]>
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" | |
50 | #include "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 | |
95e9881f KT |
144 | /* |
145 | * Maximum amount of time in micro seconds to try and enqueue to vhost. | |
146 | */ | |
147 | #define VHOST_ENQ_RETRY_USECS 100 | |
148 | ||
8a9562d2 | 149 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
150 | .rxmode = { |
151 | .mq_mode = ETH_MQ_RX_RSS, | |
152 | .split_hdr_size = 0, | |
153 | .header_split = 0, /* Header Split disabled */ | |
154 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
155 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
156 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
157 | .hw_strip_crc = 0, | |
158 | }, | |
159 | .rx_adv_conf = { | |
160 | .rss_conf = { | |
161 | .rss_key = NULL, | |
543342a4 | 162 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 163 | }, |
a28ddd11 DDP |
164 | }, |
165 | .txmode = { | |
166 | .mq_mode = ETH_MQ_TX_NONE, | |
167 | }, | |
8a9562d2 PS |
168 | }; |
169 | ||
3a100265 | 170 | enum { MAX_TX_QUEUE_LEN = 384 }; |
58f7c37b DDP |
171 | enum { DPDK_RING_SIZE = 256 }; |
172 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
173 | enum { DRAIN_TSC = 200000ULL }; |
174 | ||
58397e6c KT |
175 | enum dpdk_dev_type { |
176 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 177 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
178 | }; |
179 | ||
8a9562d2 PS |
180 | static int rte_eal_init_ret = ENODEV; |
181 | ||
182 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; | |
183 | ||
0bf765f7 IS |
184 | /* Quality of Service */ |
185 | ||
186 | /* An instance of a QoS configuration. Always associated with a particular | |
187 | * network device. | |
188 | * | |
189 | * Each QoS implementation subclasses this with whatever additional data it | |
190 | * needs. | |
191 | */ | |
192 | struct qos_conf { | |
193 | const struct dpdk_qos_ops *ops; | |
194 | }; | |
195 | ||
196 | /* A particular implementation of dpdk QoS operations. | |
197 | * | |
198 | * The functions below return 0 if successful or a positive errno value on | |
199 | * failure, except where otherwise noted. All of them must be provided, except | |
200 | * where otherwise noted. | |
201 | */ | |
202 | struct dpdk_qos_ops { | |
203 | ||
204 | /* Name of the QoS type */ | |
205 | const char *qos_name; | |
206 | ||
207 | /* Called to construct the QoS implementation on 'netdev'. The | |
208 | * implementation should make the appropriate calls to configure QoS | |
209 | * according to 'details'. The implementation may assume that any current | |
210 | * QoS configuration already installed should be destroyed before | |
211 | * constructing the new configuration. | |
212 | * | |
213 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
214 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
215 | * (which is built as ovs-vswitchd.conf.db(8)). | |
216 | * | |
217 | * This function must return 0 if and only if it sets 'netdev->qos_conf' | |
218 | * to an initialized 'struct qos_conf'. | |
219 | * | |
220 | * For all QoS implementations it should always be non-null. | |
221 | */ | |
222 | int (*qos_construct)(struct netdev *netdev, const struct smap *details); | |
223 | ||
224 | /* Destroys the data structures allocated by the implementation as part of | |
225 | * 'qos_conf. | |
226 | * | |
227 | * For all QoS implementations it should always be non-null. | |
228 | */ | |
229 | void (*qos_destruct)(struct netdev *netdev, struct qos_conf *conf); | |
230 | ||
231 | /* Retrieves details of 'netdev->qos_conf' configuration into 'details'. | |
232 | * | |
233 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
234 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
235 | * (which is built as ovs-vswitchd.conf.db(8)). | |
236 | */ | |
237 | int (*qos_get)(const struct netdev *netdev, struct smap *details); | |
238 | ||
239 | /* Reconfigures 'netdev->qos_conf' according to 'details', performing any | |
240 | * required calls to complete the reconfiguration. | |
241 | * | |
242 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
243 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
244 | * (which is built as ovs-vswitchd.conf.db(8)). | |
245 | * | |
246 | * This function may be null if 'qos_conf' is not configurable. | |
247 | */ | |
248 | int (*qos_set)(struct netdev *netdev, const struct smap *details); | |
249 | ||
250 | /* Modify an array of rte_mbufs. The modification is specific to | |
251 | * each qos implementation. | |
252 | * | |
253 | * The function should take and array of mbufs and an int representing | |
254 | * the current number of mbufs present in the array. | |
255 | * | |
256 | * After the function has performed a qos modification to the array of | |
257 | * mbufs it returns an int representing the number of mbufs now present in | |
258 | * the array. This value is can then be passed to the port send function | |
259 | * along with the modified array for transmission. | |
260 | * | |
261 | * For all QoS implementations it should always be non-null. | |
262 | */ | |
263 | int (*qos_run)(struct netdev *netdev, struct rte_mbuf **pkts, | |
264 | int pkt_cnt); | |
265 | }; | |
266 | ||
267 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
268 | static const struct dpdk_qos_ops egress_policer_ops; | |
269 | ||
270 | /* | |
271 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
272 | * operations. | |
273 | */ | |
274 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
275 | &egress_policer_ops, | |
276 | NULL | |
277 | }; | |
278 | ||
8a9562d2 | 279 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 280 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 281 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 282 | |
ca6ba700 | 283 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 284 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 285 | |
db73f716 DDP |
286 | /* This mutex must be used by non pmd threads when allocating or freeing |
287 | * mbufs through mempools. Since dpdk_queue_pkts() and dpdk_queue_flush() may | |
288 | * use mempools, a non pmd thread should hold this mutex while calling them */ | |
bce01e3a | 289 | static struct ovs_mutex nonpmd_mempool_mutex = OVS_MUTEX_INITIALIZER; |
db73f716 | 290 | |
8a9562d2 PS |
291 | struct dpdk_mp { |
292 | struct rte_mempool *mp; | |
293 | int mtu; | |
294 | int socket_id; | |
295 | int refcount; | |
ca6ba700 | 296 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
297 | }; |
298 | ||
5a034064 AW |
299 | /* There should be one 'struct dpdk_tx_queue' created for |
300 | * each cpu core. */ | |
8a9562d2 | 301 | struct dpdk_tx_queue { |
94143fc4 AW |
302 | bool flush_tx; /* Set to true to flush queue everytime */ |
303 | /* pkts are queued. */ | |
8a9562d2 | 304 | int count; |
a0cb2d66 DDP |
305 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
306 | * from concurrent access. It is used only | |
307 | * if the queue is shared among different | |
308 | * pmd threads (see 'txq_needs_locking'). */ | |
585a5bea IM |
309 | int map; /* Mapping of configured vhost-user queues |
310 | * to enabled by guest. */ | |
8a9562d2 PS |
311 | uint64_t tsc; |
312 | struct rte_mbuf *burst_pkts[MAX_TX_QUEUE_LEN]; | |
313 | }; | |
314 | ||
95fb793a | 315 | /* dpdk has no way to remove dpdk ring ethernet devices |
316 | so we have to keep them around once they've been created | |
317 | */ | |
318 | ||
ca6ba700 | 319 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 320 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 321 | |
322 | struct dpdk_ring { | |
323 | /* For the client rings */ | |
324 | struct rte_ring *cring_tx; | |
325 | struct rte_ring *cring_rx; | |
b83a2df1 | 326 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
95fb793a | 327 | int eth_port_id; /* ethernet device port id */ |
ca6ba700 | 328 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 329 | }; |
330 | ||
8a9562d2 PS |
331 | struct netdev_dpdk { |
332 | struct netdev up; | |
333 | int port_id; | |
334 | int max_packet_len; | |
58397e6c | 335 | enum dpdk_dev_type type; |
8a9562d2 | 336 | |
5a034064 | 337 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
338 | |
339 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
340 | ||
341 | struct dpdk_mp *dpdk_mp; | |
342 | int mtu; | |
343 | int socket_id; | |
344 | int buf_size; | |
8a9562d2 | 345 | struct netdev_stats stats; |
45d947c4 DDP |
346 | /* Protects stats */ |
347 | rte_spinlock_t stats_lock; | |
8a9562d2 | 348 | |
74ff3298 | 349 | struct eth_addr hwaddr; |
8a9562d2 PS |
350 | enum netdev_flags flags; |
351 | ||
352 | struct rte_eth_link link; | |
353 | int link_reset_cnt; | |
354 | ||
a0cb2d66 DDP |
355 | /* The user might request more txqs than the NIC has. We remap those |
356 | * ('up.n_txq') on these ('real_n_txq'). | |
357 | * If the numbers match, 'txq_needs_locking' is false, otherwise it is | |
358 | * true and we will take a spinlock on transmission */ | |
359 | int real_n_txq; | |
4573fbd3 | 360 | int real_n_rxq; |
a0cb2d66 DDP |
361 | bool txq_needs_locking; |
362 | ||
58397e6c KT |
363 | /* virtio-net structure for vhost device */ |
364 | OVSRCU_TYPE(struct virtio_net *) virtio_dev; | |
365 | ||
7d1ced01 CL |
366 | /* Identifier used to distinguish vhost devices from each other */ |
367 | char vhost_id[PATH_MAX]; | |
368 | ||
8a9562d2 | 369 | /* In dpdk_list. */ |
ca6ba700 | 370 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
0bf765f7 IS |
371 | |
372 | /* QoS configuration and lock for the device */ | |
373 | struct qos_conf *qos_conf; | |
374 | rte_spinlock_t qos_lock; | |
375 | ||
8a9562d2 PS |
376 | }; |
377 | ||
378 | struct netdev_rxq_dpdk { | |
379 | struct netdev_rxq up; | |
380 | int port_id; | |
381 | }; | |
382 | ||
5f17de68 | 383 | static bool dpdk_thread_is_pmd(void); |
db73f716 | 384 | |
8a9562d2 PS |
385 | static int netdev_dpdk_construct(struct netdev *); |
386 | ||
58397e6c KT |
387 | struct virtio_net * netdev_dpdk_get_virtio(const struct netdev_dpdk *dev); |
388 | ||
8a9562d2 PS |
389 | static bool |
390 | is_dpdk_class(const struct netdev_class *class) | |
391 | { | |
392 | return class->construct == netdev_dpdk_construct; | |
393 | } | |
394 | ||
4be4d22c MK |
395 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
396 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
397 | * value, insufficient buffers are allocated to accomodate the packet in its | |
398 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
399 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
400 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
401 | * behaviour, which reduces performance. To prevent this, use a buffer size that | |
402 | * is closest to 'mtu', but which satisfies the aforementioned criteria. | |
403 | */ | |
404 | static uint32_t | |
405 | dpdk_buf_size(int mtu) | |
406 | { | |
407 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
408 | NETDEV_DPDK_MBUF_ALIGN); | |
409 | } | |
410 | ||
58397e6c KT |
411 | /* XXX: use dpdk malloc for entire OVS. in fact huge page should be used |
412 | * for all other segments data, bss and text. */ | |
8a9562d2 PS |
413 | |
414 | static void * | |
415 | dpdk_rte_mzalloc(size_t sz) | |
416 | { | |
417 | void *ptr; | |
418 | ||
419 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
420 | if (ptr == NULL) { | |
421 | out_of_memory(); | |
422 | } | |
423 | return ptr; | |
424 | } | |
425 | ||
db73f716 DDP |
426 | /* XXX this function should be called only by pmd threads (or by non pmd |
427 | * threads holding the nonpmd_mempool_mutex) */ | |
8a9562d2 | 428 | void |
e14deea0 | 429 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 430 | { |
db73f716 | 431 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 432 | |
b00b4a81 | 433 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
434 | } |
435 | ||
b3cd9f9d PS |
436 | static void |
437 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
438 | void *opaque_arg OVS_UNUSED, | |
439 | void *_m, | |
440 | unsigned i OVS_UNUSED) | |
441 | { | |
442 | struct rte_mbuf *m = _m; | |
443 | ||
4be4d22c | 444 | rte_pktmbuf_init(mp, opaque_arg, _m, i); |
b3cd9f9d | 445 | |
cf62fa4c | 446 | dp_packet_init_dpdk((struct dp_packet *) m, m->buf_len); |
b3cd9f9d PS |
447 | } |
448 | ||
8a9562d2 PS |
449 | static struct dpdk_mp * |
450 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
451 | { | |
452 | struct dpdk_mp *dmp = NULL; | |
453 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
da79ce2b | 454 | unsigned mp_size; |
4be4d22c | 455 | struct rte_pktmbuf_pool_private mbp_priv; |
8a9562d2 PS |
456 | |
457 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
458 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
459 | dmp->refcount++; | |
460 | return dmp; | |
461 | } | |
462 | } | |
463 | ||
464 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
465 | dmp->socket_id = socket_id; | |
466 | dmp->mtu = mtu; | |
467 | dmp->refcount = 1; | |
4be4d22c MK |
468 | mbp_priv.mbuf_data_room_size = MBUF_SIZE(mtu) - sizeof(struct dp_packet); |
469 | mbp_priv.mbuf_priv_size = sizeof (struct dp_packet) - sizeof (struct rte_mbuf); | |
8a9562d2 | 470 | |
da79ce2b DDP |
471 | mp_size = MAX_NB_MBUF; |
472 | do { | |
473 | if (snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d_%d_%u", | |
474 | dmp->mtu, dmp->socket_id, mp_size) < 0) { | |
475 | return NULL; | |
476 | } | |
95fb793a | 477 | |
da79ce2b DDP |
478 | dmp->mp = rte_mempool_create(mp_name, mp_size, MBUF_SIZE(mtu), |
479 | MP_CACHE_SZ, | |
480 | sizeof(struct rte_pktmbuf_pool_private), | |
4be4d22c | 481 | rte_pktmbuf_pool_init, &mbp_priv, |
da79ce2b DDP |
482 | ovs_rte_pktmbuf_init, NULL, |
483 | socket_id, 0); | |
484 | } while (!dmp->mp && rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
8a9562d2 PS |
485 | |
486 | if (dmp->mp == NULL) { | |
487 | return NULL; | |
da79ce2b DDP |
488 | } else { |
489 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", mp_name, mp_size ); | |
8a9562d2 PS |
490 | } |
491 | ||
417e7e66 | 492 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); |
8a9562d2 PS |
493 | return dmp; |
494 | } | |
495 | ||
496 | static void | |
497 | dpdk_mp_put(struct dpdk_mp *dmp) | |
498 | { | |
499 | ||
500 | if (!dmp) { | |
501 | return; | |
502 | } | |
503 | ||
504 | dmp->refcount--; | |
505 | ovs_assert(dmp->refcount >= 0); | |
506 | ||
507 | #if 0 | |
508 | /* I could not find any API to destroy mp. */ | |
509 | if (dmp->refcount == 0) { | |
510 | list_delete(dmp->list_node); | |
511 | /* destroy mp-pool. */ | |
512 | } | |
513 | #endif | |
514 | } | |
515 | ||
516 | static void | |
517 | check_link_status(struct netdev_dpdk *dev) | |
518 | { | |
519 | struct rte_eth_link link; | |
520 | ||
521 | rte_eth_link_get_nowait(dev->port_id, &link); | |
522 | ||
523 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 524 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
525 | |
526 | dev->link_reset_cnt++; | |
527 | dev->link = link; | |
528 | if (dev->link.link_status) { | |
529 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
530 | dev->port_id, (unsigned)dev->link.link_speed, | |
531 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
532 | ("full-duplex") : ("half-duplex")); | |
533 | } else { | |
534 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
535 | } | |
536 | } | |
537 | } | |
538 | ||
539 | static void * | |
540 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
541 | { | |
542 | struct netdev_dpdk *dev; | |
543 | ||
544 | pthread_detach(pthread_self()); | |
545 | ||
546 | for (;;) { | |
547 | ovs_mutex_lock(&dpdk_mutex); | |
548 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
549 | ovs_mutex_lock(&dev->mutex); | |
550 | check_link_status(dev); | |
551 | ovs_mutex_unlock(&dev->mutex); | |
552 | } | |
553 | ovs_mutex_unlock(&dpdk_mutex); | |
554 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
555 | } | |
556 | ||
557 | return NULL; | |
558 | } | |
559 | ||
b98d7669 DDP |
560 | static int |
561 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
562 | { | |
563 | int diag = 0; | |
564 | int i; | |
565 | ||
566 | /* A device may report more queues than it makes available (this has | |
567 | * been observed for Intel xl710, which reserves some of them for | |
568 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
569 | * available. When this happens we can retry the configuration | |
570 | * and request less queues */ | |
571 | while (n_rxq && n_txq) { | |
572 | if (diag) { | |
573 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
574 | } | |
575 | ||
576 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &port_conf); | |
577 | if (diag) { | |
578 | break; | |
579 | } | |
580 | ||
581 | for (i = 0; i < n_txq; i++) { | |
582 | diag = rte_eth_tx_queue_setup(dev->port_id, i, NIC_PORT_TX_Q_SIZE, | |
583 | dev->socket_id, NULL); | |
584 | if (diag) { | |
585 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
586 | dev->up.name, i, rte_strerror(-diag)); | |
587 | break; | |
588 | } | |
589 | } | |
590 | ||
591 | if (i != n_txq) { | |
592 | /* Retry with less tx queues */ | |
593 | n_txq = i; | |
594 | continue; | |
595 | } | |
596 | ||
597 | for (i = 0; i < n_rxq; i++) { | |
598 | diag = rte_eth_rx_queue_setup(dev->port_id, i, NIC_PORT_RX_Q_SIZE, | |
599 | dev->socket_id, NULL, | |
600 | dev->dpdk_mp->mp); | |
601 | if (diag) { | |
602 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
603 | dev->up.name, i, rte_strerror(-diag)); | |
604 | break; | |
605 | } | |
606 | } | |
607 | ||
608 | if (i != n_rxq) { | |
609 | /* Retry with less rx queues */ | |
610 | n_rxq = i; | |
611 | continue; | |
612 | } | |
613 | ||
614 | dev->up.n_rxq = n_rxq; | |
615 | dev->real_n_txq = n_txq; | |
616 | ||
617 | return 0; | |
618 | } | |
619 | ||
620 | return diag; | |
621 | } | |
622 | ||
623 | ||
8a9562d2 PS |
624 | static int |
625 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
626 | { | |
627 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 628 | struct rte_eth_dev_info info; |
8a9562d2 PS |
629 | struct ether_addr eth_addr; |
630 | int diag; | |
b98d7669 | 631 | int n_rxq, n_txq; |
8a9562d2 PS |
632 | |
633 | if (dev->port_id < 0 || dev->port_id >= rte_eth_dev_count()) { | |
95fb793a | 634 | return ENODEV; |
8a9562d2 PS |
635 | } |
636 | ||
a0cb2d66 | 637 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 638 | |
b98d7669 DDP |
639 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
640 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
641 | ||
642 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 643 | if (diag) { |
b98d7669 DDP |
644 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
645 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 646 | return -diag; |
8a9562d2 PS |
647 | } |
648 | ||
8a9562d2 PS |
649 | diag = rte_eth_dev_start(dev->port_id); |
650 | if (diag) { | |
b98d7669 DDP |
651 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
652 | rte_strerror(-diag)); | |
95fb793a | 653 | return -diag; |
8a9562d2 PS |
654 | } |
655 | ||
656 | rte_eth_promiscuous_enable(dev->port_id); | |
657 | rte_eth_allmulticast_enable(dev->port_id); | |
658 | ||
659 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
660 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
661 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
ca92d173 | 662 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 663 | |
ca92d173 | 664 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
665 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
666 | ||
667 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
668 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
669 | ||
670 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
671 | return 0; | |
672 | } | |
673 | ||
674 | static struct netdev_dpdk * | |
675 | netdev_dpdk_cast(const struct netdev *netdev) | |
676 | { | |
677 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
678 | } | |
679 | ||
680 | static struct netdev * | |
681 | netdev_dpdk_alloc(void) | |
682 | { | |
bab69409 AC |
683 | struct netdev_dpdk *dev; |
684 | ||
685 | if (!rte_eal_init_ret) { /* Only after successful initialization */ | |
686 | dev = dpdk_rte_mzalloc(sizeof *dev); | |
687 | if (dev) { | |
688 | return &dev->up; | |
689 | } | |
690 | } | |
691 | return NULL; | |
8a9562d2 PS |
692 | } |
693 | ||
5a034064 | 694 | static void |
d46285a2 | 695 | netdev_dpdk_alloc_txq(struct netdev_dpdk *dev, unsigned int n_txqs) |
5a034064 | 696 | { |
bd5131ba | 697 | unsigned i; |
5a034064 | 698 | |
d46285a2 | 699 | dev->tx_q = dpdk_rte_mzalloc(n_txqs * sizeof *dev->tx_q); |
5a034064 | 700 | for (i = 0; i < n_txqs; i++) { |
ba0358a1 | 701 | int numa_id = ovs_numa_get_numa_id(i); |
94143fc4 | 702 | |
d46285a2 | 703 | if (!dev->txq_needs_locking) { |
a0cb2d66 DDP |
704 | /* Each index is considered as a cpu core id, since there should |
705 | * be one tx queue for each cpu core. If the corresponding core | |
d46285a2 | 706 | * is not on the same numa node as 'dev', flags the |
a0cb2d66 | 707 | * 'flush_tx'. */ |
d46285a2 | 708 | dev->tx_q[i].flush_tx = dev->socket_id == numa_id; |
a0cb2d66 DDP |
709 | } else { |
710 | /* Queues are shared among CPUs. Always flush */ | |
d46285a2 | 711 | dev->tx_q[i].flush_tx = true; |
a0cb2d66 | 712 | } |
585a5bea IM |
713 | |
714 | /* Initialize map for vhost devices. */ | |
d46285a2 DDP |
715 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
716 | rte_spinlock_init(&dev->tx_q[i].tx_lock); | |
5a034064 AW |
717 | } |
718 | } | |
719 | ||
8a9562d2 | 720 | static int |
d46285a2 | 721 | netdev_dpdk_init(struct netdev *netdev, unsigned int port_no, |
58397e6c | 722 | enum dpdk_dev_type type) |
5a034064 | 723 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 724 | { |
d46285a2 | 725 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b7a04e0 | 726 | int sid; |
95fb793a | 727 | int err = 0; |
4be4d22c | 728 | uint32_t buf_size; |
8a9562d2 | 729 | |
d46285a2 DDP |
730 | ovs_mutex_init(&dev->mutex); |
731 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 732 | |
d46285a2 | 733 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 734 | |
1b7a04e0 AW |
735 | /* If the 'sid' is negative, it means that the kernel fails |
736 | * to obtain the pci numa info. In that situation, always | |
737 | * use 'SOCKET0'. */ | |
58397e6c KT |
738 | if (type == DPDK_DEV_ETH) { |
739 | sid = rte_eth_dev_socket_id(port_no); | |
740 | } else { | |
741 | sid = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
742 | } | |
743 | ||
d46285a2 DDP |
744 | dev->socket_id = sid < 0 ? SOCKET0 : sid; |
745 | dev->port_id = port_no; | |
746 | dev->type = type; | |
747 | dev->flags = 0; | |
748 | dev->mtu = ETHER_MTU; | |
749 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
8a9562d2 | 750 | |
d46285a2 DDP |
751 | buf_size = dpdk_buf_size(dev->mtu); |
752 | dev->dpdk_mp = dpdk_mp_get(dev->socket_id, FRAME_LEN_TO_MTU(buf_size)); | |
753 | if (!dev->dpdk_mp) { | |
8a9562d2 | 754 | err = ENOMEM; |
95fb793a | 755 | goto unlock; |
8a9562d2 PS |
756 | } |
757 | ||
0bf765f7 | 758 | /* Initialise QoS configuration to NULL and qos lock to unlocked */ |
d46285a2 DDP |
759 | dev->qos_conf = NULL; |
760 | rte_spinlock_init(&dev->qos_lock); | |
0bf765f7 | 761 | |
d46285a2 DDP |
762 | netdev->n_txq = NR_QUEUE; |
763 | netdev->n_rxq = NR_QUEUE; | |
764 | netdev->requested_n_rxq = NR_QUEUE; | |
765 | dev->real_n_txq = NR_QUEUE; | |
58397e6c KT |
766 | |
767 | if (type == DPDK_DEV_ETH) { | |
d46285a2 DDP |
768 | netdev_dpdk_alloc_txq(dev, NR_QUEUE); |
769 | err = dpdk_eth_dev_init(dev); | |
1b99bb05 MG |
770 | if (err) { |
771 | goto unlock; | |
772 | } | |
585a5bea | 773 | } else { |
d46285a2 | 774 | netdev_dpdk_alloc_txq(dev, OVS_VHOST_MAX_QUEUE_NUM); |
8a9562d2 | 775 | } |
8a9562d2 | 776 | |
d46285a2 | 777 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 778 | |
95fb793a | 779 | unlock: |
5a034064 | 780 | if (err) { |
d46285a2 | 781 | rte_free(dev->tx_q); |
5a034064 | 782 | } |
d46285a2 | 783 | ovs_mutex_unlock(&dev->mutex); |
95fb793a | 784 | return err; |
785 | } | |
786 | ||
b83a2df1 MV |
787 | /* dev_name must be the prefix followed by a positive decimal number. |
788 | * (no leading + or - signs are allowed) */ | |
95fb793a | 789 | static int |
790 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
791 | unsigned int *port_no) | |
792 | { | |
793 | const char *cport; | |
794 | ||
795 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
796 | return ENODEV; | |
797 | } | |
798 | ||
799 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
800 | |
801 | if (str_to_uint(cport, 10, port_no)) { | |
802 | return 0; | |
803 | } else { | |
804 | return ENODEV; | |
805 | } | |
95fb793a | 806 | } |
807 | ||
58397e6c | 808 | static int |
d46285a2 | 809 | vhost_construct_helper(struct netdev *netdev) OVS_REQUIRES(dpdk_mutex) |
58397e6c | 810 | { |
58397e6c KT |
811 | if (rte_eal_init_ret) { |
812 | return rte_eal_init_ret; | |
813 | } | |
814 | ||
d46285a2 | 815 | return netdev_dpdk_init(netdev, -1, DPDK_DEV_VHOST); |
7d1ced01 CL |
816 | } |
817 | ||
818 | static int | |
d46285a2 | 819 | netdev_dpdk_vhost_cuse_construct(struct netdev *netdev) |
7d1ced01 | 820 | { |
d46285a2 | 821 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7d1ced01 CL |
822 | int err; |
823 | ||
bab69409 AC |
824 | if (rte_eal_init_ret) { |
825 | return rte_eal_init_ret; | |
826 | } | |
827 | ||
58397e6c | 828 | ovs_mutex_lock(&dpdk_mutex); |
d46285a2 DDP |
829 | strncpy(dev->vhost_id, netdev->name, sizeof(dev->vhost_id)); |
830 | err = vhost_construct_helper(netdev); | |
58397e6c | 831 | ovs_mutex_unlock(&dpdk_mutex); |
7d1ced01 CL |
832 | return err; |
833 | } | |
58397e6c | 834 | |
7d1ced01 | 835 | static int |
d46285a2 | 836 | netdev_dpdk_vhost_user_construct(struct netdev *netdev) |
7d1ced01 | 837 | { |
d46285a2 DDP |
838 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
839 | const char *name = netdev->name; | |
7d1ced01 | 840 | int err; |
a0cb2d66 | 841 | |
1af27e8a DDP |
842 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
843 | * the file system. '/' or '\' would traverse directories, so they're not | |
844 | * acceptable in 'name'. */ | |
845 | if (strchr(name, '/') || strchr(name, '\\')) { | |
846 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
847 | "A valid name must not include '/' or '\\'", | |
848 | name); | |
849 | return EINVAL; | |
850 | } | |
851 | ||
bab69409 AC |
852 | if (rte_eal_init_ret) { |
853 | return rte_eal_init_ret; | |
854 | } | |
855 | ||
7d1ced01 CL |
856 | ovs_mutex_lock(&dpdk_mutex); |
857 | /* Take the name of the vhost-user port and append it to the location where | |
858 | * the socket is to be created, then register the socket. | |
859 | */ | |
d46285a2 | 860 | snprintf(dev->vhost_id, sizeof(dev->vhost_id), "%s/%s", |
1af27e8a DDP |
861 | vhost_sock_dir, name); |
862 | ||
d46285a2 | 863 | err = rte_vhost_driver_register(dev->vhost_id); |
7d1ced01 CL |
864 | if (err) { |
865 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
d46285a2 | 866 | dev->vhost_id); |
e5c0f5a4 | 867 | } else { |
d46285a2 | 868 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
1a64eb93 | 869 | VLOG_INFO("Socket %s created for vhost-user port %s\n", |
d46285a2 DDP |
870 | dev->vhost_id, name); |
871 | err = vhost_construct_helper(netdev); | |
7d1ced01 | 872 | } |
e5c0f5a4 | 873 | |
7d1ced01 | 874 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
875 | return err; |
876 | } | |
877 | ||
95fb793a | 878 | static int |
879 | netdev_dpdk_construct(struct netdev *netdev) | |
880 | { | |
881 | unsigned int port_no; | |
882 | int err; | |
883 | ||
884 | if (rte_eal_init_ret) { | |
885 | return rte_eal_init_ret; | |
886 | } | |
887 | ||
888 | /* Names always start with "dpdk" */ | |
889 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
890 | if (err) { | |
891 | return err; | |
892 | } | |
893 | ||
894 | ovs_mutex_lock(&dpdk_mutex); | |
58397e6c | 895 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
8a9562d2 PS |
896 | ovs_mutex_unlock(&dpdk_mutex); |
897 | return err; | |
898 | } | |
899 | ||
900 | static void | |
d46285a2 | 901 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 902 | { |
d46285a2 | 903 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
904 | |
905 | ovs_mutex_lock(&dev->mutex); | |
906 | rte_eth_dev_stop(dev->port_id); | |
907 | ovs_mutex_unlock(&dev->mutex); | |
908 | ||
909 | ovs_mutex_lock(&dpdk_mutex); | |
5a034064 | 910 | rte_free(dev->tx_q); |
417e7e66 | 911 | ovs_list_remove(&dev->list_node); |
8a9562d2 PS |
912 | dpdk_mp_put(dev->dpdk_mp); |
913 | ovs_mutex_unlock(&dpdk_mutex); | |
58397e6c | 914 | } |
8a9562d2 | 915 | |
58397e6c | 916 | static void |
d46285a2 | 917 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 918 | { |
d46285a2 | 919 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c | 920 | |
c62da695 | 921 | /* Guest becomes an orphan if still attached. */ |
58397e6c | 922 | if (netdev_dpdk_get_virtio(dev) != NULL) { |
c62da695 | 923 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 924 | netdev->name); |
c62da695 IM |
925 | VLOG_ERR("To restore connectivity after re-adding of port, VM on socket" |
926 | " '%s' must be restarted.", | |
927 | dev->vhost_id); | |
58397e6c KT |
928 | } |
929 | ||
e04f7e4f CL |
930 | if (rte_vhost_driver_unregister(dev->vhost_id)) { |
931 | VLOG_ERR("Unable to remove vhost-user socket %s", dev->vhost_id); | |
e5c0f5a4 IM |
932 | } else { |
933 | fatal_signal_remove_file_to_unlink(dev->vhost_id); | |
e04f7e4f CL |
934 | } |
935 | ||
58397e6c | 936 | ovs_mutex_lock(&dpdk_mutex); |
6019cb63 | 937 | rte_free(dev->tx_q); |
417e7e66 | 938 | ovs_list_remove(&dev->list_node); |
58397e6c KT |
939 | dpdk_mp_put(dev->dpdk_mp); |
940 | ovs_mutex_unlock(&dpdk_mutex); | |
8a9562d2 PS |
941 | } |
942 | ||
943 | static void | |
d46285a2 | 944 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 945 | { |
d46285a2 | 946 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 947 | |
d46285a2 | 948 | rte_free(dev); |
8a9562d2 PS |
949 | } |
950 | ||
951 | static int | |
a14b8947 | 952 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 953 | { |
a14b8947 | 954 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
955 | |
956 | ovs_mutex_lock(&dev->mutex); | |
957 | ||
a14b8947 IM |
958 | smap_add_format(args, "requested_rx_queues", "%d", netdev->requested_n_rxq); |
959 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); | |
960 | smap_add_format(args, "requested_tx_queues", "%d", netdev->n_txq); | |
a0cb2d66 | 961 | smap_add_format(args, "configured_tx_queues", "%d", dev->real_n_txq); |
8a9562d2 PS |
962 | ovs_mutex_unlock(&dev->mutex); |
963 | ||
964 | return 0; | |
965 | } | |
966 | ||
a14b8947 IM |
967 | static int |
968 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args) | |
969 | { | |
970 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
971 | ||
972 | ovs_mutex_lock(&dev->mutex); | |
973 | netdev->requested_n_rxq = MAX(smap_get_int(args, "n_rxq", | |
974 | netdev->requested_n_rxq), 1); | |
975 | netdev_change_seq_changed(netdev); | |
976 | ovs_mutex_unlock(&dev->mutex); | |
977 | ||
978 | return 0; | |
979 | } | |
980 | ||
7dec44fe | 981 | static int |
d46285a2 | 982 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 983 | { |
d46285a2 | 984 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 985 | |
d46285a2 | 986 | return dev->socket_id; |
7dec44fe AW |
987 | } |
988 | ||
5496878c AW |
989 | /* Sets the number of tx queues and rx queues for the dpdk interface. |
990 | * If the configuration fails, do not try restoring its old configuration | |
991 | * and just returns the error. */ | |
992 | static int | |
d46285a2 | 993 | netdev_dpdk_set_multiq(struct netdev *netdev, unsigned int n_txq, |
5496878c AW |
994 | unsigned int n_rxq) |
995 | { | |
d46285a2 | 996 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 997 | int err = 0; |
4b8df037 | 998 | int old_rxq, old_txq; |
5496878c | 999 | |
d46285a2 | 1000 | if (netdev->n_txq == n_txq && netdev->n_rxq == n_rxq) { |
5496878c AW |
1001 | return err; |
1002 | } | |
1003 | ||
b7ccaf67 | 1004 | ovs_mutex_lock(&dpdk_mutex); |
d46285a2 | 1005 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1006 | |
d46285a2 | 1007 | rte_eth_dev_stop(dev->port_id); |
91968eb0 | 1008 | |
d46285a2 DDP |
1009 | old_txq = netdev->n_txq; |
1010 | old_rxq = netdev->n_rxq; | |
1011 | netdev->n_txq = n_txq; | |
1012 | netdev->n_rxq = n_rxq; | |
58397e6c | 1013 | |
d46285a2 DDP |
1014 | rte_free(dev->tx_q); |
1015 | err = dpdk_eth_dev_init(dev); | |
1016 | netdev_dpdk_alloc_txq(dev, dev->real_n_txq); | |
4b8df037 DDP |
1017 | if (err) { |
1018 | /* If there has been an error, it means that the requested queues | |
1019 | * have not been created. Restore the old numbers. */ | |
d46285a2 DDP |
1020 | netdev->n_txq = old_txq; |
1021 | netdev->n_rxq = old_rxq; | |
4b8df037 | 1022 | } |
a0cb2d66 | 1023 | |
d46285a2 | 1024 | dev->txq_needs_locking = dev->real_n_txq != netdev->n_txq; |
91968eb0 | 1025 | |
d46285a2 | 1026 | ovs_mutex_unlock(&dev->mutex); |
b7ccaf67 | 1027 | ovs_mutex_unlock(&dpdk_mutex); |
5496878c AW |
1028 | |
1029 | return err; | |
1030 | } | |
1031 | ||
58397e6c | 1032 | static int |
d46285a2 DDP |
1033 | netdev_dpdk_vhost_cuse_set_multiq(struct netdev *netdev, unsigned int n_txq, |
1034 | unsigned int n_rxq) | |
58397e6c | 1035 | { |
d46285a2 | 1036 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
1037 | int err = 0; |
1038 | ||
d46285a2 | 1039 | if (netdev->n_txq == n_txq && netdev->n_rxq == n_rxq) { |
58397e6c KT |
1040 | return err; |
1041 | } | |
1042 | ||
1043 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 1044 | ovs_mutex_lock(&dev->mutex); |
58397e6c | 1045 | |
d46285a2 DDP |
1046 | netdev->n_txq = n_txq; |
1047 | dev->real_n_txq = 1; | |
1048 | netdev->n_rxq = 1; | |
1049 | dev->txq_needs_locking = dev->real_n_txq != netdev->n_txq; | |
4573fbd3 | 1050 | |
d46285a2 | 1051 | ovs_mutex_unlock(&dev->mutex); |
4573fbd3 FL |
1052 | ovs_mutex_unlock(&dpdk_mutex); |
1053 | ||
1054 | return err; | |
1055 | } | |
1056 | ||
1057 | static int | |
d46285a2 | 1058 | netdev_dpdk_vhost_set_multiq(struct netdev *netdev, unsigned int n_txq, |
4573fbd3 FL |
1059 | unsigned int n_rxq) |
1060 | { | |
d46285a2 | 1061 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
4573fbd3 FL |
1062 | int err = 0; |
1063 | ||
d46285a2 | 1064 | if (netdev->n_txq == n_txq && netdev->n_rxq == n_rxq) { |
4573fbd3 FL |
1065 | return err; |
1066 | } | |
1067 | ||
1068 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 1069 | ovs_mutex_lock(&dev->mutex); |
4573fbd3 | 1070 | |
d46285a2 DDP |
1071 | netdev->n_txq = n_txq; |
1072 | netdev->n_rxq = n_rxq; | |
58397e6c | 1073 | |
d46285a2 | 1074 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
1075 | ovs_mutex_unlock(&dpdk_mutex); |
1076 | ||
1077 | return err; | |
1078 | } | |
1079 | ||
8a9562d2 PS |
1080 | static struct netdev_rxq * |
1081 | netdev_dpdk_rxq_alloc(void) | |
1082 | { | |
1083 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1084 | ||
1085 | return &rx->up; | |
1086 | } | |
1087 | ||
1088 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1089 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1090 | { |
d46285a2 | 1091 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1092 | } |
1093 | ||
1094 | static int | |
d46285a2 | 1095 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1096 | { |
d46285a2 DDP |
1097 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1098 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1099 | |
d46285a2 DDP |
1100 | ovs_mutex_lock(&dev->mutex); |
1101 | rx->port_id = dev->port_id; | |
1102 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1103 | |
1104 | return 0; | |
1105 | } | |
1106 | ||
1107 | static void | |
d46285a2 | 1108 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1109 | { |
1110 | } | |
1111 | ||
1112 | static void | |
d46285a2 | 1113 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1114 | { |
d46285a2 | 1115 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1116 | |
1117 | rte_free(rx); | |
1118 | } | |
1119 | ||
b170db2a RW |
1120 | static inline void |
1121 | dpdk_queue_flush__(struct netdev_dpdk *dev, int qid) | |
8a9562d2 PS |
1122 | { |
1123 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1304f1f8 DDP |
1124 | uint32_t nb_tx = 0; |
1125 | ||
1126 | while (nb_tx != txq->count) { | |
1127 | uint32_t ret; | |
1128 | ||
1129 | ret = rte_eth_tx_burst(dev->port_id, qid, txq->burst_pkts + nb_tx, | |
1130 | txq->count - nb_tx); | |
1131 | if (!ret) { | |
1132 | break; | |
1133 | } | |
1134 | ||
1135 | nb_tx += ret; | |
1136 | } | |
8a9562d2 | 1137 | |
b170db2a | 1138 | if (OVS_UNLIKELY(nb_tx != txq->count)) { |
db73f716 DDP |
1139 | /* free buffers, which we couldn't transmit, one at a time (each |
1140 | * packet could come from a different mempool) */ | |
1141 | int i; | |
1142 | ||
1143 | for (i = nb_tx; i < txq->count; i++) { | |
b00b4a81 | 1144 | rte_pktmbuf_free(txq->burst_pkts[i]); |
db73f716 | 1145 | } |
45d947c4 | 1146 | rte_spinlock_lock(&dev->stats_lock); |
1304f1f8 | 1147 | dev->stats.tx_dropped += txq->count-nb_tx; |
45d947c4 | 1148 | rte_spinlock_unlock(&dev->stats_lock); |
8a9562d2 | 1149 | } |
1304f1f8 | 1150 | |
8a9562d2 | 1151 | txq->count = 0; |
844f2d74 | 1152 | txq->tsc = rte_get_timer_cycles(); |
b170db2a RW |
1153 | } |
1154 | ||
1155 | static inline void | |
1156 | dpdk_queue_flush(struct netdev_dpdk *dev, int qid) | |
1157 | { | |
1158 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1159 | ||
1160 | if (txq->count == 0) { | |
1161 | return; | |
1162 | } | |
b170db2a | 1163 | dpdk_queue_flush__(dev, qid); |
8a9562d2 PS |
1164 | } |
1165 | ||
58397e6c | 1166 | static bool |
d46285a2 | 1167 | is_vhost_running(struct virtio_net *virtio_dev) |
58397e6c | 1168 | { |
d46285a2 | 1169 | return (virtio_dev != NULL && (virtio_dev->flags & VIRTIO_DEV_RUNNING)); |
58397e6c KT |
1170 | } |
1171 | ||
d6e3feb5 | 1172 | static inline void |
1173 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1174 | unsigned int packet_size) | |
1175 | { | |
1176 | /* Hard-coded search for the size bucket. */ | |
1177 | if (packet_size < 256) { | |
1178 | if (packet_size >= 128) { | |
1179 | stats->rx_128_to_255_packets++; | |
1180 | } else if (packet_size <= 64) { | |
1181 | stats->rx_1_to_64_packets++; | |
1182 | } else { | |
1183 | stats->rx_65_to_127_packets++; | |
1184 | } | |
1185 | } else { | |
1186 | if (packet_size >= 1523) { | |
1187 | stats->rx_1523_to_max_packets++; | |
1188 | } else if (packet_size >= 1024) { | |
1189 | stats->rx_1024_to_1522_packets++; | |
1190 | } else if (packet_size < 512) { | |
1191 | stats->rx_256_to_511_packets++; | |
1192 | } else { | |
1193 | stats->rx_512_to_1023_packets++; | |
1194 | } | |
1195 | } | |
1196 | } | |
1197 | ||
9e3ddd45 TP |
1198 | static inline void |
1199 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
1200 | struct dp_packet **packets, int count) | |
1201 | { | |
1202 | int i; | |
d6e3feb5 | 1203 | unsigned int packet_size; |
9e3ddd45 TP |
1204 | struct dp_packet *packet; |
1205 | ||
1206 | stats->rx_packets += count; | |
1207 | for (i = 0; i < count; i++) { | |
1208 | packet = packets[i]; | |
d6e3feb5 | 1209 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1210 | |
d6e3feb5 | 1211 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1212 | /* This only protects the following multicast counting from |
1213 | * too short packets, but it does not stop the packet from | |
1214 | * further processing. */ | |
1215 | stats->rx_errors++; | |
1216 | stats->rx_length_errors++; | |
1217 | continue; | |
1218 | } | |
1219 | ||
d6e3feb5 | 1220 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1221 | ||
9e3ddd45 TP |
1222 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1223 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1224 | stats->multicast++; | |
1225 | } | |
1226 | ||
d6e3feb5 | 1227 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1228 | } |
1229 | } | |
1230 | ||
58397e6c KT |
1231 | /* |
1232 | * The receive path for the vhost port is the TX path out from guest. | |
1233 | */ | |
1234 | static int | |
d46285a2 | 1235 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
58397e6c KT |
1236 | struct dp_packet **packets, int *c) |
1237 | { | |
d46285a2 DDP |
1238 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
1239 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(dev); | |
1240 | int qid = rxq->queue_id; | |
58397e6c KT |
1241 | uint16_t nb_rx = 0; |
1242 | ||
1243 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev))) { | |
1244 | return EAGAIN; | |
1245 | } | |
1246 | ||
d46285a2 | 1247 | if (rxq->queue_id >= dev->real_n_rxq) { |
4573fbd3 FL |
1248 | return EOPNOTSUPP; |
1249 | } | |
1250 | ||
1251 | nb_rx = rte_vhost_dequeue_burst(virtio_dev, qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
d46285a2 | 1252 | dev->dpdk_mp->mp, |
58397e6c | 1253 | (struct rte_mbuf **)packets, |
cd159f1a | 1254 | NETDEV_MAX_BURST); |
58397e6c KT |
1255 | if (!nb_rx) { |
1256 | return EAGAIN; | |
1257 | } | |
1258 | ||
d46285a2 DDP |
1259 | rte_spinlock_lock(&dev->stats_lock); |
1260 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, packets, nb_rx); | |
1261 | rte_spinlock_unlock(&dev->stats_lock); | |
45d947c4 | 1262 | |
58397e6c KT |
1263 | *c = (int) nb_rx; |
1264 | return 0; | |
1265 | } | |
1266 | ||
8a9562d2 | 1267 | static int |
d46285a2 | 1268 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet **packets, |
91088554 | 1269 | int *c) |
8a9562d2 | 1270 | { |
d46285a2 DDP |
1271 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1272 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1273 | int nb_rx; |
8a9562d2 | 1274 | |
5496878c | 1275 | /* There is only one tx queue for this core. Do not flush other |
dc6ba5dc WL |
1276 | * queues. |
1277 | * Do not flush tx queue which is shared among CPUs | |
1278 | * since it is always flushed */ | |
d46285a2 | 1279 | if (rxq->queue_id == rte_lcore_id() && |
dc6ba5dc | 1280 | OVS_LIKELY(!dev->txq_needs_locking)) { |
d46285a2 | 1281 | dpdk_queue_flush(dev, rxq->queue_id); |
5496878c | 1282 | } |
8a9562d2 | 1283 | |
d46285a2 | 1284 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
7d08d53e | 1285 | (struct rte_mbuf **) packets, |
cd159f1a | 1286 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1287 | if (!nb_rx) { |
1288 | return EAGAIN; | |
1289 | } | |
1290 | ||
8a9562d2 PS |
1291 | *c = nb_rx; |
1292 | ||
1293 | return 0; | |
1294 | } | |
1295 | ||
0bf765f7 IS |
1296 | static inline int |
1297 | netdev_dpdk_qos_run__(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1298 | int cnt) | |
1299 | { | |
1300 | struct netdev *netdev = &dev->up; | |
1301 | ||
1302 | if (dev->qos_conf != NULL) { | |
1303 | rte_spinlock_lock(&dev->qos_lock); | |
1304 | if (dev->qos_conf != NULL) { | |
1305 | cnt = dev->qos_conf->ops->qos_run(netdev, pkts, cnt); | |
1306 | } | |
1307 | rte_spinlock_unlock(&dev->qos_lock); | |
1308 | } | |
1309 | ||
1310 | return cnt; | |
1311 | } | |
1312 | ||
9e3ddd45 TP |
1313 | static inline void |
1314 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1315 | struct dp_packet **packets, | |
1316 | int attempted, | |
1317 | int dropped) | |
1318 | { | |
1319 | int i; | |
1320 | int sent = attempted - dropped; | |
1321 | ||
1322 | stats->tx_packets += sent; | |
1323 | stats->tx_dropped += dropped; | |
1324 | ||
1325 | for (i = 0; i < sent; i++) { | |
1326 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1327 | } | |
1328 | } | |
1329 | ||
58397e6c | 1330 | static void |
4573fbd3 FL |
1331 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1332 | struct dp_packet **pkts, int cnt, | |
1333 | bool may_steal) | |
58397e6c | 1334 | { |
d46285a2 DDP |
1335 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1336 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(dev); | |
95e9881f KT |
1337 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1338 | unsigned int total_pkts = cnt; | |
0bf765f7 | 1339 | unsigned int qos_pkts = cnt; |
95e9881f | 1340 | uint64_t start = 0; |
58397e6c | 1341 | |
d46285a2 | 1342 | qid = dev->tx_q[qid % dev->real_n_txq].map; |
585a5bea | 1343 | |
f3ea2ad2 | 1344 | if (OVS_UNLIKELY(!is_vhost_running(virtio_dev) || qid < 0)) { |
d46285a2 DDP |
1345 | rte_spinlock_lock(&dev->stats_lock); |
1346 | dev->stats.tx_dropped+= cnt; | |
1347 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1348 | goto out; |
58397e6c KT |
1349 | } |
1350 | ||
d46285a2 | 1351 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1352 | |
0bf765f7 | 1353 | /* Check has QoS has been configured for the netdev */ |
d46285a2 | 1354 | cnt = netdev_dpdk_qos_run__(dev, cur_pkts, cnt); |
0bf765f7 IS |
1355 | qos_pkts -= cnt; |
1356 | ||
95e9881f | 1357 | do { |
4573fbd3 | 1358 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1359 | unsigned int tx_pkts; |
1360 | ||
4573fbd3 | 1361 | tx_pkts = rte_vhost_enqueue_burst(virtio_dev, vhost_qid, |
95e9881f KT |
1362 | cur_pkts, cnt); |
1363 | if (OVS_LIKELY(tx_pkts)) { | |
1364 | /* Packets have been sent.*/ | |
1365 | cnt -= tx_pkts; | |
1366 | /* Prepare for possible next iteration.*/ | |
1367 | cur_pkts = &cur_pkts[tx_pkts]; | |
1368 | } else { | |
1369 | uint64_t timeout = VHOST_ENQ_RETRY_USECS * rte_get_timer_hz() / 1E6; | |
1370 | unsigned int expired = 0; | |
1371 | ||
1372 | if (!start) { | |
1373 | start = rte_get_timer_cycles(); | |
1374 | } | |
1375 | ||
1376 | /* | |
1377 | * Unable to enqueue packets to vhost interface. | |
1378 | * Check available entries before retrying. | |
1379 | */ | |
4573fbd3 | 1380 | while (!rte_vring_available_entries(virtio_dev, vhost_qid)) { |
95e9881f KT |
1381 | if (OVS_UNLIKELY((rte_get_timer_cycles() - start) > timeout)) { |
1382 | expired = 1; | |
1383 | break; | |
1384 | } | |
1385 | } | |
1386 | if (expired) { | |
1387 | /* break out of main loop. */ | |
1388 | break; | |
1389 | } | |
1390 | } | |
1391 | } while (cnt); | |
4573fbd3 | 1392 | |
d46285a2 | 1393 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1394 | |
d46285a2 | 1395 | rte_spinlock_lock(&dev->stats_lock); |
0bf765f7 | 1396 | cnt += qos_pkts; |
d46285a2 DDP |
1397 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, cnt); |
1398 | rte_spinlock_unlock(&dev->stats_lock); | |
58397e6c KT |
1399 | |
1400 | out: | |
1401 | if (may_steal) { | |
95e9881f KT |
1402 | int i; |
1403 | ||
1404 | for (i = 0; i < total_pkts; i++) { | |
1b99bb05 MG |
1405 | dp_packet_delete(pkts[i]); |
1406 | } | |
58397e6c KT |
1407 | } |
1408 | } | |
1409 | ||
8a9562d2 | 1410 | inline static void |
f4fd623c DDP |
1411 | dpdk_queue_pkts(struct netdev_dpdk *dev, int qid, |
1412 | struct rte_mbuf **pkts, int cnt) | |
8a9562d2 PS |
1413 | { |
1414 | struct dpdk_tx_queue *txq = &dev->tx_q[qid]; | |
1415 | uint64_t diff_tsc; | |
8a9562d2 | 1416 | |
f4fd623c DDP |
1417 | int i = 0; |
1418 | ||
f4fd623c DDP |
1419 | while (i < cnt) { |
1420 | int freeslots = MAX_TX_QUEUE_LEN - txq->count; | |
1421 | int tocopy = MIN(freeslots, cnt-i); | |
8a9562d2 | 1422 | |
f4fd623c DDP |
1423 | memcpy(&txq->burst_pkts[txq->count], &pkts[i], |
1424 | tocopy * sizeof (struct rte_mbuf *)); | |
1425 | ||
1426 | txq->count += tocopy; | |
1427 | i += tocopy; | |
1428 | ||
94143fc4 | 1429 | if (txq->count == MAX_TX_QUEUE_LEN || txq->flush_tx) { |
b170db2a | 1430 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 1431 | } |
844f2d74 | 1432 | diff_tsc = rte_get_timer_cycles() - txq->tsc; |
f4fd623c | 1433 | if (diff_tsc >= DRAIN_TSC) { |
b170db2a | 1434 | dpdk_queue_flush__(dev, qid); |
f4fd623c | 1435 | } |
8a9562d2 | 1436 | } |
8a9562d2 PS |
1437 | } |
1438 | ||
1439 | /* Tx function. Transmit packets indefinitely */ | |
1440 | static void | |
58397e6c | 1441 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet **pkts, |
2654cc33 | 1442 | int cnt) |
db73f716 | 1443 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1444 | { |
bce01e3a EJ |
1445 | #if !defined(__CHECKER__) && !defined(_WIN32) |
1446 | const size_t PKT_ARRAY_SIZE = cnt; | |
1447 | #else | |
1448 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1449 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1450 | #endif |
8a9562d2 | 1451 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
bce01e3a | 1452 | struct rte_mbuf *mbufs[PKT_ARRAY_SIZE]; |
175cf4de RW |
1453 | int dropped = 0; |
1454 | int newcnt = 0; | |
1455 | int i; | |
8a9562d2 | 1456 | |
db73f716 DDP |
1457 | /* If we are on a non pmd thread we have to use the mempool mutex, because |
1458 | * every non pmd thread shares the same mempool cache */ | |
1459 | ||
5f17de68 | 1460 | if (!dpdk_thread_is_pmd()) { |
db73f716 DDP |
1461 | ovs_mutex_lock(&nonpmd_mempool_mutex); |
1462 | } | |
1463 | ||
f4fd623c | 1464 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1465 | int size = dp_packet_size(pkts[i]); |
95fb793a | 1466 | |
f98d7864 | 1467 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c DDP |
1468 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1469 | (int)size , dev->max_packet_len); | |
1470 | ||
175cf4de | 1471 | dropped++; |
f4fd623c DDP |
1472 | continue; |
1473 | } | |
8a9562d2 | 1474 | |
f4fd623c | 1475 | mbufs[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1476 | |
f4fd623c | 1477 | if (!mbufs[newcnt]) { |
175cf4de RW |
1478 | dropped += cnt - i; |
1479 | break; | |
f4fd623c DDP |
1480 | } |
1481 | ||
1482 | /* We have to do a copy for now */ | |
b8e57534 | 1483 | memcpy(rte_pktmbuf_mtod(mbufs[newcnt], void *), dp_packet_data(pkts[i]), size); |
f4fd623c DDP |
1484 | |
1485 | rte_pktmbuf_data_len(mbufs[newcnt]) = size; | |
1486 | rte_pktmbuf_pkt_len(mbufs[newcnt]) = size; | |
1487 | ||
1488 | newcnt++; | |
1489 | } | |
8a9562d2 | 1490 | |
58397e6c | 1491 | if (dev->type == DPDK_DEV_VHOST) { |
4573fbd3 | 1492 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) mbufs, newcnt, true); |
58397e6c | 1493 | } else { |
0bf765f7 IS |
1494 | unsigned int qos_pkts = newcnt; |
1495 | ||
1496 | /* Check if QoS has been configured for this netdev. */ | |
1497 | newcnt = netdev_dpdk_qos_run__(dev, mbufs, newcnt); | |
1498 | ||
1499 | dropped += qos_pkts - newcnt; | |
58397e6c KT |
1500 | dpdk_queue_pkts(dev, qid, mbufs, newcnt); |
1501 | dpdk_queue_flush(dev, qid); | |
1502 | } | |
db73f716 | 1503 | |
0bf765f7 IS |
1504 | if (OVS_UNLIKELY(dropped)) { |
1505 | rte_spinlock_lock(&dev->stats_lock); | |
1506 | dev->stats.tx_dropped += dropped; | |
1507 | rte_spinlock_unlock(&dev->stats_lock); | |
1508 | } | |
1509 | ||
5f17de68 | 1510 | if (!dpdk_thread_is_pmd()) { |
db73f716 DDP |
1511 | ovs_mutex_unlock(&nonpmd_mempool_mutex); |
1512 | } | |
8a9562d2 PS |
1513 | } |
1514 | ||
58397e6c | 1515 | static int |
4573fbd3 | 1516 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, struct dp_packet **pkts, |
58397e6c KT |
1517 | int cnt, bool may_steal) |
1518 | { | |
1519 | if (OVS_UNLIKELY(pkts[0]->source != DPBUF_DPDK)) { | |
1520 | int i; | |
1521 | ||
1522 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); | |
1523 | if (may_steal) { | |
1524 | for (i = 0; i < cnt; i++) { | |
1525 | dp_packet_delete(pkts[i]); | |
1526 | } | |
1527 | } | |
1528 | } else { | |
4573fbd3 | 1529 | __netdev_dpdk_vhost_send(netdev, qid, pkts, cnt, may_steal); |
58397e6c KT |
1530 | } |
1531 | return 0; | |
1532 | } | |
1533 | ||
7251515e DV |
1534 | static inline void |
1535 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
e14deea0 | 1536 | struct dp_packet **pkts, int cnt, bool may_steal) |
8a9562d2 | 1537 | { |
f4fd623c | 1538 | int i; |
8a9562d2 | 1539 | |
a0cb2d66 DDP |
1540 | if (OVS_UNLIKELY(dev->txq_needs_locking)) { |
1541 | qid = qid % dev->real_n_txq; | |
1542 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); | |
1543 | } | |
1544 | ||
7251515e | 1545 | if (OVS_UNLIKELY(!may_steal || |
cf62fa4c | 1546 | pkts[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1547 | struct netdev *netdev = &dev->up; |
1548 | ||
2654cc33 | 1549 | dpdk_do_tx_copy(netdev, qid, pkts, cnt); |
b3cd9f9d PS |
1550 | |
1551 | if (may_steal) { | |
f4fd623c | 1552 | for (i = 0; i < cnt; i++) { |
e14deea0 | 1553 | dp_packet_delete(pkts[i]); |
f4fd623c | 1554 | } |
b3cd9f9d | 1555 | } |
8a9562d2 | 1556 | } else { |
f4fd623c DDP |
1557 | int next_tx_idx = 0; |
1558 | int dropped = 0; | |
0bf765f7 IS |
1559 | unsigned int qos_pkts = 0; |
1560 | unsigned int temp_cnt = 0; | |
8a9562d2 | 1561 | |
f4fd623c | 1562 | for (i = 0; i < cnt; i++) { |
cf62fa4c | 1563 | int size = dp_packet_size(pkts[i]); |
1b99bb05 | 1564 | |
f4fd623c DDP |
1565 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
1566 | if (next_tx_idx != i) { | |
0bf765f7 IS |
1567 | temp_cnt = i - next_tx_idx; |
1568 | qos_pkts = temp_cnt; | |
1569 | ||
1570 | temp_cnt = netdev_dpdk_qos_run__(dev, (struct rte_mbuf**)pkts, | |
1571 | temp_cnt); | |
1572 | dropped += qos_pkts - temp_cnt; | |
f4fd623c DDP |
1573 | dpdk_queue_pkts(dev, qid, |
1574 | (struct rte_mbuf **)&pkts[next_tx_idx], | |
0bf765f7 IS |
1575 | temp_cnt); |
1576 | ||
1ebfe1ac | 1577 | } |
f4fd623c | 1578 | |
1ebfe1ac DDP |
1579 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1580 | (int)size , dev->max_packet_len); | |
f4fd623c | 1581 | |
e14deea0 | 1582 | dp_packet_delete(pkts[i]); |
1ebfe1ac | 1583 | dropped++; |
f4fd623c DDP |
1584 | next_tx_idx = i + 1; |
1585 | } | |
1586 | } | |
1587 | if (next_tx_idx != cnt) { | |
0bf765f7 IS |
1588 | cnt -= next_tx_idx; |
1589 | qos_pkts = cnt; | |
1590 | ||
1591 | cnt = netdev_dpdk_qos_run__(dev, (struct rte_mbuf**)pkts, cnt); | |
1592 | dropped += qos_pkts - cnt; | |
1593 | dpdk_queue_pkts(dev, qid, (struct rte_mbuf **)&pkts[next_tx_idx], | |
1594 | cnt); | |
f4fd623c | 1595 | } |
8a9562d2 | 1596 | |
f4fd623c | 1597 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1598 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1599 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1600 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1601 | } |
8a9562d2 | 1602 | } |
a0cb2d66 DDP |
1603 | |
1604 | if (OVS_UNLIKELY(dev->txq_needs_locking)) { | |
1605 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); | |
1606 | } | |
7251515e DV |
1607 | } |
1608 | ||
1609 | static int | |
1610 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
e14deea0 | 1611 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e DV |
1612 | { |
1613 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1614 | |
7251515e DV |
1615 | netdev_dpdk_send__(dev, qid, pkts, cnt, may_steal); |
1616 | return 0; | |
8a9562d2 PS |
1617 | } |
1618 | ||
1619 | static int | |
74ff3298 | 1620 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1621 | { |
1622 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1623 | ||
1624 | ovs_mutex_lock(&dev->mutex); | |
1625 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1626 | dev->hwaddr = mac; |
045c0d1a | 1627 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1628 | } |
1629 | ovs_mutex_unlock(&dev->mutex); | |
1630 | ||
1631 | return 0; | |
1632 | } | |
1633 | ||
1634 | static int | |
74ff3298 | 1635 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1636 | { |
1637 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1638 | ||
1639 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1640 | *mac = dev->hwaddr; |
8a9562d2 PS |
1641 | ovs_mutex_unlock(&dev->mutex); |
1642 | ||
1643 | return 0; | |
1644 | } | |
1645 | ||
1646 | static int | |
1647 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1648 | { | |
1649 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1650 | ||
1651 | ovs_mutex_lock(&dev->mutex); | |
1652 | *mtup = dev->mtu; | |
1653 | ovs_mutex_unlock(&dev->mutex); | |
1654 | ||
1655 | return 0; | |
1656 | } | |
1657 | ||
1658 | static int | |
1659 | netdev_dpdk_set_mtu(const struct netdev *netdev, int mtu) | |
1660 | { | |
1661 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
4be4d22c | 1662 | int old_mtu, err, dpdk_mtu; |
8a9562d2 PS |
1663 | struct dpdk_mp *old_mp; |
1664 | struct dpdk_mp *mp; | |
4be4d22c | 1665 | uint32_t buf_size; |
8a9562d2 PS |
1666 | |
1667 | ovs_mutex_lock(&dpdk_mutex); | |
1668 | ovs_mutex_lock(&dev->mutex); | |
1669 | if (dev->mtu == mtu) { | |
1670 | err = 0; | |
1671 | goto out; | |
1672 | } | |
1673 | ||
4be4d22c MK |
1674 | buf_size = dpdk_buf_size(mtu); |
1675 | dpdk_mtu = FRAME_LEN_TO_MTU(buf_size); | |
1676 | ||
1677 | mp = dpdk_mp_get(dev->socket_id, dpdk_mtu); | |
8a9562d2 PS |
1678 | if (!mp) { |
1679 | err = ENOMEM; | |
1680 | goto out; | |
1681 | } | |
1682 | ||
1683 | rte_eth_dev_stop(dev->port_id); | |
1684 | ||
1685 | old_mtu = dev->mtu; | |
1686 | old_mp = dev->dpdk_mp; | |
1687 | dev->dpdk_mp = mp; | |
1688 | dev->mtu = mtu; | |
4be4d22c | 1689 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
8a9562d2 PS |
1690 | |
1691 | err = dpdk_eth_dev_init(dev); | |
1692 | if (err) { | |
8a9562d2 PS |
1693 | dpdk_mp_put(mp); |
1694 | dev->mtu = old_mtu; | |
1695 | dev->dpdk_mp = old_mp; | |
4be4d22c | 1696 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
8a9562d2 PS |
1697 | dpdk_eth_dev_init(dev); |
1698 | goto out; | |
1699 | } | |
1700 | ||
1701 | dpdk_mp_put(old_mp); | |
045c0d1a | 1702 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1703 | out: |
1704 | ovs_mutex_unlock(&dev->mutex); | |
1705 | ovs_mutex_unlock(&dpdk_mutex); | |
1706 | return err; | |
1707 | } | |
1708 | ||
1709 | static int | |
d46285a2 | 1710 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 1711 | |
58397e6c KT |
1712 | static int |
1713 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1714 | struct netdev_stats *stats) | |
1715 | { | |
1716 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1717 | ||
1718 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 1719 | |
45d947c4 | 1720 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
1721 | /* Supported Stats */ |
1722 | stats->rx_packets += dev->stats.rx_packets; | |
1723 | stats->tx_packets += dev->stats.tx_packets; | |
1724 | stats->tx_dropped += dev->stats.tx_dropped; | |
9e3ddd45 TP |
1725 | stats->multicast = dev->stats.multicast; |
1726 | stats->rx_bytes = dev->stats.rx_bytes; | |
1727 | stats->tx_bytes = dev->stats.tx_bytes; | |
1728 | stats->rx_errors = dev->stats.rx_errors; | |
1729 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 1730 | |
1731 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
1732 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
1733 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
1734 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
1735 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
1736 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
1737 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
1738 | ||
45d947c4 | 1739 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 1740 | |
58397e6c KT |
1741 | ovs_mutex_unlock(&dev->mutex); |
1742 | ||
1743 | return 0; | |
1744 | } | |
1745 | ||
d6e3feb5 | 1746 | static void |
1747 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
1748 | const struct rte_eth_xstats *xstats, | |
1749 | const unsigned int size) | |
1750 | { | |
1751 | /* XXX Current implementation is simple search through an array | |
1752 | * to find hardcoded counter names. In future DPDK release (TBD) | |
1753 | * XSTATS API will change so each counter will be represented by | |
1754 | * unique ID instead of String. */ | |
1755 | ||
1756 | for (unsigned int i = 0; i < size; i++) { | |
1757 | if (strcmp(XSTAT_RX_64_PACKETS, xstats[i].name) == 0) { | |
1758 | stats->rx_1_to_64_packets = xstats[i].value; | |
1759 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, xstats[i].name) == 0) { | |
1760 | stats->rx_65_to_127_packets = xstats[i].value; | |
1761 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, xstats[i].name) == 0) { | |
1762 | stats->rx_128_to_255_packets = xstats[i].value; | |
1763 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, xstats[i].name) == 0) { | |
1764 | stats->rx_256_to_511_packets = xstats[i].value; | |
1765 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, | |
1766 | xstats[i].name) == 0) { | |
1767 | stats->rx_512_to_1023_packets = xstats[i].value; | |
1768 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, | |
1769 | xstats[i].name) == 0) { | |
1770 | stats->rx_1024_to_1522_packets = xstats[i].value; | |
1771 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, | |
1772 | xstats[i].name) == 0) { | |
1773 | stats->rx_1523_to_max_packets = xstats[i].value; | |
1774 | } else if (strcmp(XSTAT_TX_64_PACKETS, xstats[i].name) == 0) { | |
1775 | stats->tx_1_to_64_packets = xstats[i].value; | |
1776 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, xstats[i].name) == 0) { | |
1777 | stats->tx_65_to_127_packets = xstats[i].value; | |
1778 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, xstats[i].name) == 0) { | |
1779 | stats->tx_128_to_255_packets = xstats[i].value; | |
1780 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, xstats[i].name) == 0) { | |
1781 | stats->tx_256_to_511_packets = xstats[i].value; | |
1782 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, | |
1783 | xstats[i].name) == 0) { | |
1784 | stats->tx_512_to_1023_packets = xstats[i].value; | |
1785 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, | |
1786 | xstats[i].name) == 0) { | |
1787 | stats->tx_1024_to_1522_packets = xstats[i].value; | |
1788 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, | |
1789 | xstats[i].name) == 0) { | |
1790 | stats->tx_1523_to_max_packets = xstats[i].value; | |
1791 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, xstats[i].name) == 0) { | |
1792 | stats->tx_multicast_packets = xstats[i].value; | |
1793 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, xstats[i].name) == 0) { | |
1794 | stats->rx_broadcast_packets = xstats[i].value; | |
1795 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, xstats[i].name) == 0) { | |
1796 | stats->tx_broadcast_packets = xstats[i].value; | |
1797 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, xstats[i].name) == 0) { | |
1798 | stats->rx_undersized_errors = xstats[i].value; | |
1799 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, xstats[i].name) == 0) { | |
1800 | stats->rx_fragmented_errors = xstats[i].value; | |
1801 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, xstats[i].name) == 0) { | |
1802 | stats->rx_jabber_errors = xstats[i].value; | |
1803 | } | |
1804 | } | |
1805 | } | |
1806 | ||
8a9562d2 PS |
1807 | static int |
1808 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1809 | { | |
1810 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1811 | struct rte_eth_stats rte_stats; | |
1812 | bool gg; | |
1813 | ||
1814 | netdev_dpdk_get_carrier(netdev, &gg); | |
1815 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 1816 | |
d6e3feb5 | 1817 | struct rte_eth_xstats *rte_xstats; |
1818 | int rte_xstats_len, rte_xstats_ret; | |
1819 | ||
1820 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
1821 | VLOG_ERR("Can't get ETH statistics for port: %i.", dev->port_id); | |
f9256822 | 1822 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 1823 | return EPROTO; |
1824 | } | |
1825 | ||
1826 | rte_xstats_len = rte_eth_xstats_get(dev->port_id, NULL, 0); | |
1827 | if (rte_xstats_len > 0) { | |
1828 | rte_xstats = dpdk_rte_mzalloc(sizeof(*rte_xstats) * rte_xstats_len); | |
1829 | memset(rte_xstats, 0xff, sizeof(*rte_xstats) * rte_xstats_len); | |
1830 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
1831 | rte_xstats_len); | |
1832 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
1833 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_ret); | |
1834 | } | |
1835 | rte_free(rte_xstats); | |
1836 | } else { | |
1837 | VLOG_WARN("Can't get XSTATS counters for port: %i.", dev->port_id); | |
1838 | } | |
8a9562d2 | 1839 | |
2f9dd77f PS |
1840 | stats->rx_packets = rte_stats.ipackets; |
1841 | stats->tx_packets = rte_stats.opackets; | |
1842 | stats->rx_bytes = rte_stats.ibytes; | |
1843 | stats->tx_bytes = rte_stats.obytes; | |
9e3ddd45 TP |
1844 | /* DPDK counts imissed as errors, but count them here as dropped instead */ |
1845 | stats->rx_errors = rte_stats.ierrors - rte_stats.imissed; | |
2f9dd77f PS |
1846 | stats->tx_errors = rte_stats.oerrors; |
1847 | stats->multicast = rte_stats.imcasts; | |
8a9562d2 | 1848 | |
45d947c4 | 1849 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 1850 | stats->tx_dropped = dev->stats.tx_dropped; |
45d947c4 | 1851 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
1852 | |
1853 | /* These are the available DPDK counters for packets not received due to | |
1854 | * local resource constraints in DPDK and NIC respectively. */ | |
1855 | stats->rx_dropped = rte_stats.rx_nombuf + rte_stats.imissed; | |
9e3ddd45 TP |
1856 | stats->rx_missed_errors = rte_stats.imissed; |
1857 | ||
8a9562d2 PS |
1858 | ovs_mutex_unlock(&dev->mutex); |
1859 | ||
1860 | return 0; | |
1861 | } | |
1862 | ||
1863 | static int | |
d46285a2 | 1864 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 PS |
1865 | enum netdev_features *current, |
1866 | enum netdev_features *advertised OVS_UNUSED, | |
1867 | enum netdev_features *supported OVS_UNUSED, | |
1868 | enum netdev_features *peer OVS_UNUSED) | |
1869 | { | |
d46285a2 | 1870 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1871 | struct rte_eth_link link; |
1872 | ||
1873 | ovs_mutex_lock(&dev->mutex); | |
1874 | link = dev->link; | |
1875 | ovs_mutex_unlock(&dev->mutex); | |
1876 | ||
362ca396 | 1877 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
1878 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
1879 | *current = NETDEV_F_10MB_HD; |
1880 | } | |
362ca396 | 1881 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
1882 | *current = NETDEV_F_100MB_HD; |
1883 | } | |
362ca396 | 1884 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
1885 | *current = NETDEV_F_1GB_HD; |
1886 | } | |
1887 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 1888 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
1889 | *current = NETDEV_F_10MB_FD; |
1890 | } | |
362ca396 | 1891 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
1892 | *current = NETDEV_F_100MB_FD; |
1893 | } | |
362ca396 | 1894 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
1895 | *current = NETDEV_F_1GB_FD; |
1896 | } | |
362ca396 | 1897 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
1898 | *current = NETDEV_F_10GB_FD; |
1899 | } | |
1900 | } | |
1901 | ||
362ca396 | 1902 | if (link.link_autoneg) { |
1903 | *current |= NETDEV_F_AUTONEG; | |
1904 | } | |
1905 | ||
8a9562d2 PS |
1906 | return 0; |
1907 | } | |
1908 | ||
1909 | static int | |
1910 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
1911 | { | |
1912 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1913 | int ifindex; | |
1914 | ||
1915 | ovs_mutex_lock(&dev->mutex); | |
1916 | ifindex = dev->port_id; | |
1917 | ovs_mutex_unlock(&dev->mutex); | |
1918 | ||
1919 | return ifindex; | |
1920 | } | |
1921 | ||
1922 | static int | |
d46285a2 | 1923 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 1924 | { |
d46285a2 | 1925 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1926 | |
1927 | ovs_mutex_lock(&dev->mutex); | |
1928 | check_link_status(dev); | |
1929 | *carrier = dev->link.link_status; | |
58397e6c KT |
1930 | |
1931 | ovs_mutex_unlock(&dev->mutex); | |
1932 | ||
1933 | return 0; | |
1934 | } | |
1935 | ||
1936 | static int | |
d46285a2 | 1937 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 1938 | { |
d46285a2 | 1939 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
1940 | struct virtio_net *virtio_dev = netdev_dpdk_get_virtio(dev); |
1941 | ||
1942 | ovs_mutex_lock(&dev->mutex); | |
1943 | ||
1944 | if (is_vhost_running(virtio_dev)) { | |
1945 | *carrier = 1; | |
1946 | } else { | |
1947 | *carrier = 0; | |
1948 | } | |
1949 | ||
8a9562d2 PS |
1950 | ovs_mutex_unlock(&dev->mutex); |
1951 | ||
1952 | return 0; | |
1953 | } | |
1954 | ||
1955 | static long long int | |
d46285a2 | 1956 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 1957 | { |
d46285a2 | 1958 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1959 | long long int carrier_resets; |
1960 | ||
1961 | ovs_mutex_lock(&dev->mutex); | |
1962 | carrier_resets = dev->link_reset_cnt; | |
1963 | ovs_mutex_unlock(&dev->mutex); | |
1964 | ||
1965 | return carrier_resets; | |
1966 | } | |
1967 | ||
1968 | static int | |
d46285a2 | 1969 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
1970 | long long int interval OVS_UNUSED) |
1971 | { | |
ee32150e | 1972 | return EOPNOTSUPP; |
8a9562d2 PS |
1973 | } |
1974 | ||
1975 | static int | |
1976 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
1977 | enum netdev_flags off, enum netdev_flags on, | |
95fb793a | 1978 | enum netdev_flags *old_flagsp) OVS_REQUIRES(dev->mutex) |
8a9562d2 PS |
1979 | { |
1980 | int err; | |
1981 | ||
1982 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
1983 | return EINVAL; | |
1984 | } | |
1985 | ||
1986 | *old_flagsp = dev->flags; | |
1987 | dev->flags |= on; | |
1988 | dev->flags &= ~off; | |
1989 | ||
1990 | if (dev->flags == *old_flagsp) { | |
1991 | return 0; | |
1992 | } | |
1993 | ||
58397e6c KT |
1994 | if (dev->type == DPDK_DEV_ETH) { |
1995 | if (dev->flags & NETDEV_UP) { | |
1996 | err = rte_eth_dev_start(dev->port_id); | |
1997 | if (err) | |
1998 | return -err; | |
1999 | } | |
8a9562d2 | 2000 | |
58397e6c KT |
2001 | if (dev->flags & NETDEV_PROMISC) { |
2002 | rte_eth_promiscuous_enable(dev->port_id); | |
2003 | } | |
8a9562d2 | 2004 | |
58397e6c KT |
2005 | if (!(dev->flags & NETDEV_UP)) { |
2006 | rte_eth_dev_stop(dev->port_id); | |
2007 | } | |
8a9562d2 PS |
2008 | } |
2009 | ||
2010 | return 0; | |
2011 | } | |
2012 | ||
2013 | static int | |
d46285a2 | 2014 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2015 | enum netdev_flags off, enum netdev_flags on, |
2016 | enum netdev_flags *old_flagsp) | |
2017 | { | |
d46285a2 | 2018 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2019 | int error; |
2020 | ||
d46285a2 DDP |
2021 | ovs_mutex_lock(&dev->mutex); |
2022 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2023 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2024 | |
2025 | return error; | |
2026 | } | |
2027 | ||
2028 | static int | |
d46285a2 | 2029 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2030 | { |
d46285a2 | 2031 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2032 | struct rte_eth_dev_info dev_info; |
2033 | ||
e0a801c7 | 2034 | if (dev->port_id < 0) |
8a9562d2 PS |
2035 | return ENODEV; |
2036 | ||
2037 | ovs_mutex_lock(&dev->mutex); | |
2038 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2039 | ovs_mutex_unlock(&dev->mutex); | |
2040 | ||
2041 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
2042 | ||
95fb793a | 2043 | smap_add_format(args, "port_no", "%d", dev->port_id); |
8a9562d2 PS |
2044 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); |
2045 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
2046 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2047 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2048 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2049 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2050 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
2051 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
2052 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
2053 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2054 | ||
39c2baa9 | 2055 | if (dev_info.pci_dev) { |
2056 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2057 | dev_info.pci_dev->id.vendor_id); | |
2058 | smap_add_format(args, "pci-device_id", "0x%x", | |
2059 | dev_info.pci_dev->id.device_id); | |
2060 | } | |
8a9562d2 PS |
2061 | |
2062 | return 0; | |
2063 | } | |
2064 | ||
2065 | static void | |
2066 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2067 | OVS_REQUIRES(dev->mutex) | |
2068 | { | |
2069 | enum netdev_flags old_flags; | |
2070 | ||
2071 | if (admin_state) { | |
2072 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2073 | } else { | |
2074 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2075 | } | |
2076 | } | |
2077 | ||
2078 | static void | |
2079 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2080 | const char *argv[], void *aux OVS_UNUSED) | |
2081 | { | |
2082 | bool up; | |
2083 | ||
2084 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2085 | up = true; | |
2086 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2087 | up = false; | |
2088 | } else { | |
2089 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2090 | return; | |
2091 | } | |
2092 | ||
2093 | if (argc > 2) { | |
2094 | struct netdev *netdev = netdev_from_name(argv[1]); | |
2095 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
2096 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
2097 | ||
2098 | ovs_mutex_lock(&dpdk_dev->mutex); | |
2099 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
2100 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
2101 | ||
2102 | netdev_close(netdev); | |
2103 | } else { | |
2104 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2105 | netdev_close(netdev); | |
2106 | return; | |
2107 | } | |
2108 | } else { | |
2109 | struct netdev_dpdk *netdev; | |
2110 | ||
2111 | ovs_mutex_lock(&dpdk_mutex); | |
2112 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
2113 | ovs_mutex_lock(&netdev->mutex); | |
2114 | netdev_dpdk_set_admin_state__(netdev, up); | |
2115 | ovs_mutex_unlock(&netdev->mutex); | |
2116 | } | |
2117 | ovs_mutex_unlock(&dpdk_mutex); | |
2118 | } | |
2119 | unixctl_command_reply(conn, "OK"); | |
2120 | } | |
2121 | ||
58397e6c KT |
2122 | /* |
2123 | * Set virtqueue flags so that we do not receive interrupts. | |
2124 | */ | |
2125 | static void | |
d46285a2 | 2126 | set_irq_status(struct virtio_net *virtio_dev) |
58397e6c | 2127 | { |
4573fbd3 FL |
2128 | uint32_t i; |
2129 | uint64_t idx; | |
2130 | ||
d46285a2 | 2131 | for (i = 0; i < virtio_dev->virt_qp_nb; i++) { |
4573fbd3 | 2132 | idx = i * VIRTIO_QNUM; |
d46285a2 DDP |
2133 | rte_vhost_enable_guest_notification(virtio_dev, idx + VIRTIO_RXQ, 0); |
2134 | rte_vhost_enable_guest_notification(virtio_dev, idx + VIRTIO_TXQ, 0); | |
4573fbd3 FL |
2135 | } |
2136 | } | |
2137 | ||
585a5bea IM |
2138 | /* |
2139 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
2140 | * enabling/disabling of queues and real_n_txq modifications. | |
2141 | */ | |
2142 | static void | |
d46285a2 DDP |
2143 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2144 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2145 | { |
2146 | int *enabled_queues, n_enabled = 0; | |
d46285a2 | 2147 | int i, k, total_txqs = dev->real_n_txq; |
585a5bea IM |
2148 | |
2149 | enabled_queues = dpdk_rte_mzalloc(total_txqs * sizeof *enabled_queues); | |
2150 | ||
2151 | for (i = 0; i < total_txqs; i++) { | |
2152 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2153 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2154 | enabled_queues[n_enabled++] = i; |
2155 | } | |
2156 | } | |
2157 | ||
2158 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2159 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2160 | n_enabled = 1; |
2161 | } | |
2162 | ||
2163 | k = 0; | |
2164 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2165 | if (dev->tx_q[i].map != i) { |
2166 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2167 | k = (k + 1) % n_enabled; |
2168 | } | |
2169 | } | |
2170 | ||
d46285a2 | 2171 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2172 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2173 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2174 | } |
2175 | ||
2176 | rte_free(enabled_queues); | |
2177 | } | |
4573fbd3 FL |
2178 | |
2179 | static int | |
d46285a2 DDP |
2180 | netdev_dpdk_vhost_set_queues(struct netdev_dpdk *dev, struct virtio_net *virtio_dev) |
2181 | OVS_REQUIRES(dev->mutex) | |
4573fbd3 FL |
2182 | { |
2183 | uint32_t qp_num; | |
2184 | ||
d46285a2 DDP |
2185 | qp_num = virtio_dev->virt_qp_nb; |
2186 | if (qp_num > dev->up.n_rxq) { | |
4573fbd3 | 2187 | VLOG_ERR("vHost Device '%s' %"PRIu64" can't be added - " |
d46285a2 DDP |
2188 | "too many queues %d > %d", virtio_dev->ifname, virtio_dev->device_fh, |
2189 | qp_num, dev->up.n_rxq); | |
4573fbd3 FL |
2190 | return -1; |
2191 | } | |
2192 | ||
d46285a2 DDP |
2193 | dev->real_n_rxq = qp_num; |
2194 | dev->real_n_txq = qp_num; | |
2195 | dev->txq_needs_locking = true; | |
f3ea2ad2 | 2196 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
d46285a2 DDP |
2197 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { |
2198 | dev->tx_q[0].map = 0; | |
f3ea2ad2 | 2199 | } |
585a5bea | 2200 | |
d46285a2 | 2201 | netdev_dpdk_remap_txqs(dev); |
4573fbd3 FL |
2202 | |
2203 | return 0; | |
58397e6c KT |
2204 | } |
2205 | ||
2206 | /* | |
2207 | * A new virtio-net device is added to a vhost port. | |
2208 | */ | |
2209 | static int | |
d46285a2 | 2210 | new_device(struct virtio_net *virtio_dev) |
58397e6c | 2211 | { |
d46285a2 | 2212 | struct netdev_dpdk *dev; |
58397e6c KT |
2213 | bool exists = false; |
2214 | ||
2215 | ovs_mutex_lock(&dpdk_mutex); | |
2216 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 DDP |
2217 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
2218 | if (strncmp(virtio_dev->ifname, dev->vhost_id, IF_NAME_SZ) == 0) { | |
2219 | ovs_mutex_lock(&dev->mutex); | |
2220 | if (netdev_dpdk_vhost_set_queues(dev, virtio_dev)) { | |
2221 | ovs_mutex_unlock(&dev->mutex); | |
4573fbd3 FL |
2222 | ovs_mutex_unlock(&dpdk_mutex); |
2223 | return -1; | |
2224 | } | |
d46285a2 | 2225 | ovsrcu_set(&dev->virtio_dev, virtio_dev); |
58397e6c | 2226 | exists = true; |
d46285a2 | 2227 | virtio_dev->flags |= VIRTIO_DEV_RUNNING; |
58397e6c | 2228 | /* Disable notifications. */ |
d46285a2 DDP |
2229 | set_irq_status(virtio_dev); |
2230 | ovs_mutex_unlock(&dev->mutex); | |
58397e6c KT |
2231 | break; |
2232 | } | |
2233 | } | |
2234 | ovs_mutex_unlock(&dpdk_mutex); | |
2235 | ||
2236 | if (!exists) { | |
3a8653d7 | 2237 | VLOG_INFO("vHost Device '%s' %"PRIu64" can't be added - name not " |
d46285a2 | 2238 | "found", virtio_dev->ifname, virtio_dev->device_fh); |
58397e6c KT |
2239 | |
2240 | return -1; | |
2241 | } | |
2242 | ||
d46285a2 DDP |
2243 | VLOG_INFO("vHost Device '%s' %"PRIu64" has been added", virtio_dev->ifname, |
2244 | virtio_dev->device_fh); | |
58397e6c KT |
2245 | return 0; |
2246 | } | |
2247 | ||
f3ea2ad2 IM |
2248 | /* Clears mapping for all available queues of vhost interface. */ |
2249 | static void | |
2250 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2251 | OVS_REQUIRES(dev->mutex) | |
2252 | { | |
2253 | int i; | |
2254 | ||
2255 | for (i = 0; i < dev->real_n_txq; i++) { | |
2256 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; | |
2257 | } | |
2258 | } | |
2259 | ||
58397e6c KT |
2260 | /* |
2261 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2262 | * flag to stop any more packets from being sent or received to/from a VM and | |
2263 | * ensure all currently queued packets have been sent/received before removing | |
2264 | * the device. | |
2265 | */ | |
2266 | static void | |
d46285a2 | 2267 | destroy_device(volatile struct virtio_net *virtio_dev) |
58397e6c | 2268 | { |
d46285a2 | 2269 | struct netdev_dpdk *dev; |
afee281f | 2270 | bool exists = false; |
58397e6c KT |
2271 | |
2272 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 DDP |
2273 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
2274 | if (netdev_dpdk_get_virtio(dev) == virtio_dev) { | |
58397e6c | 2275 | |
d46285a2 DDP |
2276 | ovs_mutex_lock(&dev->mutex); |
2277 | virtio_dev->flags &= ~VIRTIO_DEV_RUNNING; | |
2278 | ovsrcu_set(&dev->virtio_dev, NULL); | |
2279 | netdev_dpdk_txq_map_clear(dev); | |
afee281f | 2280 | exists = true; |
d46285a2 | 2281 | ovs_mutex_unlock(&dev->mutex); |
afee281f | 2282 | break; |
58397e6c KT |
2283 | } |
2284 | } | |
afee281f | 2285 | |
58397e6c KT |
2286 | ovs_mutex_unlock(&dpdk_mutex); |
2287 | ||
afee281f KT |
2288 | if (exists == true) { |
2289 | /* | |
2290 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2291 | * to NULL, before returning. | |
2292 | */ | |
2293 | ovsrcu_synchronize(); | |
2294 | /* | |
2295 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2296 | * put thread back into quiescent state before returning. | |
2297 | */ | |
2298 | ovsrcu_quiesce_start(); | |
d46285a2 DDP |
2299 | VLOG_INFO("vHost Device '%s' %"PRIu64" has been removed", |
2300 | virtio_dev->ifname, virtio_dev->device_fh); | |
afee281f | 2301 | } else { |
d46285a2 DDP |
2302 | VLOG_INFO("vHost Device '%s' %"PRIu64" not found", virtio_dev->ifname, |
2303 | virtio_dev->device_fh); | |
afee281f | 2304 | } |
58397e6c KT |
2305 | } |
2306 | ||
585a5bea | 2307 | static int |
d46285a2 DDP |
2308 | vring_state_changed(struct virtio_net *virtio_dev, uint16_t queue_id, |
2309 | int enable) | |
585a5bea | 2310 | { |
d46285a2 | 2311 | struct netdev_dpdk *dev; |
585a5bea IM |
2312 | bool exists = false; |
2313 | int qid = queue_id / VIRTIO_QNUM; | |
2314 | ||
2315 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2316 | return 0; | |
2317 | } | |
2318 | ||
2319 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 DDP |
2320 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
2321 | if (strncmp(virtio_dev->ifname, dev->vhost_id, IF_NAME_SZ) == 0) { | |
2322 | ovs_mutex_lock(&dev->mutex); | |
585a5bea | 2323 | if (enable) { |
d46285a2 | 2324 | dev->tx_q[qid].map = qid; |
585a5bea | 2325 | } else { |
d46285a2 | 2326 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2327 | } |
d46285a2 | 2328 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2329 | exists = true; |
d46285a2 | 2330 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2331 | break; |
2332 | } | |
2333 | } | |
2334 | ovs_mutex_unlock(&dpdk_mutex); | |
2335 | ||
2336 | if (exists) { | |
2337 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s' %" | |
d46285a2 DDP |
2338 | PRIu64" changed to \'%s\'", queue_id, qid, |
2339 | virtio_dev->ifname, virtio_dev->device_fh, | |
2340 | (enable == 1) ? "enabled" : "disabled"); | |
585a5bea | 2341 | } else { |
d46285a2 DDP |
2342 | VLOG_INFO("vHost Device '%s' %"PRIu64" not found", virtio_dev->ifname, |
2343 | virtio_dev->device_fh); | |
585a5bea IM |
2344 | return -1; |
2345 | } | |
2346 | ||
2347 | return 0; | |
2348 | } | |
2349 | ||
58397e6c KT |
2350 | struct virtio_net * |
2351 | netdev_dpdk_get_virtio(const struct netdev_dpdk *dev) | |
2352 | { | |
2353 | return ovsrcu_get(struct virtio_net *, &dev->virtio_dev); | |
2354 | } | |
2355 | ||
2356 | /* | |
2357 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
2358 | * configuration has been fully complete. | |
2359 | */ | |
bce01e3a | 2360 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
2361 | { |
2362 | .new_device = new_device, | |
2363 | .destroy_device = destroy_device, | |
585a5bea | 2364 | .vring_state_changed = vring_state_changed |
58397e6c KT |
2365 | }; |
2366 | ||
2367 | static void * | |
7d1ced01 | 2368 | start_vhost_loop(void *dummy OVS_UNUSED) |
58397e6c KT |
2369 | { |
2370 | pthread_detach(pthread_self()); | |
618f44f7 KT |
2371 | /* Put the cuse thread into quiescent state. */ |
2372 | ovsrcu_quiesce_start(); | |
58397e6c KT |
2373 | rte_vhost_driver_session_start(); |
2374 | return NULL; | |
2375 | } | |
2376 | ||
2377 | static int | |
2378 | dpdk_vhost_class_init(void) | |
7d1ced01 CL |
2379 | { |
2380 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
362ca396 | 2381 | rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4 |
2382 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
2383 | | 1ULL << VIRTIO_NET_F_CSUM); | |
2384 | ||
7d1ced01 CL |
2385 | ovs_thread_create("vhost_thread", start_vhost_loop, NULL); |
2386 | return 0; | |
2387 | } | |
2388 | ||
2389 | static int | |
2390 | dpdk_vhost_cuse_class_init(void) | |
58397e6c | 2391 | { |
7d1ced01 CL |
2392 | return 0; |
2393 | } | |
2394 | ||
2395 | static int | |
2396 | dpdk_vhost_user_class_init(void) | |
2397 | { | |
618f44f7 | 2398 | return 0; |
58397e6c KT |
2399 | } |
2400 | ||
033e9df2 DDP |
2401 | static void |
2402 | dpdk_common_init(void) | |
2403 | { | |
2404 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2405 | "[netdev] up|down", 1, 2, | |
2406 | netdev_dpdk_set_admin_state, NULL); | |
2407 | ||
033e9df2 DDP |
2408 | } |
2409 | ||
95fb793a | 2410 | /* Client Rings */ |
2411 | ||
95fb793a | 2412 | static int |
2413 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
2414 | unsigned int *eth_port_id) | |
2415 | { | |
2416 | struct dpdk_ring *ivshmem; | |
ca7e7bee | 2417 | char ring_name[RTE_RING_NAMESIZE]; |
95fb793a | 2418 | int err; |
2419 | ||
2420 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
2421 | if (ivshmem == NULL) { | |
2422 | return ENOMEM; | |
2423 | } | |
2424 | ||
7251515e | 2425 | /* XXX: Add support for multiquque ring. */ |
ca7e7bee | 2426 | err = snprintf(ring_name, sizeof(ring_name), "%s_tx", dev_name); |
95fb793a | 2427 | if (err < 0) { |
2428 | return -err; | |
2429 | } | |
2430 | ||
8f0a76c9 | 2431 | /* Create single producer tx ring, netdev does explicit locking. */ |
7251515e | 2432 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2433 | RING_F_SP_ENQ); |
95fb793a | 2434 | if (ivshmem->cring_tx == NULL) { |
2435 | rte_free(ivshmem); | |
2436 | return ENOMEM; | |
2437 | } | |
2438 | ||
ca7e7bee | 2439 | err = snprintf(ring_name, sizeof(ring_name), "%s_rx", dev_name); |
95fb793a | 2440 | if (err < 0) { |
2441 | return -err; | |
2442 | } | |
2443 | ||
8f0a76c9 | 2444 | /* Create single consumer rx ring, netdev does explicit locking. */ |
7251515e | 2445 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2446 | RING_F_SC_DEQ); |
95fb793a | 2447 | if (ivshmem->cring_rx == NULL) { |
2448 | rte_free(ivshmem); | |
2449 | return ENOMEM; | |
2450 | } | |
2451 | ||
d7310583 DDP |
2452 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
2453 | &ivshmem->cring_tx, 1, SOCKET0); | |
2454 | ||
95fb793a | 2455 | if (err < 0) { |
2456 | rte_free(ivshmem); | |
2457 | return ENODEV; | |
2458 | } | |
2459 | ||
2460 | ivshmem->user_port_id = port_no; | |
2461 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
417e7e66 | 2462 | ovs_list_push_back(&dpdk_ring_list, &ivshmem->list_node); |
95fb793a | 2463 | |
2464 | *eth_port_id = ivshmem->eth_port_id; | |
2465 | return 0; | |
2466 | } | |
2467 | ||
2468 | static int | |
2469 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) OVS_REQUIRES(dpdk_mutex) | |
2470 | { | |
2471 | struct dpdk_ring *ivshmem; | |
2472 | unsigned int port_no; | |
2473 | int err = 0; | |
2474 | ||
2475 | /* Names always start with "dpdkr" */ | |
2476 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2477 | if (err) { | |
2478 | return err; | |
2479 | } | |
2480 | ||
2481 | /* look through our list to find the device */ | |
2482 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { | |
2483 | if (ivshmem->user_port_id == port_no) { | |
58397e6c | 2484 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
95fb793a | 2485 | *eth_port_id = ivshmem->eth_port_id; /* really all that is needed */ |
2486 | return 0; | |
2487 | } | |
2488 | } | |
2489 | /* Need to create the device rings */ | |
2490 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2491 | } | |
2492 | ||
7251515e | 2493 | static int |
d46285a2 | 2494 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
e14deea0 | 2495 | struct dp_packet **pkts, int cnt, bool may_steal) |
7251515e | 2496 | { |
d46285a2 | 2497 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b99bb05 MG |
2498 | unsigned i; |
2499 | ||
2500 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that the | |
2501 | * rss hash field is clear. This is because the same mbuf may be modified by | |
2502 | * the consumer of the ring and return into the datapath without recalculating | |
2503 | * the RSS hash. */ | |
2504 | for (i = 0; i < cnt; i++) { | |
f2f44f5d | 2505 | dp_packet_rss_invalidate(pkts[i]); |
1b99bb05 | 2506 | } |
7251515e | 2507 | |
d46285a2 | 2508 | netdev_dpdk_send__(dev, qid, pkts, cnt, may_steal); |
7251515e DV |
2509 | return 0; |
2510 | } | |
2511 | ||
95fb793a | 2512 | static int |
2513 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2514 | { | |
2515 | unsigned int port_no = 0; | |
2516 | int err = 0; | |
2517 | ||
2518 | if (rte_eal_init_ret) { | |
2519 | return rte_eal_init_ret; | |
2520 | } | |
2521 | ||
2522 | ovs_mutex_lock(&dpdk_mutex); | |
2523 | ||
2524 | err = dpdk_ring_open(netdev->name, &port_no); | |
2525 | if (err) { | |
2526 | goto unlock_dpdk; | |
2527 | } | |
2528 | ||
58397e6c | 2529 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
95fb793a | 2530 | |
2531 | unlock_dpdk: | |
2532 | ovs_mutex_unlock(&dpdk_mutex); | |
2533 | return err; | |
2534 | } | |
2535 | ||
0bf765f7 IS |
2536 | /* QoS Functions */ |
2537 | ||
2538 | /* | |
2539 | * Initialize QoS configuration operations. | |
2540 | */ | |
2541 | static void | |
2542 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
2543 | { | |
2544 | conf->ops = ops; | |
2545 | } | |
2546 | ||
2547 | /* | |
2548 | * Search existing QoS operations in qos_ops and compare each set of | |
2549 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
2550 | * else return NULL | |
2551 | */ | |
2552 | static const struct dpdk_qos_ops * | |
2553 | qos_lookup_name(const char *name) | |
2554 | { | |
2555 | const struct dpdk_qos_ops *const *opsp; | |
2556 | ||
2557 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2558 | const struct dpdk_qos_ops *ops = *opsp; | |
2559 | if (!strcmp(name, ops->qos_name)) { | |
2560 | return ops; | |
2561 | } | |
2562 | } | |
2563 | return NULL; | |
2564 | } | |
2565 | ||
2566 | /* | |
2567 | * Call qos_destruct to clean up items associated with the netdevs | |
2568 | * qos_conf. Set netdevs qos_conf to NULL. | |
2569 | */ | |
2570 | static void | |
d46285a2 | 2571 | qos_delete_conf(struct netdev *netdev) |
0bf765f7 | 2572 | { |
d46285a2 | 2573 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 2574 | |
d46285a2 DDP |
2575 | rte_spinlock_lock(&dev->qos_lock); |
2576 | if (dev->qos_conf) { | |
2577 | if (dev->qos_conf->ops->qos_destruct) { | |
2578 | dev->qos_conf->ops->qos_destruct(netdev, dev->qos_conf); | |
0bf765f7 | 2579 | } |
d46285a2 | 2580 | dev->qos_conf = NULL; |
0bf765f7 | 2581 | } |
d46285a2 | 2582 | rte_spinlock_unlock(&dev->qos_lock); |
0bf765f7 IS |
2583 | } |
2584 | ||
2585 | static int | |
2586 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
2587 | struct sset *types) | |
2588 | { | |
2589 | const struct dpdk_qos_ops *const *opsp; | |
2590 | ||
2591 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2592 | const struct dpdk_qos_ops *ops = *opsp; | |
2593 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
2594 | sset_add(types, ops->qos_name); | |
2595 | } | |
2596 | } | |
2597 | return 0; | |
2598 | } | |
2599 | ||
2600 | static int | |
d46285a2 | 2601 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
2602 | const char **typep, struct smap *details) |
2603 | { | |
d46285a2 | 2604 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 IS |
2605 | int error = 0; |
2606 | ||
d46285a2 DDP |
2607 | ovs_mutex_lock(&dev->mutex); |
2608 | if(dev->qos_conf) { | |
2609 | *typep = dev->qos_conf->ops->qos_name; | |
2610 | error = (dev->qos_conf->ops->qos_get | |
2611 | ? dev->qos_conf->ops->qos_get(netdev, details): 0); | |
0bf765f7 | 2612 | } |
d46285a2 | 2613 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2614 | |
2615 | return error; | |
2616 | } | |
2617 | ||
2618 | static int | |
d46285a2 | 2619 | netdev_dpdk_set_qos(struct netdev *netdev, |
0bf765f7 IS |
2620 | const char *type, const struct smap *details) |
2621 | { | |
d46285a2 | 2622 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 IS |
2623 | const struct dpdk_qos_ops *new_ops = NULL; |
2624 | int error = 0; | |
2625 | ||
2626 | /* If type is empty or unsupported then the current QoS configuration | |
2627 | * for the dpdk-netdev can be destroyed */ | |
2628 | new_ops = qos_lookup_name(type); | |
2629 | ||
2630 | if (type[0] == '\0' || !new_ops || !new_ops->qos_construct) { | |
d46285a2 | 2631 | qos_delete_conf(netdev); |
0bf765f7 IS |
2632 | return EOPNOTSUPP; |
2633 | } | |
2634 | ||
d46285a2 | 2635 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 2636 | |
d46285a2 DDP |
2637 | if (dev->qos_conf) { |
2638 | if (new_ops == dev->qos_conf->ops) { | |
2639 | error = new_ops->qos_set ? new_ops->qos_set(netdev, details) : 0; | |
0bf765f7 IS |
2640 | } else { |
2641 | /* Delete existing QoS configuration. */ | |
d46285a2 DDP |
2642 | qos_delete_conf(netdev); |
2643 | ovs_assert(dev->qos_conf == NULL); | |
0bf765f7 IS |
2644 | |
2645 | /* Install new QoS configuration. */ | |
d46285a2 DDP |
2646 | error = new_ops->qos_construct(netdev, details); |
2647 | ovs_assert((error == 0) == (dev->qos_conf != NULL)); | |
0bf765f7 IS |
2648 | } |
2649 | } else { | |
d46285a2 DDP |
2650 | error = new_ops->qos_construct(netdev, details); |
2651 | ovs_assert((error == 0) == (dev->qos_conf != NULL)); | |
0bf765f7 IS |
2652 | } |
2653 | ||
d46285a2 | 2654 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2655 | return error; |
2656 | } | |
2657 | ||
2658 | /* egress-policer details */ | |
2659 | ||
2660 | struct egress_policer { | |
2661 | struct qos_conf qos_conf; | |
2662 | struct rte_meter_srtcm_params app_srtcm_params; | |
2663 | struct rte_meter_srtcm egress_meter; | |
2664 | }; | |
2665 | ||
2666 | static struct egress_policer * | |
d46285a2 | 2667 | egress_policer_get__(const struct netdev *netdev) |
0bf765f7 | 2668 | { |
d46285a2 DDP |
2669 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2670 | return CONTAINER_OF(dev->qos_conf, struct egress_policer, qos_conf); | |
0bf765f7 IS |
2671 | } |
2672 | ||
2673 | static int | |
d46285a2 DDP |
2674 | egress_policer_qos_construct(struct netdev *netdev, |
2675 | const struct smap *details) | |
0bf765f7 | 2676 | { |
d46285a2 | 2677 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 IS |
2678 | struct egress_policer *policer; |
2679 | const char *cir_s; | |
2680 | const char *cbs_s; | |
2681 | int err = 0; | |
2682 | ||
d46285a2 | 2683 | rte_spinlock_lock(&dev->qos_lock); |
0bf765f7 IS |
2684 | policer = xmalloc(sizeof *policer); |
2685 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
d46285a2 | 2686 | dev->qos_conf = &policer->qos_conf; |
0bf765f7 IS |
2687 | cir_s = smap_get(details, "cir"); |
2688 | cbs_s = smap_get(details, "cbs"); | |
2689 | policer->app_srtcm_params.cir = cir_s ? strtoull(cir_s, NULL, 10) : 0; | |
2690 | policer->app_srtcm_params.cbs = cbs_s ? strtoull(cbs_s, NULL, 10) : 0; | |
2691 | policer->app_srtcm_params.ebs = 0; | |
2692 | err = rte_meter_srtcm_config(&policer->egress_meter, | |
2693 | &policer->app_srtcm_params); | |
d46285a2 | 2694 | rte_spinlock_unlock(&dev->qos_lock); |
0bf765f7 IS |
2695 | |
2696 | return err; | |
2697 | } | |
2698 | ||
2699 | static void | |
d46285a2 | 2700 | egress_policer_qos_destruct(struct netdev *netdev OVS_UNUSED, |
0bf765f7 IS |
2701 | struct qos_conf *conf) |
2702 | { | |
2703 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
2704 | qos_conf); | |
2705 | free(policer); | |
2706 | } | |
2707 | ||
2708 | static int | |
2709 | egress_policer_qos_get(const struct netdev *netdev, struct smap *details) | |
2710 | { | |
2711 | struct egress_policer *policer = egress_policer_get__(netdev); | |
2712 | smap_add_format(details, "cir", "%llu", | |
2713 | 1ULL * policer->app_srtcm_params.cir); | |
2714 | smap_add_format(details, "cbs", "%llu", | |
2715 | 1ULL * policer->app_srtcm_params.cbs); | |
2716 | return 0; | |
2717 | } | |
2718 | ||
2719 | static int | |
d46285a2 | 2720 | egress_policer_qos_set(struct netdev *netdev, const struct smap *details) |
0bf765f7 IS |
2721 | { |
2722 | struct egress_policer *policer; | |
2723 | const char *cir_s; | |
2724 | const char *cbs_s; | |
2725 | int err = 0; | |
2726 | ||
d46285a2 | 2727 | policer = egress_policer_get__(netdev); |
0bf765f7 IS |
2728 | cir_s = smap_get(details, "cir"); |
2729 | cbs_s = smap_get(details, "cbs"); | |
2730 | policer->app_srtcm_params.cir = cir_s ? strtoull(cir_s, NULL, 10) : 0; | |
2731 | policer->app_srtcm_params.cbs = cbs_s ? strtoull(cbs_s, NULL, 10) : 0; | |
2732 | policer->app_srtcm_params.ebs = 0; | |
2733 | err = rte_meter_srtcm_config(&policer->egress_meter, | |
2734 | &policer->app_srtcm_params); | |
2735 | ||
2736 | return err; | |
2737 | } | |
2738 | ||
2739 | static inline bool | |
2740 | egress_policer_pkt_handle__(struct rte_meter_srtcm *meter, | |
2741 | struct rte_mbuf *pkt, uint64_t time) | |
2742 | { | |
2743 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
2744 | ||
2745 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
2746 | e_RTE_METER_GREEN; | |
2747 | } | |
2748 | ||
2749 | static int | |
d46285a2 | 2750 | egress_policer_run(struct netdev *netdev, struct rte_mbuf **pkts, |
0bf765f7 IS |
2751 | int pkt_cnt) |
2752 | { | |
2753 | int i = 0; | |
2754 | int cnt = 0; | |
d46285a2 | 2755 | struct egress_policer *policer = egress_policer_get__(netdev); |
0bf765f7 IS |
2756 | struct rte_mbuf *pkt = NULL; |
2757 | uint64_t current_time = rte_rdtsc(); | |
2758 | ||
2759 | for(i = 0; i < pkt_cnt; i++) { | |
2760 | pkt = pkts[i]; | |
2761 | /* Handle current packet */ | |
2762 | if (egress_policer_pkt_handle__(&policer->egress_meter, pkt, | |
2763 | current_time)) { | |
2764 | if (cnt != i) { | |
2765 | pkts[cnt] = pkt; | |
2766 | } | |
2767 | cnt++; | |
2768 | } else { | |
2769 | rte_pktmbuf_free(pkt); | |
2770 | } | |
2771 | } | |
2772 | ||
2773 | return cnt; | |
2774 | } | |
2775 | ||
2776 | static const struct dpdk_qos_ops egress_policer_ops = { | |
2777 | "egress-policer", /* qos_name */ | |
2778 | egress_policer_qos_construct, | |
2779 | egress_policer_qos_destruct, | |
2780 | egress_policer_qos_get, | |
2781 | egress_policer_qos_set, | |
2782 | egress_policer_run | |
2783 | }; | |
2784 | ||
58397e6c KT |
2785 | #define NETDEV_DPDK_CLASS(NAME, INIT, CONSTRUCT, DESTRUCT, MULTIQ, SEND, \ |
2786 | GET_CARRIER, GET_STATS, GET_FEATURES, GET_STATUS, RXQ_RECV) \ | |
95fb793a | 2787 | { \ |
2788 | NAME, \ | |
118c77b1 | 2789 | true, /* is_pmd */ \ |
95fb793a | 2790 | INIT, /* init */ \ |
2791 | NULL, /* netdev_dpdk_run */ \ | |
2792 | NULL, /* netdev_dpdk_wait */ \ | |
2793 | \ | |
2794 | netdev_dpdk_alloc, \ | |
2795 | CONSTRUCT, \ | |
58397e6c | 2796 | DESTRUCT, \ |
95fb793a | 2797 | netdev_dpdk_dealloc, \ |
2798 | netdev_dpdk_get_config, \ | |
a14b8947 | 2799 | netdev_dpdk_set_config, \ |
95fb793a | 2800 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
2801 | NULL, /* build header */ \ |
2802 | NULL, /* push header */ \ | |
2803 | NULL, /* pop header */ \ | |
7dec44fe | 2804 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
5496878c | 2805 | MULTIQ, /* set_multiq */ \ |
95fb793a | 2806 | \ |
7251515e | 2807 | SEND, /* send */ \ |
95fb793a | 2808 | NULL, /* send_wait */ \ |
2809 | \ | |
2810 | netdev_dpdk_set_etheraddr, \ | |
2811 | netdev_dpdk_get_etheraddr, \ | |
2812 | netdev_dpdk_get_mtu, \ | |
2813 | netdev_dpdk_set_mtu, \ | |
2814 | netdev_dpdk_get_ifindex, \ | |
58397e6c | 2815 | GET_CARRIER, \ |
95fb793a | 2816 | netdev_dpdk_get_carrier_resets, \ |
2817 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
2818 | GET_STATS, \ |
2819 | GET_FEATURES, \ | |
95fb793a | 2820 | NULL, /* set_advertisements */ \ |
2821 | \ | |
2822 | NULL, /* set_policing */ \ | |
0bf765f7 | 2823 | netdev_dpdk_get_qos_types, \ |
95fb793a | 2824 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
2825 | netdev_dpdk_get_qos, \ |
2826 | netdev_dpdk_set_qos, \ | |
95fb793a | 2827 | NULL, /* get_queue */ \ |
2828 | NULL, /* set_queue */ \ | |
2829 | NULL, /* delete_queue */ \ | |
2830 | NULL, /* get_queue_stats */ \ | |
2831 | NULL, /* queue_dump_start */ \ | |
2832 | NULL, /* queue_dump_next */ \ | |
2833 | NULL, /* queue_dump_done */ \ | |
2834 | NULL, /* dump_queue_stats */ \ | |
2835 | \ | |
95fb793a | 2836 | NULL, /* set_in4 */ \ |
a8704b50 | 2837 | NULL, /* get_addr_list */ \ |
95fb793a | 2838 | NULL, /* add_router */ \ |
2839 | NULL, /* get_next_hop */ \ | |
58397e6c | 2840 | GET_STATUS, \ |
95fb793a | 2841 | NULL, /* arp_lookup */ \ |
2842 | \ | |
2843 | netdev_dpdk_update_flags, \ | |
2844 | \ | |
2845 | netdev_dpdk_rxq_alloc, \ | |
2846 | netdev_dpdk_rxq_construct, \ | |
2847 | netdev_dpdk_rxq_destruct, \ | |
2848 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 2849 | RXQ_RECV, \ |
95fb793a | 2850 | NULL, /* rx_wait */ \ |
2851 | NULL, /* rxq_drain */ \ | |
2852 | } | |
8a9562d2 | 2853 | |
7d1ced01 CL |
2854 | static int |
2855 | process_vhost_flags(char *flag, char *default_val, int size, | |
bab69409 AC |
2856 | const struct smap *ovs_other_config, |
2857 | char **new_val) | |
7d1ced01 | 2858 | { |
bab69409 | 2859 | const char *val; |
7d1ced01 CL |
2860 | int changed = 0; |
2861 | ||
bab69409 AC |
2862 | val = smap_get(ovs_other_config, flag); |
2863 | ||
7d1ced01 | 2864 | /* Depending on which version of vhost is in use, process the vhost-specific |
bab69409 | 2865 | * flag if it is provided, otherwise resort to default value. |
7d1ced01 | 2866 | */ |
bab69409 | 2867 | if (val && (strlen(val) <= size)) { |
7d1ced01 | 2868 | changed = 1; |
bab69409 | 2869 | *new_val = xstrdup(val); |
7d1ced01 CL |
2870 | VLOG_INFO("User-provided %s in use: %s", flag, *new_val); |
2871 | } else { | |
2872 | VLOG_INFO("No %s provided - defaulting to %s", flag, default_val); | |
2873 | *new_val = default_val; | |
2874 | } | |
2875 | ||
2876 | return changed; | |
2877 | } | |
2878 | ||
bab69409 AC |
2879 | static char ** |
2880 | grow_argv(char ***argv, size_t cur_siz, size_t grow_by) | |
8a9562d2 | 2881 | { |
bab69409 AC |
2882 | return xrealloc(*argv, sizeof(char *) * (cur_siz + grow_by)); |
2883 | } | |
8a9562d2 | 2884 | |
bab69409 AC |
2885 | static void |
2886 | dpdk_option_extend(char ***argv, int argc, const char *option, | |
2887 | const char *value) | |
2888 | { | |
2889 | char **newargv = grow_argv(argv, argc, 2); | |
2890 | *argv = newargv; | |
2891 | newargv[argc] = xstrdup(option); | |
2892 | newargv[argc+1] = xstrdup(value); | |
2893 | } | |
8a9562d2 | 2894 | |
52a57e36 AC |
2895 | static char ** |
2896 | move_argv(char ***argv, size_t cur_size, char **src_argv, size_t src_argc) | |
2897 | { | |
2898 | char **newargv = grow_argv(argv, cur_size, src_argc); | |
2899 | while (src_argc--) { | |
2900 | newargv[cur_size+src_argc] = src_argv[src_argc]; | |
2901 | src_argv[src_argc] = NULL; | |
2902 | } | |
2903 | return newargv; | |
2904 | } | |
2905 | ||
eac84432 AC |
2906 | static int |
2907 | extra_dpdk_args(const char *ovs_extra_config, char ***argv, int argc) | |
2908 | { | |
2909 | int ret = argc; | |
2910 | char *release_tok = xstrdup(ovs_extra_config); | |
2911 | char *tok = release_tok, *endptr = NULL; | |
2912 | ||
2913 | for (tok = strtok_r(release_tok, " ", &endptr); tok != NULL; | |
2914 | tok = strtok_r(NULL, " ", &endptr)) { | |
2915 | char **newarg = grow_argv(argv, ret, 1); | |
2916 | *argv = newarg; | |
2917 | newarg[ret++] = xstrdup(tok); | |
2918 | } | |
2919 | free(release_tok); | |
2920 | return ret; | |
2921 | } | |
2922 | ||
52a57e36 AC |
2923 | static bool |
2924 | argv_contains(char **argv_haystack, const size_t argc_haystack, | |
2925 | const char *needle) | |
2926 | { | |
2927 | for (size_t i = 0; i < argc_haystack; ++i) { | |
2928 | if (!strcmp(argv_haystack[i], needle)) | |
2929 | return true; | |
2930 | } | |
2931 | return false; | |
2932 | } | |
2933 | ||
bab69409 AC |
2934 | static int |
2935 | construct_dpdk_options(const struct smap *ovs_other_config, | |
52a57e36 AC |
2936 | char ***argv, const int initial_size, |
2937 | char **extra_args, const size_t extra_argc) | |
bab69409 AC |
2938 | { |
2939 | struct dpdk_options_map { | |
2940 | const char *ovs_configuration; | |
2941 | const char *dpdk_option; | |
2942 | bool default_enabled; | |
2943 | const char *default_value; | |
2944 | } opts[] = { | |
2945 | {"dpdk-lcore-mask", "-c", false, NULL}, | |
2946 | {"dpdk-hugepage-dir", "--huge-dir", false, NULL}, | |
2947 | }; | |
2948 | ||
2949 | int i, ret = initial_size; | |
2950 | ||
2951 | /*First, construct from the flat-options (non-mutex)*/ | |
2952 | for (i = 0; i < ARRAY_SIZE(opts); ++i) { | |
2953 | const char *lookup = smap_get(ovs_other_config, | |
2954 | opts[i].ovs_configuration); | |
2955 | if (!lookup && opts[i].default_enabled) { | |
2956 | lookup = opts[i].default_value; | |
2957 | } | |
8a9562d2 | 2958 | |
bab69409 | 2959 | if (lookup) { |
52a57e36 AC |
2960 | if (!argv_contains(extra_args, extra_argc, opts[i].dpdk_option)) { |
2961 | dpdk_option_extend(argv, ret, opts[i].dpdk_option, lookup); | |
2962 | ret += 2; | |
2963 | } else { | |
2964 | VLOG_WARN("Ignoring database defined option '%s' due to " | |
2965 | "dpdk_extras config", opts[i].dpdk_option); | |
2966 | } | |
bab69409 AC |
2967 | } |
2968 | } | |
2969 | ||
2970 | return ret; | |
2971 | } | |
2972 | ||
2973 | #define MAX_DPDK_EXCL_OPTS 10 | |
2974 | ||
2975 | static int | |
2976 | construct_dpdk_mutex_options(const struct smap *ovs_other_config, | |
52a57e36 AC |
2977 | char ***argv, const int initial_size, |
2978 | char **extra_args, const size_t extra_argc) | |
bab69409 AC |
2979 | { |
2980 | struct dpdk_exclusive_options_map { | |
2981 | const char *category; | |
2982 | const char *ovs_dpdk_options[MAX_DPDK_EXCL_OPTS]; | |
2983 | const char *eal_dpdk_options[MAX_DPDK_EXCL_OPTS]; | |
2984 | const char *default_value; | |
2985 | int default_option; | |
2986 | } excl_opts[] = { | |
2987 | {"memory type", | |
2988 | {"dpdk-alloc-mem", "dpdk-socket-mem", NULL,}, | |
2989 | {"-m", "--socket-mem", NULL,}, | |
2990 | "1024,0", 1 | |
2991 | }, | |
2992 | }; | |
2993 | ||
2994 | int i, ret = initial_size; | |
2995 | for (i = 0; i < ARRAY_SIZE(excl_opts); ++i) { | |
2996 | int found_opts = 0, scan, found_pos = -1; | |
2997 | const char *found_value; | |
2998 | struct dpdk_exclusive_options_map *popt = &excl_opts[i]; | |
2999 | ||
3000 | for (scan = 0; scan < MAX_DPDK_EXCL_OPTS | |
3001 | && popt->ovs_dpdk_options[scan]; ++scan) { | |
3002 | const char *lookup = smap_get(ovs_other_config, | |
3003 | popt->ovs_dpdk_options[scan]); | |
3004 | if (lookup && strlen(lookup)) { | |
3005 | found_opts++; | |
3006 | found_pos = scan; | |
3007 | found_value = lookup; | |
3008 | } | |
3009 | } | |
3010 | ||
3011 | if (!found_opts) { | |
3012 | if (popt->default_option) { | |
3013 | found_pos = popt->default_option; | |
3014 | found_value = popt->default_value; | |
3015 | } else { | |
3016 | continue; | |
3017 | } | |
dfc89a55 | 3018 | } |
bab69409 AC |
3019 | |
3020 | if (found_opts > 1) { | |
3021 | VLOG_ERR("Multiple defined options for %s. Please check your" | |
3022 | " database settings and reconfigure if necessary.", | |
3023 | popt->category); | |
3024 | } | |
3025 | ||
52a57e36 AC |
3026 | if (!argv_contains(extra_args, extra_argc, |
3027 | popt->eal_dpdk_options[found_pos])) { | |
3028 | dpdk_option_extend(argv, ret, popt->eal_dpdk_options[found_pos], | |
3029 | found_value); | |
3030 | ret += 2; | |
3031 | } else { | |
3032 | VLOG_WARN("Ignoring database defined option '%s' due to " | |
3033 | "dpdk_extras config", popt->eal_dpdk_options[found_pos]); | |
3034 | } | |
bab69409 AC |
3035 | } |
3036 | ||
3037 | return ret; | |
3038 | } | |
3039 | ||
3040 | static int | |
88964e64 AC |
3041 | get_dpdk_args(const struct smap *ovs_other_config, char ***argv, |
3042 | int argc) | |
bab69409 | 3043 | { |
eac84432 | 3044 | const char *extra_configuration; |
52a57e36 AC |
3045 | char **extra_args = NULL; |
3046 | int i; | |
3047 | size_t extra_argc = 0; | |
eac84432 AC |
3048 | |
3049 | extra_configuration = smap_get(ovs_other_config, "dpdk-extra"); | |
3050 | if (extra_configuration) { | |
52a57e36 | 3051 | extra_argc = extra_dpdk_args(extra_configuration, &extra_args, 0); |
eac84432 | 3052 | } |
52a57e36 AC |
3053 | |
3054 | i = construct_dpdk_options(ovs_other_config, argv, argc, extra_args, | |
3055 | extra_argc); | |
3056 | i = construct_dpdk_mutex_options(ovs_other_config, argv, i, extra_args, | |
3057 | extra_argc); | |
3058 | ||
3059 | if (extra_configuration) { | |
3060 | *argv = move_argv(argv, i, extra_args, extra_argc); | |
3061 | } | |
3062 | ||
3063 | return i + extra_argc; | |
bab69409 AC |
3064 | } |
3065 | ||
3066 | static char **dpdk_argv; | |
3067 | static int dpdk_argc; | |
3068 | ||
3069 | static void | |
3070 | deferred_argv_release(void) | |
3071 | { | |
3072 | int result; | |
3073 | for (result = 0; result < dpdk_argc; ++result) { | |
3074 | free(dpdk_argv[result]); | |
3075 | } | |
3076 | ||
3077 | free(dpdk_argv); | |
3078 | } | |
3079 | ||
3080 | static void | |
3081 | dpdk_init__(const struct smap *ovs_other_config) | |
3082 | { | |
3083 | char **argv = NULL; | |
3084 | int result; | |
88964e64 AC |
3085 | int argc, argc_tmp; |
3086 | bool auto_determine = true; | |
bab69409 AC |
3087 | int err; |
3088 | cpu_set_t cpuset; | |
d8a8f353 AC |
3089 | #ifndef VHOST_CUSE |
3090 | char *sock_dir_subcomponent; | |
3091 | #endif | |
bab69409 AC |
3092 | |
3093 | if (!smap_get_bool(ovs_other_config, "dpdk-init", false)) { | |
3094 | VLOG_INFO("DPDK Disabled - to change this requires a restart.\n"); | |
3095 | return; | |
dfc89a55 AZ |
3096 | } |
3097 | ||
bab69409 AC |
3098 | VLOG_INFO("DPDK Enabled, initializing"); |
3099 | ||
7d1ced01 | 3100 | #ifdef VHOST_CUSE |
bab69409 AC |
3101 | if (process_vhost_flags("cuse-dev-name", xstrdup("vhost-net"), |
3102 | PATH_MAX, ovs_other_config, &cuse_dev_name)) { | |
7d1ced01 | 3103 | #else |
6eb51f9a | 3104 | if (process_vhost_flags("vhost-sock-dir", xstrdup(ovs_rundir()), |
d8a8f353 AC |
3105 | NAME_MAX, ovs_other_config, |
3106 | &sock_dir_subcomponent)) { | |
7d1ced01 | 3107 | struct stat s; |
d8a8f353 AC |
3108 | if (!strstr(sock_dir_subcomponent, "..")) { |
3109 | vhost_sock_dir = xasprintf("%s/%s", ovs_rundir(), | |
3110 | sock_dir_subcomponent); | |
3111 | ||
3112 | err = stat(vhost_sock_dir, &s); | |
3113 | if (err) { | |
3114 | VLOG_ERR("vhost-user sock directory '%s' does not exist.", | |
3115 | vhost_sock_dir); | |
3116 | } | |
3117 | } else { | |
3118 | vhost_sock_dir = xstrdup(ovs_rundir()); | |
3119 | VLOG_ERR("vhost-user sock directory request '%s/%s' has invalid" | |
3120 | "characters '..' - using %s instead.", | |
3121 | ovs_rundir(), sock_dir_subcomponent, ovs_rundir()); | |
7d1ced01 | 3122 | } |
d8a8f353 AC |
3123 | free(sock_dir_subcomponent); |
3124 | } else { | |
6eb51f9a | 3125 | vhost_sock_dir = sock_dir_subcomponent; |
7d1ced01 | 3126 | #endif |
58397e6c KT |
3127 | } |
3128 | ||
563c98d8 AC |
3129 | /* Get the main thread affinity */ |
3130 | CPU_ZERO(&cpuset); | |
bab69409 AC |
3131 | err = pthread_getaffinity_np(pthread_self(), sizeof(cpu_set_t), |
3132 | &cpuset); | |
563c98d8 AC |
3133 | if (err) { |
3134 | VLOG_ERR("Thread getaffinity error %d.", err); | |
563c98d8 AC |
3135 | } |
3136 | ||
bab69409 | 3137 | argv = grow_argv(&argv, 0, 1); |
88964e64 | 3138 | argc = 1; |
bab69409 | 3139 | argv[0] = xstrdup(ovs_get_program_name()); |
88964e64 AC |
3140 | argc_tmp = get_dpdk_args(ovs_other_config, &argv, argc); |
3141 | ||
3142 | while (argc_tmp != argc) { | |
3143 | if (!strcmp("-c", argv[argc]) || !strcmp("-l", argv[argc])) { | |
3144 | auto_determine = false; | |
3145 | break; | |
3146 | } | |
3147 | argc++; | |
3148 | } | |
3149 | argc = argc_tmp; | |
3150 | ||
3151 | /** | |
3152 | * NOTE: This is an unsophisticated mechanism for determining the DPDK | |
3153 | * lcore for the DPDK Master. | |
3154 | */ | |
3155 | if (auto_determine) { | |
3156 | int i; | |
3157 | for (i = 0; i < CPU_SETSIZE; i++) { | |
3158 | if (CPU_ISSET(i, &cpuset)) { | |
3159 | argv = grow_argv(&argv, argc, 2); | |
3160 | argv[argc++] = xstrdup("-c"); | |
3161 | argv[argc++] = xasprintf("0x%08llX", (1ULL<<i)); | |
3162 | i = CPU_SETSIZE; | |
3163 | } | |
3164 | } | |
3165 | } | |
bab69409 AC |
3166 | |
3167 | argv = grow_argv(&argv, argc, 1); | |
3168 | argv[argc] = NULL; | |
3169 | ||
3170 | optind = 1; | |
58397e6c | 3171 | |
eac84432 AC |
3172 | if (VLOG_IS_INFO_ENABLED()) { |
3173 | struct ds eal_args; | |
3174 | int opt; | |
3175 | ds_init(&eal_args); | |
3176 | ds_put_cstr(&eal_args, "EAL ARGS:"); | |
3177 | for (opt = 0; opt < argc; ++opt) { | |
3178 | ds_put_cstr(&eal_args, " "); | |
3179 | ds_put_cstr(&eal_args, argv[opt]); | |
3180 | } | |
3181 | VLOG_INFO("%s", ds_cstr_ro(&eal_args)); | |
3182 | ds_destroy(&eal_args); | |
3183 | } | |
3184 | ||
8a9562d2 PS |
3185 | /* Make sure things are initialized ... */ |
3186 | result = rte_eal_init(argc, argv); | |
451450fa | 3187 | if (result < 0) { |
58397e6c | 3188 | ovs_abort(result, "Cannot init EAL"); |
451450fa | 3189 | } |
8a9562d2 | 3190 | |
563c98d8 | 3191 | /* Set the main thread affinity back to pre rte_eal_init() value */ |
30149e29 | 3192 | if (auto_determine) { |
bab69409 AC |
3193 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), |
3194 | &cpuset); | |
3195 | if (err) { | |
3196 | VLOG_ERR("Thread setaffinity error %d", err); | |
3197 | } | |
563c98d8 AC |
3198 | } |
3199 | ||
bab69409 AC |
3200 | dpdk_argv = argv; |
3201 | dpdk_argc = argc; | |
3202 | ||
3203 | atexit(deferred_argv_release); | |
3204 | ||
d7310583 | 3205 | rte_memzone_dump(stdout); |
8a9562d2 PS |
3206 | rte_eal_init_ret = 0; |
3207 | ||
db73f716 | 3208 | /* We are called from the main thread here */ |
d5c199ea | 3209 | RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID; |
db73f716 | 3210 | |
bab69409 AC |
3211 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); |
3212 | ||
3213 | #ifdef VHOST_CUSE | |
3214 | /* Register CUSE device to handle IOCTLs. | |
3215 | * Unless otherwise specified, cuse_dev_name is set to vhost-net. | |
3216 | */ | |
3217 | err = rte_vhost_driver_register(cuse_dev_name); | |
3218 | ||
3219 | if (err != 0) { | |
3220 | VLOG_ERR("CUSE device setup failure."); | |
3221 | return; | |
3222 | } | |
3223 | #endif | |
3224 | ||
3225 | dpdk_vhost_class_init(); | |
3226 | ||
3227 | /* Finally, register the dpdk classes */ | |
3228 | netdev_dpdk_register(); | |
3229 | } | |
3230 | ||
3231 | void | |
3232 | dpdk_init(const struct smap *ovs_other_config) | |
3233 | { | |
3234 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
3235 | ||
3236 | if (ovs_other_config && ovsthread_once_start(&once)) { | |
3237 | dpdk_init__(ovs_other_config); | |
3238 | ovsthread_once_done(&once); | |
3239 | } | |
8a9562d2 PS |
3240 | } |
3241 | ||
bce01e3a | 3242 | static const struct netdev_class dpdk_class = |
95fb793a | 3243 | NETDEV_DPDK_CLASS( |
3244 | "dpdk", | |
b8e57534 | 3245 | NULL, |
5496878c | 3246 | netdev_dpdk_construct, |
58397e6c | 3247 | netdev_dpdk_destruct, |
7251515e | 3248 | netdev_dpdk_set_multiq, |
58397e6c KT |
3249 | netdev_dpdk_eth_send, |
3250 | netdev_dpdk_get_carrier, | |
3251 | netdev_dpdk_get_stats, | |
3252 | netdev_dpdk_get_features, | |
3253 | netdev_dpdk_get_status, | |
3254 | netdev_dpdk_rxq_recv); | |
95fb793a | 3255 | |
bce01e3a | 3256 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3257 | NETDEV_DPDK_CLASS( |
3258 | "dpdkr", | |
033e9df2 | 3259 | NULL, |
5496878c | 3260 | netdev_dpdk_ring_construct, |
58397e6c | 3261 | netdev_dpdk_destruct, |
a0cb2d66 | 3262 | netdev_dpdk_set_multiq, |
58397e6c KT |
3263 | netdev_dpdk_ring_send, |
3264 | netdev_dpdk_get_carrier, | |
3265 | netdev_dpdk_get_stats, | |
3266 | netdev_dpdk_get_features, | |
3267 | netdev_dpdk_get_status, | |
3268 | netdev_dpdk_rxq_recv); | |
3269 | ||
bd8baf47 | 3270 | static const struct netdev_class OVS_UNUSED dpdk_vhost_cuse_class = |
58397e6c | 3271 | NETDEV_DPDK_CLASS( |
7d1ced01 CL |
3272 | "dpdkvhostcuse", |
3273 | dpdk_vhost_cuse_class_init, | |
3274 | netdev_dpdk_vhost_cuse_construct, | |
3275 | netdev_dpdk_vhost_destruct, | |
4573fbd3 | 3276 | netdev_dpdk_vhost_cuse_set_multiq, |
7d1ced01 CL |
3277 | netdev_dpdk_vhost_send, |
3278 | netdev_dpdk_vhost_get_carrier, | |
3279 | netdev_dpdk_vhost_get_stats, | |
3280 | NULL, | |
3281 | NULL, | |
3282 | netdev_dpdk_vhost_rxq_recv); | |
3283 | ||
bd8baf47 | 3284 | static const struct netdev_class OVS_UNUSED dpdk_vhost_user_class = |
7d1ced01 CL |
3285 | NETDEV_DPDK_CLASS( |
3286 | "dpdkvhostuser", | |
3287 | dpdk_vhost_user_class_init, | |
3288 | netdev_dpdk_vhost_user_construct, | |
58397e6c KT |
3289 | netdev_dpdk_vhost_destruct, |
3290 | netdev_dpdk_vhost_set_multiq, | |
3291 | netdev_dpdk_vhost_send, | |
3292 | netdev_dpdk_vhost_get_carrier, | |
3293 | netdev_dpdk_vhost_get_stats, | |
3294 | NULL, | |
7251515e | 3295 | NULL, |
58397e6c | 3296 | netdev_dpdk_vhost_rxq_recv); |
95fb793a | 3297 | |
8a9562d2 PS |
3298 | void |
3299 | netdev_dpdk_register(void) | |
3300 | { | |
bab69409 AC |
3301 | dpdk_common_init(); |
3302 | netdev_register_provider(&dpdk_class); | |
3303 | netdev_register_provider(&dpdk_ring_class); | |
7d1ced01 | 3304 | #ifdef VHOST_CUSE |
bab69409 | 3305 | netdev_register_provider(&dpdk_vhost_cuse_class); |
7d1ced01 | 3306 | #else |
bab69409 | 3307 | netdev_register_provider(&dpdk_vhost_user_class); |
7d1ced01 | 3308 | #endif |
8a9562d2 | 3309 | } |
8617afff PS |
3310 | |
3311 | int | |
bd5131ba | 3312 | pmd_thread_setaffinity_cpu(unsigned cpu) |
8617afff PS |
3313 | { |
3314 | cpu_set_t cpuset; | |
3315 | int err; | |
3316 | ||
3317 | CPU_ZERO(&cpuset); | |
3318 | CPU_SET(cpu, &cpuset); | |
3319 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &cpuset); | |
3320 | if (err) { | |
3321 | VLOG_ERR("Thread affinity error %d",err); | |
3322 | return err; | |
3323 | } | |
abb5943d AW |
3324 | /* NON_PMD_CORE_ID is reserved for use by non pmd threads. */ |
3325 | ovs_assert(cpu != NON_PMD_CORE_ID); | |
65f13b50 | 3326 | RTE_PER_LCORE(_lcore_id) = cpu; |
8617afff PS |
3327 | |
3328 | return 0; | |
3329 | } | |
db73f716 | 3330 | |
db73f716 | 3331 | static bool |
5f17de68 | 3332 | dpdk_thread_is_pmd(void) |
db73f716 | 3333 | { |
abb5943d | 3334 | return rte_lcore_id() != NON_PMD_CORE_ID; |
db73f716 | 3335 | } |