]>
Commit | Line | Data |
---|---|---|
8a9562d2 | 1 | /* |
0d4af148 | 2 | * Copyright (c) 2014, 2015, 2016 Nicira, Inc. |
8a9562d2 PS |
3 | * |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
8a9562d2 PS |
19 | #include <string.h> |
20 | #include <signal.h> | |
21 | #include <stdlib.h> | |
22 | #include <pthread.h> | |
23 | #include <config.h> | |
24 | #include <errno.h> | |
25 | #include <sched.h> | |
26 | #include <stdlib.h> | |
27 | #include <unistd.h> | |
7d1ced01 | 28 | #include <sys/stat.h> |
8a9562d2 | 29 | #include <stdio.h> |
7d1ced01 CL |
30 | #include <sys/types.h> |
31 | #include <sys/stat.h> | |
bab69409 | 32 | #include <getopt.h> |
8a9562d2 | 33 | |
7d1ced01 | 34 | #include "dirs.h" |
e14deea0 | 35 | #include "dp-packet.h" |
8a9562d2 | 36 | #include "dpif-netdev.h" |
e5c0f5a4 | 37 | #include "fatal-signal.h" |
8a9562d2 PS |
38 | #include "netdev-dpdk.h" |
39 | #include "netdev-provider.h" | |
40 | #include "netdev-vport.h" | |
41 | #include "odp-util.h" | |
eac84432 | 42 | #include "openvswitch/dynamic-string.h" |
25d436fb BW |
43 | #include "openvswitch/list.h" |
44 | #include "openvswitch/ofp-print.h" | |
45 | #include "openvswitch/vlog.h" | |
94143fc4 | 46 | #include "ovs-numa.h" |
8a9562d2 PS |
47 | #include "ovs-thread.h" |
48 | #include "ovs-rcu.h" | |
49 | #include "packets.h" | |
ee89ea7b | 50 | #include "openvswitch/shash.h" |
0bf765f7 | 51 | #include "smap.h" |
8a9562d2 PS |
52 | #include "sset.h" |
53 | #include "unaligned.h" | |
54 | #include "timeval.h" | |
55 | #include "unixctl.h" | |
8a9562d2 | 56 | |
b8e57534 MK |
57 | #include "rte_config.h" |
58 | #include "rte_mbuf.h" | |
0bf765f7 | 59 | #include "rte_meter.h" |
4b88d678 CL |
60 | #ifdef DPDK_PDUMP |
61 | #include "rte_pdump.h" | |
62 | #endif | |
58397e6c | 63 | #include "rte_virtio_net.h" |
b8e57534 | 64 | |
8a9562d2 PS |
65 | VLOG_DEFINE_THIS_MODULE(dpdk); |
66 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); | |
67 | ||
68 | #define DPDK_PORT_WATCHDOG_INTERVAL 5 | |
69 | ||
70 | #define OVS_CACHE_LINE_SIZE CACHE_LINE_SIZE | |
71 | #define OVS_VPORT_DPDK "ovs_dpdk" | |
72 | ||
73 | /* | |
74 | * need to reserve tons of extra space in the mbufs so we can align the | |
75 | * DMA addresses to 4KB. | |
18f777b2 TP |
76 | * The minimum mbuf size is limited to avoid scatter behaviour and drop in |
77 | * performance for standard Ethernet MTU. | |
8a9562d2 | 78 | */ |
4be4d22c MK |
79 | #define ETHER_HDR_MAX_LEN (ETHER_HDR_LEN + ETHER_CRC_LEN + (2 * VLAN_HEADER_LEN)) |
80 | #define MTU_TO_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_LEN + ETHER_CRC_LEN) | |
81 | #define MTU_TO_MAX_FRAME_LEN(mtu) ((mtu) + ETHER_HDR_MAX_LEN) | |
82 | #define FRAME_LEN_TO_MTU(frame_len) ((frame_len)- ETHER_HDR_LEN - ETHER_CRC_LEN) | |
83 | #define MBUF_SIZE(mtu) ( MTU_TO_MAX_FRAME_LEN(mtu) \ | |
84 | + sizeof(struct dp_packet) \ | |
85 | + RTE_PKTMBUF_HEADROOM) | |
86 | #define NETDEV_DPDK_MBUF_ALIGN 1024 | |
0072e931 | 87 | #define NETDEV_DPDK_MAX_PKT_LEN 9728 |
8a9562d2 | 88 | |
da79ce2b DDP |
89 | /* Max and min number of packets in the mempool. OVS tries to allocate a |
90 | * mempool with MAX_NB_MBUF: if this fails (because the system doesn't have | |
91 | * enough hugepages) we keep halving the number until the allocation succeeds | |
92 | * or we reach MIN_NB_MBUF */ | |
93 | ||
94 | #define MAX_NB_MBUF (4096 * 64) | |
95 | #define MIN_NB_MBUF (4096 * 4) | |
96 | #define MP_CACHE_SZ RTE_MEMPOOL_CACHE_MAX_SIZE | |
97 | ||
98 | /* MAX_NB_MBUF can be divided by 2 many times, until MIN_NB_MBUF */ | |
99 | BUILD_ASSERT_DECL(MAX_NB_MBUF % ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF) == 0); | |
100 | ||
101 | /* The smallest possible NB_MBUF that we're going to try should be a multiple | |
102 | * of MP_CACHE_SZ. This is advised by DPDK documentation. */ | |
103 | BUILD_ASSERT_DECL((MAX_NB_MBUF / ROUND_DOWN_POW2(MAX_NB_MBUF/MIN_NB_MBUF)) | |
104 | % MP_CACHE_SZ == 0); | |
105 | ||
d6e3feb5 | 106 | /* |
107 | * DPDK XSTATS Counter names definition | |
108 | */ | |
109 | #define XSTAT_RX_64_PACKETS "rx_size_64_packets" | |
110 | #define XSTAT_RX_65_TO_127_PACKETS "rx_size_65_to_127_packets" | |
111 | #define XSTAT_RX_128_TO_255_PACKETS "rx_size_128_to_255_packets" | |
112 | #define XSTAT_RX_256_TO_511_PACKETS "rx_size_256_to_511_packets" | |
113 | #define XSTAT_RX_512_TO_1023_PACKETS "rx_size_512_to_1023_packets" | |
114 | #define XSTAT_RX_1024_TO_1522_PACKETS "rx_size_1024_to_1522_packets" | |
115 | #define XSTAT_RX_1523_TO_MAX_PACKETS "rx_size_1523_to_max_packets" | |
116 | ||
117 | #define XSTAT_TX_64_PACKETS "tx_size_64_packets" | |
118 | #define XSTAT_TX_65_TO_127_PACKETS "tx_size_65_to_127_packets" | |
119 | #define XSTAT_TX_128_TO_255_PACKETS "tx_size_128_to_255_packets" | |
120 | #define XSTAT_TX_256_TO_511_PACKETS "tx_size_256_to_511_packets" | |
121 | #define XSTAT_TX_512_TO_1023_PACKETS "tx_size_512_to_1023_packets" | |
122 | #define XSTAT_TX_1024_TO_1522_PACKETS "tx_size_1024_to_1522_packets" | |
123 | #define XSTAT_TX_1523_TO_MAX_PACKETS "tx_size_1523_to_max_packets" | |
124 | ||
125 | #define XSTAT_TX_MULTICAST_PACKETS "tx_multicast_packets" | |
126 | #define XSTAT_RX_BROADCAST_PACKETS "rx_broadcast_packets" | |
127 | #define XSTAT_TX_BROADCAST_PACKETS "tx_broadcast_packets" | |
128 | #define XSTAT_RX_UNDERSIZED_ERRORS "rx_undersized_errors" | |
129 | #define XSTAT_RX_OVERSIZE_ERRORS "rx_oversize_errors" | |
130 | #define XSTAT_RX_FRAGMENTED_ERRORS "rx_fragmented_errors" | |
131 | #define XSTAT_RX_JABBER_ERRORS "rx_jabber_errors" | |
132 | ||
8a9562d2 PS |
133 | #define SOCKET0 0 |
134 | ||
79f5354c PM |
135 | #define NIC_PORT_RX_Q_SIZE 2048 /* Size of Physical NIC RX Queue, Max (n+32<=4096)*/ |
136 | #define NIC_PORT_TX_Q_SIZE 2048 /* Size of Physical NIC TX Queue, Max (n+32<=4096)*/ | |
137 | ||
585a5bea | 138 | #define OVS_VHOST_MAX_QUEUE_NUM 1024 /* Maximum number of vHost TX queues. */ |
f3ea2ad2 IM |
139 | #define OVS_VHOST_QUEUE_MAP_UNKNOWN (-1) /* Mapping not initialized. */ |
140 | #define OVS_VHOST_QUEUE_DISABLED (-2) /* Queue was disabled by guest and not | |
141 | * yet mapped to another queue. */ | |
585a5bea | 142 | |
bd8baf47 | 143 | static char *vhost_sock_dir = NULL; /* Location of vhost-user sockets */ |
58397e6c | 144 | |
31871ee3 | 145 | #define VHOST_ENQ_RETRY_NUM 8 |
0a0f39df | 146 | #define IF_NAME_SZ (PATH_MAX > IFNAMSIZ ? PATH_MAX : IFNAMSIZ) |
95e9881f | 147 | |
8a9562d2 | 148 | static const struct rte_eth_conf port_conf = { |
a28ddd11 DDP |
149 | .rxmode = { |
150 | .mq_mode = ETH_MQ_RX_RSS, | |
151 | .split_hdr_size = 0, | |
152 | .header_split = 0, /* Header Split disabled */ | |
153 | .hw_ip_checksum = 0, /* IP checksum offload disabled */ | |
154 | .hw_vlan_filter = 0, /* VLAN filtering disabled */ | |
155 | .jumbo_frame = 0, /* Jumbo Frame Support disabled */ | |
156 | .hw_strip_crc = 0, | |
157 | }, | |
158 | .rx_adv_conf = { | |
159 | .rss_conf = { | |
160 | .rss_key = NULL, | |
543342a4 | 161 | .rss_hf = ETH_RSS_IP | ETH_RSS_UDP | ETH_RSS_TCP, |
8a9562d2 | 162 | }, |
a28ddd11 DDP |
163 | }, |
164 | .txmode = { | |
165 | .mq_mode = ETH_MQ_TX_NONE, | |
166 | }, | |
8a9562d2 PS |
167 | }; |
168 | ||
58f7c37b DDP |
169 | enum { DPDK_RING_SIZE = 256 }; |
170 | BUILD_ASSERT_DECL(IS_POW2(DPDK_RING_SIZE)); | |
8a9562d2 PS |
171 | enum { DRAIN_TSC = 200000ULL }; |
172 | ||
58397e6c KT |
173 | enum dpdk_dev_type { |
174 | DPDK_DEV_ETH = 0, | |
7d1ced01 | 175 | DPDK_DEV_VHOST = 1, |
58397e6c KT |
176 | }; |
177 | ||
8a9562d2 PS |
178 | static int rte_eal_init_ret = ENODEV; |
179 | ||
180 | static struct ovs_mutex dpdk_mutex = OVS_MUTEX_INITIALIZER; | |
181 | ||
0bf765f7 IS |
182 | /* Quality of Service */ |
183 | ||
184 | /* An instance of a QoS configuration. Always associated with a particular | |
185 | * network device. | |
186 | * | |
187 | * Each QoS implementation subclasses this with whatever additional data it | |
188 | * needs. | |
189 | */ | |
190 | struct qos_conf { | |
191 | const struct dpdk_qos_ops *ops; | |
192 | }; | |
193 | ||
194 | /* A particular implementation of dpdk QoS operations. | |
195 | * | |
196 | * The functions below return 0 if successful or a positive errno value on | |
197 | * failure, except where otherwise noted. All of them must be provided, except | |
198 | * where otherwise noted. | |
199 | */ | |
200 | struct dpdk_qos_ops { | |
201 | ||
202 | /* Name of the QoS type */ | |
203 | const char *qos_name; | |
204 | ||
205 | /* Called to construct the QoS implementation on 'netdev'. The | |
206 | * implementation should make the appropriate calls to configure QoS | |
207 | * according to 'details'. The implementation may assume that any current | |
208 | * QoS configuration already installed should be destroyed before | |
209 | * constructing the new configuration. | |
210 | * | |
211 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
212 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
213 | * (which is built as ovs-vswitchd.conf.db(8)). | |
214 | * | |
215 | * This function must return 0 if and only if it sets 'netdev->qos_conf' | |
216 | * to an initialized 'struct qos_conf'. | |
217 | * | |
218 | * For all QoS implementations it should always be non-null. | |
219 | */ | |
220 | int (*qos_construct)(struct netdev *netdev, const struct smap *details); | |
221 | ||
222 | /* Destroys the data structures allocated by the implementation as part of | |
223 | * 'qos_conf. | |
224 | * | |
225 | * For all QoS implementations it should always be non-null. | |
226 | */ | |
227 | void (*qos_destruct)(struct netdev *netdev, struct qos_conf *conf); | |
228 | ||
229 | /* Retrieves details of 'netdev->qos_conf' configuration into 'details'. | |
230 | * | |
231 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
232 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
233 | * (which is built as ovs-vswitchd.conf.db(8)). | |
234 | */ | |
235 | int (*qos_get)(const struct netdev *netdev, struct smap *details); | |
236 | ||
237 | /* Reconfigures 'netdev->qos_conf' according to 'details', performing any | |
238 | * required calls to complete the reconfiguration. | |
239 | * | |
240 | * The contents of 'details' should be documented as valid for 'ovs_name' | |
241 | * in the "other_config" column in the "QoS" table in vswitchd/vswitch.xml | |
242 | * (which is built as ovs-vswitchd.conf.db(8)). | |
243 | * | |
244 | * This function may be null if 'qos_conf' is not configurable. | |
245 | */ | |
246 | int (*qos_set)(struct netdev *netdev, const struct smap *details); | |
247 | ||
248 | /* Modify an array of rte_mbufs. The modification is specific to | |
249 | * each qos implementation. | |
250 | * | |
251 | * The function should take and array of mbufs and an int representing | |
252 | * the current number of mbufs present in the array. | |
253 | * | |
254 | * After the function has performed a qos modification to the array of | |
255 | * mbufs it returns an int representing the number of mbufs now present in | |
256 | * the array. This value is can then be passed to the port send function | |
257 | * along with the modified array for transmission. | |
258 | * | |
259 | * For all QoS implementations it should always be non-null. | |
260 | */ | |
261 | int (*qos_run)(struct netdev *netdev, struct rte_mbuf **pkts, | |
262 | int pkt_cnt); | |
263 | }; | |
264 | ||
265 | /* dpdk_qos_ops for each type of user space QoS implementation */ | |
266 | static const struct dpdk_qos_ops egress_policer_ops; | |
267 | ||
268 | /* | |
269 | * Array of dpdk_qos_ops, contains pointer to all supported QoS | |
270 | * operations. | |
271 | */ | |
272 | static const struct dpdk_qos_ops *const qos_confs[] = { | |
273 | &egress_policer_ops, | |
274 | NULL | |
275 | }; | |
276 | ||
8a9562d2 | 277 | /* Contains all 'struct dpdk_dev's. */ |
ca6ba700 | 278 | static struct ovs_list dpdk_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 279 | = OVS_LIST_INITIALIZER(&dpdk_list); |
8a9562d2 | 280 | |
ca6ba700 | 281 | static struct ovs_list dpdk_mp_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 282 | = OVS_LIST_INITIALIZER(&dpdk_mp_list); |
8a9562d2 | 283 | |
db73f716 | 284 | /* This mutex must be used by non pmd threads when allocating or freeing |
b59cc14e | 285 | * mbufs through mempools. */ |
bce01e3a | 286 | static struct ovs_mutex nonpmd_mempool_mutex = OVS_MUTEX_INITIALIZER; |
db73f716 | 287 | |
8a9562d2 PS |
288 | struct dpdk_mp { |
289 | struct rte_mempool *mp; | |
290 | int mtu; | |
291 | int socket_id; | |
292 | int refcount; | |
ca6ba700 | 293 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
8a9562d2 PS |
294 | }; |
295 | ||
5a034064 AW |
296 | /* There should be one 'struct dpdk_tx_queue' created for |
297 | * each cpu core. */ | |
8a9562d2 | 298 | struct dpdk_tx_queue { |
a0cb2d66 DDP |
299 | rte_spinlock_t tx_lock; /* Protects the members and the NIC queue |
300 | * from concurrent access. It is used only | |
301 | * if the queue is shared among different | |
324c8374 | 302 | * pmd threads (see 'concurrent_txq'). */ |
585a5bea IM |
303 | int map; /* Mapping of configured vhost-user queues |
304 | * to enabled by guest. */ | |
8a9562d2 PS |
305 | }; |
306 | ||
95fb793a | 307 | /* dpdk has no way to remove dpdk ring ethernet devices |
308 | so we have to keep them around once they've been created | |
309 | */ | |
310 | ||
ca6ba700 | 311 | static struct ovs_list dpdk_ring_list OVS_GUARDED_BY(dpdk_mutex) |
55951e15 | 312 | = OVS_LIST_INITIALIZER(&dpdk_ring_list); |
95fb793a | 313 | |
314 | struct dpdk_ring { | |
315 | /* For the client rings */ | |
316 | struct rte_ring *cring_tx; | |
317 | struct rte_ring *cring_rx; | |
b83a2df1 | 318 | unsigned int user_port_id; /* User given port no, parsed from port name */ |
95fb793a | 319 | int eth_port_id; /* ethernet device port id */ |
ca6ba700 | 320 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
95fb793a | 321 | }; |
322 | ||
9509913a IS |
323 | struct ingress_policer { |
324 | struct rte_meter_srtcm_params app_srtcm_params; | |
325 | struct rte_meter_srtcm in_policer; | |
326 | rte_spinlock_t policer_lock; | |
327 | }; | |
328 | ||
8a9562d2 PS |
329 | struct netdev_dpdk { |
330 | struct netdev up; | |
331 | int port_id; | |
332 | int max_packet_len; | |
58397e6c | 333 | enum dpdk_dev_type type; |
8a9562d2 | 334 | |
5a034064 | 335 | struct dpdk_tx_queue *tx_q; |
8a9562d2 PS |
336 | |
337 | struct ovs_mutex mutex OVS_ACQ_AFTER(dpdk_mutex); | |
338 | ||
339 | struct dpdk_mp *dpdk_mp; | |
340 | int mtu; | |
341 | int socket_id; | |
342 | int buf_size; | |
8a9562d2 | 343 | struct netdev_stats stats; |
45d947c4 DDP |
344 | /* Protects stats */ |
345 | rte_spinlock_t stats_lock; | |
8a9562d2 | 346 | |
74ff3298 | 347 | struct eth_addr hwaddr; |
8a9562d2 PS |
348 | enum netdev_flags flags; |
349 | ||
350 | struct rte_eth_link link; | |
351 | int link_reset_cnt; | |
352 | ||
0a0f39df CL |
353 | /* virtio identifier for vhost devices */ |
354 | ovsrcu_index vid; | |
355 | ||
356 | /* True if vHost device is 'up' and has been reconfigured at least once */ | |
357 | bool vhost_reconfigured; | |
58397e6c | 358 | |
c1ff66ac CL |
359 | /* Identifiers used to distinguish vhost devices from each other. */ |
360 | char vhost_server_id[PATH_MAX]; | |
361 | char vhost_client_id[PATH_MAX]; | |
7d1ced01 | 362 | |
8a9562d2 | 363 | /* In dpdk_list. */ |
ca6ba700 | 364 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
0bf765f7 IS |
365 | |
366 | /* QoS configuration and lock for the device */ | |
367 | struct qos_conf *qos_conf; | |
368 | rte_spinlock_t qos_lock; | |
369 | ||
050c60bf DDP |
370 | /* The following properties cannot be changed when a device is running, |
371 | * so we remember the request and update them next time | |
372 | * netdev_dpdk*_reconfigure() is called */ | |
0072e931 | 373 | int requested_mtu; |
050c60bf DDP |
374 | int requested_n_txq; |
375 | int requested_n_rxq; | |
9509913a | 376 | |
db8f13b0 CL |
377 | /* Socket ID detected when vHost device is brought up */ |
378 | int requested_socket_id; | |
379 | ||
c1ff66ac CL |
380 | /* Denotes whether vHost port is client/server mode */ |
381 | uint64_t vhost_driver_flags; | |
382 | ||
9509913a IS |
383 | /* Ingress Policer */ |
384 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
385 | uint32_t policer_rate; | |
386 | uint32_t policer_burst; | |
9fd39370 SC |
387 | |
388 | /* DPDK-ETH Flow control */ | |
389 | struct rte_eth_fc_conf fc_conf; | |
8a9562d2 PS |
390 | }; |
391 | ||
392 | struct netdev_rxq_dpdk { | |
393 | struct netdev_rxq up; | |
394 | int port_id; | |
395 | }; | |
396 | ||
5f17de68 | 397 | static bool dpdk_thread_is_pmd(void); |
db73f716 | 398 | |
8a9562d2 PS |
399 | static int netdev_dpdk_construct(struct netdev *); |
400 | ||
0a0f39df | 401 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 402 | |
9509913a IS |
403 | struct ingress_policer * |
404 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
405 | ||
8a9562d2 PS |
406 | static bool |
407 | is_dpdk_class(const struct netdev_class *class) | |
408 | { | |
409 | return class->construct == netdev_dpdk_construct; | |
410 | } | |
411 | ||
4be4d22c MK |
412 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
413 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
414 | * value, insufficient buffers are allocated to accomodate the packet in its | |
415 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
416 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
417 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
418 | * behaviour, which reduces performance. To prevent this, use a buffer size that | |
419 | * is closest to 'mtu', but which satisfies the aforementioned criteria. | |
420 | */ | |
421 | static uint32_t | |
422 | dpdk_buf_size(int mtu) | |
423 | { | |
424 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
425 | NETDEV_DPDK_MBUF_ALIGN); | |
426 | } | |
427 | ||
58397e6c KT |
428 | /* XXX: use dpdk malloc for entire OVS. in fact huge page should be used |
429 | * for all other segments data, bss and text. */ | |
8a9562d2 PS |
430 | |
431 | static void * | |
432 | dpdk_rte_mzalloc(size_t sz) | |
433 | { | |
434 | void *ptr; | |
435 | ||
436 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
437 | if (ptr == NULL) { | |
438 | out_of_memory(); | |
439 | } | |
440 | return ptr; | |
441 | } | |
442 | ||
db73f716 DDP |
443 | /* XXX this function should be called only by pmd threads (or by non pmd |
444 | * threads holding the nonpmd_mempool_mutex) */ | |
8a9562d2 | 445 | void |
e14deea0 | 446 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 447 | { |
db73f716 | 448 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 449 | |
b00b4a81 | 450 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
451 | } |
452 | ||
b3cd9f9d PS |
453 | static void |
454 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
455 | void *opaque_arg OVS_UNUSED, | |
456 | void *_m, | |
457 | unsigned i OVS_UNUSED) | |
458 | { | |
459 | struct rte_mbuf *m = _m; | |
460 | ||
4be4d22c | 461 | rte_pktmbuf_init(mp, opaque_arg, _m, i); |
b3cd9f9d | 462 | |
cf62fa4c | 463 | dp_packet_init_dpdk((struct dp_packet *) m, m->buf_len); |
b3cd9f9d PS |
464 | } |
465 | ||
8a9562d2 PS |
466 | static struct dpdk_mp * |
467 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
468 | { | |
469 | struct dpdk_mp *dmp = NULL; | |
470 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
da79ce2b | 471 | unsigned mp_size; |
4be4d22c | 472 | struct rte_pktmbuf_pool_private mbp_priv; |
8a9562d2 PS |
473 | |
474 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
475 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
476 | dmp->refcount++; | |
477 | return dmp; | |
478 | } | |
479 | } | |
480 | ||
481 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
482 | dmp->socket_id = socket_id; | |
483 | dmp->mtu = mtu; | |
484 | dmp->refcount = 1; | |
4be4d22c | 485 | mbp_priv.mbuf_data_room_size = MBUF_SIZE(mtu) - sizeof(struct dp_packet); |
0072e931 MK |
486 | mbp_priv.mbuf_priv_size = sizeof (struct dp_packet) |
487 | - sizeof (struct rte_mbuf); | |
488 | /* XXX: this is a really rough method of provisioning memory. | |
489 | * It's impossible to determine what the exact memory requirements are when | |
490 | * the number of ports and rxqs that utilize a particular mempool can change | |
491 | * dynamically at runtime. For the moment, use this rough heurisitic. | |
492 | */ | |
493 | if (mtu >= ETHER_MTU) { | |
494 | mp_size = MAX_NB_MBUF; | |
495 | } else { | |
496 | mp_size = MIN_NB_MBUF; | |
497 | } | |
8a9562d2 | 498 | |
da79ce2b DDP |
499 | do { |
500 | if (snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d_%d_%u", | |
501 | dmp->mtu, dmp->socket_id, mp_size) < 0) { | |
502 | return NULL; | |
503 | } | |
95fb793a | 504 | |
da79ce2b DDP |
505 | dmp->mp = rte_mempool_create(mp_name, mp_size, MBUF_SIZE(mtu), |
506 | MP_CACHE_SZ, | |
507 | sizeof(struct rte_pktmbuf_pool_private), | |
4be4d22c | 508 | rte_pktmbuf_pool_init, &mbp_priv, |
da79ce2b DDP |
509 | ovs_rte_pktmbuf_init, NULL, |
510 | socket_id, 0); | |
511 | } while (!dmp->mp && rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
8a9562d2 PS |
512 | |
513 | if (dmp->mp == NULL) { | |
514 | return NULL; | |
da79ce2b DDP |
515 | } else { |
516 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", mp_name, mp_size ); | |
8a9562d2 PS |
517 | } |
518 | ||
417e7e66 | 519 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); |
8a9562d2 PS |
520 | return dmp; |
521 | } | |
522 | ||
523 | static void | |
8d38823b | 524 | dpdk_mp_put(struct dpdk_mp *dmp) OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 525 | { |
8a9562d2 PS |
526 | if (!dmp) { |
527 | return; | |
528 | } | |
529 | ||
8d38823b | 530 | ovs_assert(dmp->refcount); |
8a9562d2 | 531 | |
8d38823b MK |
532 | if (!--dmp->refcount) { |
533 | ovs_list_remove(&dmp->list_node); | |
534 | rte_mempool_free(dmp->mp); | |
8a9562d2 | 535 | } |
8a9562d2 PS |
536 | } |
537 | ||
0072e931 MK |
538 | /* Tries to allocate new mempool on requested_socket_id with |
539 | * mbuf size corresponding to requested_mtu. | |
540 | * On success new configuration will be applied. | |
541 | * On error, device will be left unchanged. */ | |
542 | static int | |
543 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
544 | OVS_REQUIRES(dpdk_mutex) | |
545 | OVS_REQUIRES(dev->mutex) | |
546 | { | |
547 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
548 | struct dpdk_mp *mp; | |
549 | ||
550 | mp = dpdk_mp_get(dev->requested_socket_id, FRAME_LEN_TO_MTU(buf_size)); | |
551 | if (!mp) { | |
552 | VLOG_ERR("Insufficient memory to create memory pool for netdev " | |
553 | "%s, with MTU %d on socket %d\n", | |
554 | dev->up.name, dev->requested_mtu, dev->requested_socket_id); | |
555 | return ENOMEM; | |
556 | } else { | |
557 | dpdk_mp_put(dev->dpdk_mp); | |
558 | dev->dpdk_mp = mp; | |
559 | dev->mtu = dev->requested_mtu; | |
560 | dev->socket_id = dev->requested_socket_id; | |
561 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
562 | } | |
563 | ||
564 | return 0; | |
565 | } | |
566 | ||
8a9562d2 PS |
567 | static void |
568 | check_link_status(struct netdev_dpdk *dev) | |
569 | { | |
570 | struct rte_eth_link link; | |
571 | ||
572 | rte_eth_link_get_nowait(dev->port_id, &link); | |
573 | ||
574 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 575 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
576 | |
577 | dev->link_reset_cnt++; | |
578 | dev->link = link; | |
579 | if (dev->link.link_status) { | |
580 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
581 | dev->port_id, (unsigned)dev->link.link_speed, | |
582 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
583 | ("full-duplex") : ("half-duplex")); | |
584 | } else { | |
585 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
586 | } | |
587 | } | |
588 | } | |
589 | ||
590 | static void * | |
591 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
592 | { | |
593 | struct netdev_dpdk *dev; | |
594 | ||
595 | pthread_detach(pthread_self()); | |
596 | ||
597 | for (;;) { | |
598 | ovs_mutex_lock(&dpdk_mutex); | |
599 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
600 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
601 | if (dev->type == DPDK_DEV_ETH) { |
602 | check_link_status(dev); | |
603 | } | |
8a9562d2 PS |
604 | ovs_mutex_unlock(&dev->mutex); |
605 | } | |
606 | ovs_mutex_unlock(&dpdk_mutex); | |
607 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
608 | } | |
609 | ||
610 | return NULL; | |
611 | } | |
612 | ||
b98d7669 DDP |
613 | static int |
614 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
615 | { | |
616 | int diag = 0; | |
617 | int i; | |
0072e931 | 618 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 619 | |
0072e931 MK |
620 | if (dev->mtu > ETHER_MTU) { |
621 | conf.rxmode.jumbo_frame = 1; | |
622 | conf.rxmode.max_rx_pkt_len = dev->max_packet_len; | |
623 | } else { | |
624 | conf.rxmode.jumbo_frame = 0; | |
625 | conf.rxmode.max_rx_pkt_len = 0; | |
626 | } | |
b98d7669 DDP |
627 | /* A device may report more queues than it makes available (this has |
628 | * been observed for Intel xl710, which reserves some of them for | |
629 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
630 | * available. When this happens we can retry the configuration | |
631 | * and request less queues */ | |
632 | while (n_rxq && n_txq) { | |
633 | if (diag) { | |
634 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
635 | } | |
636 | ||
0072e931 | 637 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 638 | if (diag) { |
0072e931 MK |
639 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
640 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
641 | break; |
642 | } | |
643 | ||
644 | for (i = 0; i < n_txq; i++) { | |
645 | diag = rte_eth_tx_queue_setup(dev->port_id, i, NIC_PORT_TX_Q_SIZE, | |
646 | dev->socket_id, NULL); | |
647 | if (diag) { | |
648 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
649 | dev->up.name, i, rte_strerror(-diag)); | |
650 | break; | |
651 | } | |
652 | } | |
653 | ||
654 | if (i != n_txq) { | |
655 | /* Retry with less tx queues */ | |
656 | n_txq = i; | |
657 | continue; | |
658 | } | |
659 | ||
660 | for (i = 0; i < n_rxq; i++) { | |
661 | diag = rte_eth_rx_queue_setup(dev->port_id, i, NIC_PORT_RX_Q_SIZE, | |
662 | dev->socket_id, NULL, | |
663 | dev->dpdk_mp->mp); | |
664 | if (diag) { | |
665 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
666 | dev->up.name, i, rte_strerror(-diag)); | |
667 | break; | |
668 | } | |
669 | } | |
670 | ||
671 | if (i != n_rxq) { | |
672 | /* Retry with less rx queues */ | |
673 | n_rxq = i; | |
674 | continue; | |
675 | } | |
676 | ||
677 | dev->up.n_rxq = n_rxq; | |
81acebda | 678 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
679 | |
680 | return 0; | |
681 | } | |
682 | ||
683 | return diag; | |
684 | } | |
685 | ||
9fd39370 SC |
686 | static void |
687 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
688 | { | |
689 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
690 | VLOG_WARN("Failed to enable flow control on device %d", dev->port_id); | |
691 | } | |
692 | } | |
b98d7669 | 693 | |
8a9562d2 PS |
694 | static int |
695 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
696 | { | |
697 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 698 | struct rte_eth_dev_info info; |
8a9562d2 PS |
699 | struct ether_addr eth_addr; |
700 | int diag; | |
b98d7669 | 701 | int n_rxq, n_txq; |
8a9562d2 | 702 | |
7cd1261d | 703 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
95fb793a | 704 | return ENODEV; |
8a9562d2 PS |
705 | } |
706 | ||
a0cb2d66 | 707 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 708 | |
b98d7669 DDP |
709 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
710 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
711 | ||
712 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 713 | if (diag) { |
b98d7669 DDP |
714 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
715 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 716 | return -diag; |
8a9562d2 PS |
717 | } |
718 | ||
8a9562d2 PS |
719 | diag = rte_eth_dev_start(dev->port_id); |
720 | if (diag) { | |
b98d7669 DDP |
721 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
722 | rte_strerror(-diag)); | |
95fb793a | 723 | return -diag; |
8a9562d2 PS |
724 | } |
725 | ||
726 | rte_eth_promiscuous_enable(dev->port_id); | |
727 | rte_eth_allmulticast_enable(dev->port_id); | |
728 | ||
729 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
730 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
731 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
ca92d173 | 732 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 733 | |
ca92d173 | 734 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
735 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
736 | ||
737 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
738 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
739 | ||
740 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
9fd39370 SC |
741 | |
742 | /* Get the Flow control configuration for DPDK-ETH */ | |
743 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
744 | if (diag) { | |
745 | VLOG_DBG("cannot get flow control parameters on port=%d, err=%d", | |
746 | dev->port_id, diag); | |
747 | } | |
748 | ||
8a9562d2 PS |
749 | return 0; |
750 | } | |
751 | ||
752 | static struct netdev_dpdk * | |
753 | netdev_dpdk_cast(const struct netdev *netdev) | |
754 | { | |
755 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
756 | } | |
757 | ||
758 | static struct netdev * | |
759 | netdev_dpdk_alloc(void) | |
760 | { | |
bab69409 AC |
761 | struct netdev_dpdk *dev; |
762 | ||
763 | if (!rte_eal_init_ret) { /* Only after successful initialization */ | |
764 | dev = dpdk_rte_mzalloc(sizeof *dev); | |
765 | if (dev) { | |
766 | return &dev->up; | |
767 | } | |
768 | } | |
769 | return NULL; | |
8a9562d2 PS |
770 | } |
771 | ||
5a034064 | 772 | static void |
d46285a2 | 773 | netdev_dpdk_alloc_txq(struct netdev_dpdk *dev, unsigned int n_txqs) |
5a034064 | 774 | { |
bd5131ba | 775 | unsigned i; |
5a034064 | 776 | |
d46285a2 | 777 | dev->tx_q = dpdk_rte_mzalloc(n_txqs * sizeof *dev->tx_q); |
5a034064 | 778 | for (i = 0; i < n_txqs; i++) { |
585a5bea | 779 | /* Initialize map for vhost devices. */ |
d46285a2 DDP |
780 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
781 | rte_spinlock_init(&dev->tx_q[i].tx_lock); | |
5a034064 AW |
782 | } |
783 | } | |
784 | ||
8a9562d2 | 785 | static int |
d46285a2 | 786 | netdev_dpdk_init(struct netdev *netdev, unsigned int port_no, |
58397e6c | 787 | enum dpdk_dev_type type) |
5a034064 | 788 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 789 | { |
d46285a2 | 790 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b7a04e0 | 791 | int sid; |
95fb793a | 792 | int err = 0; |
8a9562d2 | 793 | |
d46285a2 DDP |
794 | ovs_mutex_init(&dev->mutex); |
795 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 796 | |
d46285a2 | 797 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 798 | |
1b7a04e0 AW |
799 | /* If the 'sid' is negative, it means that the kernel fails |
800 | * to obtain the pci numa info. In that situation, always | |
801 | * use 'SOCKET0'. */ | |
58397e6c KT |
802 | if (type == DPDK_DEV_ETH) { |
803 | sid = rte_eth_dev_socket_id(port_no); | |
804 | } else { | |
805 | sid = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
806 | } | |
807 | ||
d46285a2 | 808 | dev->socket_id = sid < 0 ? SOCKET0 : sid; |
db8f13b0 | 809 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
810 | dev->port_id = port_no; |
811 | dev->type = type; | |
812 | dev->flags = 0; | |
0072e931 | 813 | dev->requested_mtu = dev->mtu = ETHER_MTU; |
d46285a2 | 814 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
815 | ovsrcu_index_init(&dev->vid, -1); |
816 | dev->vhost_reconfigured = false; | |
c1ff66ac CL |
817 | /* initialise vHost port in server mode */ |
818 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; | |
8a9562d2 | 819 | |
0072e931 MK |
820 | err = netdev_dpdk_mempool_configure(dev); |
821 | if (err) { | |
95fb793a | 822 | goto unlock; |
8a9562d2 PS |
823 | } |
824 | ||
0bf765f7 | 825 | /* Initialise QoS configuration to NULL and qos lock to unlocked */ |
d46285a2 DDP |
826 | dev->qos_conf = NULL; |
827 | rte_spinlock_init(&dev->qos_lock); | |
0bf765f7 | 828 | |
9509913a IS |
829 | /* Initialise rcu pointer for ingress policer to NULL */ |
830 | ovsrcu_init(&dev->ingress_policer, NULL); | |
831 | dev->policer_rate = 0; | |
832 | dev->policer_burst = 0; | |
833 | ||
d46285a2 | 834 | netdev->n_rxq = NR_QUEUE; |
81acebda IM |
835 | netdev->n_txq = NR_QUEUE; |
836 | dev->requested_n_rxq = netdev->n_rxq; | |
837 | dev->requested_n_txq = netdev->n_txq; | |
58397e6c | 838 | |
9fd39370 SC |
839 | /* Initialize the flow control to NULL */ |
840 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
58397e6c | 841 | if (type == DPDK_DEV_ETH) { |
d46285a2 | 842 | err = dpdk_eth_dev_init(dev); |
1b99bb05 MG |
843 | if (err) { |
844 | goto unlock; | |
845 | } | |
81acebda | 846 | netdev_dpdk_alloc_txq(dev, netdev->n_txq); |
585a5bea | 847 | } else { |
d46285a2 | 848 | netdev_dpdk_alloc_txq(dev, OVS_VHOST_MAX_QUEUE_NUM); |
e543851d ZB |
849 | /* Enable DPDK_DEV_VHOST device and set promiscuous mode flag. */ |
850 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
8a9562d2 | 851 | } |
8a9562d2 | 852 | |
d46285a2 | 853 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 854 | |
95fb793a | 855 | unlock: |
d46285a2 | 856 | ovs_mutex_unlock(&dev->mutex); |
95fb793a | 857 | return err; |
858 | } | |
859 | ||
b83a2df1 MV |
860 | /* dev_name must be the prefix followed by a positive decimal number. |
861 | * (no leading + or - signs are allowed) */ | |
95fb793a | 862 | static int |
863 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
864 | unsigned int *port_no) | |
865 | { | |
866 | const char *cport; | |
867 | ||
868 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
869 | return ENODEV; | |
870 | } | |
871 | ||
872 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
873 | |
874 | if (str_to_uint(cport, 10, port_no)) { | |
875 | return 0; | |
876 | } else { | |
877 | return ENODEV; | |
878 | } | |
95fb793a | 879 | } |
880 | ||
c1ff66ac CL |
881 | /* Returns a pointer to the relevant vHost socket ID depending on the mode in |
882 | * use */ | |
883 | static char * | |
884 | get_vhost_id(struct netdev_dpdk *dev) | |
885 | OVS_REQUIRES(dev->mutex) | |
886 | { | |
887 | return dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT ? | |
888 | dev->vhost_client_id : dev->vhost_server_id; | |
889 | } | |
890 | ||
7d1ced01 | 891 | static int |
53f50d24 | 892 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 893 | { |
d46285a2 DDP |
894 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
895 | const char *name = netdev->name; | |
7d1ced01 | 896 | int err; |
a0cb2d66 | 897 | |
1af27e8a DDP |
898 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
899 | * the file system. '/' or '\' would traverse directories, so they're not | |
900 | * acceptable in 'name'. */ | |
901 | if (strchr(name, '/') || strchr(name, '\\')) { | |
902 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
903 | "A valid name must not include '/' or '\\'", | |
904 | name); | |
905 | return EINVAL; | |
906 | } | |
907 | ||
bab69409 AC |
908 | if (rte_eal_init_ret) { |
909 | return rte_eal_init_ret; | |
910 | } | |
911 | ||
7d1ced01 CL |
912 | ovs_mutex_lock(&dpdk_mutex); |
913 | /* Take the name of the vhost-user port and append it to the location where | |
c1ff66ac CL |
914 | * the socket is to be created, then register the socket. Sockets are |
915 | * registered initially in 'server' mode. | |
7d1ced01 | 916 | */ |
c1ff66ac | 917 | snprintf(dev->vhost_server_id, sizeof dev->vhost_server_id, "%s/%s", |
1af27e8a DDP |
918 | vhost_sock_dir, name); |
919 | ||
c1ff66ac CL |
920 | err = rte_vhost_driver_register(dev->vhost_server_id, |
921 | dev->vhost_driver_flags); | |
7d1ced01 CL |
922 | if (err) { |
923 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
c1ff66ac | 924 | dev->vhost_server_id); |
e5c0f5a4 | 925 | } else { |
c1ff66ac CL |
926 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
927 | /* OVS server mode - add this socket to list for deletion */ | |
928 | fatal_signal_add_file_to_unlink(dev->vhost_server_id); | |
929 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
930 | dev->vhost_server_id, name); | |
931 | } | |
41987644 | 932 | err = netdev_dpdk_init(netdev, -1, DPDK_DEV_VHOST); |
7d1ced01 | 933 | } |
e5c0f5a4 | 934 | |
7d1ced01 | 935 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
936 | return err; |
937 | } | |
938 | ||
95fb793a | 939 | static int |
940 | netdev_dpdk_construct(struct netdev *netdev) | |
941 | { | |
942 | unsigned int port_no; | |
943 | int err; | |
944 | ||
945 | if (rte_eal_init_ret) { | |
946 | return rte_eal_init_ret; | |
947 | } | |
948 | ||
949 | /* Names always start with "dpdk" */ | |
950 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
951 | if (err) { | |
952 | return err; | |
953 | } | |
954 | ||
955 | ovs_mutex_lock(&dpdk_mutex); | |
58397e6c | 956 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
8a9562d2 PS |
957 | ovs_mutex_unlock(&dpdk_mutex); |
958 | return err; | |
959 | } | |
960 | ||
961 | static void | |
d46285a2 | 962 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 963 | { |
d46285a2 | 964 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 965 | |
8d38823b | 966 | ovs_mutex_lock(&dpdk_mutex); |
8a9562d2 | 967 | ovs_mutex_lock(&dev->mutex); |
8d38823b | 968 | |
8a9562d2 | 969 | rte_eth_dev_stop(dev->port_id); |
9509913a IS |
970 | free(ovsrcu_get_protected(struct ingress_policer *, |
971 | &dev->ingress_policer)); | |
8a9562d2 | 972 | |
5a034064 | 973 | rte_free(dev->tx_q); |
417e7e66 | 974 | ovs_list_remove(&dev->list_node); |
8a9562d2 | 975 | dpdk_mp_put(dev->dpdk_mp); |
8d38823b MK |
976 | |
977 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 | 978 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 979 | } |
8a9562d2 | 980 | |
3f891bbe DDP |
981 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
982 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
983 | * deadlock, none of the mutexes must be held while calling this function. */ | |
984 | static int | |
c1ff66ac CL |
985 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
986 | char *vhost_id) | |
3f891bbe DDP |
987 | OVS_EXCLUDED(dpdk_mutex) |
988 | OVS_EXCLUDED(dev->mutex) | |
989 | { | |
c1ff66ac | 990 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
991 | } |
992 | ||
58397e6c | 993 | static void |
d46285a2 | 994 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 995 | { |
d46285a2 | 996 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 997 | char *vhost_id; |
58397e6c | 998 | |
8d38823b MK |
999 | ovs_mutex_lock(&dpdk_mutex); |
1000 | ovs_mutex_lock(&dev->mutex); | |
1001 | ||
c62da695 | 1002 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1003 | if (netdev_dpdk_get_vid(dev) >= 0 |
1004 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1005 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1006 | netdev->name); |
c62da695 IM |
1007 | VLOG_ERR("To restore connectivity after re-adding of port, VM on socket" |
1008 | " '%s' must be restarted.", | |
c1ff66ac | 1009 | get_vhost_id(dev)); |
58397e6c KT |
1010 | } |
1011 | ||
9509913a IS |
1012 | free(ovsrcu_get_protected(struct ingress_policer *, |
1013 | &dev->ingress_policer)); | |
9509913a | 1014 | |
6019cb63 | 1015 | rte_free(dev->tx_q); |
417e7e66 | 1016 | ovs_list_remove(&dev->list_node); |
58397e6c | 1017 | dpdk_mp_put(dev->dpdk_mp); |
8d38823b | 1018 | |
c1ff66ac CL |
1019 | vhost_id = xstrdup(get_vhost_id(dev)); |
1020 | ||
8d38823b | 1021 | ovs_mutex_unlock(&dev->mutex); |
58397e6c | 1022 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1023 | |
c1ff66ac CL |
1024 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
1025 | VLOG_ERR("Unable to remove vhost-user socket %s", vhost_id); | |
1026 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
1027 | /* OVS server mode - remove this socket from list for deletion */ | |
1028 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1029 | } |
c1ff66ac | 1030 | free(vhost_id); |
8a9562d2 PS |
1031 | } |
1032 | ||
1033 | static void | |
d46285a2 | 1034 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1035 | { |
d46285a2 | 1036 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1037 | |
d46285a2 | 1038 | rte_free(dev); |
8a9562d2 PS |
1039 | } |
1040 | ||
1041 | static int | |
a14b8947 | 1042 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1043 | { |
a14b8947 | 1044 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1045 | |
1046 | ovs_mutex_lock(&dev->mutex); | |
1047 | ||
050c60bf | 1048 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1049 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1050 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1051 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1052 | smap_add_format(args, "mtu", "%d", dev->mtu); |
8a9562d2 PS |
1053 | ovs_mutex_unlock(&dev->mutex); |
1054 | ||
1055 | return 0; | |
1056 | } | |
1057 | ||
c3d062a7 CL |
1058 | static void |
1059 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
a14b8947 | 1060 | { |
050c60bf | 1061 | int new_n_rxq; |
a14b8947 | 1062 | |
050c60bf DDP |
1063 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", dev->requested_n_rxq), 1); |
1064 | if (new_n_rxq != dev->requested_n_rxq) { | |
1065 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1066 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1067 | } |
c3d062a7 CL |
1068 | } |
1069 | ||
1070 | static int | |
1071 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args) | |
1072 | { | |
1073 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1074 | ||
1075 | ovs_mutex_lock(&dev->mutex); | |
1076 | ||
1077 | dpdk_set_rxq_config(dev, args); | |
1078 | ||
1079 | /* Flow control support is only available for DPDK Ethernet ports. */ | |
1080 | bool rx_fc_en = false; | |
1081 | bool tx_fc_en = false; | |
1082 | enum rte_eth_fc_mode fc_mode_set[2][2] = | |
1083 | {{RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1084 | {RTE_FC_RX_PAUSE, RTE_FC_FULL} | |
1085 | }; | |
1086 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); | |
1087 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
1088 | dev->fc_conf.autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); | |
1089 | dev->fc_conf.mode = fc_mode_set[tx_fc_en][rx_fc_en]; | |
1090 | ||
1091 | dpdk_eth_flow_ctrl_setup(dev); | |
9fd39370 | 1092 | |
c3d062a7 CL |
1093 | ovs_mutex_unlock(&dev->mutex); |
1094 | ||
1095 | return 0; | |
1096 | } | |
1097 | ||
1098 | static int | |
1099 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args) | |
1100 | { | |
1101 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1102 | ||
1103 | ovs_mutex_lock(&dev->mutex); | |
1104 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1105 | ovs_mutex_unlock(&dev->mutex); |
1106 | ||
1107 | return 0; | |
1108 | } | |
1109 | ||
c1ff66ac CL |
1110 | static int |
1111 | netdev_dpdk_vhost_set_config(struct netdev *netdev, const struct smap *args) | |
1112 | { | |
1113 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1114 | const char *path; | |
1115 | ||
1116 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
1117 | path = smap_get(args, "vhost-server-path"); | |
1118 | if (path && strcmp(path, dev->vhost_client_id)) { | |
1119 | strcpy(dev->vhost_client_id, path); | |
1120 | netdev_request_reconfigure(netdev); | |
1121 | } | |
1122 | } | |
1123 | ||
1124 | return 0; | |
1125 | } | |
1126 | ||
7dec44fe | 1127 | static int |
d46285a2 | 1128 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1129 | { |
d46285a2 | 1130 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1131 | |
d46285a2 | 1132 | return dev->socket_id; |
7dec44fe AW |
1133 | } |
1134 | ||
050c60bf | 1135 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1136 | static int |
050c60bf | 1137 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1138 | { |
d46285a2 | 1139 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1140 | |
d46285a2 | 1141 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1142 | |
050c60bf DDP |
1143 | if (dev->requested_n_txq == n_txq) { |
1144 | goto out; | |
4573fbd3 FL |
1145 | } |
1146 | ||
050c60bf DDP |
1147 | dev->requested_n_txq = n_txq; |
1148 | netdev_request_reconfigure(netdev); | |
58397e6c | 1149 | |
050c60bf | 1150 | out: |
d46285a2 | 1151 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1152 | return 0; |
58397e6c KT |
1153 | } |
1154 | ||
8a9562d2 PS |
1155 | static struct netdev_rxq * |
1156 | netdev_dpdk_rxq_alloc(void) | |
1157 | { | |
1158 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1159 | ||
1160 | return &rx->up; | |
1161 | } | |
1162 | ||
1163 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1164 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1165 | { |
d46285a2 | 1166 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1167 | } |
1168 | ||
1169 | static int | |
d46285a2 | 1170 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1171 | { |
d46285a2 DDP |
1172 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1173 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1174 | |
d46285a2 DDP |
1175 | ovs_mutex_lock(&dev->mutex); |
1176 | rx->port_id = dev->port_id; | |
1177 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1178 | |
1179 | return 0; | |
1180 | } | |
1181 | ||
1182 | static void | |
d46285a2 | 1183 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1184 | { |
1185 | } | |
1186 | ||
1187 | static void | |
d46285a2 | 1188 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1189 | { |
d46285a2 | 1190 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1191 | |
1192 | rte_free(rx); | |
1193 | } | |
1194 | ||
b170db2a | 1195 | static inline void |
b59cc14e IM |
1196 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
1197 | struct rte_mbuf **pkts, int cnt) | |
8a9562d2 | 1198 | { |
1304f1f8 DDP |
1199 | uint32_t nb_tx = 0; |
1200 | ||
b59cc14e | 1201 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1202 | uint32_t ret; |
1203 | ||
b59cc14e | 1204 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1205 | if (!ret) { |
1206 | break; | |
1207 | } | |
1208 | ||
1209 | nb_tx += ret; | |
1210 | } | |
8a9562d2 | 1211 | |
b59cc14e | 1212 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
db73f716 DDP |
1213 | /* free buffers, which we couldn't transmit, one at a time (each |
1214 | * packet could come from a different mempool) */ | |
1215 | int i; | |
1216 | ||
b59cc14e IM |
1217 | for (i = nb_tx; i < cnt; i++) { |
1218 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1219 | } |
45d947c4 | 1220 | rte_spinlock_lock(&dev->stats_lock); |
b59cc14e | 1221 | dev->stats.tx_dropped += cnt - nb_tx; |
45d947c4 | 1222 | rte_spinlock_unlock(&dev->stats_lock); |
8a9562d2 | 1223 | } |
8a9562d2 PS |
1224 | } |
1225 | ||
f3926f29 IS |
1226 | static inline bool |
1227 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1228 | struct rte_mbuf *pkt, uint64_t time) | |
1229 | { | |
1230 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1231 | ||
1232 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1233 | e_RTE_METER_GREEN; | |
1234 | } | |
1235 | ||
1236 | static int | |
1237 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
1238 | struct rte_mbuf **pkts, int pkt_cnt) | |
1239 | { | |
1240 | int i = 0; | |
1241 | int cnt = 0; | |
1242 | struct rte_mbuf *pkt = NULL; | |
1243 | uint64_t current_time = rte_rdtsc(); | |
1244 | ||
1245 | for (i = 0; i < pkt_cnt; i++) { | |
1246 | pkt = pkts[i]; | |
1247 | /* Handle current packet */ | |
1248 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1249 | if (cnt != i) { | |
1250 | pkts[cnt] = pkt; | |
1251 | } | |
1252 | cnt++; | |
1253 | } else { | |
1254 | rte_pktmbuf_free(pkt); | |
1255 | } | |
1256 | } | |
1257 | ||
1258 | return cnt; | |
1259 | } | |
1260 | ||
9509913a IS |
1261 | static int |
1262 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
1263 | int pkt_cnt) | |
1264 | { | |
1265 | int cnt = 0; | |
1266 | ||
1267 | rte_spinlock_lock(&policer->policer_lock); | |
1268 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, pkt_cnt); | |
1269 | rte_spinlock_unlock(&policer->policer_lock); | |
1270 | ||
1271 | return cnt; | |
1272 | } | |
1273 | ||
58397e6c | 1274 | static bool |
0a0f39df | 1275 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1276 | { |
0a0f39df | 1277 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1278 | } |
1279 | ||
d6e3feb5 | 1280 | static inline void |
1281 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1282 | unsigned int packet_size) | |
1283 | { | |
1284 | /* Hard-coded search for the size bucket. */ | |
1285 | if (packet_size < 256) { | |
1286 | if (packet_size >= 128) { | |
1287 | stats->rx_128_to_255_packets++; | |
1288 | } else if (packet_size <= 64) { | |
1289 | stats->rx_1_to_64_packets++; | |
1290 | } else { | |
1291 | stats->rx_65_to_127_packets++; | |
1292 | } | |
1293 | } else { | |
1294 | if (packet_size >= 1523) { | |
1295 | stats->rx_1523_to_max_packets++; | |
1296 | } else if (packet_size >= 1024) { | |
1297 | stats->rx_1024_to_1522_packets++; | |
1298 | } else if (packet_size < 512) { | |
1299 | stats->rx_256_to_511_packets++; | |
1300 | } else { | |
1301 | stats->rx_512_to_1023_packets++; | |
1302 | } | |
1303 | } | |
1304 | } | |
1305 | ||
9e3ddd45 TP |
1306 | static inline void |
1307 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1308 | struct dp_packet **packets, int count, |
1309 | int dropped) | |
9e3ddd45 TP |
1310 | { |
1311 | int i; | |
d6e3feb5 | 1312 | unsigned int packet_size; |
9e3ddd45 TP |
1313 | struct dp_packet *packet; |
1314 | ||
1315 | stats->rx_packets += count; | |
9509913a | 1316 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1317 | for (i = 0; i < count; i++) { |
1318 | packet = packets[i]; | |
d6e3feb5 | 1319 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1320 | |
d6e3feb5 | 1321 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1322 | /* This only protects the following multicast counting from |
1323 | * too short packets, but it does not stop the packet from | |
1324 | * further processing. */ | |
1325 | stats->rx_errors++; | |
1326 | stats->rx_length_errors++; | |
1327 | continue; | |
1328 | } | |
1329 | ||
d6e3feb5 | 1330 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1331 | ||
9e3ddd45 TP |
1332 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1333 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1334 | stats->multicast++; | |
1335 | } | |
1336 | ||
d6e3feb5 | 1337 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1338 | } |
1339 | } | |
1340 | ||
58397e6c KT |
1341 | /* |
1342 | * The receive path for the vhost port is the TX path out from guest. | |
1343 | */ | |
1344 | static int | |
d46285a2 | 1345 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1346 | struct dp_packet_batch *batch) |
58397e6c | 1347 | { |
d46285a2 | 1348 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
d46285a2 | 1349 | int qid = rxq->queue_id; |
9509913a | 1350 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1351 | uint16_t nb_rx = 0; |
9509913a | 1352 | uint16_t dropped = 0; |
58397e6c | 1353 | |
0a0f39df | 1354 | if (OVS_UNLIKELY(!is_vhost_running(dev) |
e543851d | 1355 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1356 | return EAGAIN; |
1357 | } | |
1358 | ||
0a0f39df CL |
1359 | nb_rx = rte_vhost_dequeue_burst(netdev_dpdk_get_vid(dev), |
1360 | qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
d46285a2 | 1361 | dev->dpdk_mp->mp, |
64839cf4 | 1362 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1363 | NETDEV_MAX_BURST); |
58397e6c KT |
1364 | if (!nb_rx) { |
1365 | return EAGAIN; | |
1366 | } | |
1367 | ||
9509913a IS |
1368 | if (policer) { |
1369 | dropped = nb_rx; | |
64839cf4 WT |
1370 | nb_rx = ingress_policer_run(policer, |
1371 | (struct rte_mbuf **) batch->packets, | |
1372 | nb_rx); | |
9509913a IS |
1373 | dropped -= nb_rx; |
1374 | } | |
1375 | ||
d46285a2 | 1376 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1377 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1378 | nb_rx, dropped); | |
d46285a2 | 1379 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1380 | |
64839cf4 | 1381 | batch->count = (int) nb_rx; |
58397e6c KT |
1382 | return 0; |
1383 | } | |
1384 | ||
8a9562d2 | 1385 | static int |
64839cf4 | 1386 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1387 | { |
d46285a2 DDP |
1388 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1389 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1390 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1391 | int nb_rx; |
9509913a | 1392 | int dropped = 0; |
8a9562d2 | 1393 | |
d46285a2 | 1394 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1395 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1396 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1397 | if (!nb_rx) { |
1398 | return EAGAIN; | |
1399 | } | |
1400 | ||
9509913a IS |
1401 | if (policer) { |
1402 | dropped = nb_rx; | |
64839cf4 WT |
1403 | nb_rx = ingress_policer_run(policer, |
1404 | (struct rte_mbuf **)batch->packets, | |
1405 | nb_rx); | |
9509913a IS |
1406 | dropped -= nb_rx; |
1407 | } | |
1408 | ||
1409 | /* Update stats to reflect dropped packets */ | |
1410 | if (OVS_UNLIKELY(dropped)) { | |
1411 | rte_spinlock_lock(&dev->stats_lock); | |
1412 | dev->stats.rx_dropped += dropped; | |
1413 | rte_spinlock_unlock(&dev->stats_lock); | |
1414 | } | |
1415 | ||
64839cf4 | 1416 | batch->count = nb_rx; |
8a9562d2 PS |
1417 | |
1418 | return 0; | |
1419 | } | |
1420 | ||
0bf765f7 IS |
1421 | static inline int |
1422 | netdev_dpdk_qos_run__(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1423 | int cnt) | |
1424 | { | |
1425 | struct netdev *netdev = &dev->up; | |
1426 | ||
1427 | if (dev->qos_conf != NULL) { | |
1428 | rte_spinlock_lock(&dev->qos_lock); | |
1429 | if (dev->qos_conf != NULL) { | |
1430 | cnt = dev->qos_conf->ops->qos_run(netdev, pkts, cnt); | |
1431 | } | |
1432 | rte_spinlock_unlock(&dev->qos_lock); | |
1433 | } | |
1434 | ||
1435 | return cnt; | |
1436 | } | |
1437 | ||
9e3ddd45 TP |
1438 | static inline void |
1439 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1440 | struct dp_packet **packets, | |
1441 | int attempted, | |
1442 | int dropped) | |
1443 | { | |
1444 | int i; | |
1445 | int sent = attempted - dropped; | |
1446 | ||
1447 | stats->tx_packets += sent; | |
1448 | stats->tx_dropped += dropped; | |
1449 | ||
1450 | for (i = 0; i < sent; i++) { | |
1451 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1452 | } | |
1453 | } | |
1454 | ||
58397e6c | 1455 | static void |
4573fbd3 | 1456 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 1457 | struct dp_packet **pkts, int cnt) |
58397e6c | 1458 | { |
d46285a2 | 1459 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1460 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1461 | unsigned int total_pkts = cnt; | |
dd52de45 | 1462 | unsigned int qos_pkts = 0; |
0072e931 | 1463 | unsigned int mtu_dropped = 0; |
dd52de45 | 1464 | int i, retries = 0; |
58397e6c | 1465 | |
81acebda | 1466 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1467 | |
0a0f39df | 1468 | if (OVS_UNLIKELY(!is_vhost_running(dev) || qid < 0 |
e543851d | 1469 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1470 | rte_spinlock_lock(&dev->stats_lock); |
1471 | dev->stats.tx_dropped+= cnt; | |
1472 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1473 | goto out; |
58397e6c KT |
1474 | } |
1475 | ||
d46285a2 | 1476 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1477 | |
0bf765f7 | 1478 | /* Check has QoS has been configured for the netdev */ |
d46285a2 | 1479 | cnt = netdev_dpdk_qos_run__(dev, cur_pkts, cnt); |
dd52de45 | 1480 | qos_pkts = total_pkts - cnt; |
0bf765f7 | 1481 | |
95e9881f | 1482 | do { |
4573fbd3 | 1483 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f | 1484 | unsigned int tx_pkts; |
0072e931 | 1485 | unsigned int try_tx_pkts = cnt; |
95e9881f | 1486 | |
0072e931 MK |
1487 | for (i = 0; i < cnt; i++) { |
1488 | if (cur_pkts[i]->pkt_len > dev->max_packet_len) { | |
1489 | try_tx_pkts = i; | |
1490 | break; | |
1491 | } | |
1492 | } | |
1493 | if (!try_tx_pkts) { | |
1494 | cur_pkts++; | |
1495 | mtu_dropped++; | |
1496 | cnt--; | |
1497 | continue; | |
1498 | } | |
0a0f39df | 1499 | tx_pkts = rte_vhost_enqueue_burst(netdev_dpdk_get_vid(dev), |
0072e931 | 1500 | vhost_qid, cur_pkts, try_tx_pkts); |
95e9881f KT |
1501 | if (OVS_LIKELY(tx_pkts)) { |
1502 | /* Packets have been sent.*/ | |
1503 | cnt -= tx_pkts; | |
31871ee3 | 1504 | /* Prepare for possible retry.*/ |
95e9881f | 1505 | cur_pkts = &cur_pkts[tx_pkts]; |
0072e931 MK |
1506 | if (tx_pkts != try_tx_pkts) { |
1507 | retries++; | |
1508 | } | |
95e9881f | 1509 | } else { |
31871ee3 KT |
1510 | /* No packets sent - do not retry.*/ |
1511 | break; | |
95e9881f | 1512 | } |
0072e931 | 1513 | } while (cnt && (retries <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1514 | |
d46285a2 | 1515 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1516 | |
d46285a2 | 1517 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 MK |
1518 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
1519 | cnt + mtu_dropped + qos_pkts); | |
d46285a2 | 1520 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1521 | |
1522 | out: | |
dd52de45 IM |
1523 | for (i = 0; i < total_pkts - qos_pkts; i++) { |
1524 | dp_packet_delete(pkts[i]); | |
58397e6c KT |
1525 | } |
1526 | } | |
1527 | ||
8a9562d2 PS |
1528 | /* Tx function. Transmit packets indefinitely */ |
1529 | static void | |
64839cf4 | 1530 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1531 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1532 | { |
bce01e3a | 1533 | #if !defined(__CHECKER__) && !defined(_WIN32) |
64839cf4 | 1534 | const size_t PKT_ARRAY_SIZE = batch->count; |
bce01e3a EJ |
1535 | #else |
1536 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1537 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1538 | #endif |
8a9562d2 | 1539 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
bce01e3a | 1540 | struct rte_mbuf *mbufs[PKT_ARRAY_SIZE]; |
175cf4de RW |
1541 | int dropped = 0; |
1542 | int newcnt = 0; | |
1543 | int i; | |
8a9562d2 | 1544 | |
db73f716 DDP |
1545 | /* If we are on a non pmd thread we have to use the mempool mutex, because |
1546 | * every non pmd thread shares the same mempool cache */ | |
1547 | ||
5f17de68 | 1548 | if (!dpdk_thread_is_pmd()) { |
db73f716 DDP |
1549 | ovs_mutex_lock(&nonpmd_mempool_mutex); |
1550 | } | |
1551 | ||
7d6d1a40 WT |
1552 | dp_packet_batch_apply_cutlen(batch); |
1553 | ||
64839cf4 WT |
1554 | for (i = 0; i < batch->count; i++) { |
1555 | int size = dp_packet_size(batch->packets[i]); | |
95fb793a | 1556 | |
f98d7864 | 1557 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c DDP |
1558 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1559 | (int)size , dev->max_packet_len); | |
1560 | ||
175cf4de | 1561 | dropped++; |
f4fd623c DDP |
1562 | continue; |
1563 | } | |
8a9562d2 | 1564 | |
f4fd623c | 1565 | mbufs[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1566 | |
f4fd623c | 1567 | if (!mbufs[newcnt]) { |
64839cf4 | 1568 | dropped += batch->count - i; |
175cf4de | 1569 | break; |
f4fd623c DDP |
1570 | } |
1571 | ||
1572 | /* We have to do a copy for now */ | |
64839cf4 WT |
1573 | memcpy(rte_pktmbuf_mtod(mbufs[newcnt], void *), |
1574 | dp_packet_data(batch->packets[i]), size); | |
f4fd623c DDP |
1575 | |
1576 | rte_pktmbuf_data_len(mbufs[newcnt]) = size; | |
1577 | rte_pktmbuf_pkt_len(mbufs[newcnt]) = size; | |
1578 | ||
1579 | newcnt++; | |
1580 | } | |
8a9562d2 | 1581 | |
58397e6c | 1582 | if (dev->type == DPDK_DEV_VHOST) { |
64839cf4 | 1583 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) mbufs, |
dd52de45 | 1584 | newcnt); |
58397e6c | 1585 | } else { |
0bf765f7 IS |
1586 | unsigned int qos_pkts = newcnt; |
1587 | ||
1588 | /* Check if QoS has been configured for this netdev. */ | |
1589 | newcnt = netdev_dpdk_qos_run__(dev, mbufs, newcnt); | |
1590 | ||
1591 | dropped += qos_pkts - newcnt; | |
b59cc14e | 1592 | netdev_dpdk_eth_tx_burst(dev, qid, mbufs, newcnt); |
58397e6c | 1593 | } |
db73f716 | 1594 | |
0bf765f7 IS |
1595 | if (OVS_UNLIKELY(dropped)) { |
1596 | rte_spinlock_lock(&dev->stats_lock); | |
1597 | dev->stats.tx_dropped += dropped; | |
1598 | rte_spinlock_unlock(&dev->stats_lock); | |
1599 | } | |
1600 | ||
5f17de68 | 1601 | if (!dpdk_thread_is_pmd()) { |
db73f716 DDP |
1602 | ovs_mutex_unlock(&nonpmd_mempool_mutex); |
1603 | } | |
8a9562d2 PS |
1604 | } |
1605 | ||
58397e6c | 1606 | static int |
64839cf4 WT |
1607 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1608 | struct dp_packet_batch *batch, | |
324c8374 | 1609 | bool may_steal, bool concurrent_txq OVS_UNUSED) |
58397e6c | 1610 | { |
58397e6c | 1611 | |
dd52de45 | 1612 | if (OVS_UNLIKELY(!may_steal || batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 WT |
1613 | dpdk_do_tx_copy(netdev, qid, batch); |
1614 | dp_packet_delete_batch(batch, may_steal); | |
58397e6c | 1615 | } else { |
64839cf4 | 1616 | dp_packet_batch_apply_cutlen(batch); |
dd52de45 | 1617 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
1618 | } |
1619 | return 0; | |
1620 | } | |
1621 | ||
7251515e DV |
1622 | static inline void |
1623 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
324c8374 IM |
1624 | struct dp_packet_batch *batch, bool may_steal, |
1625 | bool concurrent_txq) | |
8a9562d2 | 1626 | { |
324c8374 | 1627 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 1628 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
1629 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
1630 | } | |
1631 | ||
7251515e | 1632 | if (OVS_UNLIKELY(!may_steal || |
64839cf4 | 1633 | batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1634 | struct netdev *netdev = &dev->up; |
1635 | ||
64839cf4 WT |
1636 | dpdk_do_tx_copy(netdev, qid, batch); |
1637 | dp_packet_delete_batch(batch, may_steal); | |
8a9562d2 | 1638 | } else { |
f4fd623c DDP |
1639 | int next_tx_idx = 0; |
1640 | int dropped = 0; | |
0bf765f7 IS |
1641 | unsigned int qos_pkts = 0; |
1642 | unsigned int temp_cnt = 0; | |
64839cf4 | 1643 | int cnt = batch->count; |
8a9562d2 | 1644 | |
7d6d1a40 WT |
1645 | dp_packet_batch_apply_cutlen(batch); |
1646 | ||
64839cf4 WT |
1647 | for (int i = 0; i < cnt; i++) { |
1648 | int size = dp_packet_size(batch->packets[i]); | |
1b99bb05 | 1649 | |
f4fd623c DDP |
1650 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
1651 | if (next_tx_idx != i) { | |
0bf765f7 IS |
1652 | temp_cnt = i - next_tx_idx; |
1653 | qos_pkts = temp_cnt; | |
1654 | ||
64839cf4 WT |
1655 | temp_cnt = netdev_dpdk_qos_run__(dev, |
1656 | (struct rte_mbuf**)batch->packets, | |
1657 | temp_cnt); | |
0bf765f7 | 1658 | dropped += qos_pkts - temp_cnt; |
b59cc14e | 1659 | netdev_dpdk_eth_tx_burst(dev, qid, |
64839cf4 WT |
1660 | (struct rte_mbuf **)&batch->packets[next_tx_idx], |
1661 | temp_cnt); | |
0bf765f7 | 1662 | |
1ebfe1ac | 1663 | } |
f4fd623c | 1664 | |
1ebfe1ac DDP |
1665 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1666 | (int)size , dev->max_packet_len); | |
f4fd623c | 1667 | |
64839cf4 | 1668 | dp_packet_delete(batch->packets[i]); |
1ebfe1ac | 1669 | dropped++; |
f4fd623c DDP |
1670 | next_tx_idx = i + 1; |
1671 | } | |
1672 | } | |
1673 | if (next_tx_idx != cnt) { | |
0bf765f7 IS |
1674 | cnt -= next_tx_idx; |
1675 | qos_pkts = cnt; | |
1676 | ||
64839cf4 WT |
1677 | cnt = netdev_dpdk_qos_run__(dev, |
1678 | (struct rte_mbuf**)batch->packets, cnt); | |
0bf765f7 | 1679 | dropped += qos_pkts - cnt; |
b59cc14e | 1680 | netdev_dpdk_eth_tx_burst(dev, qid, |
64839cf4 WT |
1681 | (struct rte_mbuf **)&batch->packets[next_tx_idx], |
1682 | cnt); | |
f4fd623c | 1683 | } |
8a9562d2 | 1684 | |
f4fd623c | 1685 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1686 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1687 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1688 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1689 | } |
8a9562d2 | 1690 | } |
a0cb2d66 | 1691 | |
324c8374 | 1692 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
1693 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
1694 | } | |
7251515e DV |
1695 | } |
1696 | ||
1697 | static int | |
1698 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
324c8374 IM |
1699 | struct dp_packet_batch *batch, bool may_steal, |
1700 | bool concurrent_txq) | |
7251515e DV |
1701 | { |
1702 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1703 | |
324c8374 | 1704 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e | 1705 | return 0; |
8a9562d2 PS |
1706 | } |
1707 | ||
1708 | static int | |
74ff3298 | 1709 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1710 | { |
1711 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1712 | ||
1713 | ovs_mutex_lock(&dev->mutex); | |
1714 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1715 | dev->hwaddr = mac; |
045c0d1a | 1716 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1717 | } |
1718 | ovs_mutex_unlock(&dev->mutex); | |
1719 | ||
1720 | return 0; | |
1721 | } | |
1722 | ||
1723 | static int | |
74ff3298 | 1724 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1725 | { |
1726 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1727 | ||
1728 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1729 | *mac = dev->hwaddr; |
8a9562d2 PS |
1730 | ovs_mutex_unlock(&dev->mutex); |
1731 | ||
1732 | return 0; | |
1733 | } | |
1734 | ||
1735 | static int | |
1736 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1737 | { | |
1738 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1739 | ||
1740 | ovs_mutex_lock(&dev->mutex); | |
1741 | *mtup = dev->mtu; | |
1742 | ovs_mutex_unlock(&dev->mutex); | |
1743 | ||
1744 | return 0; | |
1745 | } | |
1746 | ||
0072e931 MK |
1747 | static int |
1748 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
1749 | { | |
1750 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1751 | ||
1752 | if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
1753 | || mtu < ETHER_MIN_MTU) { | |
1754 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
1755 | return EINVAL; | |
1756 | } | |
1757 | ||
1758 | ovs_mutex_lock(&dev->mutex); | |
1759 | if (dev->requested_mtu != mtu) { | |
1760 | dev->requested_mtu = mtu; | |
1761 | netdev_request_reconfigure(netdev); | |
1762 | } | |
1763 | ovs_mutex_unlock(&dev->mutex); | |
1764 | ||
1765 | return 0; | |
1766 | } | |
1767 | ||
8a9562d2 | 1768 | static int |
d46285a2 | 1769 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 1770 | |
58397e6c KT |
1771 | static int |
1772 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1773 | struct netdev_stats *stats) | |
1774 | { | |
1775 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1776 | ||
1777 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 1778 | |
45d947c4 | 1779 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
1780 | /* Supported Stats */ |
1781 | stats->rx_packets += dev->stats.rx_packets; | |
1782 | stats->tx_packets += dev->stats.tx_packets; | |
9509913a | 1783 | stats->rx_dropped = dev->stats.rx_dropped; |
58397e6c | 1784 | stats->tx_dropped += dev->stats.tx_dropped; |
9e3ddd45 TP |
1785 | stats->multicast = dev->stats.multicast; |
1786 | stats->rx_bytes = dev->stats.rx_bytes; | |
1787 | stats->tx_bytes = dev->stats.tx_bytes; | |
1788 | stats->rx_errors = dev->stats.rx_errors; | |
1789 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 1790 | |
1791 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
1792 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
1793 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
1794 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
1795 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
1796 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
1797 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
1798 | ||
45d947c4 | 1799 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 1800 | |
58397e6c KT |
1801 | ovs_mutex_unlock(&dev->mutex); |
1802 | ||
1803 | return 0; | |
1804 | } | |
1805 | ||
d6e3feb5 | 1806 | static void |
1807 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
1808 | const struct rte_eth_xstat *xstats, |
1809 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 1810 | const unsigned int size) |
1811 | { | |
d6e3feb5 | 1812 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 1813 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1814 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1815 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1816 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1817 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1818 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1819 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1820 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1821 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1822 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1823 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1824 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 1825 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1826 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 1827 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1828 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1829 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1830 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1831 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1832 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1833 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1834 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1835 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1836 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1837 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1838 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 1839 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1840 | stats->tx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 1841 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1842 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 1843 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1844 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 1845 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1846 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 1847 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1848 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 1849 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1850 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 1851 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1852 | stats->rx_jabber_errors = xstats[i].value; |
1853 | } | |
1854 | } | |
1855 | } | |
1856 | ||
8a9562d2 PS |
1857 | static int |
1858 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1859 | { | |
1860 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1861 | struct rte_eth_stats rte_stats; | |
1862 | bool gg; | |
1863 | ||
1864 | netdev_dpdk_get_carrier(netdev, &gg); | |
1865 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 1866 | |
0a0f39df CL |
1867 | struct rte_eth_xstat *rte_xstats = NULL; |
1868 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
1869 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 1870 | |
1871 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
1872 | VLOG_ERR("Can't get ETH statistics for port: %i.", dev->port_id); | |
f9256822 | 1873 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 1874 | return EPROTO; |
1875 | } | |
1876 | ||
0a0f39df CL |
1877 | /* Get length of statistics */ |
1878 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
1879 | if (rte_xstats_len < 0) { | |
1880 | VLOG_WARN("Cannot get XSTATS values for port: %i", dev->port_id); | |
1881 | goto out; | |
1882 | } | |
1883 | /* Reserve memory for xstats names and values */ | |
1884 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
1885 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
1886 | ||
1887 | /* Retreive xstats names */ | |
1888 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
1889 | rte_xstats_names, | |
1890 | rte_xstats_len); | |
1891 | if (rte_xstats_new_len != rte_xstats_len) { | |
1892 | VLOG_WARN("Cannot get XSTATS names for port: %i.", dev->port_id); | |
1893 | goto out; | |
1894 | } | |
1895 | /* Retreive xstats values */ | |
1896 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
1897 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
1898 | rte_xstats_len); | |
1899 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
1900 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
1901 | rte_xstats_len); | |
d6e3feb5 | 1902 | } else { |
0a0f39df | 1903 | VLOG_WARN("Cannot get XSTATS values for port: %i.", dev->port_id); |
d6e3feb5 | 1904 | } |
8a9562d2 | 1905 | |
0a0f39df CL |
1906 | out: |
1907 | free(rte_xstats); | |
1908 | free(rte_xstats_names); | |
1909 | ||
2f9dd77f PS |
1910 | stats->rx_packets = rte_stats.ipackets; |
1911 | stats->tx_packets = rte_stats.opackets; | |
1912 | stats->rx_bytes = rte_stats.ibytes; | |
1913 | stats->tx_bytes = rte_stats.obytes; | |
9e3ddd45 TP |
1914 | /* DPDK counts imissed as errors, but count them here as dropped instead */ |
1915 | stats->rx_errors = rte_stats.ierrors - rte_stats.imissed; | |
2f9dd77f | 1916 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 1917 | |
45d947c4 | 1918 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 1919 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 1920 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 1921 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
1922 | |
1923 | /* These are the available DPDK counters for packets not received due to | |
1924 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 1925 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
1926 | stats->rx_missed_errors = rte_stats.imissed; |
1927 | ||
8a9562d2 PS |
1928 | ovs_mutex_unlock(&dev->mutex); |
1929 | ||
1930 | return 0; | |
1931 | } | |
1932 | ||
1933 | static int | |
d46285a2 | 1934 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 PS |
1935 | enum netdev_features *current, |
1936 | enum netdev_features *advertised OVS_UNUSED, | |
1937 | enum netdev_features *supported OVS_UNUSED, | |
1938 | enum netdev_features *peer OVS_UNUSED) | |
1939 | { | |
d46285a2 | 1940 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1941 | struct rte_eth_link link; |
1942 | ||
1943 | ovs_mutex_lock(&dev->mutex); | |
1944 | link = dev->link; | |
1945 | ovs_mutex_unlock(&dev->mutex); | |
1946 | ||
362ca396 | 1947 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
1948 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
1949 | *current = NETDEV_F_10MB_HD; |
1950 | } | |
362ca396 | 1951 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
1952 | *current = NETDEV_F_100MB_HD; |
1953 | } | |
362ca396 | 1954 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
1955 | *current = NETDEV_F_1GB_HD; |
1956 | } | |
1957 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 1958 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
1959 | *current = NETDEV_F_10MB_FD; |
1960 | } | |
362ca396 | 1961 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
1962 | *current = NETDEV_F_100MB_FD; |
1963 | } | |
362ca396 | 1964 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
1965 | *current = NETDEV_F_1GB_FD; |
1966 | } | |
362ca396 | 1967 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
1968 | *current = NETDEV_F_10GB_FD; |
1969 | } | |
1970 | } | |
1971 | ||
362ca396 | 1972 | if (link.link_autoneg) { |
1973 | *current |= NETDEV_F_AUTONEG; | |
1974 | } | |
1975 | ||
8a9562d2 PS |
1976 | return 0; |
1977 | } | |
1978 | ||
9509913a IS |
1979 | static struct ingress_policer * |
1980 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
1981 | { | |
1982 | struct ingress_policer *policer = NULL; | |
1983 | uint64_t rate_bytes; | |
1984 | uint64_t burst_bytes; | |
1985 | int err = 0; | |
1986 | ||
1987 | policer = xmalloc(sizeof *policer); | |
1988 | rte_spinlock_init(&policer->policer_lock); | |
1989 | ||
1990 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
1991 | rate_bytes = rate * 1000/8; | |
1992 | burst_bytes = burst * 1000/8; | |
1993 | ||
1994 | policer->app_srtcm_params.cir = rate_bytes; | |
1995 | policer->app_srtcm_params.cbs = burst_bytes; | |
1996 | policer->app_srtcm_params.ebs = 0; | |
1997 | err = rte_meter_srtcm_config(&policer->in_policer, | |
1998 | &policer->app_srtcm_params); | |
1999 | if(err) { | |
2000 | VLOG_ERR("Could not create rte meter for ingress policer"); | |
2001 | return NULL; | |
2002 | } | |
2003 | ||
2004 | return policer; | |
2005 | } | |
2006 | ||
2007 | static int | |
2008 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
2009 | uint32_t policer_burst) | |
2010 | { | |
2011 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2012 | struct ingress_policer *policer; | |
2013 | ||
2014 | /* Force to 0 if no rate specified, | |
2015 | * default to 8000 kbits if burst is 0, | |
2016 | * else stick with user-specified value. | |
2017 | */ | |
2018 | policer_burst = (!policer_rate ? 0 | |
2019 | : !policer_burst ? 8000 | |
2020 | : policer_burst); | |
2021 | ||
2022 | ovs_mutex_lock(&dev->mutex); | |
2023 | ||
2024 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2025 | &dev->ingress_policer); | |
2026 | ||
2027 | if (dev->policer_rate == policer_rate && | |
2028 | dev->policer_burst == policer_burst) { | |
2029 | /* Assume that settings haven't changed since we last set them. */ | |
2030 | ovs_mutex_unlock(&dev->mutex); | |
2031 | return 0; | |
2032 | } | |
2033 | ||
2034 | /* Destroy any existing ingress policer for the device if one exists */ | |
2035 | if (policer) { | |
2036 | ovsrcu_postpone(free, policer); | |
2037 | } | |
2038 | ||
2039 | if (policer_rate != 0) { | |
2040 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2041 | } else { | |
2042 | policer = NULL; | |
2043 | } | |
2044 | ovsrcu_set(&dev->ingress_policer, policer); | |
2045 | dev->policer_rate = policer_rate; | |
2046 | dev->policer_burst = policer_burst; | |
2047 | ovs_mutex_unlock(&dev->mutex); | |
2048 | ||
2049 | return 0; | |
2050 | } | |
2051 | ||
8a9562d2 PS |
2052 | static int |
2053 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2054 | { | |
2055 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2056 | int ifindex; | |
2057 | ||
2058 | ovs_mutex_lock(&dev->mutex); | |
2059 | ifindex = dev->port_id; | |
2060 | ovs_mutex_unlock(&dev->mutex); | |
2061 | ||
2062 | return ifindex; | |
2063 | } | |
2064 | ||
2065 | static int | |
d46285a2 | 2066 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2067 | { |
d46285a2 | 2068 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2069 | |
2070 | ovs_mutex_lock(&dev->mutex); | |
2071 | check_link_status(dev); | |
2072 | *carrier = dev->link.link_status; | |
58397e6c KT |
2073 | |
2074 | ovs_mutex_unlock(&dev->mutex); | |
2075 | ||
2076 | return 0; | |
2077 | } | |
2078 | ||
2079 | static int | |
d46285a2 | 2080 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2081 | { |
d46285a2 | 2082 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2083 | |
2084 | ovs_mutex_lock(&dev->mutex); | |
2085 | ||
0a0f39df | 2086 | if (is_vhost_running(dev)) { |
58397e6c KT |
2087 | *carrier = 1; |
2088 | } else { | |
2089 | *carrier = 0; | |
2090 | } | |
2091 | ||
8a9562d2 PS |
2092 | ovs_mutex_unlock(&dev->mutex); |
2093 | ||
2094 | return 0; | |
2095 | } | |
2096 | ||
2097 | static long long int | |
d46285a2 | 2098 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2099 | { |
d46285a2 | 2100 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2101 | long long int carrier_resets; |
2102 | ||
2103 | ovs_mutex_lock(&dev->mutex); | |
2104 | carrier_resets = dev->link_reset_cnt; | |
2105 | ovs_mutex_unlock(&dev->mutex); | |
2106 | ||
2107 | return carrier_resets; | |
2108 | } | |
2109 | ||
2110 | static int | |
d46285a2 | 2111 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2112 | long long int interval OVS_UNUSED) |
2113 | { | |
ee32150e | 2114 | return EOPNOTSUPP; |
8a9562d2 PS |
2115 | } |
2116 | ||
2117 | static int | |
2118 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2119 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2120 | enum netdev_flags *old_flagsp) |
2121 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
2122 | { |
2123 | int err; | |
2124 | ||
2125 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
2126 | return EINVAL; | |
2127 | } | |
2128 | ||
2129 | *old_flagsp = dev->flags; | |
2130 | dev->flags |= on; | |
2131 | dev->flags &= ~off; | |
2132 | ||
2133 | if (dev->flags == *old_flagsp) { | |
2134 | return 0; | |
2135 | } | |
2136 | ||
58397e6c KT |
2137 | if (dev->type == DPDK_DEV_ETH) { |
2138 | if (dev->flags & NETDEV_UP) { | |
2139 | err = rte_eth_dev_start(dev->port_id); | |
2140 | if (err) | |
2141 | return -err; | |
2142 | } | |
8a9562d2 | 2143 | |
58397e6c KT |
2144 | if (dev->flags & NETDEV_PROMISC) { |
2145 | rte_eth_promiscuous_enable(dev->port_id); | |
2146 | } | |
8a9562d2 | 2147 | |
58397e6c KT |
2148 | if (!(dev->flags & NETDEV_UP)) { |
2149 | rte_eth_dev_stop(dev->port_id); | |
2150 | } | |
e543851d ZB |
2151 | } else { |
2152 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2153 | * running then change netdev's change_seq to trigger link state | |
2154 | * update. */ | |
e543851d ZB |
2155 | |
2156 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2157 | && is_vhost_running(dev)) { |
e543851d ZB |
2158 | netdev_change_seq_changed(&dev->up); |
2159 | ||
2160 | /* Clear statistics if device is getting up. */ | |
2161 | if (NETDEV_UP & on) { | |
2162 | rte_spinlock_lock(&dev->stats_lock); | |
2163 | memset(&dev->stats, 0, sizeof(dev->stats)); | |
2164 | rte_spinlock_unlock(&dev->stats_lock); | |
2165 | } | |
2166 | } | |
8a9562d2 PS |
2167 | } |
2168 | ||
2169 | return 0; | |
2170 | } | |
2171 | ||
2172 | static int | |
d46285a2 | 2173 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2174 | enum netdev_flags off, enum netdev_flags on, |
2175 | enum netdev_flags *old_flagsp) | |
2176 | { | |
d46285a2 | 2177 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2178 | int error; |
2179 | ||
d46285a2 DDP |
2180 | ovs_mutex_lock(&dev->mutex); |
2181 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2182 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2183 | |
2184 | return error; | |
2185 | } | |
2186 | ||
2187 | static int | |
d46285a2 | 2188 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2189 | { |
d46285a2 | 2190 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2191 | struct rte_eth_dev_info dev_info; |
2192 | ||
7cd1261d | 2193 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2194 | return ENODEV; |
7cd1261d | 2195 | } |
8a9562d2 PS |
2196 | |
2197 | ovs_mutex_lock(&dev->mutex); | |
2198 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2199 | ovs_mutex_unlock(&dev->mutex); | |
2200 | ||
95fb793a | 2201 | smap_add_format(args, "port_no", "%d", dev->port_id); |
8a9562d2 PS |
2202 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); |
2203 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
2204 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2205 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2206 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2207 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2208 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
2209 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
2210 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
2211 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2212 | ||
39c2baa9 | 2213 | if (dev_info.pci_dev) { |
2214 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2215 | dev_info.pci_dev->id.vendor_id); | |
2216 | smap_add_format(args, "pci-device_id", "0x%x", | |
2217 | dev_info.pci_dev->id.device_id); | |
2218 | } | |
8a9562d2 PS |
2219 | |
2220 | return 0; | |
2221 | } | |
2222 | ||
2223 | static void | |
2224 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2225 | OVS_REQUIRES(dev->mutex) | |
2226 | { | |
2227 | enum netdev_flags old_flags; | |
2228 | ||
2229 | if (admin_state) { | |
2230 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2231 | } else { | |
2232 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2233 | } | |
2234 | } | |
2235 | ||
2236 | static void | |
2237 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2238 | const char *argv[], void *aux OVS_UNUSED) | |
2239 | { | |
2240 | bool up; | |
2241 | ||
2242 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2243 | up = true; | |
2244 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2245 | up = false; | |
2246 | } else { | |
2247 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2248 | return; | |
2249 | } | |
2250 | ||
2251 | if (argc > 2) { | |
2252 | struct netdev *netdev = netdev_from_name(argv[1]); | |
2253 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
2254 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
2255 | ||
2256 | ovs_mutex_lock(&dpdk_dev->mutex); | |
2257 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
2258 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
2259 | ||
2260 | netdev_close(netdev); | |
2261 | } else { | |
2262 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2263 | netdev_close(netdev); | |
2264 | return; | |
2265 | } | |
2266 | } else { | |
2267 | struct netdev_dpdk *netdev; | |
2268 | ||
2269 | ovs_mutex_lock(&dpdk_mutex); | |
2270 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
2271 | ovs_mutex_lock(&netdev->mutex); | |
2272 | netdev_dpdk_set_admin_state__(netdev, up); | |
2273 | ovs_mutex_unlock(&netdev->mutex); | |
2274 | } | |
2275 | ovs_mutex_unlock(&dpdk_mutex); | |
2276 | } | |
2277 | unixctl_command_reply(conn, "OK"); | |
2278 | } | |
2279 | ||
58397e6c KT |
2280 | /* |
2281 | * Set virtqueue flags so that we do not receive interrupts. | |
2282 | */ | |
2283 | static void | |
0a0f39df | 2284 | set_irq_status(int vid) |
58397e6c | 2285 | { |
4573fbd3 FL |
2286 | uint32_t i; |
2287 | uint64_t idx; | |
2288 | ||
0a0f39df | 2289 | for (i = 0; i < rte_vhost_get_queue_num(vid); i++) { |
4573fbd3 | 2290 | idx = i * VIRTIO_QNUM; |
0a0f39df CL |
2291 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_RXQ, 0); |
2292 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_TXQ, 0); | |
4573fbd3 FL |
2293 | } |
2294 | } | |
2295 | ||
585a5bea IM |
2296 | /* |
2297 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2298 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2299 | */ |
2300 | static void | |
d46285a2 DDP |
2301 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2302 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2303 | { |
2304 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2305 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea IM |
2306 | |
2307 | enabled_queues = dpdk_rte_mzalloc(total_txqs * sizeof *enabled_queues); | |
2308 | ||
2309 | for (i = 0; i < total_txqs; i++) { | |
2310 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2311 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2312 | enabled_queues[n_enabled++] = i; |
2313 | } | |
2314 | } | |
2315 | ||
2316 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2317 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2318 | n_enabled = 1; |
2319 | } | |
2320 | ||
2321 | k = 0; | |
2322 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2323 | if (dev->tx_q[i].map != i) { |
2324 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2325 | k = (k + 1) % n_enabled; |
2326 | } | |
2327 | } | |
2328 | ||
c1ff66ac | 2329 | VLOG_DBG("TX queue mapping for %s\n", get_vhost_id(dev)); |
585a5bea | 2330 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2331 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2332 | } |
2333 | ||
2334 | rte_free(enabled_queues); | |
2335 | } | |
4573fbd3 | 2336 | |
58397e6c KT |
2337 | /* |
2338 | * A new virtio-net device is added to a vhost port. | |
2339 | */ | |
2340 | static int | |
0a0f39df | 2341 | new_device(int vid) |
58397e6c | 2342 | { |
d46285a2 | 2343 | struct netdev_dpdk *dev; |
58397e6c | 2344 | bool exists = false; |
db8f13b0 | 2345 | int newnode = 0; |
0a0f39df CL |
2346 | char ifname[IF_NAME_SZ]; |
2347 | ||
2348 | rte_vhost_get_ifname(vid, ifname, sizeof(ifname)); | |
58397e6c KT |
2349 | |
2350 | ovs_mutex_lock(&dpdk_mutex); | |
2351 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2352 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac CL |
2353 | ovs_mutex_lock(&dev->mutex); |
2354 | if (strncmp(ifname, get_vhost_id(dev), IF_NAME_SZ) == 0) { | |
0a0f39df | 2355 | uint32_t qp_num = rte_vhost_get_queue_num(vid); |
db8f13b0 CL |
2356 | |
2357 | /* Get NUMA information */ | |
0a0f39df CL |
2358 | newnode = rte_vhost_get_numa_node(vid); |
2359 | if (newnode == -1) { | |
db8f13b0 | 2360 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2361 | ifname); |
db8f13b0 | 2362 | newnode = dev->socket_id; |
db8f13b0 CL |
2363 | } |
2364 | ||
7f5f2bd0 IM |
2365 | if (dev->requested_n_txq != qp_num |
2366 | || dev->requested_n_rxq != qp_num | |
2367 | || dev->requested_socket_id != newnode) { | |
2368 | dev->requested_socket_id = newnode; | |
2369 | dev->requested_n_rxq = qp_num; | |
2370 | dev->requested_n_txq = qp_num; | |
2371 | netdev_request_reconfigure(&dev->up); | |
2372 | } else { | |
2373 | /* Reconfiguration not required. */ | |
2374 | dev->vhost_reconfigured = true; | |
2375 | } | |
81acebda | 2376 | |
0a0f39df | 2377 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2378 | exists = true; |
2379 | ||
58397e6c | 2380 | /* Disable notifications. */ |
0a0f39df | 2381 | set_irq_status(vid); |
e543851d | 2382 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2383 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2384 | break; |
2385 | } | |
c1ff66ac | 2386 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2387 | } |
2388 | ovs_mutex_unlock(&dpdk_mutex); | |
2389 | ||
2390 | if (!exists) { | |
0a0f39df | 2391 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2392 | |
2393 | return -1; | |
2394 | } | |
2395 | ||
0a0f39df CL |
2396 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2397 | ifname, newnode); | |
2398 | ||
58397e6c KT |
2399 | return 0; |
2400 | } | |
2401 | ||
f3ea2ad2 IM |
2402 | /* Clears mapping for all available queues of vhost interface. */ |
2403 | static void | |
2404 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2405 | OVS_REQUIRES(dev->mutex) | |
2406 | { | |
2407 | int i; | |
2408 | ||
81acebda | 2409 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
2410 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
2411 | } | |
2412 | } | |
2413 | ||
58397e6c KT |
2414 | /* |
2415 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2416 | * flag to stop any more packets from being sent or received to/from a VM and | |
2417 | * ensure all currently queued packets have been sent/received before removing | |
2418 | * the device. | |
2419 | */ | |
2420 | static void | |
0a0f39df | 2421 | destroy_device(int vid) |
58397e6c | 2422 | { |
d46285a2 | 2423 | struct netdev_dpdk *dev; |
afee281f | 2424 | bool exists = false; |
0a0f39df CL |
2425 | char ifname[IF_NAME_SZ]; |
2426 | ||
2427 | rte_vhost_get_ifname(vid, ifname, sizeof(ifname)); | |
58397e6c KT |
2428 | |
2429 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2430 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2431 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 2432 | |
d46285a2 | 2433 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
2434 | dev->vhost_reconfigured = false; |
2435 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 2436 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 2437 | |
e543851d | 2438 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2439 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 2440 | exists = true; |
afee281f | 2441 | break; |
58397e6c KT |
2442 | } |
2443 | } | |
afee281f | 2444 | |
58397e6c KT |
2445 | ovs_mutex_unlock(&dpdk_mutex); |
2446 | ||
0a0f39df | 2447 | if (exists) { |
afee281f KT |
2448 | /* |
2449 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2450 | * to NULL, before returning. | |
2451 | */ | |
2452 | ovsrcu_synchronize(); | |
2453 | /* | |
2454 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2455 | * put thread back into quiescent state before returning. | |
2456 | */ | |
2457 | ovsrcu_quiesce_start(); | |
0a0f39df | 2458 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 2459 | } else { |
0a0f39df | 2460 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 2461 | } |
58397e6c KT |
2462 | } |
2463 | ||
585a5bea | 2464 | static int |
0a0f39df | 2465 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 2466 | { |
d46285a2 | 2467 | struct netdev_dpdk *dev; |
585a5bea IM |
2468 | bool exists = false; |
2469 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
2470 | char ifname[IF_NAME_SZ]; |
2471 | ||
2472 | rte_vhost_get_ifname(vid, ifname, sizeof(ifname)); | |
585a5bea IM |
2473 | |
2474 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2475 | return 0; | |
2476 | } | |
2477 | ||
2478 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2479 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac CL |
2480 | ovs_mutex_lock(&dev->mutex); |
2481 | if (strncmp(ifname, get_vhost_id(dev), IF_NAME_SZ) == 0) { | |
585a5bea | 2482 | if (enable) { |
d46285a2 | 2483 | dev->tx_q[qid].map = qid; |
585a5bea | 2484 | } else { |
d46285a2 | 2485 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2486 | } |
d46285a2 | 2487 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2488 | exists = true; |
d46285a2 | 2489 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2490 | break; |
2491 | } | |
c1ff66ac | 2492 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2493 | } |
2494 | ovs_mutex_unlock(&dpdk_mutex); | |
2495 | ||
2496 | if (exists) { | |
0a0f39df CL |
2497 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
2498 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 2499 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 2500 | } else { |
0a0f39df | 2501 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
2502 | return -1; |
2503 | } | |
2504 | ||
2505 | return 0; | |
2506 | } | |
2507 | ||
0a0f39df CL |
2508 | int |
2509 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 2510 | { |
0a0f39df | 2511 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
2512 | } |
2513 | ||
9509913a IS |
2514 | struct ingress_policer * |
2515 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
2516 | { | |
2517 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
2518 | } | |
2519 | ||
58397e6c KT |
2520 | /* |
2521 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
2522 | * configuration has been fully complete. | |
2523 | */ | |
bce01e3a | 2524 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
2525 | { |
2526 | .new_device = new_device, | |
2527 | .destroy_device = destroy_device, | |
585a5bea | 2528 | .vring_state_changed = vring_state_changed |
58397e6c KT |
2529 | }; |
2530 | ||
2531 | static void * | |
7d1ced01 | 2532 | start_vhost_loop(void *dummy OVS_UNUSED) |
58397e6c KT |
2533 | { |
2534 | pthread_detach(pthread_self()); | |
41987644 | 2535 | /* Put the vhost thread into quiescent state. */ |
618f44f7 | 2536 | ovsrcu_quiesce_start(); |
58397e6c KT |
2537 | rte_vhost_driver_session_start(); |
2538 | return NULL; | |
2539 | } | |
2540 | ||
2541 | static int | |
2542 | dpdk_vhost_class_init(void) | |
7d1ced01 CL |
2543 | { |
2544 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
362ca396 | 2545 | rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4 |
2546 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
2547 | | 1ULL << VIRTIO_NET_F_CSUM); | |
2548 | ||
7d1ced01 CL |
2549 | ovs_thread_create("vhost_thread", start_vhost_loop, NULL); |
2550 | return 0; | |
2551 | } | |
2552 | ||
033e9df2 DDP |
2553 | static void |
2554 | dpdk_common_init(void) | |
2555 | { | |
2556 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2557 | "[netdev] up|down", 1, 2, | |
2558 | netdev_dpdk_set_admin_state, NULL); | |
2559 | ||
033e9df2 DDP |
2560 | } |
2561 | ||
95fb793a | 2562 | /* Client Rings */ |
2563 | ||
95fb793a | 2564 | static int |
2565 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
2566 | unsigned int *eth_port_id) | |
2567 | { | |
2568 | struct dpdk_ring *ivshmem; | |
ca7e7bee | 2569 | char ring_name[RTE_RING_NAMESIZE]; |
95fb793a | 2570 | int err; |
2571 | ||
2572 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
2573 | if (ivshmem == NULL) { | |
2574 | return ENOMEM; | |
2575 | } | |
2576 | ||
7251515e | 2577 | /* XXX: Add support for multiquque ring. */ |
ca7e7bee | 2578 | err = snprintf(ring_name, sizeof(ring_name), "%s_tx", dev_name); |
95fb793a | 2579 | if (err < 0) { |
2580 | return -err; | |
2581 | } | |
2582 | ||
8f0a76c9 | 2583 | /* Create single producer tx ring, netdev does explicit locking. */ |
7251515e | 2584 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2585 | RING_F_SP_ENQ); |
95fb793a | 2586 | if (ivshmem->cring_tx == NULL) { |
2587 | rte_free(ivshmem); | |
2588 | return ENOMEM; | |
2589 | } | |
2590 | ||
ca7e7bee | 2591 | err = snprintf(ring_name, sizeof(ring_name), "%s_rx", dev_name); |
95fb793a | 2592 | if (err < 0) { |
2593 | return -err; | |
2594 | } | |
2595 | ||
8f0a76c9 | 2596 | /* Create single consumer rx ring, netdev does explicit locking. */ |
7251515e | 2597 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2598 | RING_F_SC_DEQ); |
95fb793a | 2599 | if (ivshmem->cring_rx == NULL) { |
2600 | rte_free(ivshmem); | |
2601 | return ENOMEM; | |
2602 | } | |
2603 | ||
d7310583 DDP |
2604 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
2605 | &ivshmem->cring_tx, 1, SOCKET0); | |
2606 | ||
95fb793a | 2607 | if (err < 0) { |
2608 | rte_free(ivshmem); | |
2609 | return ENODEV; | |
2610 | } | |
2611 | ||
2612 | ivshmem->user_port_id = port_no; | |
2613 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
417e7e66 | 2614 | ovs_list_push_back(&dpdk_ring_list, &ivshmem->list_node); |
95fb793a | 2615 | |
2616 | *eth_port_id = ivshmem->eth_port_id; | |
2617 | return 0; | |
2618 | } | |
2619 | ||
2620 | static int | |
64839cf4 WT |
2621 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) |
2622 | OVS_REQUIRES(dpdk_mutex) | |
95fb793a | 2623 | { |
2624 | struct dpdk_ring *ivshmem; | |
2625 | unsigned int port_no; | |
2626 | int err = 0; | |
2627 | ||
2628 | /* Names always start with "dpdkr" */ | |
2629 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2630 | if (err) { | |
2631 | return err; | |
2632 | } | |
2633 | ||
2634 | /* look through our list to find the device */ | |
2635 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { | |
2636 | if (ivshmem->user_port_id == port_no) { | |
58397e6c | 2637 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
95fb793a | 2638 | *eth_port_id = ivshmem->eth_port_id; /* really all that is needed */ |
2639 | return 0; | |
2640 | } | |
2641 | } | |
2642 | /* Need to create the device rings */ | |
2643 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2644 | } | |
2645 | ||
7251515e | 2646 | static int |
d46285a2 | 2647 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
324c8374 IM |
2648 | struct dp_packet_batch *batch, bool may_steal, |
2649 | bool concurrent_txq) | |
7251515e | 2650 | { |
d46285a2 | 2651 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b99bb05 MG |
2652 | unsigned i; |
2653 | ||
2654 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that the | |
2655 | * rss hash field is clear. This is because the same mbuf may be modified by | |
2656 | * the consumer of the ring and return into the datapath without recalculating | |
2657 | * the RSS hash. */ | |
64839cf4 WT |
2658 | for (i = 0; i < batch->count; i++) { |
2659 | dp_packet_rss_invalidate(batch->packets[i]); | |
1b99bb05 | 2660 | } |
7251515e | 2661 | |
324c8374 | 2662 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e DV |
2663 | return 0; |
2664 | } | |
2665 | ||
95fb793a | 2666 | static int |
2667 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2668 | { | |
2669 | unsigned int port_no = 0; | |
2670 | int err = 0; | |
2671 | ||
2672 | if (rte_eal_init_ret) { | |
2673 | return rte_eal_init_ret; | |
2674 | } | |
2675 | ||
2676 | ovs_mutex_lock(&dpdk_mutex); | |
2677 | ||
2678 | err = dpdk_ring_open(netdev->name, &port_no); | |
2679 | if (err) { | |
2680 | goto unlock_dpdk; | |
2681 | } | |
2682 | ||
58397e6c | 2683 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
95fb793a | 2684 | |
2685 | unlock_dpdk: | |
2686 | ovs_mutex_unlock(&dpdk_mutex); | |
2687 | return err; | |
2688 | } | |
2689 | ||
0bf765f7 IS |
2690 | /* QoS Functions */ |
2691 | ||
2692 | /* | |
2693 | * Initialize QoS configuration operations. | |
2694 | */ | |
2695 | static void | |
2696 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
2697 | { | |
2698 | conf->ops = ops; | |
2699 | } | |
2700 | ||
2701 | /* | |
2702 | * Search existing QoS operations in qos_ops and compare each set of | |
2703 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
2704 | * else return NULL | |
2705 | */ | |
2706 | static const struct dpdk_qos_ops * | |
2707 | qos_lookup_name(const char *name) | |
2708 | { | |
2709 | const struct dpdk_qos_ops *const *opsp; | |
2710 | ||
2711 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2712 | const struct dpdk_qos_ops *ops = *opsp; | |
2713 | if (!strcmp(name, ops->qos_name)) { | |
2714 | return ops; | |
2715 | } | |
2716 | } | |
2717 | return NULL; | |
2718 | } | |
2719 | ||
2720 | /* | |
2721 | * Call qos_destruct to clean up items associated with the netdevs | |
2722 | * qos_conf. Set netdevs qos_conf to NULL. | |
2723 | */ | |
2724 | static void | |
d46285a2 | 2725 | qos_delete_conf(struct netdev *netdev) |
0bf765f7 | 2726 | { |
d46285a2 | 2727 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 2728 | |
d46285a2 DDP |
2729 | rte_spinlock_lock(&dev->qos_lock); |
2730 | if (dev->qos_conf) { | |
2731 | if (dev->qos_conf->ops->qos_destruct) { | |
2732 | dev->qos_conf->ops->qos_destruct(netdev, dev->qos_conf); | |
0bf765f7 | 2733 | } |
d46285a2 | 2734 | dev->qos_conf = NULL; |
0bf765f7 | 2735 | } |
d46285a2 | 2736 | rte_spinlock_unlock(&dev->qos_lock); |
0bf765f7 IS |
2737 | } |
2738 | ||
2739 | static int | |
2740 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
2741 | struct sset *types) | |
2742 | { | |
2743 | const struct dpdk_qos_ops *const *opsp; | |
2744 | ||
2745 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2746 | const struct dpdk_qos_ops *ops = *opsp; | |
2747 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
2748 | sset_add(types, ops->qos_name); | |
2749 | } | |
2750 | } | |
2751 | return 0; | |
2752 | } | |
2753 | ||
2754 | static int | |
d46285a2 | 2755 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
2756 | const char **typep, struct smap *details) |
2757 | { | |
d46285a2 | 2758 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 IS |
2759 | int error = 0; |
2760 | ||
d46285a2 DDP |
2761 | ovs_mutex_lock(&dev->mutex); |
2762 | if(dev->qos_conf) { | |
2763 | *typep = dev->qos_conf->ops->qos_name; | |
2764 | error = (dev->qos_conf->ops->qos_get | |
2765 | ? dev->qos_conf->ops->qos_get(netdev, details): 0); | |
d03603c4 MC |
2766 | } else { |
2767 | /* No QoS configuration set, return an empty string */ | |
2768 | *typep = ""; | |
0bf765f7 | 2769 | } |
d46285a2 | 2770 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2771 | |
2772 | return error; | |
2773 | } | |
2774 | ||
2775 | static int | |
d46285a2 | 2776 | netdev_dpdk_set_qos(struct netdev *netdev, |
0bf765f7 IS |
2777 | const char *type, const struct smap *details) |
2778 | { | |
d46285a2 | 2779 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 IS |
2780 | const struct dpdk_qos_ops *new_ops = NULL; |
2781 | int error = 0; | |
2782 | ||
2783 | /* If type is empty or unsupported then the current QoS configuration | |
2784 | * for the dpdk-netdev can be destroyed */ | |
2785 | new_ops = qos_lookup_name(type); | |
2786 | ||
2787 | if (type[0] == '\0' || !new_ops || !new_ops->qos_construct) { | |
d46285a2 | 2788 | qos_delete_conf(netdev); |
0bf765f7 IS |
2789 | return EOPNOTSUPP; |
2790 | } | |
2791 | ||
d46285a2 | 2792 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 2793 | |
d46285a2 DDP |
2794 | if (dev->qos_conf) { |
2795 | if (new_ops == dev->qos_conf->ops) { | |
2796 | error = new_ops->qos_set ? new_ops->qos_set(netdev, details) : 0; | |
0bf765f7 IS |
2797 | } else { |
2798 | /* Delete existing QoS configuration. */ | |
d46285a2 DDP |
2799 | qos_delete_conf(netdev); |
2800 | ovs_assert(dev->qos_conf == NULL); | |
0bf765f7 IS |
2801 | |
2802 | /* Install new QoS configuration. */ | |
d46285a2 | 2803 | error = new_ops->qos_construct(netdev, details); |
0bf765f7 IS |
2804 | } |
2805 | } else { | |
d46285a2 | 2806 | error = new_ops->qos_construct(netdev, details); |
7ea266e9 IS |
2807 | } |
2808 | ||
2809 | ovs_assert((error == 0) == (dev->qos_conf != NULL)); | |
2810 | if (error) { | |
2811 | VLOG_ERR("Failed to set QoS type %s on port %s, returned error: %s", | |
2812 | type, netdev->name, rte_strerror(-error)); | |
0bf765f7 IS |
2813 | } |
2814 | ||
d46285a2 | 2815 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2816 | return error; |
2817 | } | |
2818 | ||
2819 | /* egress-policer details */ | |
2820 | ||
2821 | struct egress_policer { | |
2822 | struct qos_conf qos_conf; | |
2823 | struct rte_meter_srtcm_params app_srtcm_params; | |
2824 | struct rte_meter_srtcm egress_meter; | |
2825 | }; | |
2826 | ||
2827 | static struct egress_policer * | |
d46285a2 | 2828 | egress_policer_get__(const struct netdev *netdev) |
0bf765f7 | 2829 | { |
d46285a2 DDP |
2830 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2831 | return CONTAINER_OF(dev->qos_conf, struct egress_policer, qos_conf); | |
0bf765f7 IS |
2832 | } |
2833 | ||
2834 | static int | |
d46285a2 DDP |
2835 | egress_policer_qos_construct(struct netdev *netdev, |
2836 | const struct smap *details) | |
0bf765f7 | 2837 | { |
d46285a2 | 2838 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 2839 | struct egress_policer *policer; |
0bf765f7 IS |
2840 | int err = 0; |
2841 | ||
d46285a2 | 2842 | rte_spinlock_lock(&dev->qos_lock); |
0bf765f7 IS |
2843 | policer = xmalloc(sizeof *policer); |
2844 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
d46285a2 | 2845 | dev->qos_conf = &policer->qos_conf; |
13c1637f BP |
2846 | policer->app_srtcm_params.cir = smap_get_ullong(details, "cir", 0); |
2847 | policer->app_srtcm_params.cbs = smap_get_ullong(details, "cbs", 0); | |
0bf765f7 IS |
2848 | policer->app_srtcm_params.ebs = 0; |
2849 | err = rte_meter_srtcm_config(&policer->egress_meter, | |
2850 | &policer->app_srtcm_params); | |
7ea266e9 IS |
2851 | |
2852 | if (err < 0) { | |
2853 | /* Error occurred during rte_meter creation, destroy the policer | |
2854 | * and set the qos configuration for the netdev dpdk to NULL | |
2855 | */ | |
2856 | free(policer); | |
2857 | dev->qos_conf = NULL; | |
2858 | err = -err; | |
2859 | } | |
d46285a2 | 2860 | rte_spinlock_unlock(&dev->qos_lock); |
0bf765f7 IS |
2861 | |
2862 | return err; | |
2863 | } | |
2864 | ||
2865 | static void | |
d46285a2 | 2866 | egress_policer_qos_destruct(struct netdev *netdev OVS_UNUSED, |
0bf765f7 IS |
2867 | struct qos_conf *conf) |
2868 | { | |
2869 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
2870 | qos_conf); | |
2871 | free(policer); | |
2872 | } | |
2873 | ||
2874 | static int | |
2875 | egress_policer_qos_get(const struct netdev *netdev, struct smap *details) | |
2876 | { | |
2877 | struct egress_policer *policer = egress_policer_get__(netdev); | |
2878 | smap_add_format(details, "cir", "%llu", | |
2879 | 1ULL * policer->app_srtcm_params.cir); | |
2880 | smap_add_format(details, "cbs", "%llu", | |
2881 | 1ULL * policer->app_srtcm_params.cbs); | |
050c60bf | 2882 | |
0bf765f7 IS |
2883 | return 0; |
2884 | } | |
2885 | ||
2886 | static int | |
d46285a2 | 2887 | egress_policer_qos_set(struct netdev *netdev, const struct smap *details) |
0bf765f7 IS |
2888 | { |
2889 | struct egress_policer *policer; | |
7ea266e9 | 2890 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 IS |
2891 | int err = 0; |
2892 | ||
d46285a2 | 2893 | policer = egress_policer_get__(netdev); |
7ea266e9 | 2894 | rte_spinlock_lock(&dev->qos_lock); |
13c1637f BP |
2895 | policer->app_srtcm_params.cir = smap_get_ullong(details, "cir", 0); |
2896 | policer->app_srtcm_params.cbs = smap_get_ullong(details, "cbs", 0); | |
0bf765f7 IS |
2897 | policer->app_srtcm_params.ebs = 0; |
2898 | err = rte_meter_srtcm_config(&policer->egress_meter, | |
2899 | &policer->app_srtcm_params); | |
2900 | ||
7ea266e9 IS |
2901 | if (err < 0) { |
2902 | /* Error occurred during rte_meter creation, destroy the policer | |
2903 | * and set the qos configuration for the netdev dpdk to NULL | |
2904 | */ | |
2905 | free(policer); | |
2906 | dev->qos_conf = NULL; | |
2907 | err = -err; | |
2908 | } | |
2909 | rte_spinlock_unlock(&dev->qos_lock); | |
2910 | ||
0bf765f7 IS |
2911 | return err; |
2912 | } | |
2913 | ||
0bf765f7 | 2914 | static int |
f3926f29 | 2915 | egress_policer_run(struct netdev *netdev, struct rte_mbuf **pkts, int pkt_cnt) |
0bf765f7 | 2916 | { |
0bf765f7 | 2917 | int cnt = 0; |
d46285a2 | 2918 | struct egress_policer *policer = egress_policer_get__(netdev); |
0bf765f7 | 2919 | |
f3926f29 | 2920 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, pkt_cnt); |
0bf765f7 IS |
2921 | |
2922 | return cnt; | |
2923 | } | |
2924 | ||
2925 | static const struct dpdk_qos_ops egress_policer_ops = { | |
2926 | "egress-policer", /* qos_name */ | |
2927 | egress_policer_qos_construct, | |
2928 | egress_policer_qos_destruct, | |
2929 | egress_policer_qos_get, | |
2930 | egress_policer_qos_set, | |
2931 | egress_policer_run | |
2932 | }; | |
2933 | ||
050c60bf DDP |
2934 | static int |
2935 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
2936 | { | |
2937 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2938 | int err = 0; | |
2939 | ||
2940 | ovs_mutex_lock(&dpdk_mutex); | |
2941 | ovs_mutex_lock(&dev->mutex); | |
2942 | ||
2943 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 MK |
2944 | && netdev->n_rxq == dev->requested_n_rxq |
2945 | && dev->mtu == dev->requested_mtu) { | |
050c60bf DDP |
2946 | /* Reconfiguration is unnecessary */ |
2947 | ||
2948 | goto out; | |
2949 | } | |
2950 | ||
2951 | rte_eth_dev_stop(dev->port_id); | |
2952 | ||
0072e931 MK |
2953 | if (dev->mtu != dev->requested_mtu) { |
2954 | netdev_dpdk_mempool_configure(dev); | |
2955 | } | |
2956 | ||
050c60bf DDP |
2957 | netdev->n_txq = dev->requested_n_txq; |
2958 | netdev->n_rxq = dev->requested_n_rxq; | |
2959 | ||
2960 | rte_free(dev->tx_q); | |
2961 | err = dpdk_eth_dev_init(dev); | |
81acebda | 2962 | netdev_dpdk_alloc_txq(dev, netdev->n_txq); |
050c60bf | 2963 | |
0072e931 MK |
2964 | netdev_change_seq_changed(netdev); |
2965 | ||
050c60bf DDP |
2966 | out: |
2967 | ||
2968 | ovs_mutex_unlock(&dev->mutex); | |
2969 | ovs_mutex_unlock(&dpdk_mutex); | |
2970 | ||
2971 | return err; | |
2972 | } | |
2973 | ||
2974 | static int | |
53f50d24 | 2975 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) |
050c60bf DDP |
2976 | { |
2977 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
c1ff66ac | 2978 | int err = 0; |
050c60bf DDP |
2979 | |
2980 | ovs_mutex_lock(&dpdk_mutex); | |
2981 | ovs_mutex_lock(&dev->mutex); | |
2982 | ||
2983 | netdev->n_txq = dev->requested_n_txq; | |
2984 | netdev->n_rxq = dev->requested_n_rxq; | |
2985 | ||
81acebda IM |
2986 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
2987 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
2988 | dev->tx_q[0].map = 0; | |
2989 | } | |
2990 | ||
2991 | netdev_dpdk_remap_txqs(dev); | |
2992 | ||
0072e931 MK |
2993 | if (dev->requested_socket_id != dev->socket_id |
2994 | || dev->requested_mtu != dev->mtu) { | |
2995 | if (!netdev_dpdk_mempool_configure(dev)) { | |
2996 | netdev_change_seq_changed(netdev); | |
db8f13b0 CL |
2997 | } |
2998 | } | |
2999 | ||
0a0f39df CL |
3000 | if (netdev_dpdk_get_vid(dev) >= 0) { |
3001 | dev->vhost_reconfigured = true; | |
81acebda IM |
3002 | } |
3003 | ||
c1ff66ac CL |
3004 | /* Configure vHost client mode if requested and if the following criteria |
3005 | * are met: | |
3006 | * 1. Device is currently in 'server' mode. | |
3007 | * 2. Device is currently not active. | |
3008 | * 3. A path has been specified. | |
3009 | */ | |
3010 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
3011 | && !(netdev_dpdk_get_vid(dev) >= 0) | |
3012 | && strlen(dev->vhost_client_id)) { | |
3013 | /* Unregister server-mode device */ | |
3014 | char *vhost_id = xstrdup(get_vhost_id(dev)); | |
3015 | ||
3016 | ovs_mutex_unlock(&dev->mutex); | |
3017 | ovs_mutex_unlock(&dpdk_mutex); | |
3018 | err = dpdk_vhost_driver_unregister(dev, vhost_id); | |
3019 | free(vhost_id); | |
3020 | ovs_mutex_lock(&dpdk_mutex); | |
3021 | ovs_mutex_lock(&dev->mutex); | |
3022 | if (err) { | |
3023 | VLOG_ERR("Unable to remove vhost-user socket %s", | |
3024 | get_vhost_id(dev)); | |
3025 | } else { | |
3026 | fatal_signal_remove_file_to_unlink(get_vhost_id(dev)); | |
3027 | /* Register client-mode device */ | |
3028 | err = rte_vhost_driver_register(dev->vhost_client_id, | |
3029 | RTE_VHOST_USER_CLIENT); | |
3030 | if (err) { | |
3031 | VLOG_ERR("vhost-user device setup failure for device %s\n", | |
3032 | dev->vhost_client_id); | |
3033 | } else { | |
3034 | /* Configuration successful */ | |
3035 | dev->vhost_driver_flags |= RTE_VHOST_USER_CLIENT; | |
3036 | VLOG_INFO("vHost User device '%s' changed to 'client' mode, " | |
3037 | "using client socket '%s'", | |
3038 | dev->up.name, get_vhost_id(dev)); | |
3039 | } | |
3040 | } | |
3041 | } | |
3042 | ||
050c60bf DDP |
3043 | ovs_mutex_unlock(&dev->mutex); |
3044 | ovs_mutex_unlock(&dpdk_mutex); | |
3045 | ||
0072e931 | 3046 | return 0; |
050c60bf DDP |
3047 | } |
3048 | ||
53f50d24 | 3049 | #define NETDEV_DPDK_CLASS(NAME, CONSTRUCT, DESTRUCT, \ |
81acebda IM |
3050 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
3051 | GET_CARRIER, GET_STATS, \ | |
3052 | GET_FEATURES, GET_STATUS, \ | |
3053 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3054 | { \ |
3055 | NAME, \ | |
118c77b1 | 3056 | true, /* is_pmd */ \ |
53f50d24 | 3057 | NULL, /* init */ \ |
95fb793a | 3058 | NULL, /* netdev_dpdk_run */ \ |
3059 | NULL, /* netdev_dpdk_wait */ \ | |
3060 | \ | |
3061 | netdev_dpdk_alloc, \ | |
3062 | CONSTRUCT, \ | |
58397e6c | 3063 | DESTRUCT, \ |
95fb793a | 3064 | netdev_dpdk_dealloc, \ |
3065 | netdev_dpdk_get_config, \ | |
81acebda | 3066 | SET_CONFIG, \ |
95fb793a | 3067 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3068 | NULL, /* build header */ \ |
3069 | NULL, /* push header */ \ | |
3070 | NULL, /* pop header */ \ | |
7dec44fe | 3071 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3072 | SET_TX_MULTIQ, \ |
95fb793a | 3073 | \ |
7251515e | 3074 | SEND, /* send */ \ |
95fb793a | 3075 | NULL, /* send_wait */ \ |
3076 | \ | |
3077 | netdev_dpdk_set_etheraddr, \ | |
3078 | netdev_dpdk_get_etheraddr, \ | |
3079 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3080 | netdev_dpdk_set_mtu, \ |
95fb793a | 3081 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3082 | GET_CARRIER, \ |
95fb793a | 3083 | netdev_dpdk_get_carrier_resets, \ |
3084 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
3085 | GET_STATS, \ |
3086 | GET_FEATURES, \ | |
95fb793a | 3087 | NULL, /* set_advertisements */ \ |
3088 | \ | |
9509913a | 3089 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3090 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3091 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3092 | netdev_dpdk_get_qos, \ |
3093 | netdev_dpdk_set_qos, \ | |
95fb793a | 3094 | NULL, /* get_queue */ \ |
3095 | NULL, /* set_queue */ \ | |
3096 | NULL, /* delete_queue */ \ | |
3097 | NULL, /* get_queue_stats */ \ | |
3098 | NULL, /* queue_dump_start */ \ | |
3099 | NULL, /* queue_dump_next */ \ | |
3100 | NULL, /* queue_dump_done */ \ | |
3101 | NULL, /* dump_queue_stats */ \ | |
3102 | \ | |
95fb793a | 3103 | NULL, /* set_in4 */ \ |
a8704b50 | 3104 | NULL, /* get_addr_list */ \ |
95fb793a | 3105 | NULL, /* add_router */ \ |
3106 | NULL, /* get_next_hop */ \ | |
58397e6c | 3107 | GET_STATUS, \ |
95fb793a | 3108 | NULL, /* arp_lookup */ \ |
3109 | \ | |
3110 | netdev_dpdk_update_flags, \ | |
050c60bf | 3111 | RECONFIGURE, \ |
95fb793a | 3112 | \ |
3113 | netdev_dpdk_rxq_alloc, \ | |
3114 | netdev_dpdk_rxq_construct, \ | |
3115 | netdev_dpdk_rxq_destruct, \ | |
3116 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3117 | RXQ_RECV, \ |
95fb793a | 3118 | NULL, /* rx_wait */ \ |
3119 | NULL, /* rxq_drain */ \ | |
3120 | } | |
8a9562d2 | 3121 | |
7d1ced01 CL |
3122 | static int |
3123 | process_vhost_flags(char *flag, char *default_val, int size, | |
bab69409 AC |
3124 | const struct smap *ovs_other_config, |
3125 | char **new_val) | |
7d1ced01 | 3126 | { |
bab69409 | 3127 | const char *val; |
7d1ced01 CL |
3128 | int changed = 0; |
3129 | ||
bab69409 AC |
3130 | val = smap_get(ovs_other_config, flag); |
3131 | ||
41987644 CL |
3132 | /* Process the vhost-sock-dir flag if it is provided, otherwise resort to |
3133 | * default value. | |
7d1ced01 | 3134 | */ |
bab69409 | 3135 | if (val && (strlen(val) <= size)) { |
7d1ced01 | 3136 | changed = 1; |
bab69409 | 3137 | *new_val = xstrdup(val); |
7d1ced01 CL |
3138 | VLOG_INFO("User-provided %s in use: %s", flag, *new_val); |
3139 | } else { | |
3140 | VLOG_INFO("No %s provided - defaulting to %s", flag, default_val); | |
3141 | *new_val = default_val; | |
3142 | } | |
3143 | ||
3144 | return changed; | |
3145 | } | |
3146 | ||
bab69409 AC |
3147 | static char ** |
3148 | grow_argv(char ***argv, size_t cur_siz, size_t grow_by) | |
8a9562d2 | 3149 | { |
bab69409 AC |
3150 | return xrealloc(*argv, sizeof(char *) * (cur_siz + grow_by)); |
3151 | } | |
8a9562d2 | 3152 | |
bab69409 AC |
3153 | static void |
3154 | dpdk_option_extend(char ***argv, int argc, const char *option, | |
3155 | const char *value) | |
3156 | { | |
3157 | char **newargv = grow_argv(argv, argc, 2); | |
3158 | *argv = newargv; | |
3159 | newargv[argc] = xstrdup(option); | |
3160 | newargv[argc+1] = xstrdup(value); | |
3161 | } | |
8a9562d2 | 3162 | |
52a57e36 AC |
3163 | static char ** |
3164 | move_argv(char ***argv, size_t cur_size, char **src_argv, size_t src_argc) | |
3165 | { | |
3166 | char **newargv = grow_argv(argv, cur_size, src_argc); | |
3167 | while (src_argc--) { | |
3168 | newargv[cur_size+src_argc] = src_argv[src_argc]; | |
3169 | src_argv[src_argc] = NULL; | |
3170 | } | |
3171 | return newargv; | |
3172 | } | |
3173 | ||
eac84432 AC |
3174 | static int |
3175 | extra_dpdk_args(const char *ovs_extra_config, char ***argv, int argc) | |
3176 | { | |
3177 | int ret = argc; | |
3178 | char *release_tok = xstrdup(ovs_extra_config); | |
27373420 | 3179 | char *tok, *endptr = NULL; |
eac84432 AC |
3180 | |
3181 | for (tok = strtok_r(release_tok, " ", &endptr); tok != NULL; | |
3182 | tok = strtok_r(NULL, " ", &endptr)) { | |
3183 | char **newarg = grow_argv(argv, ret, 1); | |
3184 | *argv = newarg; | |
3185 | newarg[ret++] = xstrdup(tok); | |
3186 | } | |
3187 | free(release_tok); | |
3188 | return ret; | |
3189 | } | |
3190 | ||
52a57e36 AC |
3191 | static bool |
3192 | argv_contains(char **argv_haystack, const size_t argc_haystack, | |
3193 | const char *needle) | |
3194 | { | |
3195 | for (size_t i = 0; i < argc_haystack; ++i) { | |
3196 | if (!strcmp(argv_haystack[i], needle)) | |
3197 | return true; | |
3198 | } | |
3199 | return false; | |
3200 | } | |
3201 | ||
bab69409 AC |
3202 | static int |
3203 | construct_dpdk_options(const struct smap *ovs_other_config, | |
52a57e36 AC |
3204 | char ***argv, const int initial_size, |
3205 | char **extra_args, const size_t extra_argc) | |
bab69409 AC |
3206 | { |
3207 | struct dpdk_options_map { | |
3208 | const char *ovs_configuration; | |
3209 | const char *dpdk_option; | |
3210 | bool default_enabled; | |
3211 | const char *default_value; | |
3212 | } opts[] = { | |
3213 | {"dpdk-lcore-mask", "-c", false, NULL}, | |
3214 | {"dpdk-hugepage-dir", "--huge-dir", false, NULL}, | |
3215 | }; | |
3216 | ||
3217 | int i, ret = initial_size; | |
3218 | ||
3219 | /*First, construct from the flat-options (non-mutex)*/ | |
3220 | for (i = 0; i < ARRAY_SIZE(opts); ++i) { | |
3221 | const char *lookup = smap_get(ovs_other_config, | |
3222 | opts[i].ovs_configuration); | |
3223 | if (!lookup && opts[i].default_enabled) { | |
3224 | lookup = opts[i].default_value; | |
3225 | } | |
8a9562d2 | 3226 | |
bab69409 | 3227 | if (lookup) { |
52a57e36 AC |
3228 | if (!argv_contains(extra_args, extra_argc, opts[i].dpdk_option)) { |
3229 | dpdk_option_extend(argv, ret, opts[i].dpdk_option, lookup); | |
3230 | ret += 2; | |
3231 | } else { | |
3232 | VLOG_WARN("Ignoring database defined option '%s' due to " | |
3233 | "dpdk_extras config", opts[i].dpdk_option); | |
3234 | } | |
bab69409 AC |
3235 | } |
3236 | } | |
3237 | ||
3238 | return ret; | |
3239 | } | |
3240 | ||
3241 | #define MAX_DPDK_EXCL_OPTS 10 | |
3242 | ||
3243 | static int | |
3244 | construct_dpdk_mutex_options(const struct smap *ovs_other_config, | |
52a57e36 AC |
3245 | char ***argv, const int initial_size, |
3246 | char **extra_args, const size_t extra_argc) | |
bab69409 AC |
3247 | { |
3248 | struct dpdk_exclusive_options_map { | |
3249 | const char *category; | |
3250 | const char *ovs_dpdk_options[MAX_DPDK_EXCL_OPTS]; | |
3251 | const char *eal_dpdk_options[MAX_DPDK_EXCL_OPTS]; | |
3252 | const char *default_value; | |
3253 | int default_option; | |
3254 | } excl_opts[] = { | |
3255 | {"memory type", | |
3256 | {"dpdk-alloc-mem", "dpdk-socket-mem", NULL,}, | |
3257 | {"-m", "--socket-mem", NULL,}, | |
3258 | "1024,0", 1 | |
3259 | }, | |
3260 | }; | |
3261 | ||
3262 | int i, ret = initial_size; | |
3263 | for (i = 0; i < ARRAY_SIZE(excl_opts); ++i) { | |
3264 | int found_opts = 0, scan, found_pos = -1; | |
3265 | const char *found_value; | |
3266 | struct dpdk_exclusive_options_map *popt = &excl_opts[i]; | |
3267 | ||
3268 | for (scan = 0; scan < MAX_DPDK_EXCL_OPTS | |
3269 | && popt->ovs_dpdk_options[scan]; ++scan) { | |
3270 | const char *lookup = smap_get(ovs_other_config, | |
3271 | popt->ovs_dpdk_options[scan]); | |
3272 | if (lookup && strlen(lookup)) { | |
3273 | found_opts++; | |
3274 | found_pos = scan; | |
3275 | found_value = lookup; | |
3276 | } | |
3277 | } | |
3278 | ||
3279 | if (!found_opts) { | |
3280 | if (popt->default_option) { | |
3281 | found_pos = popt->default_option; | |
3282 | found_value = popt->default_value; | |
3283 | } else { | |
3284 | continue; | |
3285 | } | |
dfc89a55 | 3286 | } |
bab69409 AC |
3287 | |
3288 | if (found_opts > 1) { | |
3289 | VLOG_ERR("Multiple defined options for %s. Please check your" | |
3290 | " database settings and reconfigure if necessary.", | |
3291 | popt->category); | |
3292 | } | |
3293 | ||
52a57e36 AC |
3294 | if (!argv_contains(extra_args, extra_argc, |
3295 | popt->eal_dpdk_options[found_pos])) { | |
3296 | dpdk_option_extend(argv, ret, popt->eal_dpdk_options[found_pos], | |
3297 | found_value); | |
3298 | ret += 2; | |
3299 | } else { | |
3300 | VLOG_WARN("Ignoring database defined option '%s' due to " | |
3301 | "dpdk_extras config", popt->eal_dpdk_options[found_pos]); | |
3302 | } | |
bab69409 AC |
3303 | } |
3304 | ||
3305 | return ret; | |
3306 | } | |
3307 | ||
3308 | static int | |
88964e64 AC |
3309 | get_dpdk_args(const struct smap *ovs_other_config, char ***argv, |
3310 | int argc) | |
bab69409 | 3311 | { |
eac84432 | 3312 | const char *extra_configuration; |
52a57e36 AC |
3313 | char **extra_args = NULL; |
3314 | int i; | |
3315 | size_t extra_argc = 0; | |
eac84432 AC |
3316 | |
3317 | extra_configuration = smap_get(ovs_other_config, "dpdk-extra"); | |
3318 | if (extra_configuration) { | |
52a57e36 | 3319 | extra_argc = extra_dpdk_args(extra_configuration, &extra_args, 0); |
eac84432 | 3320 | } |
52a57e36 AC |
3321 | |
3322 | i = construct_dpdk_options(ovs_other_config, argv, argc, extra_args, | |
3323 | extra_argc); | |
3324 | i = construct_dpdk_mutex_options(ovs_other_config, argv, i, extra_args, | |
3325 | extra_argc); | |
3326 | ||
3327 | if (extra_configuration) { | |
3328 | *argv = move_argv(argv, i, extra_args, extra_argc); | |
3329 | } | |
3330 | ||
3331 | return i + extra_argc; | |
bab69409 AC |
3332 | } |
3333 | ||
3334 | static char **dpdk_argv; | |
3335 | static int dpdk_argc; | |
3336 | ||
3337 | static void | |
3338 | deferred_argv_release(void) | |
3339 | { | |
3340 | int result; | |
3341 | for (result = 0; result < dpdk_argc; ++result) { | |
3342 | free(dpdk_argv[result]); | |
3343 | } | |
3344 | ||
3345 | free(dpdk_argv); | |
3346 | } | |
3347 | ||
3348 | static void | |
3349 | dpdk_init__(const struct smap *ovs_other_config) | |
3350 | { | |
3351 | char **argv = NULL; | |
3352 | int result; | |
88964e64 AC |
3353 | int argc, argc_tmp; |
3354 | bool auto_determine = true; | |
d8e2f4cc | 3355 | int err = 0; |
bab69409 | 3356 | cpu_set_t cpuset; |
d8a8f353 | 3357 | char *sock_dir_subcomponent; |
bab69409 AC |
3358 | |
3359 | if (!smap_get_bool(ovs_other_config, "dpdk-init", false)) { | |
3360 | VLOG_INFO("DPDK Disabled - to change this requires a restart.\n"); | |
3361 | return; | |
dfc89a55 AZ |
3362 | } |
3363 | ||
bab69409 | 3364 | VLOG_INFO("DPDK Enabled, initializing"); |
6eb51f9a | 3365 | if (process_vhost_flags("vhost-sock-dir", xstrdup(ovs_rundir()), |
d8a8f353 AC |
3366 | NAME_MAX, ovs_other_config, |
3367 | &sock_dir_subcomponent)) { | |
7d1ced01 | 3368 | struct stat s; |
d8a8f353 AC |
3369 | if (!strstr(sock_dir_subcomponent, "..")) { |
3370 | vhost_sock_dir = xasprintf("%s/%s", ovs_rundir(), | |
3371 | sock_dir_subcomponent); | |
3372 | ||
3373 | err = stat(vhost_sock_dir, &s); | |
3374 | if (err) { | |
3375 | VLOG_ERR("vhost-user sock directory '%s' does not exist.", | |
3376 | vhost_sock_dir); | |
3377 | } | |
3378 | } else { | |
3379 | vhost_sock_dir = xstrdup(ovs_rundir()); | |
3380 | VLOG_ERR("vhost-user sock directory request '%s/%s' has invalid" | |
3381 | "characters '..' - using %s instead.", | |
3382 | ovs_rundir(), sock_dir_subcomponent, ovs_rundir()); | |
7d1ced01 | 3383 | } |
d8a8f353 AC |
3384 | free(sock_dir_subcomponent); |
3385 | } else { | |
6eb51f9a | 3386 | vhost_sock_dir = sock_dir_subcomponent; |
58397e6c KT |
3387 | } |
3388 | ||
bab69409 | 3389 | argv = grow_argv(&argv, 0, 1); |
88964e64 | 3390 | argc = 1; |
bab69409 | 3391 | argv[0] = xstrdup(ovs_get_program_name()); |
88964e64 AC |
3392 | argc_tmp = get_dpdk_args(ovs_other_config, &argv, argc); |
3393 | ||
3394 | while (argc_tmp != argc) { | |
3395 | if (!strcmp("-c", argv[argc]) || !strcmp("-l", argv[argc])) { | |
3396 | auto_determine = false; | |
3397 | break; | |
3398 | } | |
3399 | argc++; | |
3400 | } | |
3401 | argc = argc_tmp; | |
3402 | ||
3403 | /** | |
3404 | * NOTE: This is an unsophisticated mechanism for determining the DPDK | |
3405 | * lcore for the DPDK Master. | |
3406 | */ | |
3407 | if (auto_determine) { | |
3408 | int i; | |
d8e2f4cc KT |
3409 | /* Get the main thread affinity */ |
3410 | CPU_ZERO(&cpuset); | |
3411 | err = pthread_getaffinity_np(pthread_self(), sizeof(cpu_set_t), | |
3412 | &cpuset); | |
3413 | if (!err) { | |
3414 | for (i = 0; i < CPU_SETSIZE; i++) { | |
3415 | if (CPU_ISSET(i, &cpuset)) { | |
3416 | argv = grow_argv(&argv, argc, 2); | |
3417 | argv[argc++] = xstrdup("-c"); | |
3418 | argv[argc++] = xasprintf("0x%08llX", (1ULL<<i)); | |
3419 | i = CPU_SETSIZE; | |
3420 | } | |
88964e64 | 3421 | } |
d8e2f4cc KT |
3422 | } else { |
3423 | VLOG_ERR("Thread getaffinity error %d. Using core 0x1", err); | |
3424 | /* User did not set dpdk-lcore-mask and unable to get current | |
3425 | * thread affintity - default to core 0x1 */ | |
3426 | argv = grow_argv(&argv, argc, 2); | |
3427 | argv[argc++] = xstrdup("-c"); | |
3428 | argv[argc++] = xasprintf("0x%X", 1); | |
88964e64 AC |
3429 | } |
3430 | } | |
bab69409 AC |
3431 | |
3432 | argv = grow_argv(&argv, argc, 1); | |
3433 | argv[argc] = NULL; | |
3434 | ||
3435 | optind = 1; | |
58397e6c | 3436 | |
eac84432 AC |
3437 | if (VLOG_IS_INFO_ENABLED()) { |
3438 | struct ds eal_args; | |
3439 | int opt; | |
3440 | ds_init(&eal_args); | |
3441 | ds_put_cstr(&eal_args, "EAL ARGS:"); | |
3442 | for (opt = 0; opt < argc; ++opt) { | |
3443 | ds_put_cstr(&eal_args, " "); | |
3444 | ds_put_cstr(&eal_args, argv[opt]); | |
3445 | } | |
3446 | VLOG_INFO("%s", ds_cstr_ro(&eal_args)); | |
3447 | ds_destroy(&eal_args); | |
3448 | } | |
3449 | ||
8a9562d2 PS |
3450 | /* Make sure things are initialized ... */ |
3451 | result = rte_eal_init(argc, argv); | |
451450fa | 3452 | if (result < 0) { |
58397e6c | 3453 | ovs_abort(result, "Cannot init EAL"); |
451450fa | 3454 | } |
8a9562d2 | 3455 | |
563c98d8 | 3456 | /* Set the main thread affinity back to pre rte_eal_init() value */ |
d8e2f4cc | 3457 | if (auto_determine && !err) { |
bab69409 AC |
3458 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), |
3459 | &cpuset); | |
3460 | if (err) { | |
3461 | VLOG_ERR("Thread setaffinity error %d", err); | |
3462 | } | |
563c98d8 AC |
3463 | } |
3464 | ||
bab69409 AC |
3465 | dpdk_argv = argv; |
3466 | dpdk_argc = argc; | |
3467 | ||
3468 | atexit(deferred_argv_release); | |
3469 | ||
d7310583 | 3470 | rte_memzone_dump(stdout); |
8a9562d2 PS |
3471 | rte_eal_init_ret = 0; |
3472 | ||
db73f716 | 3473 | /* We are called from the main thread here */ |
d5c199ea | 3474 | RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID; |
db73f716 | 3475 | |
bab69409 AC |
3476 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); |
3477 | ||
bab69409 AC |
3478 | dpdk_vhost_class_init(); |
3479 | ||
4b88d678 CL |
3480 | #ifdef DPDK_PDUMP |
3481 | VLOG_INFO("DPDK pdump packet capture enabled"); | |
3482 | err = rte_pdump_init(ovs_rundir()); | |
3483 | if (err) { | |
3484 | VLOG_INFO("Error initialising DPDK pdump"); | |
3485 | rte_pdump_uninit(); | |
3486 | } else { | |
3487 | char *server_socket_path; | |
3488 | ||
3489 | server_socket_path = xasprintf("%s/%s", ovs_rundir(), | |
3490 | "pdump_server_socket"); | |
3491 | fatal_signal_add_file_to_unlink(server_socket_path); | |
3492 | free(server_socket_path); | |
3493 | } | |
3494 | #endif | |
3495 | ||
bab69409 AC |
3496 | /* Finally, register the dpdk classes */ |
3497 | netdev_dpdk_register(); | |
3498 | } | |
3499 | ||
3500 | void | |
3501 | dpdk_init(const struct smap *ovs_other_config) | |
3502 | { | |
3503 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
3504 | ||
3505 | if (ovs_other_config && ovsthread_once_start(&once)) { | |
3506 | dpdk_init__(ovs_other_config); | |
3507 | ovsthread_once_done(&once); | |
3508 | } | |
8a9562d2 PS |
3509 | } |
3510 | ||
bce01e3a | 3511 | static const struct netdev_class dpdk_class = |
95fb793a | 3512 | NETDEV_DPDK_CLASS( |
3513 | "dpdk", | |
5496878c | 3514 | netdev_dpdk_construct, |
58397e6c | 3515 | netdev_dpdk_destruct, |
81acebda IM |
3516 | netdev_dpdk_set_config, |
3517 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3518 | netdev_dpdk_eth_send, |
3519 | netdev_dpdk_get_carrier, | |
3520 | netdev_dpdk_get_stats, | |
3521 | netdev_dpdk_get_features, | |
3522 | netdev_dpdk_get_status, | |
050c60bf | 3523 | netdev_dpdk_reconfigure, |
58397e6c | 3524 | netdev_dpdk_rxq_recv); |
95fb793a | 3525 | |
bce01e3a | 3526 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3527 | NETDEV_DPDK_CLASS( |
3528 | "dpdkr", | |
5496878c | 3529 | netdev_dpdk_ring_construct, |
58397e6c | 3530 | netdev_dpdk_destruct, |
c3d062a7 | 3531 | netdev_dpdk_ring_set_config, |
81acebda | 3532 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3533 | netdev_dpdk_ring_send, |
3534 | netdev_dpdk_get_carrier, | |
3535 | netdev_dpdk_get_stats, | |
3536 | netdev_dpdk_get_features, | |
3537 | netdev_dpdk_get_status, | |
050c60bf | 3538 | netdev_dpdk_reconfigure, |
58397e6c KT |
3539 | netdev_dpdk_rxq_recv); |
3540 | ||
53f50d24 | 3541 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3542 | NETDEV_DPDK_CLASS( |
3543 | "dpdkvhostuser", | |
53f50d24 | 3544 | netdev_dpdk_vhost_construct, |
58397e6c | 3545 | netdev_dpdk_vhost_destruct, |
c1ff66ac | 3546 | netdev_dpdk_vhost_set_config, |
81acebda | 3547 | NULL, |
58397e6c KT |
3548 | netdev_dpdk_vhost_send, |
3549 | netdev_dpdk_vhost_get_carrier, | |
3550 | netdev_dpdk_vhost_get_stats, | |
3551 | NULL, | |
7251515e | 3552 | NULL, |
53f50d24 | 3553 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3554 | netdev_dpdk_vhost_rxq_recv); |
95fb793a | 3555 | |
8a9562d2 PS |
3556 | void |
3557 | netdev_dpdk_register(void) | |
3558 | { | |
bab69409 AC |
3559 | dpdk_common_init(); |
3560 | netdev_register_provider(&dpdk_class); | |
3561 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3562 | netdev_register_provider(&dpdk_vhost_class); |
8a9562d2 | 3563 | } |
8617afff | 3564 | |
6930c7e0 DDP |
3565 | void |
3566 | dpdk_set_lcore_id(unsigned cpu) | |
8617afff | 3567 | { |
abb5943d AW |
3568 | /* NON_PMD_CORE_ID is reserved for use by non pmd threads. */ |
3569 | ovs_assert(cpu != NON_PMD_CORE_ID); | |
65f13b50 | 3570 | RTE_PER_LCORE(_lcore_id) = cpu; |
8617afff | 3571 | } |
db73f716 | 3572 | |
db73f716 | 3573 | static bool |
5f17de68 | 3574 | dpdk_thread_is_pmd(void) |
db73f716 | 3575 | { |
abb5943d | 3576 | return rte_lcore_id() != NON_PMD_CORE_ID; |
db73f716 | 3577 | } |