name, |
852 | sizeof dev->vhost_id); | |
d46285a2 | 853 | err = vhost_construct_helper(netdev); |
58397e6c | 854 | ovs_mutex_unlock(&dpdk_mutex); |
7d1ced01 CL |
855 | return err; |
856 | } | |
58397e6c | 857 | |
7d1ced01 | 858 | static int |
d46285a2 | 859 | netdev_dpdk_vhost_user_construct(struct netdev *netdev) |
7d1ced01 | 860 | { |
d46285a2 DDP |
861 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
862 | const char *name = netdev->name; | |
7d1ced01 | 863 | int err; |
0a0f39df | 864 | uint64_t flags = 0; |
a0cb2d66 | 865 | |
1af27e8a DDP |
866 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
867 | * the file system. '/' or '\' would traverse directories, so they're not | |
868 | * acceptable in 'name'. */ | |
869 | if (strchr(name, '/') || strchr(name, '\\')) { | |
870 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
871 | "A valid name must not include '/' or '\\'", | |
872 | name); | |
873 | return EINVAL; | |
874 | } | |
875 | ||
bab69409 AC |
876 | if (rte_eal_init_ret) { |
877 | return rte_eal_init_ret; | |
878 | } | |
879 | ||
7d1ced01 CL |
880 | ovs_mutex_lock(&dpdk_mutex); |
881 | /* Take the name of the vhost-user port and append it to the location where | |
882 | * the socket is to be created, then register the socket. | |
883 | */ | |
3f891bbe | 884 | snprintf(CONST_CAST(char *, dev->vhost_id), sizeof dev->vhost_id, "%s/%s", |
1af27e8a DDP |
885 | vhost_sock_dir, name); |
886 | ||
0a0f39df | 887 | err = rte_vhost_driver_register(dev->vhost_id, flags); |
7d1ced01 CL |
888 | if (err) { |
889 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
d46285a2 | 890 | dev->vhost_id); |
e5c0f5a4 | 891 | } else { |
d46285a2 | 892 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
1a64eb93 | 893 | VLOG_INFO("Socket %s created for vhost-user port %s\n", |
d46285a2 DDP |
894 | dev->vhost_id, name); |
895 | err = vhost_construct_helper(netdev); | |
7d1ced01 | 896 | } |
e5c0f5a4 | 897 | |
7d1ced01 | 898 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
899 | return err; |
900 | } | |
901 | ||
95fb793a | 902 | static int |
903 | netdev_dpdk_construct(struct netdev *netdev) | |
904 | { | |
905 | unsigned int port_no; | |
906 | int err; | |
907 | ||
908 | if (rte_eal_init_ret) { | |
909 | return rte_eal_init_ret; | |
910 | } | |
911 | ||
912 | /* Names always start with "dpdk" */ | |
913 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
914 | if (err) { | |
915 | return err; | |
916 | } | |
917 | ||
918 | ovs_mutex_lock(&dpdk_mutex); | |
58397e6c | 919 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
8a9562d2 PS |
920 | ovs_mutex_unlock(&dpdk_mutex); |
921 | return err; | |
922 | } | |
923 | ||
924 | static void | |
d46285a2 | 925 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 926 | { |
d46285a2 | 927 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 928 | |
8d38823b | 929 | ovs_mutex_lock(&dpdk_mutex); |
8a9562d2 | 930 | ovs_mutex_lock(&dev->mutex); |
8d38823b | 931 | |
8a9562d2 | 932 | rte_eth_dev_stop(dev->port_id); |
9509913a IS |
933 | free(ovsrcu_get_protected(struct ingress_policer *, |
934 | &dev->ingress_policer)); | |
8a9562d2 | 935 | |
5a034064 | 936 | rte_free(dev->tx_q); |
417e7e66 | 937 | ovs_list_remove(&dev->list_node); |
8a9562d2 | 938 | dpdk_mp_put(dev->dpdk_mp); |
8d38823b MK |
939 | |
940 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 | 941 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 942 | } |
8a9562d2 | 943 | |
3f891bbe DDP |
944 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
945 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
946 | * deadlock, none of the mutexes must be held while calling this function. */ | |
947 | static int | |
948 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev) | |
949 | OVS_EXCLUDED(dpdk_mutex) | |
950 | OVS_EXCLUDED(dev->mutex) | |
951 | { | |
952 | return rte_vhost_driver_unregister(dev->vhost_id); | |
953 | } | |
954 | ||
58397e6c | 955 | static void |
d46285a2 | 956 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 957 | { |
d46285a2 | 958 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c | 959 | |
8d38823b MK |
960 | ovs_mutex_lock(&dpdk_mutex); |
961 | ovs_mutex_lock(&dev->mutex); | |
962 | ||
c62da695 | 963 | /* Guest becomes an orphan if still attached. */ |
0a0f39df | 964 | if (netdev_dpdk_get_vid(dev) >= 0) { |
c62da695 | 965 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 966 | netdev->name); |
c62da695 IM |
967 | VLOG_ERR("To restore connectivity after re-adding of port, VM on socket" |
968 | " '%s' must be restarted.", | |
969 | dev->vhost_id); | |
58397e6c KT |
970 | } |
971 | ||
9509913a IS |
972 | free(ovsrcu_get_protected(struct ingress_policer *, |
973 | &dev->ingress_policer)); | |
9509913a | 974 | |
6019cb63 | 975 | rte_free(dev->tx_q); |
417e7e66 | 976 | ovs_list_remove(&dev->list_node); |
58397e6c | 977 | dpdk_mp_put(dev->dpdk_mp); |
8d38823b MK |
978 | |
979 | ovs_mutex_unlock(&dev->mutex); | |
58397e6c | 980 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe DDP |
981 | |
982 | if (dpdk_vhost_driver_unregister(dev)) { | |
983 | VLOG_ERR("Unable to remove vhost-user socket %s", dev->vhost_id); | |
984 | } else { | |
985 | fatal_signal_remove_file_to_unlink(dev->vhost_id); | |
986 | } | |
8a9562d2 PS |
987 | } |
988 | ||
989 | static void | |
d46285a2 | 990 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 991 | { |
d46285a2 | 992 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 993 | |
d46285a2 | 994 | rte_free(dev); |
8a9562d2 PS |
995 | } |
996 | ||
997 | static int | |
a14b8947 | 998 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 999 | { |
a14b8947 | 1000 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1001 | |
1002 | ovs_mutex_lock(&dev->mutex); | |
1003 | ||
050c60bf | 1004 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1005 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1006 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1007 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
8a9562d2 PS |
1008 | ovs_mutex_unlock(&dev->mutex); |
1009 | ||
1010 | return 0; | |
1011 | } | |
1012 | ||
a14b8947 IM |
1013 | static int |
1014 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args) | |
1015 | { | |
1016 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
050c60bf | 1017 | int new_n_rxq; |
a14b8947 IM |
1018 | |
1019 | ovs_mutex_lock(&dev->mutex); | |
050c60bf DDP |
1020 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", dev->requested_n_rxq), 1); |
1021 | if (new_n_rxq != dev->requested_n_rxq) { | |
1022 | dev->requested_n_rxq = new_n_rxq; | |
1023 | netdev_request_reconfigure(netdev); | |
1024 | } | |
9fd39370 SC |
1025 | |
1026 | /* Flow control configuration for DPDK Ethernet ports. */ | |
1027 | if (dev->type == DPDK_DEV_ETH) { | |
1028 | bool rx_fc_en = false; | |
1029 | bool tx_fc_en = false; | |
1030 | enum rte_eth_fc_mode fc_mode_set[2][2] = | |
1031 | {{RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1032 | {RTE_FC_RX_PAUSE, RTE_FC_FULL} | |
1033 | }; | |
1034 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); | |
1035 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
1036 | dev->fc_conf.autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); | |
1037 | dev->fc_conf.mode = fc_mode_set[tx_fc_en][rx_fc_en]; | |
1038 | ||
1039 | dpdk_eth_flow_ctrl_setup(dev); | |
1040 | } | |
a14b8947 IM |
1041 | ovs_mutex_unlock(&dev->mutex); |
1042 | ||
1043 | return 0; | |
1044 | } | |
1045 | ||
7dec44fe | 1046 | static int |
d46285a2 | 1047 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1048 | { |
d46285a2 | 1049 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1050 | |
d46285a2 | 1051 | return dev->socket_id; |
7dec44fe AW |
1052 | } |
1053 | ||
050c60bf | 1054 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1055 | static int |
050c60bf | 1056 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1057 | { |
d46285a2 | 1058 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1059 | |
d46285a2 | 1060 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1061 | |
050c60bf DDP |
1062 | if (dev->requested_n_txq == n_txq) { |
1063 | goto out; | |
4573fbd3 FL |
1064 | } |
1065 | ||
050c60bf DDP |
1066 | dev->requested_n_txq = n_txq; |
1067 | netdev_request_reconfigure(netdev); | |
58397e6c | 1068 | |
050c60bf | 1069 | out: |
d46285a2 | 1070 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1071 | return 0; |
58397e6c KT |
1072 | } |
1073 | ||
8a9562d2 PS |
1074 | static struct netdev_rxq * |
1075 | netdev_dpdk_rxq_alloc(void) | |
1076 | { | |
1077 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1078 | ||
1079 | return &rx->up; | |
1080 | } | |
1081 | ||
1082 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1083 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1084 | { |
d46285a2 | 1085 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1086 | } |
1087 | ||
1088 | static int | |
d46285a2 | 1089 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1090 | { |
d46285a2 DDP |
1091 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1092 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1093 | |
d46285a2 DDP |
1094 | ovs_mutex_lock(&dev->mutex); |
1095 | rx->port_id = dev->port_id; | |
1096 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1097 | |
1098 | return 0; | |
1099 | } | |
1100 | ||
1101 | static void | |
d46285a2 | 1102 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1103 | { |
1104 | } | |
1105 | ||
1106 | static void | |
d46285a2 | 1107 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1108 | { |
d46285a2 | 1109 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1110 | |
1111 | rte_free(rx); | |
1112 | } | |
1113 | ||
b170db2a | 1114 | static inline void |
b59cc14e IM |
1115 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
1116 | struct rte_mbuf **pkts, int cnt) | |
8a9562d2 | 1117 | { |
1304f1f8 DDP |
1118 | uint32_t nb_tx = 0; |
1119 | ||
b59cc14e | 1120 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1121 | uint32_t ret; |
1122 | ||
b59cc14e | 1123 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1124 | if (!ret) { |
1125 | break; | |
1126 | } | |
1127 | ||
1128 | nb_tx += ret; | |
1129 | } | |
8a9562d2 | 1130 | |
b59cc14e | 1131 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
db73f716 DDP |
1132 | /* free buffers, which we couldn't transmit, one at a time (each |
1133 | * packet could come from a different mempool) */ | |
1134 | int i; | |
1135 | ||
b59cc14e IM |
1136 | for (i = nb_tx; i < cnt; i++) { |
1137 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1138 | } |
45d947c4 | 1139 | rte_spinlock_lock(&dev->stats_lock); |
b59cc14e | 1140 | dev->stats.tx_dropped += cnt - nb_tx; |
45d947c4 | 1141 | rte_spinlock_unlock(&dev->stats_lock); |
8a9562d2 | 1142 | } |
8a9562d2 PS |
1143 | } |
1144 | ||
f3926f29 IS |
1145 | static inline bool |
1146 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1147 | struct rte_mbuf *pkt, uint64_t time) | |
1148 | { | |
1149 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1150 | ||
1151 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1152 | e_RTE_METER_GREEN; | |
1153 | } | |
1154 | ||
1155 | static int | |
1156 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
1157 | struct rte_mbuf **pkts, int pkt_cnt) | |
1158 | { | |
1159 | int i = 0; | |
1160 | int cnt = 0; | |
1161 | struct rte_mbuf *pkt = NULL; | |
1162 | uint64_t current_time = rte_rdtsc(); | |
1163 | ||
1164 | for (i = 0; i < pkt_cnt; i++) { | |
1165 | pkt = pkts[i]; | |
1166 | /* Handle current packet */ | |
1167 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1168 | if (cnt != i) { | |
1169 | pkts[cnt] = pkt; | |
1170 | } | |
1171 | cnt++; | |
1172 | } else { | |
1173 | rte_pktmbuf_free(pkt); | |
1174 | } | |
1175 | } | |
1176 | ||
1177 | return cnt; | |
1178 | } | |
1179 | ||
9509913a IS |
1180 | static int |
1181 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
1182 | int pkt_cnt) | |
1183 | { | |
1184 | int cnt = 0; | |
1185 | ||
1186 | rte_spinlock_lock(&policer->policer_lock); | |
1187 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, pkt_cnt); | |
1188 | rte_spinlock_unlock(&policer->policer_lock); | |
1189 | ||
1190 | return cnt; | |
1191 | } | |
1192 | ||
58397e6c | 1193 | static bool |
0a0f39df | 1194 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1195 | { |
0a0f39df | 1196 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1197 | } |
1198 | ||
d6e3feb5 | 1199 | static inline void |
1200 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1201 | unsigned int packet_size) | |
1202 | { | |
1203 | /* Hard-coded search for the size bucket. */ | |
1204 | if (packet_size < 256) { | |
1205 | if (packet_size >= 128) { | |
1206 | stats->rx_128_to_255_packets++; | |
1207 | } else if (packet_size <= 64) { | |
1208 | stats->rx_1_to_64_packets++; | |
1209 | } else { | |
1210 | stats->rx_65_to_127_packets++; | |
1211 | } | |
1212 | } else { | |
1213 | if (packet_size >= 1523) { | |
1214 | stats->rx_1523_to_max_packets++; | |
1215 | } else if (packet_size >= 1024) { | |
1216 | stats->rx_1024_to_1522_packets++; | |
1217 | } else if (packet_size < 512) { | |
1218 | stats->rx_256_to_511_packets++; | |
1219 | } else { | |
1220 | stats->rx_512_to_1023_packets++; | |
1221 | } | |
1222 | } | |
1223 | } | |
1224 | ||
9e3ddd45 TP |
1225 | static inline void |
1226 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1227 | struct dp_packet **packets, int count, |
1228 | int dropped) | |
9e3ddd45 TP |
1229 | { |
1230 | int i; | |
d6e3feb5 | 1231 | unsigned int packet_size; |
9e3ddd45 TP |
1232 | struct dp_packet *packet; |
1233 | ||
1234 | stats->rx_packets += count; | |
9509913a | 1235 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1236 | for (i = 0; i < count; i++) { |
1237 | packet = packets[i]; | |
d6e3feb5 | 1238 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1239 | |
d6e3feb5 | 1240 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1241 | /* This only protects the following multicast counting from |
1242 | * too short packets, but it does not stop the packet from | |
1243 | * further processing. */ | |
1244 | stats->rx_errors++; | |
1245 | stats->rx_length_errors++; | |
1246 | continue; | |
1247 | } | |
1248 | ||
d6e3feb5 | 1249 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1250 | ||
9e3ddd45 TP |
1251 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1252 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1253 | stats->multicast++; | |
1254 | } | |
1255 | ||
d6e3feb5 | 1256 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1257 | } |
1258 | } | |
1259 | ||
58397e6c KT |
1260 | /* |
1261 | * The receive path for the vhost port is the TX path out from guest. | |
1262 | */ | |
1263 | static int | |
d46285a2 | 1264 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1265 | struct dp_packet_batch *batch) |
58397e6c | 1266 | { |
d46285a2 | 1267 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
d46285a2 | 1268 | int qid = rxq->queue_id; |
9509913a | 1269 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1270 | uint16_t nb_rx = 0; |
9509913a | 1271 | uint16_t dropped = 0; |
58397e6c | 1272 | |
0a0f39df | 1273 | if (OVS_UNLIKELY(!is_vhost_running(dev) |
e543851d | 1274 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1275 | return EAGAIN; |
1276 | } | |
1277 | ||
0a0f39df CL |
1278 | nb_rx = rte_vhost_dequeue_burst(netdev_dpdk_get_vid(dev), |
1279 | qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
d46285a2 | 1280 | dev->dpdk_mp->mp, |
64839cf4 | 1281 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1282 | NETDEV_MAX_BURST); |
58397e6c KT |
1283 | if (!nb_rx) { |
1284 | return EAGAIN; | |
1285 | } | |
1286 | ||
9509913a IS |
1287 | if (policer) { |
1288 | dropped = nb_rx; | |
64839cf4 WT |
1289 | nb_rx = ingress_policer_run(policer, |
1290 | (struct rte_mbuf **) batch->packets, | |
1291 | nb_rx); | |
9509913a IS |
1292 | dropped -= nb_rx; |
1293 | } | |
1294 | ||
d46285a2 | 1295 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1296 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1297 | nb_rx, dropped); | |
d46285a2 | 1298 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1299 | |
64839cf4 | 1300 | batch->count = (int) nb_rx; |
58397e6c KT |
1301 | return 0; |
1302 | } | |
1303 | ||
8a9562d2 | 1304 | static int |
64839cf4 | 1305 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1306 | { |
d46285a2 DDP |
1307 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1308 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1309 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1310 | int nb_rx; |
9509913a | 1311 | int dropped = 0; |
8a9562d2 | 1312 | |
d46285a2 | 1313 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1314 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1315 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1316 | if (!nb_rx) { |
1317 | return EAGAIN; | |
1318 | } | |
1319 | ||
9509913a IS |
1320 | if (policer) { |
1321 | dropped = nb_rx; | |
64839cf4 WT |
1322 | nb_rx = ingress_policer_run(policer, |
1323 | (struct rte_mbuf **)batch->packets, | |
1324 | nb_rx); | |
9509913a IS |
1325 | dropped -= nb_rx; |
1326 | } | |
1327 | ||
1328 | /* Update stats to reflect dropped packets */ | |
1329 | if (OVS_UNLIKELY(dropped)) { | |
1330 | rte_spinlock_lock(&dev->stats_lock); | |
1331 | dev->stats.rx_dropped += dropped; | |
1332 | rte_spinlock_unlock(&dev->stats_lock); | |
1333 | } | |
1334 | ||
64839cf4 | 1335 | batch->count = nb_rx; |
8a9562d2 PS |
1336 | |
1337 | return 0; | |
1338 | } | |
1339 | ||
0bf765f7 IS |
1340 | static inline int |
1341 | netdev_dpdk_qos_run__(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1342 | int cnt) | |
1343 | { | |
1344 | struct netdev *netdev = &dev->up; | |
1345 | ||
1346 | if (dev->qos_conf != NULL) { | |
1347 | rte_spinlock_lock(&dev->qos_lock); | |
1348 | if (dev->qos_conf != NULL) { | |
1349 | cnt = dev->qos_conf->ops->qos_run(netdev, pkts, cnt); | |
1350 | } | |
1351 | rte_spinlock_unlock(&dev->qos_lock); | |
1352 | } | |
1353 | ||
1354 | return cnt; | |
1355 | } | |
1356 | ||
9e3ddd45 TP |
1357 | static inline void |
1358 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1359 | struct dp_packet **packets, | |
1360 | int attempted, | |
1361 | int dropped) | |
1362 | { | |
1363 | int i; | |
1364 | int sent = attempted - dropped; | |
1365 | ||
1366 | stats->tx_packets += sent; | |
1367 | stats->tx_dropped += dropped; | |
1368 | ||
1369 | for (i = 0; i < sent; i++) { | |
1370 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1371 | } | |
1372 | } | |
1373 | ||
58397e6c | 1374 | static void |
4573fbd3 FL |
1375 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1376 | struct dp_packet **pkts, int cnt, | |
1377 | bool may_steal) | |
58397e6c | 1378 | { |
d46285a2 | 1379 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1380 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1381 | unsigned int total_pkts = cnt; | |
0bf765f7 | 1382 | unsigned int qos_pkts = cnt; |
31871ee3 | 1383 | int retries = 0; |
58397e6c | 1384 | |
81acebda | 1385 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1386 | |
0a0f39df | 1387 | if (OVS_UNLIKELY(!is_vhost_running(dev) || qid < 0 |
e543851d | 1388 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1389 | rte_spinlock_lock(&dev->stats_lock); |
1390 | dev->stats.tx_dropped+= cnt; | |
1391 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1392 | goto out; |
58397e6c KT |
1393 | } |
1394 | ||
d46285a2 | 1395 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1396 | |
0bf765f7 | 1397 | /* Check has QoS has been configured for the netdev */ |
d46285a2 | 1398 | cnt = netdev_dpdk_qos_run__(dev, cur_pkts, cnt); |
0bf765f7 IS |
1399 | qos_pkts -= cnt; |
1400 | ||
95e9881f | 1401 | do { |
4573fbd3 | 1402 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1403 | unsigned int tx_pkts; |
1404 | ||
0a0f39df CL |
1405 | tx_pkts = rte_vhost_enqueue_burst(netdev_dpdk_get_vid(dev), |
1406 | vhost_qid, cur_pkts, cnt); | |
95e9881f KT |
1407 | if (OVS_LIKELY(tx_pkts)) { |
1408 | /* Packets have been sent.*/ | |
1409 | cnt -= tx_pkts; | |
31871ee3 | 1410 | /* Prepare for possible retry.*/ |
95e9881f KT |
1411 | cur_pkts = &cur_pkts[tx_pkts]; |
1412 | } else { | |
31871ee3 KT |
1413 | /* No packets sent - do not retry.*/ |
1414 | break; | |
95e9881f | 1415 | } |
31871ee3 | 1416 | } while (cnt && (retries++ < VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1417 | |
d46285a2 | 1418 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1419 | |
d46285a2 | 1420 | rte_spinlock_lock(&dev->stats_lock); |
0bf765f7 | 1421 | cnt += qos_pkts; |
d46285a2 DDP |
1422 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, cnt); |
1423 | rte_spinlock_unlock(&dev->stats_lock); | |
58397e6c KT |
1424 | |
1425 | out: | |
1426 | if (may_steal) { | |
95e9881f KT |
1427 | int i; |
1428 | ||
1429 | for (i = 0; i < total_pkts; i++) { | |
1b99bb05 MG |
1430 | dp_packet_delete(pkts[i]); |
1431 | } | |
58397e6c KT |
1432 | } |
1433 | } | |
1434 | ||
8a9562d2 PS |
1435 | /* Tx function. Transmit packets indefinitely */ |
1436 | static void | |
64839cf4 | 1437 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1438 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1439 | { |
bce01e3a | 1440 | #if !defined(__CHECKER__) && !defined(_WIN32) |
64839cf4 | 1441 | const size_t PKT_ARRAY_SIZE = batch->count; |
bce01e3a EJ |
1442 | #else |
1443 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1444 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1445 | #endif |
8a9562d2 | 1446 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
bce01e3a | 1447 | struct rte_mbuf *mbufs[PKT_ARRAY_SIZE]; |
175cf4de RW |
1448 | int dropped = 0; |
1449 | int newcnt = 0; | |
1450 | int i; | |
8a9562d2 | 1451 | |
db73f716 DDP |
1452 | /* If we are on a non pmd thread we have to use the mempool mutex, because |
1453 | * every non pmd thread shares the same mempool cache */ | |
1454 | ||
5f17de68 | 1455 | if (!dpdk_thread_is_pmd()) { |
db73f716 DDP |
1456 | ovs_mutex_lock(&nonpmd_mempool_mutex); |
1457 | } | |
1458 | ||
7d6d1a40 WT |
1459 | dp_packet_batch_apply_cutlen(batch); |
1460 | ||
64839cf4 WT |
1461 | for (i = 0; i < batch->count; i++) { |
1462 | int size = dp_packet_size(batch->packets[i]); | |
95fb793a | 1463 | |
f98d7864 | 1464 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c DDP |
1465 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1466 | (int)size , dev->max_packet_len); | |
1467 | ||
175cf4de | 1468 | dropped++; |
f4fd623c DDP |
1469 | continue; |
1470 | } | |
8a9562d2 | 1471 | |
f4fd623c | 1472 | mbufs[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1473 | |
f4fd623c | 1474 | if (!mbufs[newcnt]) { |
64839cf4 | 1475 | dropped += batch->count - i; |
175cf4de | 1476 | break; |
f4fd623c DDP |
1477 | } |
1478 | ||
1479 | /* We have to do a copy for now */ | |
64839cf4 WT |
1480 | memcpy(rte_pktmbuf_mtod(mbufs[newcnt], void *), |
1481 | dp_packet_data(batch->packets[i]), size); | |
f4fd623c DDP |
1482 | |
1483 | rte_pktmbuf_data_len(mbufs[newcnt]) = size; | |
1484 | rte_pktmbuf_pkt_len(mbufs[newcnt]) = size; | |
1485 | ||
1486 | newcnt++; | |
1487 | } | |
8a9562d2 | 1488 | |
58397e6c | 1489 | if (dev->type == DPDK_DEV_VHOST) { |
64839cf4 WT |
1490 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) mbufs, |
1491 | newcnt, true); | |
58397e6c | 1492 | } else { |
0bf765f7 IS |
1493 | unsigned int qos_pkts = newcnt; |
1494 | ||
1495 | /* Check if QoS has been configured for this netdev. */ | |
1496 | newcnt = netdev_dpdk_qos_run__(dev, mbufs, newcnt); | |
1497 | ||
1498 | dropped += qos_pkts - newcnt; | |
b59cc14e | 1499 | netdev_dpdk_eth_tx_burst(dev, qid, mbufs, newcnt); |
58397e6c | 1500 | } |
db73f716 | 1501 | |
0bf765f7 IS |
1502 | if (OVS_UNLIKELY(dropped)) { |
1503 | rte_spinlock_lock(&dev->stats_lock); | |
1504 | dev->stats.tx_dropped += dropped; | |
1505 | rte_spinlock_unlock(&dev->stats_lock); | |
1506 | } | |
1507 | ||
5f17de68 | 1508 | if (!dpdk_thread_is_pmd()) { |
db73f716 DDP |
1509 | ovs_mutex_unlock(&nonpmd_mempool_mutex); |
1510 | } | |
8a9562d2 PS |
1511 | } |
1512 | ||
58397e6c | 1513 | static int |
64839cf4 WT |
1514 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1515 | struct dp_packet_batch *batch, | |
324c8374 | 1516 | bool may_steal, bool concurrent_txq OVS_UNUSED) |
58397e6c | 1517 | { |
58397e6c | 1518 | |
64839cf4 WT |
1519 | if (OVS_UNLIKELY(batch->packets[0]->source != DPBUF_DPDK)) { |
1520 | dpdk_do_tx_copy(netdev, qid, batch); | |
1521 | dp_packet_delete_batch(batch, may_steal); | |
58397e6c | 1522 | } else { |
64839cf4 WT |
1523 | dp_packet_batch_apply_cutlen(batch); |
1524 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count, | |
1525 | may_steal); | |
58397e6c KT |
1526 | } |
1527 | return 0; | |
1528 | } | |
1529 | ||
7251515e DV |
1530 | static inline void |
1531 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
324c8374 IM |
1532 | struct dp_packet_batch *batch, bool may_steal, |
1533 | bool concurrent_txq) | |
8a9562d2 | 1534 | { |
324c8374 | 1535 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 1536 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
1537 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
1538 | } | |
1539 | ||
7251515e | 1540 | if (OVS_UNLIKELY(!may_steal || |
64839cf4 | 1541 | batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1542 | struct netdev *netdev = &dev->up; |
1543 | ||
64839cf4 WT |
1544 | dpdk_do_tx_copy(netdev, qid, batch); |
1545 | dp_packet_delete_batch(batch, may_steal); | |
8a9562d2 | 1546 | } else { |
f4fd623c DDP |
1547 | int next_tx_idx = 0; |
1548 | int dropped = 0; | |
0bf765f7 IS |
1549 | unsigned int qos_pkts = 0; |
1550 | unsigned int temp_cnt = 0; | |
64839cf4 | 1551 | int cnt = batch->count; |
8a9562d2 | 1552 | |
7d6d1a40 WT |
1553 | dp_packet_batch_apply_cutlen(batch); |
1554 | ||
64839cf4 WT |
1555 | for (int i = 0; i < cnt; i++) { |
1556 | int size = dp_packet_size(batch->packets[i]); | |
1b99bb05 | 1557 | |
f4fd623c DDP |
1558 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
1559 | if (next_tx_idx != i) { | |
0bf765f7 IS |
1560 | temp_cnt = i - next_tx_idx; |
1561 | qos_pkts = temp_cnt; | |
1562 | ||
64839cf4 WT |
1563 | temp_cnt = netdev_dpdk_qos_run__(dev, |
1564 | (struct rte_mbuf**)batch->packets, | |
1565 | temp_cnt); | |
0bf765f7 | 1566 | dropped += qos_pkts - temp_cnt; |
b59cc14e | 1567 | netdev_dpdk_eth_tx_burst(dev, qid, |
64839cf4 WT |
1568 | (struct rte_mbuf **)&batch->packets[next_tx_idx], |
1569 | temp_cnt); | |
0bf765f7 | 1570 | |
1ebfe1ac | 1571 | } |
f4fd623c | 1572 | |
1ebfe1ac DDP |
1573 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1574 | (int)size , dev->max_packet_len); | |
f4fd623c | 1575 | |
64839cf4 | 1576 | dp_packet_delete(batch->packets[i]); |
1ebfe1ac | 1577 | dropped++; |
f4fd623c DDP |
1578 | next_tx_idx = i + 1; |
1579 | } | |
1580 | } | |
1581 | if (next_tx_idx != cnt) { | |
0bf765f7 IS |
1582 | cnt -= next_tx_idx; |
1583 | qos_pkts = cnt; | |
1584 | ||
64839cf4 WT |
1585 | cnt = netdev_dpdk_qos_run__(dev, |
1586 | (struct rte_mbuf**)batch->packets, cnt); | |
0bf765f7 | 1587 | dropped += qos_pkts - cnt; |
b59cc14e | 1588 | netdev_dpdk_eth_tx_burst(dev, qid, |
64839cf4 WT |
1589 | (struct rte_mbuf **)&batch->packets[next_tx_idx], |
1590 | cnt); | |
f4fd623c | 1591 | } |
8a9562d2 | 1592 | |
f4fd623c | 1593 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1594 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1595 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1596 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1597 | } |
8a9562d2 | 1598 | } |
a0cb2d66 | 1599 | |
324c8374 | 1600 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
1601 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
1602 | } | |
7251515e DV |
1603 | } |
1604 | ||
1605 | static int | |
1606 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
324c8374 IM |
1607 | struct dp_packet_batch *batch, bool may_steal, |
1608 | bool concurrent_txq) | |
7251515e DV |
1609 | { |
1610 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1611 | |
324c8374 | 1612 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e | 1613 | return 0; |
8a9562d2 PS |
1614 | } |
1615 | ||
1616 | static int | |
74ff3298 | 1617 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1618 | { |
1619 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1620 | ||
1621 | ovs_mutex_lock(&dev->mutex); | |
1622 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1623 | dev->hwaddr = mac; |
045c0d1a | 1624 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1625 | } |
1626 | ovs_mutex_unlock(&dev->mutex); | |
1627 | ||
1628 | return 0; | |
1629 | } | |
1630 | ||
1631 | static int | |
74ff3298 | 1632 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1633 | { |
1634 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1635 | ||
1636 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1637 | *mac = dev->hwaddr; |
8a9562d2 PS |
1638 | ovs_mutex_unlock(&dev->mutex); |
1639 | ||
1640 | return 0; | |
1641 | } | |
1642 | ||
1643 | static int | |
1644 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1645 | { | |
1646 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1647 | ||
1648 | ovs_mutex_lock(&dev->mutex); | |
1649 | *mtup = dev->mtu; | |
1650 | ovs_mutex_unlock(&dev->mutex); | |
1651 | ||
1652 | return 0; | |
1653 | } | |
1654 | ||
1655 | static int | |
1656 | netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu) | |
1657 | { | |
1658 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
4be4d22c | 1659 | int old_mtu, err, dpdk_mtu; |
8a9562d2 PS |
1660 | struct dpdk_mp *old_mp; |
1661 | struct dpdk_mp *mp; | |
4be4d22c | 1662 | uint32_t buf_size; |
8a9562d2 PS |
1663 | |
1664 | ovs_mutex_lock(&dpdk_mutex); | |
1665 | ovs_mutex_lock(&dev->mutex); | |
1666 | if (dev->mtu == mtu) { | |
1667 | err = 0; | |
1668 | goto out; | |
1669 | } | |
1670 | ||
4be4d22c MK |
1671 | buf_size = dpdk_buf_size(mtu); |
1672 | dpdk_mtu = FRAME_LEN_TO_MTU(buf_size); | |
1673 | ||
1674 | mp = dpdk_mp_get(dev->socket_id, dpdk_mtu); | |
8a9562d2 PS |
1675 | if (!mp) { |
1676 | err = ENOMEM; | |
1677 | goto out; | |
1678 | } | |
1679 | ||
1680 | rte_eth_dev_stop(dev->port_id); | |
1681 | ||
1682 | old_mtu = dev->mtu; | |
1683 | old_mp = dev->dpdk_mp; | |
1684 | dev->dpdk_mp = mp; | |
1685 | dev->mtu = mtu; | |
4be4d22c | 1686 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
8a9562d2 PS |
1687 | |
1688 | err = dpdk_eth_dev_init(dev); | |
1689 | if (err) { | |
8a9562d2 PS |
1690 | dpdk_mp_put(mp); |
1691 | dev->mtu = old_mtu; | |
1692 | dev->dpdk_mp = old_mp; | |
4be4d22c | 1693 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
8a9562d2 PS |
1694 | dpdk_eth_dev_init(dev); |
1695 | goto out; | |
1696 | } | |
1697 | ||
1698 | dpdk_mp_put(old_mp); | |
045c0d1a | 1699 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1700 | out: |
1701 | ovs_mutex_unlock(&dev->mutex); | |
1702 | ovs_mutex_unlock(&dpdk_mutex); | |
1703 | return err; | |
1704 | } | |
1705 | ||
1706 | static int | |
d46285a2 | 1707 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 1708 | |
58397e6c KT |
1709 | static int |
1710 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1711 | struct netdev_stats *stats) | |
1712 | { | |
1713 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1714 | ||
1715 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 1716 | |
45d947c4 | 1717 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
1718 | /* Supported Stats */ |
1719 | stats->rx_packets += dev->stats.rx_packets; | |
1720 | stats->tx_packets += dev->stats.tx_packets; | |
9509913a | 1721 | stats->rx_dropped = dev->stats.rx_dropped; |
58397e6c | 1722 | stats->tx_dropped += dev->stats.tx_dropped; |
9e3ddd45 TP |
1723 | stats->multicast = dev->stats.multicast; |
1724 | stats->rx_bytes = dev->stats.rx_bytes; | |
1725 | stats->tx_bytes = dev->stats.tx_bytes; | |
1726 | stats->rx_errors = dev->stats.rx_errors; | |
1727 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 1728 | |
1729 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
1730 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
1731 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
1732 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
1733 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
1734 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
1735 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
1736 | ||
45d947c4 | 1737 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 1738 | |
58397e6c KT |
1739 | ovs_mutex_unlock(&dev->mutex); |
1740 | ||
1741 | return 0; | |
1742 | } | |
1743 | ||
d6e3feb5 | 1744 | static void |
1745 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
1746 | const struct rte_eth_xstat *xstats, |
1747 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 1748 | const unsigned int size) |
1749 | { | |
d6e3feb5 | 1750 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 1751 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1752 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1753 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1754 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1755 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1756 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1757 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1758 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1759 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1760 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1761 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1762 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 1763 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1764 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 1765 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1766 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1767 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1768 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1769 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1770 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1771 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1772 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1773 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1774 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1775 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1776 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 1777 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1778 | stats->tx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 1779 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1780 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 1781 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1782 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 1783 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1784 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 1785 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1786 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 1787 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1788 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 1789 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1790 | stats->rx_jabber_errors = xstats[i].value; |
1791 | } | |
1792 | } | |
1793 | } | |
1794 | ||
8a9562d2 PS |
1795 | static int |
1796 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1797 | { | |
1798 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1799 | struct rte_eth_stats rte_stats; | |
1800 | bool gg; | |
1801 | ||
1802 | netdev_dpdk_get_carrier(netdev, &gg); | |
1803 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 1804 | |
0a0f39df CL |
1805 | struct rte_eth_xstat *rte_xstats = NULL; |
1806 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
1807 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 1808 | |
1809 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
1810 | VLOG_ERR("Can't get ETH statistics for port: %i.", dev->port_id); | |
f9256822 | 1811 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 1812 | return EPROTO; |
1813 | } | |
1814 | ||
0a0f39df CL |
1815 | /* Get length of statistics */ |
1816 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
1817 | if (rte_xstats_len < 0) { | |
1818 | VLOG_WARN("Cannot get XSTATS values for port: %i", dev->port_id); | |
1819 | goto out; | |
1820 | } | |
1821 | /* Reserve memory for xstats names and values */ | |
1822 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
1823 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
1824 | ||
1825 | /* Retreive xstats names */ | |
1826 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
1827 | rte_xstats_names, | |
1828 | rte_xstats_len); | |
1829 | if (rte_xstats_new_len != rte_xstats_len) { | |
1830 | VLOG_WARN("Cannot get XSTATS names for port: %i.", dev->port_id); | |
1831 | goto out; | |
1832 | } | |
1833 | /* Retreive xstats values */ | |
1834 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
1835 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
1836 | rte_xstats_len); | |
1837 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
1838 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
1839 | rte_xstats_len); | |
d6e3feb5 | 1840 | } else { |
0a0f39df | 1841 | VLOG_WARN("Cannot get XSTATS values for port: %i.", dev->port_id); |
d6e3feb5 | 1842 | } |
8a9562d2 | 1843 | |
0a0f39df CL |
1844 | out: |
1845 | free(rte_xstats); | |
1846 | free(rte_xstats_names); | |
1847 | ||
2f9dd77f PS |
1848 | stats->rx_packets = rte_stats.ipackets; |
1849 | stats->tx_packets = rte_stats.opackets; | |
1850 | stats->rx_bytes = rte_stats.ibytes; | |
1851 | stats->tx_bytes = rte_stats.obytes; | |
9e3ddd45 TP |
1852 | /* DPDK counts imissed as errors, but count them here as dropped instead */ |
1853 | stats->rx_errors = rte_stats.ierrors - rte_stats.imissed; | |
2f9dd77f | 1854 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 1855 | |
45d947c4 | 1856 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 1857 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 1858 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 1859 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
1860 | |
1861 | /* These are the available DPDK counters for packets not received due to | |
1862 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 1863 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
1864 | stats->rx_missed_errors = rte_stats.imissed; |
1865 | ||
8a9562d2 PS |
1866 | ovs_mutex_unlock(&dev->mutex); |
1867 | ||
1868 | return 0; | |
1869 | } | |
1870 | ||
1871 | static int | |
d46285a2 | 1872 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 PS |
1873 | enum netdev_features *current, |
1874 | enum netdev_features *advertised OVS_UNUSED, | |
1875 | enum netdev_features *supported OVS_UNUSED, | |
1876 | enum netdev_features *peer OVS_UNUSED) | |
1877 | { | |
d46285a2 | 1878 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1879 | struct rte_eth_link link; |
1880 | ||
1881 | ovs_mutex_lock(&dev->mutex); | |
1882 | link = dev->link; | |
1883 | ovs_mutex_unlock(&dev->mutex); | |
1884 | ||
362ca396 | 1885 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
1886 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
1887 | *current = NETDEV_F_10MB_HD; |
1888 | } | |
362ca396 | 1889 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
1890 | *current = NETDEV_F_100MB_HD; |
1891 | } | |
362ca396 | 1892 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
1893 | *current = NETDEV_F_1GB_HD; |
1894 | } | |
1895 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 1896 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
1897 | *current = NETDEV_F_10MB_FD; |
1898 | } | |
362ca396 | 1899 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
1900 | *current = NETDEV_F_100MB_FD; |
1901 | } | |
362ca396 | 1902 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
1903 | *current = NETDEV_F_1GB_FD; |
1904 | } | |
362ca396 | 1905 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
1906 | *current = NETDEV_F_10GB_FD; |
1907 | } | |
1908 | } | |
1909 | ||
362ca396 | 1910 | if (link.link_autoneg) { |
1911 | *current |= NETDEV_F_AUTONEG; | |
1912 | } | |
1913 | ||
8a9562d2 PS |
1914 | return 0; |
1915 | } | |
1916 | ||
9509913a IS |
1917 | static struct ingress_policer * |
1918 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
1919 | { | |
1920 | struct ingress_policer *policer = NULL; | |
1921 | uint64_t rate_bytes; | |
1922 | uint64_t burst_bytes; | |
1923 | int err = 0; | |
1924 | ||
1925 | policer = xmalloc(sizeof *policer); | |
1926 | rte_spinlock_init(&policer->policer_lock); | |
1927 | ||
1928 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
1929 | rate_bytes = rate * 1000/8; | |
1930 | burst_bytes = burst * 1000/8; | |
1931 | ||
1932 | policer->app_srtcm_params.cir = rate_bytes; | |
1933 | policer->app_srtcm_params.cbs = burst_bytes; | |
1934 | policer->app_srtcm_params.ebs = 0; | |
1935 | err = rte_meter_srtcm_config(&policer->in_policer, | |
1936 | &policer->app_srtcm_params); | |
1937 | if(err) { | |
1938 | VLOG_ERR("Could not create rte meter for ingress policer"); | |
1939 | return NULL; | |
1940 | } | |
1941 | ||
1942 | return policer; | |
1943 | } | |
1944 | ||
1945 | static int | |
1946 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
1947 | uint32_t policer_burst) | |
1948 | { | |
1949 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1950 | struct ingress_policer *policer; | |
1951 | ||
1952 | /* Force to 0 if no rate specified, | |
1953 | * default to 8000 kbits if burst is 0, | |
1954 | * else stick with user-specified value. | |
1955 | */ | |
1956 | policer_burst = (!policer_rate ? 0 | |
1957 | : !policer_burst ? 8000 | |
1958 | : policer_burst); | |
1959 | ||
1960 | ovs_mutex_lock(&dev->mutex); | |
1961 | ||
1962 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
1963 | &dev->ingress_policer); | |
1964 | ||
1965 | if (dev->policer_rate == policer_rate && | |
1966 | dev->policer_burst == policer_burst) { | |
1967 | /* Assume that settings haven't changed since we last set them. */ | |
1968 | ovs_mutex_unlock(&dev->mutex); | |
1969 | return 0; | |
1970 | } | |
1971 | ||
1972 | /* Destroy any existing ingress policer for the device if one exists */ | |
1973 | if (policer) { | |
1974 | ovsrcu_postpone(free, policer); | |
1975 | } | |
1976 | ||
1977 | if (policer_rate != 0) { | |
1978 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
1979 | } else { | |
1980 | policer = NULL; | |
1981 | } | |
1982 | ovsrcu_set(&dev->ingress_policer, policer); | |
1983 | dev->policer_rate = policer_rate; | |
1984 | dev->policer_burst = policer_burst; | |
1985 | ovs_mutex_unlock(&dev->mutex); | |
1986 | ||
1987 | return 0; | |
1988 | } | |
1989 | ||
8a9562d2 PS |
1990 | static int |
1991 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
1992 | { | |
1993 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1994 | int ifindex; | |
1995 | ||
1996 | ovs_mutex_lock(&dev->mutex); | |
1997 | ifindex = dev->port_id; | |
1998 | ovs_mutex_unlock(&dev->mutex); | |
1999 | ||
2000 | return ifindex; | |
2001 | } | |
2002 | ||
2003 | static int | |
d46285a2 | 2004 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2005 | { |
d46285a2 | 2006 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2007 | |
2008 | ovs_mutex_lock(&dev->mutex); | |
2009 | check_link_status(dev); | |
2010 | *carrier = dev->link.link_status; | |
58397e6c KT |
2011 | |
2012 | ovs_mutex_unlock(&dev->mutex); | |
2013 | ||
2014 | return 0; | |
2015 | } | |
2016 | ||
2017 | static int | |
d46285a2 | 2018 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2019 | { |
d46285a2 | 2020 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2021 | |
2022 | ovs_mutex_lock(&dev->mutex); | |
2023 | ||
0a0f39df | 2024 | if (is_vhost_running(dev)) { |
58397e6c KT |
2025 | *carrier = 1; |
2026 | } else { | |
2027 | *carrier = 0; | |
2028 | } | |
2029 | ||
8a9562d2 PS |
2030 | ovs_mutex_unlock(&dev->mutex); |
2031 | ||
2032 | return 0; | |
2033 | } | |
2034 | ||
2035 | static long long int | |
d46285a2 | 2036 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2037 | { |
d46285a2 | 2038 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2039 | long long int carrier_resets; |
2040 | ||
2041 | ovs_mutex_lock(&dev->mutex); | |
2042 | carrier_resets = dev->link_reset_cnt; | |
2043 | ovs_mutex_unlock(&dev->mutex); | |
2044 | ||
2045 | return carrier_resets; | |
2046 | } | |
2047 | ||
2048 | static int | |
d46285a2 | 2049 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2050 | long long int interval OVS_UNUSED) |
2051 | { | |
ee32150e | 2052 | return EOPNOTSUPP; |
8a9562d2 PS |
2053 | } |
2054 | ||
2055 | static int | |
2056 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2057 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2058 | enum netdev_flags *old_flagsp) |
2059 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
2060 | { |
2061 | int err; | |
2062 | ||
2063 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
2064 | return EINVAL; | |
2065 | } | |
2066 | ||
2067 | *old_flagsp = dev->flags; | |
2068 | dev->flags |= on; | |
2069 | dev->flags &= ~off; | |
2070 | ||
2071 | if (dev->flags == *old_flagsp) { | |
2072 | return 0; | |
2073 | } | |
2074 | ||
58397e6c KT |
2075 | if (dev->type == DPDK_DEV_ETH) { |
2076 | if (dev->flags & NETDEV_UP) { | |
2077 | err = rte_eth_dev_start(dev->port_id); | |
2078 | if (err) | |
2079 | return -err; | |
2080 | } | |
8a9562d2 | 2081 | |
58397e6c KT |
2082 | if (dev->flags & NETDEV_PROMISC) { |
2083 | rte_eth_promiscuous_enable(dev->port_id); | |
2084 | } | |
8a9562d2 | 2085 | |
58397e6c KT |
2086 | if (!(dev->flags & NETDEV_UP)) { |
2087 | rte_eth_dev_stop(dev->port_id); | |
2088 | } | |
e543851d ZB |
2089 | } else { |
2090 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2091 | * running then change netdev's change_seq to trigger link state | |
2092 | * update. */ | |
e543851d ZB |
2093 | |
2094 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2095 | && is_vhost_running(dev)) { |
e543851d ZB |
2096 | netdev_change_seq_changed(&dev->up); |
2097 | ||
2098 | /* Clear statistics if device is getting up. */ | |
2099 | if (NETDEV_UP & on) { | |
2100 | rte_spinlock_lock(&dev->stats_lock); | |
2101 | memset(&dev->stats, 0, sizeof(dev->stats)); | |
2102 | rte_spinlock_unlock(&dev->stats_lock); | |
2103 | } | |
2104 | } | |
8a9562d2 PS |
2105 | } |
2106 | ||
2107 | return 0; | |
2108 | } | |
2109 | ||
2110 | static int | |
d46285a2 | 2111 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2112 | enum netdev_flags off, enum netdev_flags on, |
2113 | enum netdev_flags *old_flagsp) | |
2114 | { | |
d46285a2 | 2115 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2116 | int error; |
2117 | ||
d46285a2 DDP |
2118 | ovs_mutex_lock(&dev->mutex); |
2119 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2120 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2121 | |
2122 | return error; | |
2123 | } | |
2124 | ||
2125 | static int | |
d46285a2 | 2126 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2127 | { |
d46285a2 | 2128 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2129 | struct rte_eth_dev_info dev_info; |
2130 | ||
e0a801c7 | 2131 | if (dev->port_id < 0) |
8a9562d2 PS |
2132 | return ENODEV; |
2133 | ||
2134 | ovs_mutex_lock(&dev->mutex); | |
2135 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2136 | ovs_mutex_unlock(&dev->mutex); | |
2137 | ||
95fb793a | 2138 | smap_add_format(args, "port_no", "%d", dev->port_id); |
8a9562d2 PS |
2139 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); |
2140 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
2141 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2142 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2143 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2144 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2145 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
2146 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
2147 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
2148 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2149 | ||
39c2baa9 | 2150 | if (dev_info.pci_dev) { |
2151 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2152 | dev_info.pci_dev->id.vendor_id); | |
2153 | smap_add_format(args, "pci-device_id", "0x%x", | |
2154 | dev_info.pci_dev->id.device_id); | |
2155 | } | |
8a9562d2 PS |
2156 | |
2157 | return 0; | |
2158 | } | |
2159 | ||
2160 | static void | |
2161 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2162 | OVS_REQUIRES(dev->mutex) | |
2163 | { | |
2164 | enum netdev_flags old_flags; | |
2165 | ||
2166 | if (admin_state) { | |
2167 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2168 | } else { | |
2169 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2170 | } | |
2171 | } | |
2172 | ||
2173 | static void | |
2174 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2175 | const char *argv[], void *aux OVS_UNUSED) | |
2176 | { | |
2177 | bool up; | |
2178 | ||
2179 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2180 | up = true; | |
2181 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2182 | up = false; | |
2183 | } else { | |
2184 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2185 | return; | |
2186 | } | |
2187 | ||
2188 | if (argc > 2) { | |
2189 | struct netdev *netdev = netdev_from_name(argv[1]); | |
2190 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
2191 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
2192 | ||
2193 | ovs_mutex_lock(&dpdk_dev->mutex); | |
2194 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
2195 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
2196 | ||
2197 | netdev_close(netdev); | |
2198 | } else { | |
2199 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2200 | netdev_close(netdev); | |
2201 | return; | |
2202 | } | |
2203 | } else { | |
2204 | struct netdev_dpdk *netdev; | |
2205 | ||
2206 | ovs_mutex_lock(&dpdk_mutex); | |
2207 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
2208 | ovs_mutex_lock(&netdev->mutex); | |
2209 | netdev_dpdk_set_admin_state__(netdev, up); | |
2210 | ovs_mutex_unlock(&netdev->mutex); | |
2211 | } | |
2212 | ovs_mutex_unlock(&dpdk_mutex); | |
2213 | } | |
2214 | unixctl_command_reply(conn, "OK"); | |
2215 | } | |
2216 | ||
58397e6c KT |
2217 | /* |
2218 | * Set virtqueue flags so that we do not receive interrupts. | |
2219 | */ | |
2220 | static void | |
0a0f39df | 2221 | set_irq_status(int vid) |
58397e6c | 2222 | { |
4573fbd3 FL |
2223 | uint32_t i; |
2224 | uint64_t idx; | |
2225 | ||
0a0f39df | 2226 | for (i = 0; i < rte_vhost_get_queue_num(vid); i++) { |
4573fbd3 | 2227 | idx = i * VIRTIO_QNUM; |
0a0f39df CL |
2228 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_RXQ, 0); |
2229 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_TXQ, 0); | |
4573fbd3 FL |
2230 | } |
2231 | } | |
2232 | ||
585a5bea IM |
2233 | /* |
2234 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2235 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2236 | */ |
2237 | static void | |
d46285a2 DDP |
2238 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2239 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2240 | { |
2241 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2242 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea IM |
2243 | |
2244 | enabled_queues = dpdk_rte_mzalloc(total_txqs * sizeof *enabled_queues); | |
2245 | ||
2246 | for (i = 0; i < total_txqs; i++) { | |
2247 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2248 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2249 | enabled_queues[n_enabled++] = i; |
2250 | } | |
2251 | } | |
2252 | ||
2253 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2254 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2255 | n_enabled = 1; |
2256 | } | |
2257 | ||
2258 | k = 0; | |
2259 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2260 | if (dev->tx_q[i].map != i) { |
2261 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2262 | k = (k + 1) % n_enabled; |
2263 | } | |
2264 | } | |
2265 | ||
d46285a2 | 2266 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2267 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2268 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2269 | } |
2270 | ||
2271 | rte_free(enabled_queues); | |
2272 | } | |
4573fbd3 | 2273 | |
58397e6c KT |
2274 | /* |
2275 | * A new virtio-net device is added to a vhost port. | |
2276 | */ | |
2277 | static int | |
0a0f39df | 2278 | new_device(int vid) |
58397e6c | 2279 | { |
d46285a2 | 2280 | struct netdev_dpdk *dev; |
58397e6c | 2281 | bool exists = false; |
db8f13b0 | 2282 | int newnode = 0; |
0a0f39df CL |
2283 | char ifname[IF_NAME_SZ]; |
2284 | ||
2285 | rte_vhost_get_ifname(vid, ifname, sizeof(ifname)); | |
58397e6c KT |
2286 | |
2287 | ovs_mutex_lock(&dpdk_mutex); | |
2288 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2289 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
0a0f39df CL |
2290 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
2291 | uint32_t qp_num = rte_vhost_get_queue_num(vid); | |
db8f13b0 | 2292 | |
81acebda | 2293 | ovs_mutex_lock(&dev->mutex); |
db8f13b0 | 2294 | /* Get NUMA information */ |
0a0f39df CL |
2295 | newnode = rte_vhost_get_numa_node(vid); |
2296 | if (newnode == -1) { | |
db8f13b0 | 2297 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2298 | ifname); |
db8f13b0 | 2299 | newnode = dev->socket_id; |
db8f13b0 CL |
2300 | } |
2301 | ||
81acebda IM |
2302 | dev->requested_socket_id = newnode; |
2303 | dev->requested_n_rxq = qp_num; | |
2304 | dev->requested_n_txq = qp_num; | |
2305 | netdev_request_reconfigure(&dev->up); | |
2306 | ||
0a0f39df | 2307 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2308 | exists = true; |
2309 | ||
58397e6c | 2310 | /* Disable notifications. */ |
0a0f39df | 2311 | set_irq_status(vid); |
e543851d | 2312 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2313 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2314 | break; |
2315 | } | |
2316 | } | |
2317 | ovs_mutex_unlock(&dpdk_mutex); | |
2318 | ||
2319 | if (!exists) { | |
0a0f39df | 2320 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2321 | |
2322 | return -1; | |
2323 | } | |
2324 | ||
0a0f39df CL |
2325 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2326 | ifname, newnode); | |
2327 | ||
58397e6c KT |
2328 | return 0; |
2329 | } | |
2330 | ||
f3ea2ad2 IM |
2331 | /* Clears mapping for all available queues of vhost interface. */ |
2332 | static void | |
2333 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2334 | OVS_REQUIRES(dev->mutex) | |
2335 | { | |
2336 | int i; | |
2337 | ||
81acebda | 2338 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
2339 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
2340 | } | |
2341 | } | |
2342 | ||
58397e6c KT |
2343 | /* |
2344 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2345 | * flag to stop any more packets from being sent or received to/from a VM and | |
2346 | * ensure all currently queued packets have been sent/received before removing | |
2347 | * the device. | |
2348 | */ | |
2349 | static void | |
0a0f39df | 2350 | destroy_device(int vid) |
58397e6c | 2351 | { |
d46285a2 | 2352 | struct netdev_dpdk *dev; |
afee281f | 2353 | bool exists = false; |
0a0f39df CL |
2354 | char ifname[IF_NAME_SZ]; |
2355 | ||
2356 | rte_vhost_get_ifname(vid, ifname, sizeof(ifname)); | |
58397e6c KT |
2357 | |
2358 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2359 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2360 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 2361 | |
d46285a2 | 2362 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
2363 | dev->vhost_reconfigured = false; |
2364 | ovsrcu_index_set(&dev->vid, -1); | |
81acebda | 2365 | /* Clear tx/rx queue settings. */ |
d46285a2 | 2366 | netdev_dpdk_txq_map_clear(dev); |
81acebda IM |
2367 | dev->requested_n_rxq = NR_QUEUE; |
2368 | dev->requested_n_txq = NR_QUEUE; | |
2369 | netdev_request_reconfigure(&dev->up); | |
2370 | ||
e543851d | 2371 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2372 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 2373 | exists = true; |
afee281f | 2374 | break; |
58397e6c KT |
2375 | } |
2376 | } | |
afee281f | 2377 | |
58397e6c KT |
2378 | ovs_mutex_unlock(&dpdk_mutex); |
2379 | ||
0a0f39df | 2380 | if (exists) { |
afee281f KT |
2381 | /* |
2382 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2383 | * to NULL, before returning. | |
2384 | */ | |
2385 | ovsrcu_synchronize(); | |
2386 | /* | |
2387 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2388 | * put thread back into quiescent state before returning. | |
2389 | */ | |
2390 | ovsrcu_quiesce_start(); | |
0a0f39df | 2391 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 2392 | } else { |
0a0f39df | 2393 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 2394 | } |
58397e6c KT |
2395 | } |
2396 | ||
585a5bea | 2397 | static int |
0a0f39df | 2398 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 2399 | { |
d46285a2 | 2400 | struct netdev_dpdk *dev; |
585a5bea IM |
2401 | bool exists = false; |
2402 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
2403 | char ifname[IF_NAME_SZ]; |
2404 | ||
2405 | rte_vhost_get_ifname(vid, ifname, sizeof(ifname)); | |
585a5bea IM |
2406 | |
2407 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2408 | return 0; | |
2409 | } | |
2410 | ||
2411 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2412 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2413 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
d46285a2 | 2414 | ovs_mutex_lock(&dev->mutex); |
585a5bea | 2415 | if (enable) { |
d46285a2 | 2416 | dev->tx_q[qid].map = qid; |
585a5bea | 2417 | } else { |
d46285a2 | 2418 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2419 | } |
d46285a2 | 2420 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2421 | exists = true; |
d46285a2 | 2422 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2423 | break; |
2424 | } | |
2425 | } | |
2426 | ovs_mutex_unlock(&dpdk_mutex); | |
2427 | ||
2428 | if (exists) { | |
0a0f39df CL |
2429 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
2430 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 2431 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 2432 | } else { |
0a0f39df | 2433 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
2434 | return -1; |
2435 | } | |
2436 | ||
2437 | return 0; | |
2438 | } | |
2439 | ||
0a0f39df CL |
2440 | int |
2441 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 2442 | { |
0a0f39df | 2443 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
2444 | } |
2445 | ||
9509913a IS |
2446 | struct ingress_policer * |
2447 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
2448 | { | |
2449 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
2450 | } | |
2451 | ||
58397e6c KT |
2452 | /* |
2453 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
2454 | * configuration has been fully complete. | |
2455 | */ | |
bce01e3a | 2456 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
2457 | { |
2458 | .new_device = new_device, | |
2459 | .destroy_device = destroy_device, | |
585a5bea | 2460 | .vring_state_changed = vring_state_changed |
58397e6c KT |
2461 | }; |
2462 | ||
2463 | static void * | |
7d1ced01 | 2464 | start_vhost_loop(void *dummy OVS_UNUSED) |
58397e6c KT |
2465 | { |
2466 | pthread_detach(pthread_self()); | |
618f44f7 KT |
2467 | /* Put the cuse thread into quiescent state. */ |
2468 | ovsrcu_quiesce_start(); | |
58397e6c KT |
2469 | rte_vhost_driver_session_start(); |
2470 | return NULL; | |
2471 | } | |
2472 | ||
2473 | static int | |
2474 | dpdk_vhost_class_init(void) | |
7d1ced01 CL |
2475 | { |
2476 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
362ca396 | 2477 | rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4 |
2478 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
2479 | | 1ULL << VIRTIO_NET_F_CSUM); | |
2480 | ||
7d1ced01 CL |
2481 | ovs_thread_create("vhost_thread", start_vhost_loop, NULL); |
2482 | return 0; | |
2483 | } | |
2484 | ||
2485 | static int | |
2486 | dpdk_vhost_cuse_class_init(void) | |
58397e6c | 2487 | { |
7d1ced01 CL |
2488 | return 0; |
2489 | } | |
2490 | ||
2491 | static int | |
2492 | dpdk_vhost_user_class_init(void) | |
2493 | { | |
618f44f7 | 2494 | return 0; |
58397e6c KT |
2495 | } |
2496 | ||
033e9df2 DDP |
2497 | static void |
2498 | dpdk_common_init(void) | |
2499 | { | |
2500 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2501 | "[netdev] up|down", 1, 2, | |
2502 | netdev_dpdk_set_admin_state, NULL); | |
2503 | ||
033e9df2 DDP |
2504 | } |
2505 | ||
95fb793a | 2506 | /* Client Rings */ |
2507 | ||
95fb793a | 2508 | static int |
2509 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
2510 | unsigned int *eth_port_id) | |
2511 | { | |
2512 | struct dpdk_ring *ivshmem; | |
ca7e7bee | 2513 | char ring_name[RTE_RING_NAMESIZE]; |
95fb793a | 2514 | int err; |
2515 | ||
2516 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
2517 | if (ivshmem == NULL) { | |
2518 | return ENOMEM; | |
2519 | } | |
2520 | ||
7251515e | 2521 | /* XXX: Add support for multiquque ring. */ |
ca7e7bee | 2522 | err = snprintf(ring_name, sizeof(ring_name), "%s_tx", dev_name); |
95fb793a | 2523 | if (err < 0) { |
2524 | return -err; | |
2525 | } | |
2526 | ||
8f0a76c9 | 2527 | /* Create single producer tx ring, netdev does explicit locking. */ |
7251515e | 2528 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2529 | RING_F_SP_ENQ); |
95fb793a | 2530 | if (ivshmem->cring_tx == NULL) { |
2531 | rte_free(ivshmem); | |
2532 | return ENOMEM; | |
2533 | } | |
2534 | ||
ca7e7bee | 2535 | err = snprintf(ring_name, sizeof(ring_name), "%s_rx", dev_name); |
95fb793a | 2536 | if (err < 0) { |
2537 | return -err; | |
2538 | } | |
2539 | ||
8f0a76c9 | 2540 | /* Create single consumer rx ring, netdev does explicit locking. */ |
7251515e | 2541 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2542 | RING_F_SC_DEQ); |
95fb793a | 2543 | if (ivshmem->cring_rx == NULL) { |
2544 | rte_free(ivshmem); | |
2545 | return ENOMEM; | |
2546 | } | |
2547 | ||
d7310583 DDP |
2548 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
2549 | &ivshmem->cring_tx, 1, SOCKET0); | |
2550 | ||
95fb793a | 2551 | if (err < 0) { |
2552 | rte_free(ivshmem); | |
2553 | return ENODEV; | |
2554 | } | |
2555 | ||
2556 | ivshmem->user_port_id = port_no; | |
2557 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
417e7e66 | 2558 | ovs_list_push_back(&dpdk_ring_list, &ivshmem->list_node); |
95fb793a | 2559 | |
2560 | *eth_port_id = ivshmem->eth_port_id; | |
2561 | return 0; | |
2562 | } | |
2563 | ||
2564 | static int | |
64839cf4 WT |
2565 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) |
2566 | OVS_REQUIRES(dpdk_mutex) | |
95fb793a | 2567 | { |
2568 | struct dpdk_ring *ivshmem; | |
2569 | unsigned int port_no; | |
2570 | int err = 0; | |
2571 | ||
2572 | /* Names always start with "dpdkr" */ | |
2573 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2574 | if (err) { | |
2575 | return err; | |
2576 | } | |
2577 | ||
2578 | /* look through our list to find the device */ | |
2579 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { | |
2580 | if (ivshmem->user_port_id == port_no) { | |
58397e6c | 2581 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
95fb793a | 2582 | *eth_port_id = ivshmem->eth_port_id; /* really all that is needed */ |
2583 | return 0; | |
2584 | } | |
2585 | } | |
2586 | /* Need to create the device rings */ | |
2587 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2588 | } | |
2589 | ||
7251515e | 2590 | static int |
d46285a2 | 2591 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
324c8374 IM |
2592 | struct dp_packet_batch *batch, bool may_steal, |
2593 | bool concurrent_txq) | |
7251515e | 2594 | { |
d46285a2 | 2595 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b99bb05 MG |
2596 | unsigned i; |
2597 | ||
2598 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that the | |
2599 | * rss hash field is clear. This is because the same mbuf may be modified by | |
2600 | * the consumer of the ring and return into the datapath without recalculating | |
2601 | * the RSS hash. */ | |
64839cf4 WT |
2602 | for (i = 0; i < batch->count; i++) { |
2603 | dp_packet_rss_invalidate(batch->packets[i]); | |
1b99bb05 | 2604 | } |
7251515e | 2605 | |
324c8374 | 2606 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e DV |
2607 | return 0; |
2608 | } | |
2609 | ||
95fb793a | 2610 | static int |
2611 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2612 | { | |
2613 | unsigned int port_no = 0; | |
2614 | int err = 0; | |
2615 | ||
2616 | if (rte_eal_init_ret) { | |
2617 | return rte_eal_init_ret; | |
2618 | } | |
2619 | ||
2620 | ovs_mutex_lock(&dpdk_mutex); | |
2621 | ||
2622 | err = dpdk_ring_open(netdev->name, &port_no); | |
2623 | if (err) { | |
2624 | goto unlock_dpdk; | |
2625 | } | |
2626 | ||
58397e6c | 2627 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
95fb793a | 2628 | |
2629 | unlock_dpdk: | |
2630 | ovs_mutex_unlock(&dpdk_mutex); | |
2631 | return err; | |
2632 | } | |
2633 | ||
0bf765f7 IS |
2634 | /* QoS Functions */ |
2635 | ||
2636 | /* | |
2637 | * Initialize QoS configuration operations. | |
2638 | */ | |
2639 | static void | |
2640 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
2641 | { | |
2642 | conf->ops = ops; | |
2643 | } | |
2644 | ||
2645 | /* | |
2646 | * Search existing QoS operations in qos_ops and compare each set of | |
2647 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
2648 | * else return NULL | |
2649 | */ | |
2650 | static const struct dpdk_qos_ops * | |
2651 | qos_lookup_name(const char *name) | |
2652 | { | |
2653 | const struct dpdk_qos_ops *const *opsp; | |
2654 | ||
2655 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2656 | const struct dpdk_qos_ops *ops = *opsp; | |
2657 | if (!strcmp(name, ops->qos_name)) { | |
2658 | return ops; | |
2659 | } | |
2660 | } | |
2661 | return NULL; | |
2662 | } | |
2663 | ||
2664 | /* | |
2665 | * Call qos_destruct to clean up items associated with the netdevs | |
2666 | * qos_conf. Set netdevs qos_conf to NULL. | |
2667 | */ | |
2668 | static void | |
d46285a2 | 2669 | qos_delete_conf(struct netdev *netdev) |
0bf765f7 | 2670 | { |
d46285a2 | 2671 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 2672 | |
d46285a2 DDP |
2673 | rte_spinlock_lock(&dev->qos_lock); |
2674 | if (dev->qos_conf) { | |
2675 | if (dev->qos_conf->ops->qos_destruct) { | |
2676 | dev->qos_conf->ops->qos_destruct(netdev, dev->qos_conf); | |
0bf765f7 | 2677 | } |
d46285a2 | 2678 | dev->qos_conf = NULL; |
0bf765f7 | 2679 | } |
d46285a2 | 2680 | rte_spinlock_unlock(&dev->qos_lock); |
0bf765f7 IS |
2681 | } |
2682 | ||
2683 | static int | |
2684 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
2685 | struct sset *types) | |
2686 | { | |
2687 | const struct dpdk_qos_ops *const *opsp; | |
2688 | ||
2689 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2690 | const struct dpdk_qos_ops *ops = *opsp; | |
2691 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
2692 | sset_add(types, ops->qos_name); | |
2693 | } | |
2694 | } | |
2695 | return 0; | |
2696 | } | |
2697 | ||
2698 | static int | |
d46285a2 | 2699 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
2700 | const char **typep, struct smap *details) |
2701 | { | |
d46285a2 | 2702 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 IS |
2703 | int error = 0; |
2704 | ||
d46285a2 DDP |
2705 | ovs_mutex_lock(&dev->mutex); |
2706 | if(dev->qos_conf) { | |
2707 | *typep = dev->qos_conf->ops->qos_name; | |
2708 | error = (dev->qos_conf->ops->qos_get | |
2709 | ? dev->qos_conf->ops->qos_get(netdev, details): 0); | |
d03603c4 MC |
2710 | } else { |
2711 | /* No QoS configuration set, return an empty string */ | |
2712 | *typep = ""; | |
0bf765f7 | 2713 | } |
d46285a2 | 2714 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2715 | |
2716 | return error; | |
2717 | } | |
2718 | ||
2719 | static int | |
d46285a2 | 2720 | netdev_dpdk_set_qos(struct netdev *netdev, |
0bf765f7 IS |
2721 | const char *type, const struct smap *details) |
2722 | { | |
d46285a2 | 2723 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 IS |
2724 | const struct dpdk_qos_ops *new_ops = NULL; |
2725 | int error = 0; | |
2726 | ||
2727 | /* If type is empty or unsupported then the current QoS configuration | |
2728 | * for the dpdk-netdev can be destroyed */ | |
2729 | new_ops = qos_lookup_name(type); | |
2730 | ||
2731 | if (type[0] == '\0' || !new_ops || !new_ops->qos_construct) { | |
d46285a2 | 2732 | qos_delete_conf(netdev); |
0bf765f7 IS |
2733 | return EOPNOTSUPP; |
2734 | } | |
2735 | ||
d46285a2 | 2736 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 2737 | |
d46285a2 DDP |
2738 | if (dev->qos_conf) { |
2739 | if (new_ops == dev->qos_conf->ops) { | |
2740 | error = new_ops->qos_set ? new_ops->qos_set(netdev, details) : 0; | |
0bf765f7 IS |
2741 | } else { |
2742 | /* Delete existing QoS configuration. */ | |
d46285a2 DDP |
2743 | qos_delete_conf(netdev); |
2744 | ovs_assert(dev->qos_conf == NULL); | |
0bf765f7 IS |
2745 | |
2746 | /* Install new QoS configuration. */ | |
d46285a2 DDP |
2747 | error = new_ops->qos_construct(netdev, details); |
2748 | ovs_assert((error == 0) == (dev->qos_conf != NULL)); | |
0bf765f7 IS |
2749 | } |
2750 | } else { | |
d46285a2 DDP |
2751 | error = new_ops->qos_construct(netdev, details); |
2752 | ovs_assert((error == 0) == (dev->qos_conf != NULL)); | |
0bf765f7 IS |
2753 | } |
2754 | ||
d46285a2 | 2755 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2756 | return error; |
2757 | } | |
2758 | ||
2759 | /* egress-policer details */ | |
2760 | ||
2761 | struct egress_policer { | |
2762 | struct qos_conf qos_conf; | |
2763 | struct rte_meter_srtcm_params app_srtcm_params; | |
2764 | struct rte_meter_srtcm egress_meter; | |
2765 | }; | |
2766 | ||
2767 | static struct egress_policer * | |
d46285a2 | 2768 | egress_policer_get__(const struct netdev *netdev) |
0bf765f7 | 2769 | { |
d46285a2 DDP |
2770 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2771 | return CONTAINER_OF(dev->qos_conf, struct egress_policer, qos_conf); | |
0bf765f7 IS |
2772 | } |
2773 | ||
2774 | static int | |
d46285a2 DDP |
2775 | egress_policer_qos_construct(struct netdev *netdev, |
2776 | const struct smap *details) | |
0bf765f7 | 2777 | { |
d46285a2 | 2778 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 2779 | struct egress_policer *policer; |
0bf765f7 IS |
2780 | int err = 0; |
2781 | ||
d46285a2 | 2782 | rte_spinlock_lock(&dev->qos_lock); |
0bf765f7 IS |
2783 | policer = xmalloc(sizeof *policer); |
2784 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
d46285a2 | 2785 | dev->qos_conf = &policer->qos_conf; |
13c1637f BP |
2786 | policer->app_srtcm_params.cir = smap_get_ullong(details, "cir", 0); |
2787 | policer->app_srtcm_params.cbs = smap_get_ullong(details, "cbs", 0); | |
0bf765f7 IS |
2788 | policer->app_srtcm_params.ebs = 0; |
2789 | err = rte_meter_srtcm_config(&policer->egress_meter, | |
2790 | &policer->app_srtcm_params); | |
d46285a2 | 2791 | rte_spinlock_unlock(&dev->qos_lock); |
0bf765f7 IS |
2792 | |
2793 | return err; | |
2794 | } | |
2795 | ||
2796 | static void | |
d46285a2 | 2797 | egress_policer_qos_destruct(struct netdev *netdev OVS_UNUSED, |
0bf765f7 IS |
2798 | struct qos_conf *conf) |
2799 | { | |
2800 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
2801 | qos_conf); | |
2802 | free(policer); | |
2803 | } | |
2804 | ||
2805 | static int | |
2806 | egress_policer_qos_get(const struct netdev *netdev, struct smap *details) | |
2807 | { | |
2808 | struct egress_policer *policer = egress_policer_get__(netdev); | |
2809 | smap_add_format(details, "cir", "%llu", | |
2810 | 1ULL * policer->app_srtcm_params.cir); | |
2811 | smap_add_format(details, "cbs", "%llu", | |
2812 | 1ULL * policer->app_srtcm_params.cbs); | |
050c60bf | 2813 | |
0bf765f7 IS |
2814 | return 0; |
2815 | } | |
2816 | ||
2817 | static int | |
d46285a2 | 2818 | egress_policer_qos_set(struct netdev *netdev, const struct smap *details) |
0bf765f7 IS |
2819 | { |
2820 | struct egress_policer *policer; | |
0bf765f7 IS |
2821 | int err = 0; |
2822 | ||
d46285a2 | 2823 | policer = egress_policer_get__(netdev); |
13c1637f BP |
2824 | policer->app_srtcm_params.cir = smap_get_ullong(details, "cir", 0); |
2825 | policer->app_srtcm_params.cbs = smap_get_ullong(details, "cbs", 0); | |
0bf765f7 IS |
2826 | policer->app_srtcm_params.ebs = 0; |
2827 | err = rte_meter_srtcm_config(&policer->egress_meter, | |
2828 | &policer->app_srtcm_params); | |
2829 | ||
2830 | return err; | |
2831 | } | |
2832 | ||
0bf765f7 | 2833 | static int |
f3926f29 | 2834 | egress_policer_run(struct netdev *netdev, struct rte_mbuf **pkts, int pkt_cnt) |
0bf765f7 | 2835 | { |
0bf765f7 | 2836 | int cnt = 0; |
d46285a2 | 2837 | struct egress_policer *policer = egress_policer_get__(netdev); |
0bf765f7 | 2838 | |
f3926f29 | 2839 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, pkt_cnt); |
0bf765f7 IS |
2840 | |
2841 | return cnt; | |
2842 | } | |
2843 | ||
2844 | static const struct dpdk_qos_ops egress_policer_ops = { | |
2845 | "egress-policer", /* qos_name */ | |
2846 | egress_policer_qos_construct, | |
2847 | egress_policer_qos_destruct, | |
2848 | egress_policer_qos_get, | |
2849 | egress_policer_qos_set, | |
2850 | egress_policer_run | |
2851 | }; | |
2852 | ||
050c60bf DDP |
2853 | static int |
2854 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
2855 | { | |
2856 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2857 | int err = 0; | |
2858 | ||
2859 | ovs_mutex_lock(&dpdk_mutex); | |
2860 | ovs_mutex_lock(&dev->mutex); | |
2861 | ||
2862 | if (netdev->n_txq == dev->requested_n_txq | |
2863 | && netdev->n_rxq == dev->requested_n_rxq) { | |
2864 | /* Reconfiguration is unnecessary */ | |
2865 | ||
2866 | goto out; | |
2867 | } | |
2868 | ||
2869 | rte_eth_dev_stop(dev->port_id); | |
2870 | ||
2871 | netdev->n_txq = dev->requested_n_txq; | |
2872 | netdev->n_rxq = dev->requested_n_rxq; | |
2873 | ||
2874 | rte_free(dev->tx_q); | |
2875 | err = dpdk_eth_dev_init(dev); | |
81acebda | 2876 | netdev_dpdk_alloc_txq(dev, netdev->n_txq); |
050c60bf | 2877 | |
050c60bf DDP |
2878 | out: |
2879 | ||
2880 | ovs_mutex_unlock(&dev->mutex); | |
2881 | ovs_mutex_unlock(&dpdk_mutex); | |
2882 | ||
2883 | return err; | |
2884 | } | |
2885 | ||
2886 | static int | |
2887 | netdev_dpdk_vhost_user_reconfigure(struct netdev *netdev) | |
2888 | { | |
2889 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
db8f13b0 | 2890 | int err = 0; |
050c60bf DDP |
2891 | |
2892 | ovs_mutex_lock(&dpdk_mutex); | |
2893 | ovs_mutex_lock(&dev->mutex); | |
2894 | ||
2895 | netdev->n_txq = dev->requested_n_txq; | |
2896 | netdev->n_rxq = dev->requested_n_rxq; | |
2897 | ||
81acebda IM |
2898 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
2899 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
2900 | dev->tx_q[0].map = 0; | |
2901 | } | |
2902 | ||
2903 | netdev_dpdk_remap_txqs(dev); | |
2904 | ||
db8f13b0 CL |
2905 | if (dev->requested_socket_id != dev->socket_id) { |
2906 | dev->socket_id = dev->requested_socket_id; | |
2907 | /* Change mempool to new NUMA Node */ | |
2908 | dpdk_mp_put(dev->dpdk_mp); | |
2909 | dev->dpdk_mp = dpdk_mp_get(dev->socket_id, dev->mtu); | |
2910 | if (!dev->dpdk_mp) { | |
2911 | err = ENOMEM; | |
2912 | } | |
2913 | } | |
2914 | ||
0a0f39df CL |
2915 | if (netdev_dpdk_get_vid(dev) >= 0) { |
2916 | dev->vhost_reconfigured = true; | |
81acebda IM |
2917 | } |
2918 | ||
050c60bf DDP |
2919 | ovs_mutex_unlock(&dev->mutex); |
2920 | ovs_mutex_unlock(&dpdk_mutex); | |
2921 | ||
db8f13b0 | 2922 | return err; |
050c60bf DDP |
2923 | } |
2924 | ||
2925 | static int | |
2926 | netdev_dpdk_vhost_cuse_reconfigure(struct netdev *netdev) | |
2927 | { | |
2928 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2929 | ||
2930 | ovs_mutex_lock(&dpdk_mutex); | |
2931 | ovs_mutex_lock(&dev->mutex); | |
2932 | ||
2933 | netdev->n_txq = dev->requested_n_txq; | |
050c60bf | 2934 | netdev->n_rxq = 1; |
050c60bf DDP |
2935 | |
2936 | ovs_mutex_unlock(&dev->mutex); | |
2937 | ovs_mutex_unlock(&dpdk_mutex); | |
2938 | ||
2939 | return 0; | |
2940 | } | |
2941 | ||
81acebda IM |
2942 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, \ |
2943 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ | |
2944 | GET_CARRIER, GET_STATS, \ | |
2945 | GET_FEATURES, GET_STATUS, \ | |
2946 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 2947 | { \ |
2948 | NAME, \ | |
118c77b1 | 2949 | true, /* is_pmd */ \ |
95fb793a | 2950 | INIT, /* init */ \ |
2951 | NULL, /* netdev_dpdk_run */ \ | |
2952 | NULL, /* netdev_dpdk_wait */ \ | |
2953 | \ | |
2954 | netdev_dpdk_alloc, \ | |
2955 | CONSTRUCT, \ | |
58397e6c | 2956 | DESTRUCT, \ |
95fb793a | 2957 | netdev_dpdk_dealloc, \ |
2958 | netdev_dpdk_get_config, \ | |
81acebda | 2959 | SET_CONFIG, \ |
95fb793a | 2960 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
2961 | NULL, /* build header */ \ |
2962 | NULL, /* push header */ \ | |
2963 | NULL, /* pop header */ \ | |
7dec44fe | 2964 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 2965 | SET_TX_MULTIQ, \ |
95fb793a | 2966 | \ |
7251515e | 2967 | SEND, /* send */ \ |
95fb793a | 2968 | NULL, /* send_wait */ \ |
2969 | \ | |
2970 | netdev_dpdk_set_etheraddr, \ | |
2971 | netdev_dpdk_get_etheraddr, \ | |
2972 | netdev_dpdk_get_mtu, \ | |
2973 | netdev_dpdk_set_mtu, \ | |
2974 | netdev_dpdk_get_ifindex, \ | |
58397e6c | 2975 | GET_CARRIER, \ |
95fb793a | 2976 | netdev_dpdk_get_carrier_resets, \ |
2977 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
2978 | GET_STATS, \ |
2979 | GET_FEATURES, \ | |
95fb793a | 2980 | NULL, /* set_advertisements */ \ |
2981 | \ | |
9509913a | 2982 | netdev_dpdk_set_policing, \ |
0bf765f7 | 2983 | netdev_dpdk_get_qos_types, \ |
95fb793a | 2984 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
2985 | netdev_dpdk_get_qos, \ |
2986 | netdev_dpdk_set_qos, \ | |
95fb793a | 2987 | NULL, /* get_queue */ \ |
2988 | NULL, /* set_queue */ \ | |
2989 | NULL, /* delete_queue */ \ | |
2990 | NULL, /* get_queue_stats */ \ | |
2991 | NULL, /* queue_dump_start */ \ | |
2992 | NULL, /* queue_dump_next */ \ | |
2993 | NULL, /* queue_dump_done */ \ | |
2994 | NULL, /* dump_queue_stats */ \ | |
2995 | \ | |
95fb793a | 2996 | NULL, /* set_in4 */ \ |
a8704b50 | 2997 | NULL, /* get_addr_list */ \ |
95fb793a | 2998 | NULL, /* add_router */ \ |
2999 | NULL, /* get_next_hop */ \ | |
58397e6c | 3000 | GET_STATUS, \ |
95fb793a | 3001 | NULL, /* arp_lookup */ \ |
3002 | \ | |
3003 | netdev_dpdk_update_flags, \ | |
050c60bf | 3004 | RECONFIGURE, \ |
95fb793a | 3005 | \ |
3006 | netdev_dpdk_rxq_alloc, \ | |
3007 | netdev_dpdk_rxq_construct, \ | |
3008 | netdev_dpdk_rxq_destruct, \ | |
3009 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3010 | RXQ_RECV, \ |
95fb793a | 3011 | NULL, /* rx_wait */ \ |
3012 | NULL, /* rxq_drain */ \ | |
3013 | } | |
8a9562d2 | 3014 | |
7d1ced01 CL |
3015 | static int |
3016 | process_vhost_flags(char *flag, char *default_val, int size, | |
bab69409 AC |
3017 | const struct smap *ovs_other_config, |
3018 | char **new_val) | |
7d1ced01 | 3019 | { |
bab69409 | 3020 | const char *val; |
7d1ced01 CL |
3021 | int changed = 0; |
3022 | ||
bab69409 AC |
3023 | val = smap_get(ovs_other_config, flag); |
3024 | ||
7d1ced01 | 3025 | /* Depending on which version of vhost is in use, process the vhost-specific |
bab69409 | 3026 | * flag if it is provided, otherwise resort to default value. |
7d1ced01 | 3027 | */ |
bab69409 | 3028 | if (val && (strlen(val) <= size)) { |
7d1ced01 | 3029 | changed = 1; |
bab69409 | 3030 | *new_val = xstrdup(val); |
7d1ced01 CL |
3031 | VLOG_INFO("User-provided %s in use: %s", flag, *new_val); |
3032 | } else { | |
3033 | VLOG_INFO("No %s provided - defaulting to %s", flag, default_val); | |
3034 | *new_val = default_val; | |
3035 | } | |
3036 | ||
3037 | return changed; | |
3038 | } | |
3039 | ||
bab69409 AC |
3040 | static char ** |
3041 | grow_argv(char ***argv, size_t cur_siz, size_t grow_by) | |
8a9562d2 | 3042 | { |
bab69409 AC |
3043 | return xrealloc(*argv, sizeof(char *) * (cur_siz + grow_by)); |
3044 | } | |
8a9562d2 | 3045 | |
bab69409 AC |
3046 | static void |
3047 | dpdk_option_extend(char ***argv, int argc, const char *option, | |
3048 | const char *value) | |
3049 | { | |
3050 | char **newargv = grow_argv(argv, argc, 2); | |
3051 | *argv = newargv; | |
3052 | newargv[argc] = xstrdup(option); | |
3053 | newargv[argc+1] = xstrdup(value); | |
3054 | } | |
8a9562d2 | 3055 | |
52a57e36 AC |
3056 | static char ** |
3057 | move_argv(char ***argv, size_t cur_size, char **src_argv, size_t src_argc) | |
3058 | { | |
3059 | char **newargv = grow_argv(argv, cur_size, src_argc); | |
3060 | while (src_argc--) { | |
3061 | newargv[cur_size+src_argc] = src_argv[src_argc]; | |
3062 | src_argv[src_argc] = NULL; | |
3063 | } | |
3064 | return newargv; | |
3065 | } | |
3066 | ||
eac84432 AC |
3067 | static int |
3068 | extra_dpdk_args(const char *ovs_extra_config, char ***argv, int argc) | |
3069 | { | |
3070 | int ret = argc; | |
3071 | char *release_tok = xstrdup(ovs_extra_config); | |
3072 | char *tok = release_tok, *endptr = NULL; | |
3073 | ||
3074 | for (tok = strtok_r(release_tok, " ", &endptr); tok != NULL; | |
3075 | tok = strtok_r(NULL, " ", &endptr)) { | |
3076 | char **newarg = grow_argv(argv, ret, 1); | |
3077 | *argv = newarg; | |
3078 | newarg[ret++] = xstrdup(tok); | |
3079 | } | |
3080 | free(release_tok); | |
3081 | return ret; | |
3082 | } | |
3083 | ||
52a57e36 AC |
3084 | static bool |
3085 | argv_contains(char **argv_haystack, const size_t argc_haystack, | |
3086 | const char *needle) | |
3087 | { | |
3088 | for (size_t i = 0; i < argc_haystack; ++i) { | |
3089 | if (!strcmp(argv_haystack[i], needle)) | |
3090 | return true; | |
3091 | } | |
3092 | return false; | |
3093 | } | |
3094 | ||
bab69409 AC |
3095 | static int |
3096 | construct_dpdk_options(const struct smap *ovs_other_config, | |
52a57e36 AC |
3097 | char ***argv, const int initial_size, |
3098 | char **extra_args, const size_t extra_argc) | |
bab69409 AC |
3099 | { |
3100 | struct dpdk_options_map { | |
3101 | const char *ovs_configuration; | |
3102 | const char *dpdk_option; | |
3103 | bool default_enabled; | |
3104 | const char *default_value; | |
3105 | } opts[] = { | |
3106 | {"dpdk-lcore-mask", "-c", false, NULL}, | |
3107 | {"dpdk-hugepage-dir", "--huge-dir", false, NULL}, | |
3108 | }; | |
3109 | ||
3110 | int i, ret = initial_size; | |
3111 | ||
3112 | /*First, construct from the flat-options (non-mutex)*/ | |
3113 | for (i = 0; i < ARRAY_SIZE(opts); ++i) { | |
3114 | const char *lookup = smap_get(ovs_other_config, | |
3115 | opts[i].ovs_configuration); | |
3116 | if (!lookup && opts[i].default_enabled) { | |
3117 | lookup = opts[i].default_value; | |
3118 | } | |
8a9562d2 | 3119 | |
bab69409 | 3120 | if (lookup) { |
52a57e36 AC |
3121 | if (!argv_contains(extra_args, extra_argc, opts[i].dpdk_option)) { |
3122 | dpdk_option_extend(argv, ret, opts[i].dpdk_option, lookup); | |
3123 | ret += 2; | |
3124 | } else { | |
3125 | VLOG_WARN("Ignoring database defined option '%s' due to " | |
3126 | "dpdk_extras config", opts[i].dpdk_option); | |
3127 | } | |
bab69409 AC |
3128 | } |
3129 | } | |
3130 | ||
3131 | return ret; | |
3132 | } | |
3133 | ||
3134 | #define MAX_DPDK_EXCL_OPTS 10 | |
3135 | ||
3136 | static int | |
3137 | construct_dpdk_mutex_options(const struct smap *ovs_other_config, | |
52a57e36 AC |
3138 | char ***argv, const int initial_size, |
3139 | char **extra_args, const size_t extra_argc) | |
bab69409 AC |
3140 | { |
3141 | struct dpdk_exclusive_options_map { | |
3142 | const char *category; | |
3143 | const char *ovs_dpdk_options[MAX_DPDK_EXCL_OPTS]; | |
3144 | const char *eal_dpdk_options[MAX_DPDK_EXCL_OPTS]; | |
3145 | const char *default_value; | |
3146 | int default_option; | |
3147 | } excl_opts[] = { | |
3148 | {"memory type", | |
3149 | {"dpdk-alloc-mem", "dpdk-socket-mem", NULL,}, | |
3150 | {"-m", "--socket-mem", NULL,}, | |
3151 | "1024,0", 1 | |
3152 | }, | |
3153 | }; | |
3154 | ||
3155 | int i, ret = initial_size; | |
3156 | for (i = 0; i < ARRAY_SIZE(excl_opts); ++i) { | |
3157 | int found_opts = 0, scan, found_pos = -1; | |
3158 | const char *found_value; | |
3159 | struct dpdk_exclusive_options_map *popt = &excl_opts[i]; | |
3160 | ||
3161 | for (scan = 0; scan < MAX_DPDK_EXCL_OPTS | |
3162 | && popt->ovs_dpdk_options[scan]; ++scan) { | |
3163 | const char *lookup = smap_get(ovs_other_config, | |
3164 | popt->ovs_dpdk_options[scan]); | |
3165 | if (lookup && strlen(lookup)) { | |
3166 | found_opts++; | |
3167 | found_pos = scan; | |
3168 | found_value = lookup; | |
3169 | } | |
3170 | } | |
3171 | ||
3172 | if (!found_opts) { | |
3173 | if (popt->default_option) { | |
3174 | found_pos = popt->default_option; | |
3175 | found_value = popt->default_value; | |
3176 | } else { | |
3177 | continue; | |
3178 | } | |
dfc89a55 | 3179 | } |
bab69409 AC |
3180 | |
3181 | if (found_opts > 1) { | |
3182 | VLOG_ERR("Multiple defined options for %s. Please check your" | |
3183 | " database settings and reconfigure if necessary.", | |
3184 | popt->category); | |
3185 | } | |
3186 | ||
52a57e36 AC |
3187 | if (!argv_contains(extra_args, extra_argc, |
3188 | popt->eal_dpdk_options[found_pos])) { | |
3189 | dpdk_option_extend(argv, ret, popt->eal_dpdk_options[found_pos], | |
3190 | found_value); | |
3191 | ret += 2; | |
3192 | } else { | |
3193 | VLOG_WARN("Ignoring database defined option '%s' due to " | |
3194 | "dpdk_extras config", popt->eal_dpdk_options[found_pos]); | |
3195 | } | |
bab69409 AC |
3196 | } |
3197 | ||
3198 | return ret; | |
3199 | } | |
3200 | ||
3201 | static int | |
88964e64 AC |
3202 | get_dpdk_args(const struct smap *ovs_other_config, char ***argv, |
3203 | int argc) | |
bab69409 | 3204 | { |
eac84432 | 3205 | const char *extra_configuration; |
52a57e36 AC |
3206 | char **extra_args = NULL; |
3207 | int i; | |
3208 | size_t extra_argc = 0; | |
eac84432 AC |
3209 | |
3210 | extra_configuration = smap_get(ovs_other_config, "dpdk-extra"); | |
3211 | if (extra_configuration) { | |
52a57e36 | 3212 | extra_argc = extra_dpdk_args(extra_configuration, &extra_args, 0); |
eac84432 | 3213 | } |
52a57e36 AC |
3214 | |
3215 | i = construct_dpdk_options(ovs_other_config, argv, argc, extra_args, | |
3216 | extra_argc); | |
3217 | i = construct_dpdk_mutex_options(ovs_other_config, argv, i, extra_args, | |
3218 | extra_argc); | |
3219 | ||
3220 | if (extra_configuration) { | |
3221 | *argv = move_argv(argv, i, extra_args, extra_argc); | |
3222 | } | |
3223 | ||
3224 | return i + extra_argc; | |
bab69409 AC |
3225 | } |
3226 | ||
3227 | static char **dpdk_argv; | |
3228 | static int dpdk_argc; | |
3229 | ||
3230 | static void | |
3231 | deferred_argv_release(void) | |
3232 | { | |
3233 | int result; | |
3234 | for (result = 0; result < dpdk_argc; ++result) { | |
3235 | free(dpdk_argv[result]); | |
3236 | } | |
3237 | ||
3238 | free(dpdk_argv); | |
3239 | } | |
3240 | ||
3241 | static void | |
3242 | dpdk_init__(const struct smap *ovs_other_config) | |
3243 | { | |
3244 | char **argv = NULL; | |
3245 | int result; | |
88964e64 AC |
3246 | int argc, argc_tmp; |
3247 | bool auto_determine = true; | |
d8e2f4cc | 3248 | int err = 0; |
bab69409 | 3249 | cpu_set_t cpuset; |
d8a8f353 AC |
3250 | #ifndef VHOST_CUSE |
3251 | char *sock_dir_subcomponent; | |
3252 | #endif | |
bab69409 AC |
3253 | |
3254 | if (!smap_get_bool(ovs_other_config, "dpdk-init", false)) { | |
3255 | VLOG_INFO("DPDK Disabled - to change this requires a restart.\n"); | |
3256 | return; | |
dfc89a55 AZ |
3257 | } |
3258 | ||
bab69409 AC |
3259 | VLOG_INFO("DPDK Enabled, initializing"); |
3260 | ||
7d1ced01 | 3261 | #ifdef VHOST_CUSE |
bab69409 AC |
3262 | if (process_vhost_flags("cuse-dev-name", xstrdup("vhost-net"), |
3263 | PATH_MAX, ovs_other_config, &cuse_dev_name)) { | |
7d1ced01 | 3264 | #else |
6eb51f9a | 3265 | if (process_vhost_flags("vhost-sock-dir", xstrdup(ovs_rundir()), |
d8a8f353 AC |
3266 | NAME_MAX, ovs_other_config, |
3267 | &sock_dir_subcomponent)) { | |
7d1ced01 | 3268 | struct stat s; |
d8a8f353 AC |
3269 | if (!strstr(sock_dir_subcomponent, "..")) { |
3270 | vhost_sock_dir = xasprintf("%s/%s", ovs_rundir(), | |
3271 | sock_dir_subcomponent); | |
3272 | ||
3273 | err = stat(vhost_sock_dir, &s); | |
3274 | if (err) { | |
3275 | VLOG_ERR("vhost-user sock directory '%s' does not exist.", | |
3276 | vhost_sock_dir); | |
3277 | } | |
3278 | } else { | |
3279 | vhost_sock_dir = xstrdup(ovs_rundir()); | |
3280 | VLOG_ERR("vhost-user sock directory request '%s/%s' has invalid" | |
3281 | "characters '..' - using %s instead.", | |
3282 | ovs_rundir(), sock_dir_subcomponent, ovs_rundir()); | |
7d1ced01 | 3283 | } |
d8a8f353 AC |
3284 | free(sock_dir_subcomponent); |
3285 | } else { | |
6eb51f9a | 3286 | vhost_sock_dir = sock_dir_subcomponent; |
7d1ced01 | 3287 | #endif |
58397e6c KT |
3288 | } |
3289 | ||
bab69409 | 3290 | argv = grow_argv(&argv, 0, 1); |
88964e64 | 3291 | argc = 1; |
bab69409 | 3292 | argv[0] = xstrdup(ovs_get_program_name()); |
88964e64 AC |
3293 | argc_tmp = get_dpdk_args(ovs_other_config, &argv, argc); |
3294 | ||
3295 | while (argc_tmp != argc) { | |
3296 | if (!strcmp("-c", argv[argc]) || !strcmp("-l", argv[argc])) { | |
3297 | auto_determine = false; | |
3298 | break; | |
3299 | } | |
3300 | argc++; | |
3301 | } | |
3302 | argc = argc_tmp; | |
3303 | ||
3304 | /** | |
3305 | * NOTE: This is an unsophisticated mechanism for determining the DPDK | |
3306 | * lcore for the DPDK Master. | |
3307 | */ | |
3308 | if (auto_determine) { | |
3309 | int i; | |
d8e2f4cc KT |
3310 | /* Get the main thread affinity */ |
3311 | CPU_ZERO(&cpuset); | |
3312 | err = pthread_getaffinity_np(pthread_self(), sizeof(cpu_set_t), | |
3313 | &cpuset); | |
3314 | if (!err) { | |
3315 | for (i = 0; i < CPU_SETSIZE; i++) { | |
3316 | if (CPU_ISSET(i, &cpuset)) { | |
3317 | argv = grow_argv(&argv, argc, 2); | |
3318 | argv[argc++] = xstrdup("-c"); | |
3319 | argv[argc++] = xasprintf("0x%08llX", (1ULL<<i)); | |
3320 | i = CPU_SETSIZE; | |
3321 | } | |
88964e64 | 3322 | } |
d8e2f4cc KT |
3323 | } else { |
3324 | VLOG_ERR("Thread getaffinity error %d. Using core 0x1", err); | |
3325 | /* User did not set dpdk-lcore-mask and unable to get current | |
3326 | * thread affintity - default to core 0x1 */ | |
3327 | argv = grow_argv(&argv, argc, 2); | |
3328 | argv[argc++] = xstrdup("-c"); | |
3329 | argv[argc++] = xasprintf("0x%X", 1); | |
88964e64 AC |
3330 | } |
3331 | } | |
bab69409 AC |
3332 | |
3333 | argv = grow_argv(&argv, argc, 1); | |
3334 | argv[argc] = NULL; | |
3335 | ||
3336 | optind = 1; | |
58397e6c | 3337 | |
eac84432 AC |
3338 | if (VLOG_IS_INFO_ENABLED()) { |
3339 | struct ds eal_args; | |
3340 | int opt; | |
3341 | ds_init(&eal_args); | |
3342 | ds_put_cstr(&eal_args, "EAL ARGS:"); | |
3343 | for (opt = 0; opt < argc; ++opt) { | |
3344 | ds_put_cstr(&eal_args, " "); | |
3345 | ds_put_cstr(&eal_args, argv[opt]); | |
3346 | } | |
3347 | VLOG_INFO("%s", ds_cstr_ro(&eal_args)); | |
3348 | ds_destroy(&eal_args); | |
3349 | } | |
3350 | ||
8a9562d2 PS |
3351 | /* Make sure things are initialized ... */ |
3352 | result = rte_eal_init(argc, argv); | |
451450fa | 3353 | if (result < 0) { |
58397e6c | 3354 | ovs_abort(result, "Cannot init EAL"); |
451450fa | 3355 | } |
8a9562d2 | 3356 | |
563c98d8 | 3357 | /* Set the main thread affinity back to pre rte_eal_init() value */ |
d8e2f4cc | 3358 | if (auto_determine && !err) { |
bab69409 AC |
3359 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), |
3360 | &cpuset); | |
3361 | if (err) { | |
3362 | VLOG_ERR("Thread setaffinity error %d", err); | |
3363 | } | |
563c98d8 AC |
3364 | } |
3365 | ||
bab69409 AC |
3366 | dpdk_argv = argv; |
3367 | dpdk_argc = argc; | |
3368 | ||
3369 | atexit(deferred_argv_release); | |
3370 | ||
d7310583 | 3371 | rte_memzone_dump(stdout); |
8a9562d2 PS |
3372 | rte_eal_init_ret = 0; |
3373 | ||
db73f716 | 3374 | /* We are called from the main thread here */ |
d5c199ea | 3375 | RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID; |
db73f716 | 3376 | |
bab69409 AC |
3377 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); |
3378 | ||
3379 | #ifdef VHOST_CUSE | |
3380 | /* Register CUSE device to handle IOCTLs. | |
3381 | * Unless otherwise specified, cuse_dev_name is set to vhost-net. | |
3382 | */ | |
0a0f39df | 3383 | err = rte_vhost_driver_register(cuse_dev_name, 0); |
bab69409 AC |
3384 | |
3385 | if (err != 0) { | |
3386 | VLOG_ERR("CUSE device setup failure."); | |
3387 | return; | |
3388 | } | |
3389 | #endif | |
3390 | ||
3391 | dpdk_vhost_class_init(); | |
3392 | ||
3393 | /* Finally, register the dpdk classes */ | |
3394 | netdev_dpdk_register(); | |
3395 | } | |
3396 | ||
3397 | void | |
3398 | dpdk_init(const struct smap *ovs_other_config) | |
3399 | { | |
3400 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
3401 | ||
3402 | if (ovs_other_config && ovsthread_once_start(&once)) { | |
3403 | dpdk_init__(ovs_other_config); | |
3404 | ovsthread_once_done(&once); | |
3405 | } | |
8a9562d2 PS |
3406 | } |
3407 | ||
bce01e3a | 3408 | static const struct netdev_class dpdk_class = |
95fb793a | 3409 | NETDEV_DPDK_CLASS( |
3410 | "dpdk", | |
b8e57534 | 3411 | NULL, |
5496878c | 3412 | netdev_dpdk_construct, |
58397e6c | 3413 | netdev_dpdk_destruct, |
81acebda IM |
3414 | netdev_dpdk_set_config, |
3415 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3416 | netdev_dpdk_eth_send, |
3417 | netdev_dpdk_get_carrier, | |
3418 | netdev_dpdk_get_stats, | |
3419 | netdev_dpdk_get_features, | |
3420 | netdev_dpdk_get_status, | |
050c60bf | 3421 | netdev_dpdk_reconfigure, |
58397e6c | 3422 | netdev_dpdk_rxq_recv); |
95fb793a | 3423 | |
bce01e3a | 3424 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3425 | NETDEV_DPDK_CLASS( |
3426 | "dpdkr", | |
033e9df2 | 3427 | NULL, |
5496878c | 3428 | netdev_dpdk_ring_construct, |
58397e6c | 3429 | netdev_dpdk_destruct, |
81acebda IM |
3430 | netdev_dpdk_set_config, |
3431 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3432 | netdev_dpdk_ring_send, |
3433 | netdev_dpdk_get_carrier, | |
3434 | netdev_dpdk_get_stats, | |
3435 | netdev_dpdk_get_features, | |
3436 | netdev_dpdk_get_status, | |
050c60bf | 3437 | netdev_dpdk_reconfigure, |
58397e6c KT |
3438 | netdev_dpdk_rxq_recv); |
3439 | ||
bd8baf47 | 3440 | static const struct netdev_class OVS_UNUSED dpdk_vhost_cuse_class = |
58397e6c | 3441 | NETDEV_DPDK_CLASS( |
7d1ced01 CL |
3442 | "dpdkvhostcuse", |
3443 | dpdk_vhost_cuse_class_init, | |
3444 | netdev_dpdk_vhost_cuse_construct, | |
3445 | netdev_dpdk_vhost_destruct, | |
81acebda IM |
3446 | NULL, |
3447 | NULL, | |
7d1ced01 CL |
3448 | netdev_dpdk_vhost_send, |
3449 | netdev_dpdk_vhost_get_carrier, | |
3450 | netdev_dpdk_vhost_get_stats, | |
3451 | NULL, | |
3452 | NULL, | |
050c60bf | 3453 | netdev_dpdk_vhost_cuse_reconfigure, |
7d1ced01 CL |
3454 | netdev_dpdk_vhost_rxq_recv); |
3455 | ||
bd8baf47 | 3456 | static const struct netdev_class OVS_UNUSED dpdk_vhost_user_class = |
7d1ced01 CL |
3457 | NETDEV_DPDK_CLASS( |
3458 | "dpdkvhostuser", | |
3459 | dpdk_vhost_user_class_init, | |
3460 | netdev_dpdk_vhost_user_construct, | |
58397e6c | 3461 | netdev_dpdk_vhost_destruct, |
81acebda IM |
3462 | NULL, |
3463 | NULL, | |
58397e6c KT |
3464 | netdev_dpdk_vhost_send, |
3465 | netdev_dpdk_vhost_get_carrier, | |
3466 | netdev_dpdk_vhost_get_stats, | |
3467 | NULL, | |
7251515e | 3468 | NULL, |
050c60bf | 3469 | netdev_dpdk_vhost_user_reconfigure, |
58397e6c | 3470 | netdev_dpdk_vhost_rxq_recv); |
95fb793a | 3471 | |
8a9562d2 PS |
3472 | void |
3473 | netdev_dpdk_register(void) | |
3474 | { | |
bab69409 AC |
3475 | dpdk_common_init(); |
3476 | netdev_register_provider(&dpdk_class); | |
3477 | netdev_register_provider(&dpdk_ring_class); | |
7d1ced01 | 3478 | #ifdef VHOST_CUSE |
bab69409 | 3479 | netdev_register_provider(&dpdk_vhost_cuse_class); |
7d1ced01 | 3480 | #else |
bab69409 | 3481 | netdev_register_provider(&dpdk_vhost_user_class); |
7d1ced01 | 3482 | #endif |
8a9562d2 | 3483 | } |
8617afff | 3484 | |
6930c7e0 DDP |
3485 | void |
3486 | dpdk_set_lcore_id(unsigned cpu) | |
8617afff | 3487 | { |
abb5943d AW |
3488 | /* NON_PMD_CORE_ID is reserved for use by non pmd threads. */ |
3489 | ovs_assert(cpu != NON_PMD_CORE_ID); | |
65f13b50 | 3490 | RTE_PER_LCORE(_lcore_id) = cpu; |
8617afff | 3491 | } |
db73f716 | 3492 | |
db73f716 | 3493 | static bool |
5f17de68 | 3494 | dpdk_thread_is_pmd(void) |
db73f716 | 3495 | { |
abb5943d | 3496 | return rte_lcore_id() != NON_PMD_CORE_ID; |
db73f716 | 3497 | } |